Re: Eliminating Ventricular Noise from DIPY Free Water Corrected Scalars
Hi Linda, With your permission, I am adding the DIPY mailing list, so others can weigh in and/or benefit from the discussion. My hunch is that the noise you are seeing in the ventricles is due to artifacts/noise. Do you do any removal of Gibbs ringing artifacts or any denoising of the data before analyzing it with fwdti? Cheers, Ariel On Wed, Jul 1, 2020 at 12:22 PM Linda Jasmine Hoffman <tuf72977@temple.edu> wrote:
Good afternoon DIPY experts,
My name is Linda Hoffman, and I'm the lab manager for Dr. Ingrid Olson's Cognitive Neuroscience Lab at Temple University. I have been working on implementing a DIPY-based free-water elimination (FWE) pipeline that my labmate, Katie Jobson, adapted from your website <https://dipy.org/documentation/1.0.0./examples_built/reconst_fwdti/> in order to extract free-water corrected (FWC) scalar maps from a HYDI dataset that I'm analyzing. For your reference, I am ultimately planning to calculate FWC DTI metrics for the fornix and genu of the corpus callosum after performing probabilistic tractography. I have preprocessed my data using FSL version 6.0 and MRtrix3 on a linux machine.
While I have successfully extracted FWC FA, MD, RD, and AD maps from my data using this pipeline, there still seems to be a disproportionate amount of noise in the ventricles, especially when comparing my output to your examples on the website linked above. This is the case even after eliminating voxels with a water volume fraction (WVF) exceeding 70%. In light of this, I was wondering if you may be able to address the following questions:
- Is the amount of ventricular noise post-FWE in my scalar maps within a normal range? Will this preclude me from extracting valid FWC DTI metrics from the fornix and the genu? Here are some screenshots from a representative subject's scalar maps:
*FA map with WVF elimination at a threshold of 70%* [image: fa_70.png] *MD map with WVF elimination at a threshold of 70%* [image: md_70.png] *RD map with WVF elimination at a threshold of 70%* [image: rd_70.png] *AD map with WVF elimination at a threshold of 70%* [image: ad_70.png]
- If this noise is not within an acceptable range, how might I be able optimize our DIPY script so that I can perform a better FWE? I tried comparing the results from using a stricter WVF threshold of 60% as well as using no WVF thresholding to the above results. Using a stricter threshold did not completely eliminate the noise problem, but it did help a little bit. However, I'm not sure if there is a precedent for this level of thresholding in the literature, or if it is actually appropriate. Screenshots from a representative subject are listed below:
*FA map with WVF elimination at a threshold of 60%* [image: fa_60.png]
*MD map with WVF elimination at a threshold of 60%* [image: md_60.png] *FA map with No WVF elimination threshold* [image: fa_none.png] *MD map with No WVF elimination threshold* [image: md_none.png]
I have attached a zip file with the following information for your reference:
1. Input data from a representative subject. This includes DWI volumes collected at b values between 0 to 2000. This is contained in the *subject_data *subfolder. 2. Scalar maps collected with a WVF thresholding rate of 70% (*F>.7*), 60% (*F>.6*), and with no thresholding (*no_F_threshold*). 3. Three versions of the DIPY script I've been using - each one accounts for a different rate of WVF thresholding. These scripts are contained in the *dipy_fwe_script_versions* subfolder.
I sincerely appreciate all of your time and consideration, and look forward to hearing from you soon!
Kind regards, Linda
dipyfwe.zip <https://drive.google.com/a/temple.edu/file/d/1yvLYeB-cUNqBoce-43ler0-zf_vG8b...> -- *Lab Manager* *Cognitive Neuroscience Lab* Temple University 1701 N. 13th St. Philadelphia, PA 19122
*Pronouns: * She/Her *Phone*: (215) 204-1708 *Email*: tuf72977@temple.edu
Hi Ariel, Our preprocessing pipeline includes the following steps for noise reduction in FSL: - topup - correct for the susceptibility induced field and movement - eddy - correct for eddy current distortions and movement We don't have a step in our pipeline to correct for Gibbs artifacts. Do you think this particular type of artifact is what's underpinning this issue with the FWC scalar maps? If so, I found a command in MRtrix3 (mrdegibbs) that eliminates ringing, but it will necessitate that I go back and redo a large amount of preprocessing. Do you know of an alternative route to mitigate this problem that may obviate my need to reprocess my data? Thank you so much for your help! Kind regards, Linda On Thu, Jul 2, 2020 at 12:45 AM Ariel Rokem <arokem@gmail.com> wrote:
Hi Linda,
With your permission, I am adding the DIPY mailing list, so others can weigh in and/or benefit from the discussion.
My hunch is that the noise you are seeing in the ventricles is due to artifacts/noise. Do you do any removal of Gibbs ringing artifacts or any denoising of the data before analyzing it with fwdti?
Cheers,
Ariel
On Wed, Jul 1, 2020 at 12:22 PM Linda Jasmine Hoffman <tuf72977@temple.edu> wrote:
Good afternoon DIPY experts,
My name is Linda Hoffman, and I'm the lab manager for Dr. Ingrid Olson's Cognitive Neuroscience Lab at Temple University. I have been working on implementing a DIPY-based free-water elimination (FWE) pipeline that my labmate, Katie Jobson, adapted from your website <https://dipy.org/documentation/1.0.0./examples_built/reconst_fwdti/> in order to extract free-water corrected (FWC) scalar maps from a HYDI dataset that I'm analyzing. For your reference, I am ultimately planning to calculate FWC DTI metrics for the fornix and genu of the corpus callosum after performing probabilistic tractography. I have preprocessed my data using FSL version 6.0 and MRtrix3 on a linux machine.
While I have successfully extracted FWC FA, MD, RD, and AD maps from my data using this pipeline, there still seems to be a disproportionate amount of noise in the ventricles, especially when comparing my output to your examples on the website linked above. This is the case even after eliminating voxels with a water volume fraction (WVF) exceeding 70%. In light of this, I was wondering if you may be able to address the following questions:
- Is the amount of ventricular noise post-FWE in my scalar maps within a normal range? Will this preclude me from extracting valid FWC DTI metrics from the fornix and the genu? Here are some screenshots from a representative subject's scalar maps:
*FA map with WVF elimination at a threshold of 70%* [image: fa_70.png] *MD map with WVF elimination at a threshold of 70%* [image: md_70.png] *RD map with WVF elimination at a threshold of 70%* [image: rd_70.png] *AD map with WVF elimination at a threshold of 70%* [image: ad_70.png]
- If this noise is not within an acceptable range, how might I be able optimize our DIPY script so that I can perform a better FWE? I tried comparing the results from using a stricter WVF threshold of 60% as well as using no WVF thresholding to the above results. Using a stricter threshold did not completely eliminate the noise problem, but it did help a little bit. However, I'm not sure if there is a precedent for this level of thresholding in the literature, or if it is actually appropriate. Screenshots from a representative subject are listed below:
*FA map with WVF elimination at a threshold of 60%* [image: fa_60.png]
*MD map with WVF elimination at a threshold of 60%* [image: md_60.png] *FA map with No WVF elimination threshold* [image: fa_none.png] *MD map with No WVF elimination threshold* [image: md_none.png]
I have attached a zip file with the following information for your reference:
1. Input data from a representative subject. This includes DWI volumes collected at b values between 0 to 2000. This is contained in the *subject_data *subfolder. 2. Scalar maps collected with a WVF thresholding rate of 70% (*F>.7*), 60% (*F>.6*), and with no thresholding (*no_F_threshold*). 3. Three versions of the DIPY script I've been using - each one accounts for a different rate of WVF thresholding. These scripts are contained in the *dipy_fwe_script_versions* subfolder.
I sincerely appreciate all of your time and consideration, and look forward to hearing from you soon!
Kind regards, Linda
dipyfwe.zip <https://drive.google.com/a/temple.edu/file/d/1yvLYeB-cUNqBoce-43ler0-zf_vG8b...> -- *Lab Manager* *Cognitive Neuroscience Lab* Temple University 1701 N. 13th St. Philadelphia, PA 19122
*Pronouns: * She/Her *Phone*: (215) 204-1708 *Email*: tuf72977@temple.edu
-- *Lab Manager* *Cognitive Neuroscience Lab* Temple University 1701 N. 13th St. Philadelphia, PA 19122 *Pronouns: * She/Her *Phone*: (215) 204-1708 *Email*: tuf72977@temple.edu
Hi everyone, I just wanted to touch base with you to see if you've had the opportunity to give my previous email some consideration. Please let me know what my next steps should be re: denoising my DWI data to eliminate excessive ventricular artifacts post-fwc. Thank you! Linda On Thu, Jul 2, 2020 at 12:41 PM Linda Jasmine Hoffman <tuf72977@temple.edu> wrote:
Hi Ariel,
Our preprocessing pipeline includes the following steps for noise reduction in FSL:
- topup - correct for the susceptibility induced field and movement - eddy - correct for eddy current distortions and movement
We don't have a step in our pipeline to correct for Gibbs artifacts. Do you think this particular type of artifact is what's underpinning this issue with the FWC scalar maps? If so, I found a command in MRtrix3 (mrdegibbs) that eliminates ringing, but it will necessitate that I go back and redo a large amount of preprocessing. Do you know of an alternative route to mitigate this problem that may obviate my need to reprocess my data?
Thank you so much for your help! Kind regards, Linda
On Thu, Jul 2, 2020 at 12:45 AM Ariel Rokem <arokem@gmail.com> wrote:
Hi Linda,
With your permission, I am adding the DIPY mailing list, so others can weigh in and/or benefit from the discussion.
My hunch is that the noise you are seeing in the ventricles is due to artifacts/noise. Do you do any removal of Gibbs ringing artifacts or any denoising of the data before analyzing it with fwdti?
Cheers,
Ariel
On Wed, Jul 1, 2020 at 12:22 PM Linda Jasmine Hoffman < tuf72977@temple.edu> wrote:
Good afternoon DIPY experts,
My name is Linda Hoffman, and I'm the lab manager for Dr. Ingrid Olson's Cognitive Neuroscience Lab at Temple University. I have been working on implementing a DIPY-based free-water elimination (FWE) pipeline that my labmate, Katie Jobson, adapted from your website <https://dipy.org/documentation/1.0.0./examples_built/reconst_fwdti/> in order to extract free-water corrected (FWC) scalar maps from a HYDI dataset that I'm analyzing. For your reference, I am ultimately planning to calculate FWC DTI metrics for the fornix and genu of the corpus callosum after performing probabilistic tractography. I have preprocessed my data using FSL version 6.0 and MRtrix3 on a linux machine.
While I have successfully extracted FWC FA, MD, RD, and AD maps from my data using this pipeline, there still seems to be a disproportionate amount of noise in the ventricles, especially when comparing my output to your examples on the website linked above. This is the case even after eliminating voxels with a water volume fraction (WVF) exceeding 70%. In light of this, I was wondering if you may be able to address the following questions:
- Is the amount of ventricular noise post-FWE in my scalar maps within a normal range? Will this preclude me from extracting valid FWC DTI metrics from the fornix and the genu? Here are some screenshots from a representative subject's scalar maps:
*FA map with WVF elimination at a threshold of 70%* [image: fa_70.png] *MD map with WVF elimination at a threshold of 70%* [image: md_70.png] *RD map with WVF elimination at a threshold of 70%* [image: rd_70.png] *AD map with WVF elimination at a threshold of 70%* [image: ad_70.png]
- If this noise is not within an acceptable range, how might I be able optimize our DIPY script so that I can perform a better FWE? I tried comparing the results from using a stricter WVF threshold of 60% as well as using no WVF thresholding to the above results. Using a stricter threshold did not completely eliminate the noise problem, but it did help a little bit. However, I'm not sure if there is a precedent for this level of thresholding in the literature, or if it is actually appropriate. Screenshots from a representative subject are listed below:
*FA map with WVF elimination at a threshold of 60%* [image: fa_60.png]
*MD map with WVF elimination at a threshold of 60%* [image: md_60.png] *FA map with No WVF elimination threshold* [image: fa_none.png] *MD map with No WVF elimination threshold* [image: md_none.png]
I have attached a zip file with the following information for your reference:
1. Input data from a representative subject. This includes DWI volumes collected at b values between 0 to 2000. This is contained in the *subject_data *subfolder. 2. Scalar maps collected with a WVF thresholding rate of 70% (*F>.7*), 60% (*F>.6*), and with no thresholding (*no_F_threshold*). 3. Three versions of the DIPY script I've been using - each one accounts for a different rate of WVF thresholding. These scripts are contained in the *dipy_fwe_script_versions* subfolder.
I sincerely appreciate all of your time and consideration, and look forward to hearing from you soon!
Kind regards, Linda
dipyfwe.zip <https://drive.google.com/a/temple.edu/file/d/1yvLYeB-cUNqBoce-43ler0-zf_vG8b...> -- *Lab Manager* *Cognitive Neuroscience Lab* Temple University 1701 N. 13th St. Philadelphia, PA 19122
*Pronouns: * She/Her *Phone*: (215) 204-1708 *Email*: tuf72977@temple.edu
-- *Lab Manager* *Cognitive Neuroscience Lab* Temple University 1701 N. 13th St. Philadelphia, PA 19122
*Pronouns: * She/Her *Phone*: (215) 204-1708 *Email*: tuf72977@temple.edu
-- *Lab Manager* *Cognitive Neuroscience Lab* Temple University 1701 N. 13th St. Philadelphia, PA 19122 *Pronouns: * She/Her *Phone*: (215) 204-1708 *Email*: tuf72977@temple.edu
Hi Linda, Have you had a chance to try Gibbs ringing removal or and/or denoising on at least one subject? Cheers, Ariel On Thu, Jul 9, 2020 at 2:34 PM Linda Jasmine Hoffman <tuf72977@temple.edu> wrote:
Hi everyone,
I just wanted to touch base with you to see if you've had the opportunity to give my previous email some consideration. Please let me know what my next steps should be re: denoising my DWI data to eliminate excessive ventricular artifacts post-fwc.
Thank you! Linda
On Thu, Jul 2, 2020 at 12:41 PM Linda Jasmine Hoffman <tuf72977@temple.edu> wrote:
Hi Ariel,
Our preprocessing pipeline includes the following steps for noise reduction in FSL:
- topup - correct for the susceptibility induced field and movement - eddy - correct for eddy current distortions and movement
We don't have a step in our pipeline to correct for Gibbs artifacts. Do you think this particular type of artifact is what's underpinning this issue with the FWC scalar maps? If so, I found a command in MRtrix3 (mrdegibbs) that eliminates ringing, but it will necessitate that I go back and redo a large amount of preprocessing. Do you know of an alternative route to mitigate this problem that may obviate my need to reprocess my data?
Thank you so much for your help! Kind regards, Linda
On Thu, Jul 2, 2020 at 12:45 AM Ariel Rokem <arokem@gmail.com> wrote:
Hi Linda,
With your permission, I am adding the DIPY mailing list, so others can weigh in and/or benefit from the discussion.
My hunch is that the noise you are seeing in the ventricles is due to artifacts/noise. Do you do any removal of Gibbs ringing artifacts or any denoising of the data before analyzing it with fwdti?
Cheers,
Ariel
On Wed, Jul 1, 2020 at 12:22 PM Linda Jasmine Hoffman < tuf72977@temple.edu> wrote:
Good afternoon DIPY experts,
My name is Linda Hoffman, and I'm the lab manager for Dr. Ingrid Olson's Cognitive Neuroscience Lab at Temple University. I have been working on implementing a DIPY-based free-water elimination (FWE) pipeline that my labmate, Katie Jobson, adapted from your website <https://dipy.org/documentation/1.0.0./examples_built/reconst_fwdti/> in order to extract free-water corrected (FWC) scalar maps from a HYDI dataset that I'm analyzing. For your reference, I am ultimately planning to calculate FWC DTI metrics for the fornix and genu of the corpus callosum after performing probabilistic tractography. I have preprocessed my data using FSL version 6.0 and MRtrix3 on a linux machine.
While I have successfully extracted FWC FA, MD, RD, and AD maps from my data using this pipeline, there still seems to be a disproportionate amount of noise in the ventricles, especially when comparing my output to your examples on the website linked above. This is the case even after eliminating voxels with a water volume fraction (WVF) exceeding 70%. In light of this, I was wondering if you may be able to address the following questions:
- Is the amount of ventricular noise post-FWE in my scalar maps within a normal range? Will this preclude me from extracting valid FWC DTI metrics from the fornix and the genu? Here are some screenshots from a representative subject's scalar maps:
*FA map with WVF elimination at a threshold of 70%* [image: fa_70.png] *MD map with WVF elimination at a threshold of 70%* [image: md_70.png] *RD map with WVF elimination at a threshold of 70%* [image: rd_70.png] *AD map with WVF elimination at a threshold of 70%* [image: ad_70.png]
- If this noise is not within an acceptable range, how might I be able optimize our DIPY script so that I can perform a better FWE? I tried comparing the results from using a stricter WVF threshold of 60% as well as using no WVF thresholding to the above results. Using a stricter threshold did not completely eliminate the noise problem, but it did help a little bit. However, I'm not sure if there is a precedent for this level of thresholding in the literature, or if it is actually appropriate. Screenshots from a representative subject are listed below:
*FA map with WVF elimination at a threshold of 60%* [image: fa_60.png]
*MD map with WVF elimination at a threshold of 60%* [image: md_60.png] *FA map with No WVF elimination threshold* [image: fa_none.png] *MD map with No WVF elimination threshold* [image: md_none.png]
I have attached a zip file with the following information for your reference:
1. Input data from a representative subject. This includes DWI volumes collected at b values between 0 to 2000. This is contained in the *subject_data *subfolder. 2. Scalar maps collected with a WVF thresholding rate of 70% (*F>.7*), 60% (*F>.6*), and with no thresholding (*no_F_threshold*). 3. Three versions of the DIPY script I've been using - each one accounts for a different rate of WVF thresholding. These scripts are contained in the *dipy_fwe_script_versions* subfolder.
I sincerely appreciate all of your time and consideration, and look forward to hearing from you soon!
Kind regards, Linda
dipyfwe.zip <https://drive.google.com/a/temple.edu/file/d/1yvLYeB-cUNqBoce-43ler0-zf_vG8b...> -- *Lab Manager* *Cognitive Neuroscience Lab* Temple University 1701 N. 13th St. Philadelphia, PA 19122
*Pronouns: * She/Her *Phone*: (215) 204-1708 *Email*: tuf72977@temple.edu
-- *Lab Manager* *Cognitive Neuroscience Lab* Temple University 1701 N. 13th St. Philadelphia, PA 19122
*Pronouns: * She/Her *Phone*: (215) 204-1708 *Email*: tuf72977@temple.edu
-- *Lab Manager* *Cognitive Neuroscience Lab* Temple University 1701 N. 13th St. Philadelphia, PA 19122
*Pronouns: * She/Her *Phone*: (215) 204-1708 *Email*: tuf72977@temple.edu
I haven't; I'll try that now. Thank you! Linda On Thu, Jul 9, 2020 at 5:38 PM Ariel Rokem <arokem@gmail.com> wrote:
Hi Linda,
Have you had a chance to try Gibbs ringing removal or and/or denoising on at least one subject?
Cheers,
Ariel
On Thu, Jul 9, 2020 at 2:34 PM Linda Jasmine Hoffman <tuf72977@temple.edu> wrote:
Hi everyone,
I just wanted to touch base with you to see if you've had the opportunity to give my previous email some consideration. Please let me know what my next steps should be re: denoising my DWI data to eliminate excessive ventricular artifacts post-fwc.
Thank you! Linda
On Thu, Jul 2, 2020 at 12:41 PM Linda Jasmine Hoffman < tuf72977@temple.edu> wrote:
Hi Ariel,
Our preprocessing pipeline includes the following steps for noise reduction in FSL:
- topup - correct for the susceptibility induced field and movement - eddy - correct for eddy current distortions and movement
We don't have a step in our pipeline to correct for Gibbs artifacts. Do you think this particular type of artifact is what's underpinning this issue with the FWC scalar maps? If so, I found a command in MRtrix3 (mrdegibbs) that eliminates ringing, but it will necessitate that I go back and redo a large amount of preprocessing. Do you know of an alternative route to mitigate this problem that may obviate my need to reprocess my data?
Thank you so much for your help! Kind regards, Linda
On Thu, Jul 2, 2020 at 12:45 AM Ariel Rokem <arokem@gmail.com> wrote:
Hi Linda,
With your permission, I am adding the DIPY mailing list, so others can weigh in and/or benefit from the discussion.
My hunch is that the noise you are seeing in the ventricles is due to artifacts/noise. Do you do any removal of Gibbs ringing artifacts or any denoising of the data before analyzing it with fwdti?
Cheers,
Ariel
On Wed, Jul 1, 2020 at 12:22 PM Linda Jasmine Hoffman < tuf72977@temple.edu> wrote:
Good afternoon DIPY experts,
My name is Linda Hoffman, and I'm the lab manager for Dr. Ingrid Olson's Cognitive Neuroscience Lab at Temple University. I have been working on implementing a DIPY-based free-water elimination (FWE) pipeline that my labmate, Katie Jobson, adapted from your website <https://dipy.org/documentation/1.0.0./examples_built/reconst_fwdti/> in order to extract free-water corrected (FWC) scalar maps from a HYDI dataset that I'm analyzing. For your reference, I am ultimately planning to calculate FWC DTI metrics for the fornix and genu of the corpus callosum after performing probabilistic tractography. I have preprocessed my data using FSL version 6.0 and MRtrix3 on a linux machine.
While I have successfully extracted FWC FA, MD, RD, and AD maps from my data using this pipeline, there still seems to be a disproportionate amount of noise in the ventricles, especially when comparing my output to your examples on the website linked above. This is the case even after eliminating voxels with a water volume fraction (WVF) exceeding 70%. In light of this, I was wondering if you may be able to address the following questions:
- Is the amount of ventricular noise post-FWE in my scalar maps within a normal range? Will this preclude me from extracting valid FWC DTI metrics from the fornix and the genu? Here are some screenshots from a representative subject's scalar maps:
*FA map with WVF elimination at a threshold of 70%* [image: fa_70.png] *MD map with WVF elimination at a threshold of 70%* [image: md_70.png] *RD map with WVF elimination at a threshold of 70%* [image: rd_70.png] *AD map with WVF elimination at a threshold of 70%* [image: ad_70.png]
- If this noise is not within an acceptable range, how might I be able optimize our DIPY script so that I can perform a better FWE? I tried comparing the results from using a stricter WVF threshold of 60% as well as using no WVF thresholding to the above results. Using a stricter threshold did not completely eliminate the noise problem, but it did help a little bit. However, I'm not sure if there is a precedent for this level of thresholding in the literature, or if it is actually appropriate. Screenshots from a representative subject are listed below:
*FA map with WVF elimination at a threshold of 60%* [image: fa_60.png]
*MD map with WVF elimination at a threshold of 60%* [image: md_60.png] *FA map with No WVF elimination threshold* [image: fa_none.png] *MD map with No WVF elimination threshold* [image: md_none.png]
I have attached a zip file with the following information for your reference:
1. Input data from a representative subject. This includes DWI volumes collected at b values between 0 to 2000. This is contained in the *subject_data *subfolder. 2. Scalar maps collected with a WVF thresholding rate of 70% ( *F>.7*), 60% (*F>.6*), and with no thresholding (*no_F_threshold*). 3. Three versions of the DIPY script I've been using - each one accounts for a different rate of WVF thresholding. These scripts are contained in the *dipy_fwe_script_versions* subfolder.
I sincerely appreciate all of your time and consideration, and look forward to hearing from you soon!
Kind regards, Linda
dipyfwe.zip <https://drive.google.com/a/temple.edu/file/d/1yvLYeB-cUNqBoce-43ler0-zf_vG8b...> -- *Lab Manager* *Cognitive Neuroscience Lab* Temple University 1701 N. 13th St. Philadelphia, PA 19122
*Pronouns: * She/Her *Phone*: (215) 204-1708 *Email*: tuf72977@temple.edu
-- *Lab Manager* *Cognitive Neuroscience Lab* Temple University 1701 N. 13th St. Philadelphia, PA 19122
*Pronouns: * She/Her *Phone*: (215) 204-1708 *Email*: tuf72977@temple.edu
-- *Lab Manager* *Cognitive Neuroscience Lab* Temple University 1701 N. 13th St. Philadelphia, PA 19122
*Pronouns: * She/Her *Phone*: (215) 204-1708 *Email*: tuf72977@temple.edu
-- *Lab Manager* *Cognitive Neuroscience Lab* Temple University 1701 N. 13th St. Philadelphia, PA 19122 *Pronouns: * She/Her *Phone*: (215) 204-1708 *Email*: tuf72977@temple.edu
Good evening DIPY experts, I have developed a denoising protocol for my HYDI data, and it has afforded me some success in eliminating a portion of the excess ventricular noise that I have been finding in my free-water-corrected (FWC) scalars. Below is an example from a representative subject (i.e. "Subject 1") for whom this course of actions seems to have worked quite well: *Subject 1: Original MD map (no denoising of DWI data):* [image: 5022_md.png] *Subject 1: New MD map (with denoising of DWI data):* [image: 5022_md_denoised.png] However, I have a few concerns. First, my data is still not as clean as I would like it to be, given the persisting residual noise that is still present in the sagittal view. Second, the denoising protocol that I have implemented did not work consistently well for all subjects. Here is an example from a second representative subject (i.e. "Subject 2") to illustrate this issue: *Subject 2: Original MD map (no denoising of DWI data):* [image: 5216_md.png] *Subject 2: New MD map (with denoising of DWI data):* [image: 5216_md_denoised.png] What is particularly concerning about this is that the resultant image for Subject 2 is still not as clean as what is presented on your DIPY free-water elimination page <https://dipy.org/documentation/1.0.0./examples_built/reconst_fwdti/>. It is worth noting that quality assurance measures have been taken for all of our data, and this subject did not exhibit inordinate imaging artifacts. For your reference, my denoising pipeline utilized the *dwidenoise* and *mrdegibbs* functions in MRtrix3. I incorporated these steps into my processing protocol in the following order: 1. FSL - topup 2. MRtrix3 - dwidenoise 3. MRtrix3 - mrdegibbs 4. FSL - eddy Note that I completed *topup* first since this step does not affect the raw, DICOM-to-NIfTI-converted DWI volumes in any way, and it is necessary for yielding a hifi brain mask. The scripts that I used for denoising/degibbing are delineated below: *#dMRI noise level estimation and denoising using Marchenko-Pastur PCA:* for n in 5022 5216 5302 5391 do dwidenoise -mask /data/projects/tbi/denoise/${n}/topup_output/my_hifi_b0_Tcollapsed_brain_mask.nii.gz -noise /data/projects/tbi/denoise/${n}/dwidenoise/noise_hifi_map.nii /data/projects/tbi/denoise/${n}/6-cmrr_mb3hydi_ipat2_64ch/output.nii /data/projects/tbi/denoise/${n}/dwidenoise/denoised_hifi_vol.nii done *#Remove Gibbs Ringing Artifacts:* for n in 5022 5216 5302 5391 do mrdegibbs /data/projects/tbi/denoise/${n}/dwidenoise/denoised_hifi_vol.nii /data/projects/tbi/denoise/${n}/mrdegibbs/denoised_degibbs_hifi_vol.nii done What are your thoughts on the scripts I have implemented? Might I have done something incorrectly, or is there something further I should do to optimize this denoising pipeline? Is there anything I can do in addition to denoising to eliminate these undue levels of post-FWC ventricular noise in my scalars? Finally, do you recommend denoising and degibbing DWI data as a canonical part of my pipeline? I ask because I know there is a tradeoff between SNR and spatial resolution following noise reduction procedures, so I'm curious to know what best-practices are in this regard. At the very least it seems like an important step if one intends to pursue FWE. I sincerely appreciate all of your time and consideration on this matter. Kind regards, Linda On Thu, Jul 9, 2020 at 5:40 PM Linda Jasmine Hoffman <tuf72977@temple.edu> wrote:
I haven't; I'll try that now.
Thank you! Linda
On Thu, Jul 9, 2020 at 5:38 PM Ariel Rokem <arokem@gmail.com> wrote:
Hi Linda,
Have you had a chance to try Gibbs ringing removal or and/or denoising on at least one subject?
Cheers,
Ariel
On Thu, Jul 9, 2020 at 2:34 PM Linda Jasmine Hoffman <tuf72977@temple.edu> wrote:
Hi everyone,
I just wanted to touch base with you to see if you've had the opportunity to give my previous email some consideration. Please let me know what my next steps should be re: denoising my DWI data to eliminate excessive ventricular artifacts post-fwc.
Thank you! Linda
On Thu, Jul 2, 2020 at 12:41 PM Linda Jasmine Hoffman < tuf72977@temple.edu> wrote:
Hi Ariel,
Our preprocessing pipeline includes the following steps for noise reduction in FSL:
- topup - correct for the susceptibility induced field and movement - eddy - correct for eddy current distortions and movement
We don't have a step in our pipeline to correct for Gibbs artifacts. Do you think this particular type of artifact is what's underpinning this issue with the FWC scalar maps? If so, I found a command in MRtrix3 (mrdegibbs) that eliminates ringing, but it will necessitate that I go back and redo a large amount of preprocessing. Do you know of an alternative route to mitigate this problem that may obviate my need to reprocess my data?
Thank you so much for your help! Kind regards, Linda
On Thu, Jul 2, 2020 at 12:45 AM Ariel Rokem <arokem@gmail.com> wrote:
Hi Linda,
With your permission, I am adding the DIPY mailing list, so others can weigh in and/or benefit from the discussion.
My hunch is that the noise you are seeing in the ventricles is due to artifacts/noise. Do you do any removal of Gibbs ringing artifacts or any denoising of the data before analyzing it with fwdti?
Cheers,
Ariel
On Wed, Jul 1, 2020 at 12:22 PM Linda Jasmine Hoffman < tuf72977@temple.edu> wrote:
Good afternoon DIPY experts,
My name is Linda Hoffman, and I'm the lab manager for Dr. Ingrid Olson's Cognitive Neuroscience Lab at Temple University. I have been working on implementing a DIPY-based free-water elimination (FWE) pipeline that my labmate, Katie Jobson, adapted from your website <https://dipy.org/documentation/1.0.0./examples_built/reconst_fwdti/> in order to extract free-water corrected (FWC) scalar maps from a HYDI dataset that I'm analyzing. For your reference, I am ultimately planning to calculate FWC DTI metrics for the fornix and genu of the corpus callosum after performing probabilistic tractography. I have preprocessed my data using FSL version 6.0 and MRtrix3 on a linux machine.
While I have successfully extracted FWC FA, MD, RD, and AD maps from my data using this pipeline, there still seems to be a disproportionate amount of noise in the ventricles, especially when comparing my output to your examples on the website linked above. This is the case even after eliminating voxels with a water volume fraction (WVF) exceeding 70%. In light of this, I was wondering if you may be able to address the following questions:
- Is the amount of ventricular noise post-FWE in my scalar maps within a normal range? Will this preclude me from extracting valid FWC DTI metrics from the fornix and the genu? Here are some screenshots from a representative subject's scalar maps:
*FA map with WVF elimination at a threshold of 70%* [image: fa_70.png] *MD map with WVF elimination at a threshold of 70%* [image: md_70.png] *RD map with WVF elimination at a threshold of 70%* [image: rd_70.png] *AD map with WVF elimination at a threshold of 70%* [image: ad_70.png]
- If this noise is not within an acceptable range, how might I be able optimize our DIPY script so that I can perform a better FWE? I tried comparing the results from using a stricter WVF threshold of 60% as well as using no WVF thresholding to the above results. Using a stricter threshold did not completely eliminate the noise problem, but it did help a little bit. However, I'm not sure if there is a precedent for this level of thresholding in the literature, or if it is actually appropriate. Screenshots from a representative subject are listed below:
*FA map with WVF elimination at a threshold of 60%* [image: fa_60.png]
*MD map with WVF elimination at a threshold of 60%* [image: md_60.png] *FA map with No WVF elimination threshold* [image: fa_none.png] *MD map with No WVF elimination threshold* [image: md_none.png]
I have attached a zip file with the following information for your reference:
1. Input data from a representative subject. This includes DWI volumes collected at b values between 0 to 2000. This is contained in the *subject_data *subfolder. 2. Scalar maps collected with a WVF thresholding rate of 70% ( *F>.7*), 60% (*F>.6*), and with no thresholding (*no_F_threshold* ). 3. Three versions of the DIPY script I've been using - each one accounts for a different rate of WVF thresholding. These scripts are contained in the *dipy_fwe_script_versions* subfolder.
I sincerely appreciate all of your time and consideration, and look forward to hearing from you soon!
Kind regards, Linda
dipyfwe.zip <https://drive.google.com/a/temple.edu/file/d/1yvLYeB-cUNqBoce-43ler0-zf_vG8b...> -- *Lab Manager* *Cognitive Neuroscience Lab* Temple University 1701 N. 13th St. Philadelphia, PA 19122
*Pronouns: * She/Her *Phone*: (215) 204-1708 *Email*: tuf72977@temple.edu
-- *Lab Manager* *Cognitive Neuroscience Lab* Temple University 1701 N. 13th St. Philadelphia, PA 19122
*Pronouns: * She/Her *Phone*: (215) 204-1708 *Email*: tuf72977@temple.edu
-- *Lab Manager* *Cognitive Neuroscience Lab* Temple University 1701 N. 13th St. Philadelphia, PA 19122
*Pronouns: * She/Her *Phone*: (215) 204-1708 *Email*: tuf72977@temple.edu
-- *Lab Manager* *Cognitive Neuroscience Lab* Temple University 1701 N. 13th St. Philadelphia, PA 19122
*Pronouns: * She/Her *Phone*: (215) 204-1708 *Email*: tuf72977@temple.edu
-- *Lab Manager* *Cognitive Neuroscience Lab* Temple University 1701 N. 13th St. Philadelphia, PA 19122 *Pronouns: * She/Her *Phone*: (215) 204-1708 *Email*: tuf72977@temple.edu
Good evening DIPY experts, I just wanted to follow up with you all as per my last email to see if you've had the opportunity to give my questions some consideration. Please let me know! I look forward to hearing from you soon! Kind regards, Linda On Tue, Jul 28, 2020 at 10:36 PM Linda Jasmine Hoffman <tuf72977@temple.edu> wrote:
Good evening DIPY experts,
I have developed a denoising protocol for my HYDI data, and it has afforded me some success in eliminating a portion of the excess ventricular noise that I have been finding in my free-water-corrected (FWC) scalars. Below is an example from a representative subject (i.e. "Subject 1") for whom this course of actions seems to have worked quite well:
*Subject 1: Original MD map (no denoising of DWI data):* [image: 5022_md.png]
*Subject 1: New MD map (with denoising of DWI data):* [image: 5022_md_denoised.png]
However, I have a few concerns. First, my data is still not as clean as I would like it to be, given the persisting residual noise that is still present in the sagittal view. Second, the denoising protocol that I have implemented did not work consistently well for all subjects. Here is an example from a second representative subject (i.e. "Subject 2") to illustrate this issue:
*Subject 2: Original MD map (no denoising of DWI data):* [image: 5216_md.png]
*Subject 2: New MD map (with denoising of DWI data):* [image: 5216_md_denoised.png]
What is particularly concerning about this is that the resultant image for Subject 2 is still not as clean as what is presented on your DIPY free-water elimination page <https://dipy.org/documentation/1.0.0./examples_built/reconst_fwdti/>. It is worth noting that quality assurance measures have been taken for all of our data, and this subject did not exhibit inordinate imaging artifacts.
For your reference, my denoising pipeline utilized the *dwidenoise* and *mrdegibbs* functions in MRtrix3. I incorporated these steps into my processing protocol in the following order:
1. FSL - topup 2. MRtrix3 - dwidenoise 3. MRtrix3 - mrdegibbs 4. FSL - eddy
Note that I completed *topup* first since this step does not affect the raw, DICOM-to-NIfTI-converted DWI volumes in any way, and it is necessary for yielding a hifi brain mask. The scripts that I used for denoising/degibbing are delineated below:
*#dMRI noise level estimation and denoising using Marchenko-Pastur PCA:* for n in 5022 5216 5302 5391 do
dwidenoise
-mask /data/projects/tbi/denoise/${n}/topup_output/my_hifi_b0_Tcollapsed_brain_mask.nii.gz -noise /data/projects/tbi/denoise/${n}/dwidenoise/noise_hifi_map.nii /data/projects/tbi/denoise/${n}/6-cmrr_mb3hydi_ipat2_64ch/output.nii /data/projects/tbi/denoise/${n}/dwidenoise/denoised_hifi_vol.nii
done
*#Remove Gibbs Ringing Artifacts:* for n in 5022 5216 5302 5391 do
mrdegibbs
/data/projects/tbi/denoise/${n}/dwidenoise/denoised_hifi_vol.nii /data/projects/tbi/denoise/${n}/mrdegibbs/denoised_degibbs_hifi_vol.nii
done
What are your thoughts on the scripts I have implemented? Might I have done something incorrectly, or is there something further I should do to optimize this denoising pipeline? Is there anything I can do in addition to denoising to eliminate these undue levels of post-FWC ventricular noise in my scalars?
Finally, do you recommend denoising and degibbing DWI data as a canonical part of my pipeline? I ask because I know there is a tradeoff between SNR and spatial resolution following noise reduction procedures, so I'm curious to know what best-practices are in this regard. At the very least it seems like an important step if one intends to pursue FWE.
I sincerely appreciate all of your time and consideration on this matter.
Kind regards, Linda
On Thu, Jul 9, 2020 at 5:40 PM Linda Jasmine Hoffman <tuf72977@temple.edu> wrote:
I haven't; I'll try that now.
Thank you! Linda
On Thu, Jul 9, 2020 at 5:38 PM Ariel Rokem <arokem@gmail.com> wrote:
Hi Linda,
Have you had a chance to try Gibbs ringing removal or and/or denoising on at least one subject?
Cheers,
Ariel
On Thu, Jul 9, 2020 at 2:34 PM Linda Jasmine Hoffman < tuf72977@temple.edu> wrote:
Hi everyone,
I just wanted to touch base with you to see if you've had the opportunity to give my previous email some consideration. Please let me know what my next steps should be re: denoising my DWI data to eliminate excessive ventricular artifacts post-fwc.
Thank you! Linda
On Thu, Jul 2, 2020 at 12:41 PM Linda Jasmine Hoffman < tuf72977@temple.edu> wrote:
Hi Ariel,
Our preprocessing pipeline includes the following steps for noise reduction in FSL:
- topup - correct for the susceptibility induced field and movement - eddy - correct for eddy current distortions and movement
We don't have a step in our pipeline to correct for Gibbs artifacts. Do you think this particular type of artifact is what's underpinning this issue with the FWC scalar maps? If so, I found a command in MRtrix3 (mrdegibbs) that eliminates ringing, but it will necessitate that I go back and redo a large amount of preprocessing. Do you know of an alternative route to mitigate this problem that may obviate my need to reprocess my data?
Thank you so much for your help! Kind regards, Linda
On Thu, Jul 2, 2020 at 12:45 AM Ariel Rokem <arokem@gmail.com> wrote:
Hi Linda,
With your permission, I am adding the DIPY mailing list, so others can weigh in and/or benefit from the discussion.
My hunch is that the noise you are seeing in the ventricles is due to artifacts/noise. Do you do any removal of Gibbs ringing artifacts or any denoising of the data before analyzing it with fwdti?
Cheers,
Ariel
On Wed, Jul 1, 2020 at 12:22 PM Linda Jasmine Hoffman < tuf72977@temple.edu> wrote:
> Good afternoon DIPY experts, > > My name is Linda Hoffman, and I'm the lab manager for Dr. Ingrid > Olson's Cognitive Neuroscience Lab at Temple University. I have been > working on implementing a DIPY-based free-water elimination (FWE) pipeline > that my labmate, Katie Jobson, adapted from your website > <https://dipy.org/documentation/1.0.0./examples_built/reconst_fwdti/> > in order to extract free-water corrected (FWC) scalar maps from a HYDI > dataset that I'm analyzing. For your reference, I am ultimately > planning to calculate FWC DTI metrics for the fornix and genu of the corpus > callosum after performing probabilistic tractography. I have preprocessed > my data using FSL version 6.0 and MRtrix3 on a linux machine. > > While I have successfully extracted FWC FA, MD, RD, and AD maps from > my data using this pipeline, there still seems to be a disproportionate > amount of noise in the ventricles, especially when comparing my output to > your examples on the website linked above. This is the case even after > eliminating voxels with a water volume fraction (WVF) exceeding 70%. In > light of this, I was wondering if you may be able to address the following > questions: > > - Is the amount of ventricular noise post-FWE in my scalar maps > within a normal range? Will this preclude me from extracting valid FWC DTI > metrics from the fornix and the genu? Here are some screenshots from a > representative subject's scalar maps: > > *FA map with WVF elimination at a threshold of 70%* > [image: fa_70.png] > *MD map with WVF elimination at a threshold of 70%* > [image: md_70.png] > *RD map with WVF elimination at a threshold of 70%* > [image: rd_70.png] > *AD map with WVF elimination at a threshold of 70%* > [image: ad_70.png] > > > - If this noise is not within an acceptable range, how might I > be able optimize our DIPY script so that I can perform a better FWE? I > tried comparing the results from using a stricter WVF threshold of 60% as > well as using no WVF thresholding to the above results. Using a stricter > threshold did not completely eliminate the noise problem, but it did help a > little bit. However, I'm not sure if there is a precedent for this level > of thresholding in the literature, or if it is actually appropriate. > Screenshots from a representative subject are listed below: > > *FA map with WVF elimination at a threshold of 60%* > [image: fa_60.png] > > *MD map with WVF elimination at a threshold of 60%* > [image: md_60.png] > *FA map with No WVF elimination threshold* > [image: fa_none.png] > *MD map with No WVF elimination threshold* > [image: md_none.png] > > I have attached a zip file with the following information for your > reference: > > 1. Input data from a representative subject. This includes DWI > volumes collected at b values between 0 to 2000. This is contained in the *subject_data > *subfolder. > 2. Scalar maps collected with a WVF thresholding rate of 70% ( > *F>.7*), 60% (*F>.6*), and with no thresholding (*no_F_threshold* > ). > 3. Three versions of the DIPY script I've been using - each one > accounts for a different rate of WVF thresholding. These > scripts are contained in the *dipy_fwe_script_versions* > subfolder. > > I sincerely appreciate all of your time and consideration, and look > forward to hearing from you soon! > > Kind regards, > Linda > > dipyfwe.zip > <https://drive.google.com/a/temple.edu/file/d/1yvLYeB-cUNqBoce-43ler0-zf_vG8b...> > -- > *Lab Manager* > *Cognitive Neuroscience Lab* > Temple University > 1701 N. 13th St. > Philadelphia, PA 19122 > > *Pronouns: * She/Her > *Phone*: (215) 204-1708 > *Email*: tuf72977@temple.edu >
-- *Lab Manager* *Cognitive Neuroscience Lab* Temple University 1701 N. 13th St. Philadelphia, PA 19122
*Pronouns: * She/Her *Phone*: (215) 204-1708 *Email*: tuf72977@temple.edu
-- *Lab Manager* *Cognitive Neuroscience Lab* Temple University 1701 N. 13th St. Philadelphia, PA 19122
*Pronouns: * She/Her *Phone*: (215) 204-1708 *Email*: tuf72977@temple.edu
-- *Lab Manager* *Cognitive Neuroscience Lab* Temple University 1701 N. 13th St. Philadelphia, PA 19122
*Pronouns: * She/Her *Phone*: (215) 204-1708 *Email*: tuf72977@temple.edu
-- *Lab Manager* *Cognitive Neuroscience Lab* Temple University 1701 N. 13th St. Philadelphia, PA 19122
*Pronouns: * She/Her *Phone*: (215) 204-1708 *Email*: tuf72977@temple.edu
-- *Lab Manager* *Cognitive Neuroscience Lab* Temple University 1701 N. 13th St. Philadelphia, PA 19122 *Pronouns: * She/Her *Phone*: (215) 204-1708 *Email*: tuf72977@temple.edu
Good morning DIPY experts, I hope you have all been doing well! I just wanted to follow up with you again as per my latest update re: persisting ventricular noise post-denoising & FWC. Please let me know if you can shed any light on why this noise may still be an issue, even after implementing Ariel's denoising/degibbing suggestion. I look forward to hearing from you soon! Kind regards, Linda On Mon, Aug 3, 2020 at 7:33 PM Linda Jasmine Hoffman <tuf72977@temple.edu> wrote:
Good evening DIPY experts,
I just wanted to follow up with you all as per my last email to see if you've had the opportunity to give my questions some consideration.
Please let me know! I look forward to hearing from you soon!
Kind regards, Linda
On Tue, Jul 28, 2020 at 10:36 PM Linda Jasmine Hoffman < tuf72977@temple.edu> wrote:
Good evening DIPY experts,
I have developed a denoising protocol for my HYDI data, and it has afforded me some success in eliminating a portion of the excess ventricular noise that I have been finding in my free-water-corrected (FWC) scalars. Below is an example from a representative subject (i.e. "Subject 1") for whom this course of actions seems to have worked quite well:
*Subject 1: Original MD map (no denoising of DWI data):* [image: 5022_md.png]
*Subject 1: New MD map (with denoising of DWI data):* [image: 5022_md_denoised.png]
However, I have a few concerns. First, my data is still not as clean as I would like it to be, given the persisting residual noise that is still present in the sagittal view. Second, the denoising protocol that I have implemented did not work consistently well for all subjects. Here is an example from a second representative subject (i.e. "Subject 2") to illustrate this issue:
*Subject 2: Original MD map (no denoising of DWI data):* [image: 5216_md.png]
*Subject 2: New MD map (with denoising of DWI data):* [image: 5216_md_denoised.png]
What is particularly concerning about this is that the resultant image for Subject 2 is still not as clean as what is presented on your DIPY free-water elimination page <https://dipy.org/documentation/1.0.0./examples_built/reconst_fwdti/>. It is worth noting that quality assurance measures have been taken for all of our data, and this subject did not exhibit inordinate imaging artifacts.
For your reference, my denoising pipeline utilized the *dwidenoise* and *mrdegibbs* functions in MRtrix3. I incorporated these steps into my processing protocol in the following order:
1. FSL - topup 2. MRtrix3 - dwidenoise 3. MRtrix3 - mrdegibbs 4. FSL - eddy
Note that I completed *topup* first since this step does not affect the raw, DICOM-to-NIfTI-converted DWI volumes in any way, and it is necessary for yielding a hifi brain mask. The scripts that I used for denoising/degibbing are delineated below:
*#dMRI noise level estimation and denoising using Marchenko-Pastur PCA:* for n in 5022 5216 5302 5391 do
dwidenoise
-mask /data/projects/tbi/denoise/${n}/topup_output/my_hifi_b0_Tcollapsed_brain_mask.nii.gz -noise /data/projects/tbi/denoise/${n}/dwidenoise/noise_hifi_map.nii /data/projects/tbi/denoise/${n}/6-cmrr_mb3hydi_ipat2_64ch/output.nii /data/projects/tbi/denoise/${n}/dwidenoise/denoised_hifi_vol.nii
done
*#Remove Gibbs Ringing Artifacts:* for n in 5022 5216 5302 5391 do
mrdegibbs
/data/projects/tbi/denoise/${n}/dwidenoise/denoised_hifi_vol.nii /data/projects/tbi/denoise/${n}/mrdegibbs/denoised_degibbs_hifi_vol.nii
done
What are your thoughts on the scripts I have implemented? Might I have done something incorrectly, or is there something further I should do to optimize this denoising pipeline? Is there anything I can do in addition to denoising to eliminate these undue levels of post-FWC ventricular noise in my scalars?
Finally, do you recommend denoising and degibbing DWI data as a canonical part of my pipeline? I ask because I know there is a tradeoff between SNR and spatial resolution following noise reduction procedures, so I'm curious to know what best-practices are in this regard. At the very least it seems like an important step if one intends to pursue FWE.
I sincerely appreciate all of your time and consideration on this matter.
Kind regards, Linda
On Thu, Jul 9, 2020 at 5:40 PM Linda Jasmine Hoffman <tuf72977@temple.edu> wrote:
I haven't; I'll try that now.
Thank you! Linda
On Thu, Jul 9, 2020 at 5:38 PM Ariel Rokem <arokem@gmail.com> wrote:
Hi Linda,
Have you had a chance to try Gibbs ringing removal or and/or denoising on at least one subject?
Cheers,
Ariel
On Thu, Jul 9, 2020 at 2:34 PM Linda Jasmine Hoffman < tuf72977@temple.edu> wrote:
Hi everyone,
I just wanted to touch base with you to see if you've had the opportunity to give my previous email some consideration. Please let me know what my next steps should be re: denoising my DWI data to eliminate excessive ventricular artifacts post-fwc.
Thank you! Linda
On Thu, Jul 2, 2020 at 12:41 PM Linda Jasmine Hoffman < tuf72977@temple.edu> wrote:
Hi Ariel,
Our preprocessing pipeline includes the following steps for noise reduction in FSL:
- topup - correct for the susceptibility induced field and movement - eddy - correct for eddy current distortions and movement
We don't have a step in our pipeline to correct for Gibbs artifacts. Do you think this particular type of artifact is what's underpinning this issue with the FWC scalar maps? If so, I found a command in MRtrix3 (mrdegibbs) that eliminates ringing, but it will necessitate that I go back and redo a large amount of preprocessing. Do you know of an alternative route to mitigate this problem that may obviate my need to reprocess my data?
Thank you so much for your help! Kind regards, Linda
On Thu, Jul 2, 2020 at 12:45 AM Ariel Rokem <arokem@gmail.com> wrote:
> Hi Linda, > > With your permission, I am adding the DIPY mailing list, so others > can weigh in and/or benefit from the discussion. > > My hunch is that the noise you are seeing in the ventricles is due > to artifacts/noise. Do you do any removal of Gibbs ringing artifacts or any > denoising of the data before analyzing it with fwdti? > > Cheers, > > Ariel > > > On Wed, Jul 1, 2020 at 12:22 PM Linda Jasmine Hoffman < > tuf72977@temple.edu> wrote: > >> Good afternoon DIPY experts, >> >> My name is Linda Hoffman, and I'm the lab manager for Dr. Ingrid >> Olson's Cognitive Neuroscience Lab at Temple University. I have been >> working on implementing a DIPY-based free-water elimination (FWE) pipeline >> that my labmate, Katie Jobson, adapted from your website >> <https://dipy.org/documentation/1.0.0./examples_built/reconst_fwdti/> >> in order to extract free-water corrected (FWC) scalar maps from a HYDI >> dataset that I'm analyzing. For your reference, I am ultimately >> planning to calculate FWC DTI metrics for the fornix and genu of the corpus >> callosum after performing probabilistic tractography. I have preprocessed >> my data using FSL version 6.0 and MRtrix3 on a linux machine. >> >> While I have successfully extracted FWC FA, MD, RD, and AD maps >> from my data using this pipeline, there still seems to be a >> disproportionate amount of noise in the ventricles, especially when >> comparing my output to your examples on the website linked above. This is >> the case even after eliminating voxels with a water volume fraction (WVF) >> exceeding 70%. In light of this, I was wondering if you may be able to >> address the following questions: >> >> - Is the amount of ventricular noise post-FWE in my scalar maps >> within a normal range? Will this preclude me from extracting valid FWC DTI >> metrics from the fornix and the genu? Here are some screenshots from a >> representative subject's scalar maps: >> >> *FA map with WVF elimination at a threshold of 70%* >> [image: fa_70.png] >> *MD map with WVF elimination at a threshold of 70%* >> [image: md_70.png] >> *RD map with WVF elimination at a threshold of 70%* >> [image: rd_70.png] >> *AD map with WVF elimination at a threshold of 70%* >> [image: ad_70.png] >> >> >> - If this noise is not within an acceptable range, how might I >> be able optimize our DIPY script so that I can perform a better FWE? I >> tried comparing the results from using a stricter WVF threshold of 60% as >> well as using no WVF thresholding to the above results. Using a stricter >> threshold did not completely eliminate the noise problem, but it did help a >> little bit. However, I'm not sure if there is a precedent for this level >> of thresholding in the literature, or if it is actually appropriate. >> Screenshots from a representative subject are listed below: >> >> *FA map with WVF elimination at a threshold of 60%* >> [image: fa_60.png] >> >> *MD map with WVF elimination at a threshold of 60%* >> [image: md_60.png] >> *FA map with No WVF elimination threshold* >> [image: fa_none.png] >> *MD map with No WVF elimination threshold* >> [image: md_none.png] >> >> I have attached a zip file with the following information for your >> reference: >> >> 1. Input data from a representative subject. This includes DWI >> volumes collected at b values between 0 to 2000. This is contained in the *subject_data >> *subfolder. >> 2. Scalar maps collected with a WVF thresholding rate of 70% ( >> *F>.7*), 60% (*F>.6*), and with no thresholding ( >> *no_F_threshold*). >> 3. Three versions of the DIPY script I've been using - each one >> accounts for a different rate of WVF thresholding. These >> scripts are contained in the *dipy_fwe_script_versions* >> subfolder. >> >> I sincerely appreciate all of your time and consideration, and look >> forward to hearing from you soon! >> >> Kind regards, >> Linda >> >> dipyfwe.zip >> <https://drive.google.com/a/temple.edu/file/d/1yvLYeB-cUNqBoce-43ler0-zf_vG8b...> >> -- >> *Lab Manager* >> *Cognitive Neuroscience Lab* >> Temple University >> 1701 N. 13th St. >> Philadelphia, PA 19122 >> >> *Pronouns: * She/Her >> *Phone*: (215) 204-1708 >> *Email*: tuf72977@temple.edu >> >
-- *Lab Manager* *Cognitive Neuroscience Lab* Temple University 1701 N. 13th St. Philadelphia, PA 19122
*Pronouns: * She/Her *Phone*: (215) 204-1708 *Email*: tuf72977@temple.edu
-- *Lab Manager* *Cognitive Neuroscience Lab* Temple University 1701 N. 13th St. Philadelphia, PA 19122
*Pronouns: * She/Her *Phone*: (215) 204-1708 *Email*: tuf72977@temple.edu
-- *Lab Manager* *Cognitive Neuroscience Lab* Temple University 1701 N. 13th St. Philadelphia, PA 19122
*Pronouns: * She/Her *Phone*: (215) 204-1708 *Email*: tuf72977@temple.edu
-- *Lab Manager* *Cognitive Neuroscience Lab* Temple University 1701 N. 13th St. Philadelphia, PA 19122
*Pronouns: * She/Her *Phone*: (215) 204-1708 *Email*: tuf72977@temple.edu
-- *Lab Manager* *Cognitive Neuroscience Lab* Temple University 1701 N. 13th St. Philadelphia, PA 19122
*Pronouns: * She/Her *Phone*: (215) 204-1708 *Email*: tuf72977@temple.edu
-- *Lab Manager* *Cognitive Neuroscience Lab* Temple University 1701 N. 13th St. Philadelphia, PA 19122 *Pronouns: * She/Her *Phone*: (215) 204-1708 *Email*: tuf72977@temple.edu
Hi Linda, Sorry for the slowness here... It's been... challenging. Two thoughts: 1. Sorry if I wasn't clear about this before: It is usually recommended that denoising and Gibbs ringing removal be done *before *other steps in preprocessing. To be on the safe side, I would recommend using https://qsiprep.readthedocs.io/en/latest/ for preprocessing. It implements the state of the art, and can be run as a docker/singularity container, which simplifies installation issues. 2. I am wondering what the signal is like in these voxels that still appear with very high MD values. Is there something unusual about their B0 signal? Or are the other data so low as to be indistinguishable from the noise floor? If you could find the coordinate of one of these voxels, and then us that to share with us the signal values in this voxel (as well as b-values and b-vectors) it would help diagnose this. Cheers, Ariel On Mon, Aug 17, 2020 at 9:14 AM Linda Jasmine Hoffman <tuf72977@temple.edu> wrote:
Good morning DIPY experts,
I hope you have all been doing well! I just wanted to follow up with you again as per my latest update re: persisting ventricular noise post-denoising & FWC. Please let me know if you can shed any light on why this noise may still be an issue, even after implementing Ariel's denoising/degibbing suggestion.
I look forward to hearing from you soon!
Kind regards, Linda
On Mon, Aug 3, 2020 at 7:33 PM Linda Jasmine Hoffman <tuf72977@temple.edu> wrote:
Good evening DIPY experts,
I just wanted to follow up with you all as per my last email to see if you've had the opportunity to give my questions some consideration.
Please let me know! I look forward to hearing from you soon!
Kind regards, Linda
On Tue, Jul 28, 2020 at 10:36 PM Linda Jasmine Hoffman < tuf72977@temple.edu> wrote:
Good evening DIPY experts,
I have developed a denoising protocol for my HYDI data, and it has afforded me some success in eliminating a portion of the excess ventricular noise that I have been finding in my free-water-corrected (FWC) scalars. Below is an example from a representative subject (i.e. "Subject 1") for whom this course of actions seems to have worked quite well:
*Subject 1: Original MD map (no denoising of DWI data):* [image: 5022_md.png]
*Subject 1: New MD map (with denoising of DWI data):* [image: 5022_md_denoised.png]
However, I have a few concerns. First, my data is still not as clean as I would like it to be, given the persisting residual noise that is still present in the sagittal view. Second, the denoising protocol that I have implemented did not work consistently well for all subjects. Here is an example from a second representative subject (i.e. "Subject 2") to illustrate this issue:
*Subject 2: Original MD map (no denoising of DWI data):* [image: 5216_md.png]
*Subject 2: New MD map (with denoising of DWI data):* [image: 5216_md_denoised.png]
What is particularly concerning about this is that the resultant image for Subject 2 is still not as clean as what is presented on your DIPY free-water elimination page <https://dipy.org/documentation/1.0.0./examples_built/reconst_fwdti/>. It is worth noting that quality assurance measures have been taken for all of our data, and this subject did not exhibit inordinate imaging artifacts.
For your reference, my denoising pipeline utilized the *dwidenoise* and *mrdegibbs* functions in MRtrix3. I incorporated these steps into my processing protocol in the following order:
1. FSL - topup 2. MRtrix3 - dwidenoise 3. MRtrix3 - mrdegibbs 4. FSL - eddy
Note that I completed *topup* first since this step does not affect the raw, DICOM-to-NIfTI-converted DWI volumes in any way, and it is necessary for yielding a hifi brain mask. The scripts that I used for denoising/degibbing are delineated below:
*#dMRI noise level estimation and denoising using Marchenko-Pastur PCA:* for n in 5022 5216 5302 5391 do
dwidenoise
-mask /data/projects/tbi/denoise/${n}/topup_output/my_hifi_b0_Tcollapsed_brain_mask.nii.gz -noise /data/projects/tbi/denoise/${n}/dwidenoise/noise_hifi_map.nii /data/projects/tbi/denoise/${n}/6-cmrr_mb3hydi_ipat2_64ch/output.nii /data/projects/tbi/denoise/${n}/dwidenoise/denoised_hifi_vol.nii
done
*#Remove Gibbs Ringing Artifacts:* for n in 5022 5216 5302 5391 do
mrdegibbs
/data/projects/tbi/denoise/${n}/dwidenoise/denoised_hifi_vol.nii /data/projects/tbi/denoise/${n}/mrdegibbs/denoised_degibbs_hifi_vol.nii
done
What are your thoughts on the scripts I have implemented? Might I have done something incorrectly, or is there something further I should do to optimize this denoising pipeline? Is there anything I can do in addition to denoising to eliminate these undue levels of post-FWC ventricular noise in my scalars?
Finally, do you recommend denoising and degibbing DWI data as a canonical part of my pipeline? I ask because I know there is a tradeoff between SNR and spatial resolution following noise reduction procedures, so I'm curious to know what best-practices are in this regard. At the very least it seems like an important step if one intends to pursue FWE.
I sincerely appreciate all of your time and consideration on this matter.
Kind regards, Linda
On Thu, Jul 9, 2020 at 5:40 PM Linda Jasmine Hoffman < tuf72977@temple.edu> wrote:
I haven't; I'll try that now.
Thank you! Linda
On Thu, Jul 9, 2020 at 5:38 PM Ariel Rokem <arokem@gmail.com> wrote:
Hi Linda,
Have you had a chance to try Gibbs ringing removal or and/or denoising on at least one subject?
Cheers,
Ariel
On Thu, Jul 9, 2020 at 2:34 PM Linda Jasmine Hoffman < tuf72977@temple.edu> wrote:
Hi everyone,
I just wanted to touch base with you to see if you've had the opportunity to give my previous email some consideration. Please let me know what my next steps should be re: denoising my DWI data to eliminate excessive ventricular artifacts post-fwc.
Thank you! Linda
On Thu, Jul 2, 2020 at 12:41 PM Linda Jasmine Hoffman < tuf72977@temple.edu> wrote:
> Hi Ariel, > > Our preprocessing pipeline includes the following steps for noise > reduction in FSL: > > - topup - correct for the susceptibility induced field and > movement > - eddy - correct for eddy current distortions and movement > > We don't have a step in our pipeline to correct for Gibbs > artifacts. Do you think this particular type of artifact is what's > underpinning this issue with the FWC scalar maps? If so, I found a command > in MRtrix3 (mrdegibbs) that eliminates ringing, but it will necessitate > that I go back and redo a large amount of preprocessing. Do you know of an > alternative route to mitigate this problem that may obviate my need to > reprocess my data? > > Thank you so much for your help! > Kind regards, > Linda > > On Thu, Jul 2, 2020 at 12:45 AM Ariel Rokem <arokem@gmail.com> > wrote: > >> Hi Linda, >> >> With your permission, I am adding the DIPY mailing list, so others >> can weigh in and/or benefit from the discussion. >> >> My hunch is that the noise you are seeing in the ventricles is due >> to artifacts/noise. Do you do any removal of Gibbs ringing artifacts or any >> denoising of the data before analyzing it with fwdti? >> >> Cheers, >> >> Ariel >> >> >> On Wed, Jul 1, 2020 at 12:22 PM Linda Jasmine Hoffman < >> tuf72977@temple.edu> wrote: >> >>> Good afternoon DIPY experts, >>> >>> My name is Linda Hoffman, and I'm the lab manager for Dr. Ingrid >>> Olson's Cognitive Neuroscience Lab at Temple University. I have been >>> working on implementing a DIPY-based free-water elimination (FWE) pipeline >>> that my labmate, Katie Jobson, adapted from your website >>> <https://dipy.org/documentation/1.0.0./examples_built/reconst_fwdti/> >>> in order to extract free-water corrected (FWC) scalar maps from a HYDI >>> dataset that I'm analyzing. For your reference, I am ultimately >>> planning to calculate FWC DTI metrics for the fornix and genu of the corpus >>> callosum after performing probabilistic tractography. I have preprocessed >>> my data using FSL version 6.0 and MRtrix3 on a linux machine. >>> >>> While I have successfully extracted FWC FA, MD, RD, and AD maps >>> from my data using this pipeline, there still seems to be a >>> disproportionate amount of noise in the ventricles, especially when >>> comparing my output to your examples on the website linked above. This is >>> the case even after eliminating voxels with a water volume fraction (WVF) >>> exceeding 70%. In light of this, I was wondering if you may be able to >>> address the following questions: >>> >>> - Is the amount of ventricular noise post-FWE in my scalar >>> maps within a normal range? Will this preclude me from extracting valid >>> FWC DTI metrics from the fornix and the genu? Here are some screenshots >>> from a representative subject's scalar maps: >>> >>> *FA map with WVF elimination at a threshold of 70%* >>> [image: fa_70.png] >>> *MD map with WVF elimination at a threshold of 70%* >>> [image: md_70.png] >>> *RD map with WVF elimination at a threshold of 70%* >>> [image: rd_70.png] >>> *AD map with WVF elimination at a threshold of 70%* >>> [image: ad_70.png] >>> >>> >>> - If this noise is not within an acceptable range, how might I >>> be able optimize our DIPY script so that I can perform a better FWE? I >>> tried comparing the results from using a stricter WVF threshold of 60% as >>> well as using no WVF thresholding to the above results. Using a stricter >>> threshold did not completely eliminate the noise problem, but it did help a >>> little bit. However, I'm not sure if there is a precedent for this level >>> of thresholding in the literature, or if it is actually appropriate. >>> Screenshots from a representative subject are listed below: >>> >>> *FA map with WVF elimination at a threshold of 60%* >>> [image: fa_60.png] >>> >>> *MD map with WVF elimination at a threshold of 60%* >>> [image: md_60.png] >>> *FA map with No WVF elimination threshold* >>> [image: fa_none.png] >>> *MD map with No WVF elimination threshold* >>> [image: md_none.png] >>> >>> I have attached a zip file with the following information for your >>> reference: >>> >>> 1. Input data from a representative subject. This includes >>> DWI volumes collected at b values between 0 to 2000. This is contained in >>> the *subject_data *subfolder. >>> 2. Scalar maps collected with a WVF thresholding rate of 70% ( >>> *F>.7*), 60% (*F>.6*), and with no thresholding ( >>> *no_F_threshold*). >>> 3. Three versions of the DIPY script I've been using - each >>> one accounts for a different rate of WVF thresholding. These >>> scripts are contained in the *dipy_fwe_script_versions* >>> subfolder. >>> >>> I sincerely appreciate all of your time and consideration, and >>> look forward to hearing from you soon! >>> >>> Kind regards, >>> Linda >>> >>> dipyfwe.zip >>> <https://drive.google.com/a/temple.edu/file/d/1yvLYeB-cUNqBoce-43ler0-zf_vG8b...> >>> -- >>> *Lab Manager* >>> *Cognitive Neuroscience Lab* >>> Temple University >>> 1701 N. 13th St. >>> Philadelphia, PA 19122 >>> >>> *Pronouns: * She/Her >>> *Phone*: (215) 204-1708 >>> *Email*: tuf72977@temple.edu >>> >> > > -- > *Lab Manager* > *Cognitive Neuroscience Lab* > Temple University > 1701 N. 13th St. > Philadelphia, PA 19122 > > *Pronouns: * She/Her > *Phone*: (215) 204-1708 > *Email*: tuf72977@temple.edu >
-- *Lab Manager* *Cognitive Neuroscience Lab* Temple University 1701 N. 13th St. Philadelphia, PA 19122
*Pronouns: * She/Her *Phone*: (215) 204-1708 *Email*: tuf72977@temple.edu
-- *Lab Manager* *Cognitive Neuroscience Lab* Temple University 1701 N. 13th St. Philadelphia, PA 19122
*Pronouns: * She/Her *Phone*: (215) 204-1708 *Email*: tuf72977@temple.edu
-- *Lab Manager* *Cognitive Neuroscience Lab* Temple University 1701 N. 13th St. Philadelphia, PA 19122
*Pronouns: * She/Her *Phone*: (215) 204-1708 *Email*: tuf72977@temple.edu
-- *Lab Manager* *Cognitive Neuroscience Lab* Temple University 1701 N. 13th St. Philadelphia, PA 19122
*Pronouns: * She/Her *Phone*: (215) 204-1708 *Email*: tuf72977@temple.edu
-- *Lab Manager* *Cognitive Neuroscience Lab* Temple University 1701 N. 13th St. Philadelphia, PA 19122
*Pronouns: * She/Her *Phone*: (215) 204-1708 *Email*: tuf72977@temple.edu
Hi Ariel, No worries at all! I completely understand you've been very busy, especially with Neurohackademy going on. I appreciate your response! Our lab has tried to implement a QSIprep pipeline in the past, and it presented a number of issues for us that we were unable to resolve, particularly given the nascent nature of the software. In light of this, I would prefer to incorporate the denoising/degibbing procedures into our existing preprocessing protocol if at all possible. I understand that denoising/degibbing must occur before any motion correction or other preprocessing is performed, and I don't believe that performing topup first violates this rule. To clarify, our topup script is executed as follows: *#FSL topup script* for n in 5022 5216 5302 5391 do topup --imain=/data/projects/tbi/dti/${n}/*a2p_p2a_b0.nii.gz * --datain=/data/projects/tbi/dti/acqp.txt --config=b02b0_1.cnf --out=/data/projects/tbi/dti/${n}/topup_output/topup_output --iout=/data/projects/tbi/dti/${n}/topup_output/my_hifi_b0 --fout=/data/projects/tbi/dti/${n}/topup_output/displacement done Note that the only data that goes into the *topup* command are our concatenated anterior-to-posterior and posterior-to-anterior b0 fieldmaps (i.e. *a2p_p2a_b0.nii.gz*). I thought it would be best to do *topup* first since it... 1. ...does not affect our DWI volumes directly - it merely gives us further information concerning motion and the susceptibility-induced field to feed into *eddy* - our most critical preprocessing step. 2. ...yields a high-fidelity brain mask that I was unable to obtain through other means (mainly through unsuccessfully running *bet* on my 4D DWI volumes, and through acquiring a suboptimal brain mask using *dwi2mask* in MRtrix3). I wanted to be sure to include a brain mask in my denoising pipeline since I didn't want the inclusion of skull matter to affect how MRtrix3 estimated the noise structure of my data. Did I go wrong by failing to denoise/degibb my fieldmaps in addition to my DWI volumes? As for problematic noise voxels in my MD image, I have taken the following screenshots for your reference from a representative subject (i.e. 5216): *Noise voxel #1 signal: ~0.05* [image: Screen Shot 2020-08-17 at 1.34.46 PM.png] *Noise voxel #2 singal: ~0.01* [image: Screen Shot 2020-08-17 at 1.35.35 PM.png] I have attached the free-water corrected scalars for this subject, as well as their MRtrix3 extracted bvals/bvecs to this email for your reference. I sincerely appreciate all of your continued time and consideration on this matter! I look forward to hearing from you soon! Kind regards, Linda On Mon, Aug 17, 2020 at 12:43 PM Ariel Rokem <arokem@gmail.com> wrote:
Hi Linda,
Sorry for the slowness here... It's been... challenging.
Two thoughts:
1. Sorry if I wasn't clear about this before: It is usually recommended that denoising and Gibbs ringing removal be done *before *other steps in preprocessing. To be on the safe side, I would recommend using https://qsiprep.readthedocs.io/en/latest/ for preprocessing. It implements the state of the art, and can be run as a docker/singularity container, which simplifies installation issues.
2. I am wondering what the signal is like in these voxels that still appear with very high MD values. Is there something unusual about their B0 signal? Or are the other data so low as to be indistinguishable from the noise floor? If you could find the coordinate of one of these voxels, and then us that to share with us the signal values in this voxel (as well as b-values and b-vectors) it would help diagnose this.
Cheers,
Ariel
On Mon, Aug 17, 2020 at 9:14 AM Linda Jasmine Hoffman <tuf72977@temple.edu> wrote:
Good morning DIPY experts,
I hope you have all been doing well! I just wanted to follow up with you again as per my latest update re: persisting ventricular noise post-denoising & FWC. Please let me know if you can shed any light on why this noise may still be an issue, even after implementing Ariel's denoising/degibbing suggestion.
I look forward to hearing from you soon!
Kind regards, Linda
On Mon, Aug 3, 2020 at 7:33 PM Linda Jasmine Hoffman <tuf72977@temple.edu> wrote:
Good evening DIPY experts,
I just wanted to follow up with you all as per my last email to see if you've had the opportunity to give my questions some consideration.
Please let me know! I look forward to hearing from you soon!
Kind regards, Linda
On Tue, Jul 28, 2020 at 10:36 PM Linda Jasmine Hoffman < tuf72977@temple.edu> wrote:
Good evening DIPY experts,
I have developed a denoising protocol for my HYDI data, and it has afforded me some success in eliminating a portion of the excess ventricular noise that I have been finding in my free-water-corrected (FWC) scalars. Below is an example from a representative subject (i.e. "Subject 1") for whom this course of actions seems to have worked quite well:
*Subject 1: Original MD map (no denoising of DWI data):* [image: 5022_md.png]
*Subject 1: New MD map (with denoising of DWI data):* [image: 5022_md_denoised.png]
However, I have a few concerns. First, my data is still not as clean as I would like it to be, given the persisting residual noise that is still present in the sagittal view. Second, the denoising protocol that I have implemented did not work consistently well for all subjects. Here is an example from a second representative subject (i.e. "Subject 2") to illustrate this issue:
*Subject 2: Original MD map (no denoising of DWI data):* [image: 5216_md.png]
*Subject 2: New MD map (with denoising of DWI data):* [image: 5216_md_denoised.png]
What is particularly concerning about this is that the resultant image for Subject 2 is still not as clean as what is presented on your DIPY free-water elimination page <https://dipy.org/documentation/1.0.0./examples_built/reconst_fwdti/>. It is worth noting that quality assurance measures have been taken for all of our data, and this subject did not exhibit inordinate imaging artifacts.
For your reference, my denoising pipeline utilized the *dwidenoise* and *mrdegibbs* functions in MRtrix3. I incorporated these steps into my processing protocol in the following order:
1. FSL - topup 2. MRtrix3 - dwidenoise 3. MRtrix3 - mrdegibbs 4. FSL - eddy
Note that I completed *topup* first since this step does not affect the raw, DICOM-to-NIfTI-converted DWI volumes in any way, and it is necessary for yielding a hifi brain mask. The scripts that I used for denoising/degibbing are delineated below:
*#dMRI noise level estimation and denoising using Marchenko-Pastur PCA:* for n in 5022 5216 5302 5391 do
dwidenoise
-mask /data/projects/tbi/denoise/${n}/topup_output/my_hifi_b0_Tcollapsed_brain_mask.nii.gz -noise /data/projects/tbi/denoise/${n}/dwidenoise/noise_hifi_map.nii /data/projects/tbi/denoise/${n}/6-cmrr_mb3hydi_ipat2_64ch/output.nii /data/projects/tbi/denoise/${n}/dwidenoise/denoised_hifi_vol.nii
done
*#Remove Gibbs Ringing Artifacts:* for n in 5022 5216 5302 5391 do
mrdegibbs
/data/projects/tbi/denoise/${n}/dwidenoise/denoised_hifi_vol.nii /data/projects/tbi/denoise/${n}/mrdegibbs/denoised_degibbs_hifi_vol.nii
done
What are your thoughts on the scripts I have implemented? Might I have done something incorrectly, or is there something further I should do to optimize this denoising pipeline? Is there anything I can do in addition to denoising to eliminate these undue levels of post-FWC ventricular noise in my scalars?
Finally, do you recommend denoising and degibbing DWI data as a canonical part of my pipeline? I ask because I know there is a tradeoff between SNR and spatial resolution following noise reduction procedures, so I'm curious to know what best-practices are in this regard. At the very least it seems like an important step if one intends to pursue FWE.
I sincerely appreciate all of your time and consideration on this matter.
Kind regards, Linda
On Thu, Jul 9, 2020 at 5:40 PM Linda Jasmine Hoffman < tuf72977@temple.edu> wrote:
I haven't; I'll try that now.
Thank you! Linda
On Thu, Jul 9, 2020 at 5:38 PM Ariel Rokem <arokem@gmail.com> wrote:
Hi Linda,
Have you had a chance to try Gibbs ringing removal or and/or denoising on at least one subject?
Cheers,
Ariel
On Thu, Jul 9, 2020 at 2:34 PM Linda Jasmine Hoffman < tuf72977@temple.edu> wrote:
> Hi everyone, > > I just wanted to touch base with you to see if you've had the > opportunity to give my previous email some consideration. Please let me > know what my next steps should be re: denoising my DWI data to > eliminate excessive ventricular artifacts post-fwc. > > Thank you! > Linda > > On Thu, Jul 2, 2020 at 12:41 PM Linda Jasmine Hoffman < > tuf72977@temple.edu> wrote: > >> Hi Ariel, >> >> Our preprocessing pipeline includes the following steps for noise >> reduction in FSL: >> >> - topup - correct for the susceptibility induced field and >> movement >> - eddy - correct for eddy current distortions and movement >> >> We don't have a step in our pipeline to correct for Gibbs >> artifacts. Do you think this particular type of artifact is what's >> underpinning this issue with the FWC scalar maps? If so, I found a command >> in MRtrix3 (mrdegibbs) that eliminates ringing, but it will necessitate >> that I go back and redo a large amount of preprocessing. Do you know of an >> alternative route to mitigate this problem that may obviate my need to >> reprocess my data? >> >> Thank you so much for your help! >> Kind regards, >> Linda >> >> On Thu, Jul 2, 2020 at 12:45 AM Ariel Rokem <arokem@gmail.com> >> wrote: >> >>> Hi Linda, >>> >>> With your permission, I am adding the DIPY mailing list, so others >>> can weigh in and/or benefit from the discussion. >>> >>> My hunch is that the noise you are seeing in the ventricles is due >>> to artifacts/noise. Do you do any removal of Gibbs ringing artifacts or any >>> denoising of the data before analyzing it with fwdti? >>> >>> Cheers, >>> >>> Ariel >>> >>> >>> On Wed, Jul 1, 2020 at 12:22 PM Linda Jasmine Hoffman < >>> tuf72977@temple.edu> wrote: >>> >>>> Good afternoon DIPY experts, >>>> >>>> My name is Linda Hoffman, and I'm the lab manager for Dr. Ingrid >>>> Olson's Cognitive Neuroscience Lab at Temple University. I have been >>>> working on implementing a DIPY-based free-water elimination (FWE) pipeline >>>> that my labmate, Katie Jobson, adapted from your website >>>> <https://dipy.org/documentation/1.0.0./examples_built/reconst_fwdti/> >>>> in order to extract free-water corrected (FWC) scalar maps from a HYDI >>>> dataset that I'm analyzing. For your reference, I am ultimately >>>> planning to calculate FWC DTI metrics for the fornix and genu of the corpus >>>> callosum after performing probabilistic tractography. I have preprocessed >>>> my data using FSL version 6.0 and MRtrix3 on a linux machine. >>>> >>>> While I have successfully extracted FWC FA, MD, RD, and AD maps >>>> from my data using this pipeline, there still seems to be a >>>> disproportionate amount of noise in the ventricles, especially when >>>> comparing my output to your examples on the website linked above. This is >>>> the case even after eliminating voxels with a water volume fraction (WVF) >>>> exceeding 70%. In light of this, I was wondering if you may be able to >>>> address the following questions: >>>> >>>> - Is the amount of ventricular noise post-FWE in my scalar >>>> maps within a normal range? Will this preclude me from extracting valid >>>> FWC DTI metrics from the fornix and the genu? Here are some screenshots >>>> from a representative subject's scalar maps: >>>> >>>> *FA map with WVF elimination at a threshold of 70%* >>>> [image: fa_70.png] >>>> *MD map with WVF elimination at a threshold of 70%* >>>> [image: md_70.png] >>>> *RD map with WVF elimination at a threshold of 70%* >>>> [image: rd_70.png] >>>> *AD map with WVF elimination at a threshold of 70%* >>>> [image: ad_70.png] >>>> >>>> >>>> - If this noise is not within an acceptable range, how might >>>> I be able optimize our DIPY script so that I can perform a better FWE? I >>>> tried comparing the results from using a stricter WVF threshold of 60% as >>>> well as using no WVF thresholding to the above results. Using a stricter >>>> threshold did not completely eliminate the noise problem, but it did help a >>>> little bit. However, I'm not sure if there is a precedent for this level >>>> of thresholding in the literature, or if it is actually appropriate. >>>> Screenshots from a representative subject are listed below: >>>> >>>> *FA map with WVF elimination at a threshold of 60%* >>>> [image: fa_60.png] >>>> >>>> *MD map with WVF elimination at a threshold of 60%* >>>> [image: md_60.png] >>>> *FA map with No WVF elimination threshold* >>>> [image: fa_none.png] >>>> *MD map with No WVF elimination threshold* >>>> [image: md_none.png] >>>> >>>> I have attached a zip file with the following information for >>>> your reference: >>>> >>>> 1. Input data from a representative subject. This includes >>>> DWI volumes collected at b values between 0 to 2000. This is contained in >>>> the *subject_data *subfolder. >>>> 2. Scalar maps collected with a WVF thresholding rate of 70% ( >>>> *F>.7*), 60% (*F>.6*), and with no thresholding ( >>>> *no_F_threshold*). >>>> 3. Three versions of the DIPY script I've been using - each >>>> one accounts for a different rate of WVF thresholding. These >>>> scripts are contained in the *dipy_fwe_script_versions* >>>> subfolder. >>>> >>>> I sincerely appreciate all of your time and consideration, and >>>> look forward to hearing from you soon! >>>> >>>> Kind regards, >>>> Linda >>>> >>>> dipyfwe.zip >>>> <https://drive.google.com/a/temple.edu/file/d/1yvLYeB-cUNqBoce-43ler0-zf_vG8b...> >>>> -- >>>> *Lab Manager* >>>> *Cognitive Neuroscience Lab* >>>> Temple University >>>> 1701 N. 13th St. >>>> Philadelphia, PA 19122 >>>> >>>> *Pronouns: * She/Her >>>> *Phone*: (215) 204-1708 >>>> *Email*: tuf72977@temple.edu >>>> >>> >> >> -- >> *Lab Manager* >> *Cognitive Neuroscience Lab* >> Temple University >> 1701 N. 13th St. >> Philadelphia, PA 19122 >> >> *Pronouns: * She/Her >> *Phone*: (215) 204-1708 >> *Email*: tuf72977@temple.edu >> > > > -- > *Lab Manager* > *Cognitive Neuroscience Lab* > Temple University > 1701 N. 13th St. > Philadelphia, PA 19122 > > *Pronouns: * She/Her > *Phone*: (215) 204-1708 > *Email*: tuf72977@temple.edu >
-- *Lab Manager* *Cognitive Neuroscience Lab* Temple University 1701 N. 13th St. Philadelphia, PA 19122
*Pronouns: * She/Her *Phone*: (215) 204-1708 *Email*: tuf72977@temple.edu
-- *Lab Manager* *Cognitive Neuroscience Lab* Temple University 1701 N. 13th St. Philadelphia, PA 19122
*Pronouns: * She/Her *Phone*: (215) 204-1708 *Email*: tuf72977@temple.edu
-- *Lab Manager* *Cognitive Neuroscience Lab* Temple University 1701 N. 13th St. Philadelphia, PA 19122
*Pronouns: * She/Her *Phone*: (215) 204-1708 *Email*: tuf72977@temple.edu
-- *Lab Manager* *Cognitive Neuroscience Lab* Temple University 1701 N. 13th St. Philadelphia, PA 19122
*Pronouns: * She/Her *Phone*: (215) 204-1708 *Email*: tuf72977@temple.edu
-- *Lab Manager* *Cognitive Neuroscience Lab* Temple University 1701 N. 13th St. Philadelphia, PA 19122 *Pronouns: * She/Her *Phone*: (215) 204-1708 *Email*: tuf72977@temple.edu
Good afternoon Ariel, I just wanted to follow up with you as per my last email to see if you had any thoughts. Thank you so much! Kind regards, Linda On Mon, Aug 17, 2020 at 1:51 PM Linda Jasmine Hoffman <tuf72977@temple.edu> wrote:
Hi Ariel,
No worries at all! I completely understand you've been very busy, especially with Neurohackademy going on. I appreciate your response!
Our lab has tried to implement a QSIprep pipeline in the past, and it presented a number of issues for us that we were unable to resolve, particularly given the nascent nature of the software. In light of this, I would prefer to incorporate the denoising/degibbing procedures into our existing preprocessing protocol if at all possible. I understand that denoising/degibbing must occur before any motion correction or other preprocessing is performed, and I don't believe that performing topup first violates this rule. To clarify, our topup script is executed as follows:
*#FSL topup script*
for n in 5022 5216 5302 5391 do
topup
--imain=/data/projects/tbi/dti/${n}/*a2p_p2a_b0.nii.gz *
--datain=/data/projects/tbi/dti/acqp.txt
--config=b02b0_1.cnf
--out=/data/projects/tbi/dti/${n}/topup_output/topup_output
--iout=/data/projects/tbi/dti/${n}/topup_output/my_hifi_b0
--fout=/data/projects/tbi/dti/${n}/topup_output/displacement
done
Note that the only data that goes into the *topup* command are our concatenated anterior-to-posterior and posterior-to-anterior b0 fieldmaps (i.e. *a2p_p2a_b0.nii.gz*). I thought it would be best to do *topup* first since it...
1. ...does not affect our DWI volumes directly - it merely gives us further information concerning motion and the susceptibility-induced field to feed into *eddy* - our most critical preprocessing step. 2. ...yields a high-fidelity brain mask that I was unable to obtain through other means (mainly through unsuccessfully running *bet* on my 4D DWI volumes, and through acquiring a suboptimal brain mask using *dwi2mask* in MRtrix3).
I wanted to be sure to include a brain mask in my denoising pipeline since I didn't want the inclusion of skull matter to affect how MRtrix3 estimated the noise structure of my data. Did I go wrong by failing to denoise/degibb my fieldmaps in addition to my DWI volumes?
As for problematic noise voxels in my MD image, I have taken the following screenshots for your reference from a representative subject (i.e. 5216):
*Noise voxel #1 signal: ~0.05* [image: Screen Shot 2020-08-17 at 1.34.46 PM.png]
*Noise voxel #2 singal: ~0.01* [image: Screen Shot 2020-08-17 at 1.35.35 PM.png]
I have attached the free-water corrected scalars for this subject, as well as their MRtrix3 extracted bvals/bvecs to this email for your reference. I sincerely appreciate all of your continued time and consideration on this matter!
I look forward to hearing from you soon!
Kind regards, Linda
On Mon, Aug 17, 2020 at 12:43 PM Ariel Rokem <arokem@gmail.com> wrote:
Hi Linda,
Sorry for the slowness here... It's been... challenging.
Two thoughts:
1. Sorry if I wasn't clear about this before: It is usually recommended that denoising and Gibbs ringing removal be done *before *other steps in preprocessing. To be on the safe side, I would recommend using https://qsiprep.readthedocs.io/en/latest/ for preprocessing. It implements the state of the art, and can be run as a docker/singularity container, which simplifies installation issues.
2. I am wondering what the signal is like in these voxels that still appear with very high MD values. Is there something unusual about their B0 signal? Or are the other data so low as to be indistinguishable from the noise floor? If you could find the coordinate of one of these voxels, and then us that to share with us the signal values in this voxel (as well as b-values and b-vectors) it would help diagnose this.
Cheers,
Ariel
On Mon, Aug 17, 2020 at 9:14 AM Linda Jasmine Hoffman < tuf72977@temple.edu> wrote:
Good morning DIPY experts,
I hope you have all been doing well! I just wanted to follow up with you again as per my latest update re: persisting ventricular noise post-denoising & FWC. Please let me know if you can shed any light on why this noise may still be an issue, even after implementing Ariel's denoising/degibbing suggestion.
I look forward to hearing from you soon!
Kind regards, Linda
On Mon, Aug 3, 2020 at 7:33 PM Linda Jasmine Hoffman < tuf72977@temple.edu> wrote:
Good evening DIPY experts,
I just wanted to follow up with you all as per my last email to see if you've had the opportunity to give my questions some consideration.
Please let me know! I look forward to hearing from you soon!
Kind regards, Linda
On Tue, Jul 28, 2020 at 10:36 PM Linda Jasmine Hoffman < tuf72977@temple.edu> wrote:
Good evening DIPY experts,
I have developed a denoising protocol for my HYDI data, and it has afforded me some success in eliminating a portion of the excess ventricular noise that I have been finding in my free-water-corrected (FWC) scalars. Below is an example from a representative subject (i.e. "Subject 1") for whom this course of actions seems to have worked quite well:
*Subject 1: Original MD map (no denoising of DWI data):* [image: 5022_md.png]
*Subject 1: New MD map (with denoising of DWI data):* [image: 5022_md_denoised.png]
However, I have a few concerns. First, my data is still not as clean as I would like it to be, given the persisting residual noise that is still present in the sagittal view. Second, the denoising protocol that I have implemented did not work consistently well for all subjects. Here is an example from a second representative subject (i.e. "Subject 2") to illustrate this issue:
*Subject 2: Original MD map (no denoising of DWI data):* [image: 5216_md.png]
*Subject 2: New MD map (with denoising of DWI data):* [image: 5216_md_denoised.png]
What is particularly concerning about this is that the resultant image for Subject 2 is still not as clean as what is presented on your DIPY free-water elimination page <https://dipy.org/documentation/1.0.0./examples_built/reconst_fwdti/>. It is worth noting that quality assurance measures have been taken for all of our data, and this subject did not exhibit inordinate imaging artifacts.
For your reference, my denoising pipeline utilized the *dwidenoise* and *mrdegibbs* functions in MRtrix3. I incorporated these steps into my processing protocol in the following order:
1. FSL - topup 2. MRtrix3 - dwidenoise 3. MRtrix3 - mrdegibbs 4. FSL - eddy
Note that I completed *topup* first since this step does not affect the raw, DICOM-to-NIfTI-converted DWI volumes in any way, and it is necessary for yielding a hifi brain mask. The scripts that I used for denoising/degibbing are delineated below:
*#dMRI noise level estimation and denoising using Marchenko-Pastur PCA:* for n in 5022 5216 5302 5391 do
dwidenoise
-mask /data/projects/tbi/denoise/${n}/topup_output/my_hifi_b0_Tcollapsed_brain_mask.nii.gz -noise /data/projects/tbi/denoise/${n}/dwidenoise/noise_hifi_map.nii /data/projects/tbi/denoise/${n}/6-cmrr_mb3hydi_ipat2_64ch/output.nii /data/projects/tbi/denoise/${n}/dwidenoise/denoised_hifi_vol.nii
done
*#Remove Gibbs Ringing Artifacts:* for n in 5022 5216 5302 5391 do
mrdegibbs
/data/projects/tbi/denoise/${n}/dwidenoise/denoised_hifi_vol.nii /data/projects/tbi/denoise/${n}/mrdegibbs/denoised_degibbs_hifi_vol.nii
done
What are your thoughts on the scripts I have implemented? Might I have done something incorrectly, or is there something further I should do to optimize this denoising pipeline? Is there anything I can do in addition to denoising to eliminate these undue levels of post-FWC ventricular noise in my scalars?
Finally, do you recommend denoising and degibbing DWI data as a canonical part of my pipeline? I ask because I know there is a tradeoff between SNR and spatial resolution following noise reduction procedures, so I'm curious to know what best-practices are in this regard. At the very least it seems like an important step if one intends to pursue FWE.
I sincerely appreciate all of your time and consideration on this matter.
Kind regards, Linda
On Thu, Jul 9, 2020 at 5:40 PM Linda Jasmine Hoffman < tuf72977@temple.edu> wrote:
I haven't; I'll try that now.
Thank you! Linda
On Thu, Jul 9, 2020 at 5:38 PM Ariel Rokem <arokem@gmail.com> wrote:
> Hi Linda, > > Have you had a chance to try Gibbs ringing removal or and/or > denoising on at least one subject? > > Cheers, > > Ariel > > > > On Thu, Jul 9, 2020 at 2:34 PM Linda Jasmine Hoffman < > tuf72977@temple.edu> wrote: > >> Hi everyone, >> >> I just wanted to touch base with you to see if you've had the >> opportunity to give my previous email some consideration. Please let me >> know what my next steps should be re: denoising my DWI data to >> eliminate excessive ventricular artifacts post-fwc. >> >> Thank you! >> Linda >> >> On Thu, Jul 2, 2020 at 12:41 PM Linda Jasmine Hoffman < >> tuf72977@temple.edu> wrote: >> >>> Hi Ariel, >>> >>> Our preprocessing pipeline includes the following steps for noise >>> reduction in FSL: >>> >>> - topup - correct for the susceptibility induced field and >>> movement >>> - eddy - correct for eddy current distortions and movement >>> >>> We don't have a step in our pipeline to correct for Gibbs >>> artifacts. Do you think this particular type of artifact is what's >>> underpinning this issue with the FWC scalar maps? If so, I found a command >>> in MRtrix3 (mrdegibbs) that eliminates ringing, but it will necessitate >>> that I go back and redo a large amount of preprocessing. Do you know of an >>> alternative route to mitigate this problem that may obviate my need to >>> reprocess my data? >>> >>> Thank you so much for your help! >>> Kind regards, >>> Linda >>> >>> On Thu, Jul 2, 2020 at 12:45 AM Ariel Rokem <arokem@gmail.com> >>> wrote: >>> >>>> Hi Linda, >>>> >>>> With your permission, I am adding the DIPY mailing list, so >>>> others can weigh in and/or benefit from the discussion. >>>> >>>> My hunch is that the noise you are seeing in the ventricles is >>>> due to artifacts/noise. Do you do any removal of Gibbs ringing artifacts or >>>> any denoising of the data before analyzing it with fwdti? >>>> >>>> Cheers, >>>> >>>> Ariel >>>> >>>> >>>> On Wed, Jul 1, 2020 at 12:22 PM Linda Jasmine Hoffman < >>>> tuf72977@temple.edu> wrote: >>>> >>>>> Good afternoon DIPY experts, >>>>> >>>>> My name is Linda Hoffman, and I'm the lab manager for Dr. Ingrid >>>>> Olson's Cognitive Neuroscience Lab at Temple University. I have been >>>>> working on implementing a DIPY-based free-water elimination (FWE) pipeline >>>>> that my labmate, Katie Jobson, adapted from your website >>>>> <https://dipy.org/documentation/1.0.0./examples_built/reconst_fwdti/> >>>>> in order to extract free-water corrected (FWC) scalar maps from a HYDI >>>>> dataset that I'm analyzing. For your reference, I am ultimately >>>>> planning to calculate FWC DTI metrics for the fornix and genu of the corpus >>>>> callosum after performing probabilistic tractography. I have preprocessed >>>>> my data using FSL version 6.0 and MRtrix3 on a linux machine. >>>>> >>>>> While I have successfully extracted FWC FA, MD, RD, and AD maps >>>>> from my data using this pipeline, there still seems to be a >>>>> disproportionate amount of noise in the ventricles, especially when >>>>> comparing my output to your examples on the website linked above. This is >>>>> the case even after eliminating voxels with a water volume fraction (WVF) >>>>> exceeding 70%. In light of this, I was wondering if you may be able to >>>>> address the following questions: >>>>> >>>>> - Is the amount of ventricular noise post-FWE in my scalar >>>>> maps within a normal range? Will this preclude me from extracting valid >>>>> FWC DTI metrics from the fornix and the genu? Here are some screenshots >>>>> from a representative subject's scalar maps: >>>>> >>>>> *FA map with WVF elimination at a threshold of 70%* >>>>> [image: fa_70.png] >>>>> *MD map with WVF elimination at a threshold of 70%* >>>>> [image: md_70.png] >>>>> *RD map with WVF elimination at a threshold of 70%* >>>>> [image: rd_70.png] >>>>> *AD map with WVF elimination at a threshold of 70%* >>>>> [image: ad_70.png] >>>>> >>>>> >>>>> - If this noise is not within an acceptable range, how might >>>>> I be able optimize our DIPY script so that I can perform a better FWE? I >>>>> tried comparing the results from using a stricter WVF threshold of 60% as >>>>> well as using no WVF thresholding to the above results. Using a stricter >>>>> threshold did not completely eliminate the noise problem, but it did help a >>>>> little bit. However, I'm not sure if there is a precedent for this level >>>>> of thresholding in the literature, or if it is actually appropriate. >>>>> Screenshots from a representative subject are listed below: >>>>> >>>>> *FA map with WVF elimination at a threshold of 60%* >>>>> [image: fa_60.png] >>>>> >>>>> *MD map with WVF elimination at a threshold of 60%* >>>>> [image: md_60.png] >>>>> *FA map with No WVF elimination threshold* >>>>> [image: fa_none.png] >>>>> *MD map with No WVF elimination threshold* >>>>> [image: md_none.png] >>>>> >>>>> I have attached a zip file with the following information for >>>>> your reference: >>>>> >>>>> 1. Input data from a representative subject. This includes >>>>> DWI volumes collected at b values between 0 to 2000. This is contained in >>>>> the *subject_data *subfolder. >>>>> 2. Scalar maps collected with a WVF thresholding rate of 70% >>>>> (*F>.7*), 60% (*F>.6*), and with no thresholding ( >>>>> *no_F_threshold*). >>>>> 3. Three versions of the DIPY script I've been using - each >>>>> one accounts for a different rate of WVF thresholding. >>>>> These scripts are contained in the *dipy_fwe_script_versions* >>>>> subfolder. >>>>> >>>>> I sincerely appreciate all of your time and consideration, and >>>>> look forward to hearing from you soon! >>>>> >>>>> Kind regards, >>>>> Linda >>>>> >>>>> dipyfwe.zip >>>>> <https://drive.google.com/a/temple.edu/file/d/1yvLYeB-cUNqBoce-43ler0-zf_vG8b...> >>>>> -- >>>>> *Lab Manager* >>>>> *Cognitive Neuroscience Lab* >>>>> Temple University >>>>> 1701 N. 13th St. >>>>> Philadelphia, PA 19122 >>>>> >>>>> *Pronouns: * She/Her >>>>> *Phone*: (215) 204-1708 >>>>> *Email*: tuf72977@temple.edu >>>>> >>>> >>> >>> -- >>> *Lab Manager* >>> *Cognitive Neuroscience Lab* >>> Temple University >>> 1701 N. 13th St. >>> Philadelphia, PA 19122 >>> >>> *Pronouns: * She/Her >>> *Phone*: (215) 204-1708 >>> *Email*: tuf72977@temple.edu >>> >> >> >> -- >> *Lab Manager* >> *Cognitive Neuroscience Lab* >> Temple University >> 1701 N. 13th St. >> Philadelphia, PA 19122 >> >> *Pronouns: * She/Her >> *Phone*: (215) 204-1708 >> *Email*: tuf72977@temple.edu >> >
-- *Lab Manager* *Cognitive Neuroscience Lab* Temple University 1701 N. 13th St. Philadelphia, PA 19122
*Pronouns: * She/Her *Phone*: (215) 204-1708 *Email*: tuf72977@temple.edu
-- *Lab Manager* *Cognitive Neuroscience Lab* Temple University 1701 N. 13th St. Philadelphia, PA 19122
*Pronouns: * She/Her *Phone*: (215) 204-1708 *Email*: tuf72977@temple.edu
-- *Lab Manager* *Cognitive Neuroscience Lab* Temple University 1701 N. 13th St. Philadelphia, PA 19122
*Pronouns: * She/Her *Phone*: (215) 204-1708 *Email*: tuf72977@temple.edu
-- *Lab Manager* *Cognitive Neuroscience Lab* Temple University 1701 N. 13th St. Philadelphia, PA 19122
*Pronouns: * She/Her *Phone*: (215) 204-1708 *Email*: tuf72977@temple.edu
-- *Lab Manager* *Cognitive Neuroscience Lab* Temple University 1701 N. 13th St. Philadelphia, PA 19122
*Pronouns: * She/Her *Phone*: (215) 204-1708 *Email*: tuf72977@temple.edu
-- *Lab Manager* *Cognitive Neuroscience Lab* Temple University 1701 N. 13th St. Philadelphia, PA 19122 *Pronouns: * She/Her *Phone*: (215) 204-1708 *Email*: tuf72977@temple.edu
Hi Linda, On Mon, Aug 17, 2020 at 10:51 AM Linda Jasmine Hoffman <tuf72977@temple.edu> wrote:
Hi Ariel,
No worries at all! I completely understand you've been very busy, especially with Neurohackademy going on. I appreciate your response!
Our lab has tried to implement a QSIprep pipeline in the past, and it presented a number of issues for us that we were unable to resolve, particularly given the nascent nature of the software. In light of this, I would prefer to incorporate the denoising/degibbing procedures into our existing preprocessing protocol if at all possible. I understand that denoising/degibbing must occur before any motion correction or other preprocessing is performed, and I don't believe that performing topup first violates this rule. To clarify, our topup script is executed as follows:
*#FSL topup script*
for n in 5022 5216 5302 5391 do
topup
--imain=/data/projects/tbi/dti/${n}/*a2p_p2a_b0.nii.gz *
--datain=/data/projects/tbi/dti/acqp.txt
--config=b02b0_1.cnf
--out=/data/projects/tbi/dti/${n}/topup_output/topup_output
--iout=/data/projects/tbi/dti/${n}/topup_output/my_hifi_b0
--fout=/data/projects/tbi/dti/${n}/topup_output/displacement
done
Note that the only data that goes into the *topup* command are our concatenated anterior-to-posterior and posterior-to-anterior b0 fieldmaps (i.e. *a2p_p2a_b0.nii.gz*). I thought it would be best to do *topup* first since it...
1. ...does not affect our DWI volumes directly - it merely gives us further information concerning motion and the susceptibility-induced field to feed into *eddy* - our most critical preprocessing step. 2. ...yields a high-fidelity brain mask that I was unable to obtain through other means (mainly through unsuccessfully running *bet* on my 4D DWI volumes, and through acquiring a suboptimal brain mask using *dwi2mask* in MRtrix3).
That makes sense.
I wanted to be sure to include a brain mask in my denoising pipeline since I didn't want the inclusion of skull matter to affect how MRtrix3 estimated the noise structure of my data. Did I go wrong by failing to denoise/degibb my fieldmaps in addition to my DWI volumes?
I don't think the fieldmaps need to be denoised.
As for problematic noise voxels in my MD image, I have taken the following screenshots for your reference from a representative subject (i.e. 5216):
*Noise voxel #1 signal: ~0.05* [image: Screen Shot 2020-08-17 at 1.34.46 PM.png]
*Noise voxel #2 singal: ~0.01* [image: Screen Shot 2020-08-17 at 1.35.35 PM.png]
I have attached the free-water corrected scalars for this subject, as well as their MRtrix3 extracted bvals/bvecs to this email for your reference. I sincerely appreciate all of your continued time and consideration on this matter!
Any chance you could share the preprocessed data for the whole volume for
this individual? It's hard to say without looking at the values in these voxels. My hypothesis is that these are simply voxels where the signal is very low. In which case, I don't think there is any harm in masking them out.
I look forward to hearing from you soon!
Kind regards, Linda
On Mon, Aug 17, 2020 at 12:43 PM Ariel Rokem <arokem@gmail.com> wrote:
Hi Linda,
Sorry for the slowness here... It's been... challenging.
Two thoughts:
1. Sorry if I wasn't clear about this before: It is usually recommended that denoising and Gibbs ringing removal be done *before *other steps in preprocessing. To be on the safe side, I would recommend using https://qsiprep.readthedocs.io/en/latest/ for preprocessing. It implements the state of the art, and can be run as a docker/singularity container, which simplifies installation issues.
2. I am wondering what the signal is like in these voxels that still appear with very high MD values. Is there something unusual about their B0 signal? Or are the other data so low as to be indistinguishable from the noise floor? If you could find the coordinate of one of these voxels, and then us that to share with us the signal values in this voxel (as well as b-values and b-vectors) it would help diagnose this.
Cheers,
Ariel
On Mon, Aug 17, 2020 at 9:14 AM Linda Jasmine Hoffman < tuf72977@temple.edu> wrote:
Good morning DIPY experts,
I hope you have all been doing well! I just wanted to follow up with you again as per my latest update re: persisting ventricular noise post-denoising & FWC. Please let me know if you can shed any light on why this noise may still be an issue, even after implementing Ariel's denoising/degibbing suggestion.
I look forward to hearing from you soon!
Kind regards, Linda
On Mon, Aug 3, 2020 at 7:33 PM Linda Jasmine Hoffman < tuf72977@temple.edu> wrote:
Good evening DIPY experts,
I just wanted to follow up with you all as per my last email to see if you've had the opportunity to give my questions some consideration.
Please let me know! I look forward to hearing from you soon!
Kind regards, Linda
On Tue, Jul 28, 2020 at 10:36 PM Linda Jasmine Hoffman < tuf72977@temple.edu> wrote:
Good evening DIPY experts,
I have developed a denoising protocol for my HYDI data, and it has afforded me some success in eliminating a portion of the excess ventricular noise that I have been finding in my free-water-corrected (FWC) scalars. Below is an example from a representative subject (i.e. "Subject 1") for whom this course of actions seems to have worked quite well:
*Subject 1: Original MD map (no denoising of DWI data):* [image: 5022_md.png]
*Subject 1: New MD map (with denoising of DWI data):* [image: 5022_md_denoised.png]
However, I have a few concerns. First, my data is still not as clean as I would like it to be, given the persisting residual noise that is still present in the sagittal view. Second, the denoising protocol that I have implemented did not work consistently well for all subjects. Here is an example from a second representative subject (i.e. "Subject 2") to illustrate this issue:
*Subject 2: Original MD map (no denoising of DWI data):* [image: 5216_md.png]
*Subject 2: New MD map (with denoising of DWI data):* [image: 5216_md_denoised.png]
What is particularly concerning about this is that the resultant image for Subject 2 is still not as clean as what is presented on your DIPY free-water elimination page <https://dipy.org/documentation/1.0.0./examples_built/reconst_fwdti/>. It is worth noting that quality assurance measures have been taken for all of our data, and this subject did not exhibit inordinate imaging artifacts.
For your reference, my denoising pipeline utilized the *dwidenoise* and *mrdegibbs* functions in MRtrix3. I incorporated these steps into my processing protocol in the following order:
1. FSL - topup 2. MRtrix3 - dwidenoise 3. MRtrix3 - mrdegibbs 4. FSL - eddy
Note that I completed *topup* first since this step does not affect the raw, DICOM-to-NIfTI-converted DWI volumes in any way, and it is necessary for yielding a hifi brain mask. The scripts that I used for denoising/degibbing are delineated below:
*#dMRI noise level estimation and denoising using Marchenko-Pastur PCA:* for n in 5022 5216 5302 5391 do
dwidenoise
-mask /data/projects/tbi/denoise/${n}/topup_output/my_hifi_b0_Tcollapsed_brain_mask.nii.gz -noise /data/projects/tbi/denoise/${n}/dwidenoise/noise_hifi_map.nii /data/projects/tbi/denoise/${n}/6-cmrr_mb3hydi_ipat2_64ch/output.nii /data/projects/tbi/denoise/${n}/dwidenoise/denoised_hifi_vol.nii
done
*#Remove Gibbs Ringing Artifacts:* for n in 5022 5216 5302 5391 do
mrdegibbs
/data/projects/tbi/denoise/${n}/dwidenoise/denoised_hifi_vol.nii /data/projects/tbi/denoise/${n}/mrdegibbs/denoised_degibbs_hifi_vol.nii
done
What are your thoughts on the scripts I have implemented? Might I have done something incorrectly, or is there something further I should do to optimize this denoising pipeline? Is there anything I can do in addition to denoising to eliminate these undue levels of post-FWC ventricular noise in my scalars?
Finally, do you recommend denoising and degibbing DWI data as a canonical part of my pipeline? I ask because I know there is a tradeoff between SNR and spatial resolution following noise reduction procedures, so I'm curious to know what best-practices are in this regard. At the very least it seems like an important step if one intends to pursue FWE.
I sincerely appreciate all of your time and consideration on this matter.
Kind regards, Linda
On Thu, Jul 9, 2020 at 5:40 PM Linda Jasmine Hoffman < tuf72977@temple.edu> wrote:
I haven't; I'll try that now.
Thank you! Linda
On Thu, Jul 9, 2020 at 5:38 PM Ariel Rokem <arokem@gmail.com> wrote:
> Hi Linda, > > Have you had a chance to try Gibbs ringing removal or and/or > denoising on at least one subject? > > Cheers, > > Ariel > > > > On Thu, Jul 9, 2020 at 2:34 PM Linda Jasmine Hoffman < > tuf72977@temple.edu> wrote: > >> Hi everyone, >> >> I just wanted to touch base with you to see if you've had the >> opportunity to give my previous email some consideration. Please let me >> know what my next steps should be re: denoising my DWI data to >> eliminate excessive ventricular artifacts post-fwc. >> >> Thank you! >> Linda >> >> On Thu, Jul 2, 2020 at 12:41 PM Linda Jasmine Hoffman < >> tuf72977@temple.edu> wrote: >> >>> Hi Ariel, >>> >>> Our preprocessing pipeline includes the following steps for noise >>> reduction in FSL: >>> >>> - topup - correct for the susceptibility induced field and >>> movement >>> - eddy - correct for eddy current distortions and movement >>> >>> We don't have a step in our pipeline to correct for Gibbs >>> artifacts. Do you think this particular type of artifact is what's >>> underpinning this issue with the FWC scalar maps? If so, I found a command >>> in MRtrix3 (mrdegibbs) that eliminates ringing, but it will necessitate >>> that I go back and redo a large amount of preprocessing. Do you know of an >>> alternative route to mitigate this problem that may obviate my need to >>> reprocess my data? >>> >>> Thank you so much for your help! >>> Kind regards, >>> Linda >>> >>> On Thu, Jul 2, 2020 at 12:45 AM Ariel Rokem <arokem@gmail.com> >>> wrote: >>> >>>> Hi Linda, >>>> >>>> With your permission, I am adding the DIPY mailing list, so >>>> others can weigh in and/or benefit from the discussion. >>>> >>>> My hunch is that the noise you are seeing in the ventricles is >>>> due to artifacts/noise. Do you do any removal of Gibbs ringing artifacts or >>>> any denoising of the data before analyzing it with fwdti? >>>> >>>> Cheers, >>>> >>>> Ariel >>>> >>>> >>>> On Wed, Jul 1, 2020 at 12:22 PM Linda Jasmine Hoffman < >>>> tuf72977@temple.edu> wrote: >>>> >>>>> Good afternoon DIPY experts, >>>>> >>>>> My name is Linda Hoffman, and I'm the lab manager for Dr. Ingrid >>>>> Olson's Cognitive Neuroscience Lab at Temple University. I have been >>>>> working on implementing a DIPY-based free-water elimination (FWE) pipeline >>>>> that my labmate, Katie Jobson, adapted from your website >>>>> <https://dipy.org/documentation/1.0.0./examples_built/reconst_fwdti/> >>>>> in order to extract free-water corrected (FWC) scalar maps from a HYDI >>>>> dataset that I'm analyzing. For your reference, I am ultimately >>>>> planning to calculate FWC DTI metrics for the fornix and genu of the corpus >>>>> callosum after performing probabilistic tractography. I have preprocessed >>>>> my data using FSL version 6.0 and MRtrix3 on a linux machine. >>>>> >>>>> While I have successfully extracted FWC FA, MD, RD, and AD maps >>>>> from my data using this pipeline, there still seems to be a >>>>> disproportionate amount of noise in the ventricles, especially when >>>>> comparing my output to your examples on the website linked above. This is >>>>> the case even after eliminating voxels with a water volume fraction (WVF) >>>>> exceeding 70%. In light of this, I was wondering if you may be able to >>>>> address the following questions: >>>>> >>>>> - Is the amount of ventricular noise post-FWE in my scalar >>>>> maps within a normal range? Will this preclude me from extracting valid >>>>> FWC DTI metrics from the fornix and the genu? Here are some screenshots >>>>> from a representative subject's scalar maps: >>>>> >>>>> *FA map with WVF elimination at a threshold of 70%* >>>>> [image: fa_70.png] >>>>> *MD map with WVF elimination at a threshold of 70%* >>>>> [image: md_70.png] >>>>> *RD map with WVF elimination at a threshold of 70%* >>>>> [image: rd_70.png] >>>>> *AD map with WVF elimination at a threshold of 70%* >>>>> [image: ad_70.png] >>>>> >>>>> >>>>> - If this noise is not within an acceptable range, how might >>>>> I be able optimize our DIPY script so that I can perform a better FWE? I >>>>> tried comparing the results from using a stricter WVF threshold of 60% as >>>>> well as using no WVF thresholding to the above results. Using a stricter >>>>> threshold did not completely eliminate the noise problem, but it did help a >>>>> little bit. However, I'm not sure if there is a precedent for this level >>>>> of thresholding in the literature, or if it is actually appropriate. >>>>> Screenshots from a representative subject are listed below: >>>>> >>>>> *FA map with WVF elimination at a threshold of 60%* >>>>> [image: fa_60.png] >>>>> >>>>> *MD map with WVF elimination at a threshold of 60%* >>>>> [image: md_60.png] >>>>> *FA map with No WVF elimination threshold* >>>>> [image: fa_none.png] >>>>> *MD map with No WVF elimination threshold* >>>>> [image: md_none.png] >>>>> >>>>> I have attached a zip file with the following information for >>>>> your reference: >>>>> >>>>> 1. Input data from a representative subject. This includes >>>>> DWI volumes collected at b values between 0 to 2000. This is contained in >>>>> the *subject_data *subfolder. >>>>> 2. Scalar maps collected with a WVF thresholding rate of 70% >>>>> (*F>.7*), 60% (*F>.6*), and with no thresholding ( >>>>> *no_F_threshold*). >>>>> 3. Three versions of the DIPY script I've been using - each >>>>> one accounts for a different rate of WVF thresholding. >>>>> These scripts are contained in the *dipy_fwe_script_versions* >>>>> subfolder. >>>>> >>>>> I sincerely appreciate all of your time and consideration, and >>>>> look forward to hearing from you soon! >>>>> >>>>> Kind regards, >>>>> Linda >>>>> >>>>> dipyfwe.zip >>>>> <https://drive.google.com/a/temple.edu/file/d/1yvLYeB-cUNqBoce-43ler0-zf_vG8b...> >>>>> -- >>>>> *Lab Manager* >>>>> *Cognitive Neuroscience Lab* >>>>> Temple University >>>>> 1701 N. 13th St. >>>>> Philadelphia, PA 19122 >>>>> >>>>> *Pronouns: * She/Her >>>>> *Phone*: (215) 204-1708 >>>>> *Email*: tuf72977@temple.edu >>>>> >>>> >>> >>> -- >>> *Lab Manager* >>> *Cognitive Neuroscience Lab* >>> Temple University >>> 1701 N. 13th St. >>> Philadelphia, PA 19122 >>> >>> *Pronouns: * She/Her >>> *Phone*: (215) 204-1708 >>> *Email*: tuf72977@temple.edu >>> >> >> >> -- >> *Lab Manager* >> *Cognitive Neuroscience Lab* >> Temple University >> 1701 N. 13th St. >> Philadelphia, PA 19122 >> >> *Pronouns: * She/Her >> *Phone*: (215) 204-1708 >> *Email*: tuf72977@temple.edu >> >
-- *Lab Manager* *Cognitive Neuroscience Lab* Temple University 1701 N. 13th St. Philadelphia, PA 19122
*Pronouns: * She/Her *Phone*: (215) 204-1708 *Email*: tuf72977@temple.edu
-- *Lab Manager* *Cognitive Neuroscience Lab* Temple University 1701 N. 13th St. Philadelphia, PA 19122
*Pronouns: * She/Her *Phone*: (215) 204-1708 *Email*: tuf72977@temple.edu
-- *Lab Manager* *Cognitive Neuroscience Lab* Temple University 1701 N. 13th St. Philadelphia, PA 19122
*Pronouns: * She/Her *Phone*: (215) 204-1708 *Email*: tuf72977@temple.edu
-- *Lab Manager* *Cognitive Neuroscience Lab* Temple University 1701 N. 13th St. Philadelphia, PA 19122
*Pronouns: * She/Her *Phone*: (215) 204-1708 *Email*: tuf72977@temple.edu
-- *Lab Manager* *Cognitive Neuroscience Lab* Temple University 1701 N. 13th St. Philadelphia, PA 19122
*Pronouns: * She/Her *Phone*: (215) 204-1708 *Email*: tuf72977@temple.edu
Good afternoon Ariel, Thank you so much for your response; this is extremely helpful! I just sent you preprocessed data for subject 5216 (from the example above) through Temple University's secure file transfer system (i.e. TUSafesend). Please let me know if you encounter any issues picking up the data, and if you have any questions about how the data are organized. For your reference, you can find the eddy-corrected DWI volumes in the *eddy_output* folder, and the relevant file is titled "*eddy_corrrected_data.nii.gz*". However, the data volumes that I fed into the DIPY FWC script are located in the *dwiextract_2000* folder, and the relevant file is titled " *data_250_1000_2000.nii.gz*" The FWC scalar maps themselves are located in the *fwe* folder. Finally, if your hypothesis is indeed correct, should I mask-out the ventricles by creating a native space ventricular map for each subject and then subtract the mask out of each of my scalars? Might this have negative consequences for the goodness-of-fit of DIPY's free-water corrected tensor model? Thank you again for all of your continued support; it is deeply appreciated! Kind regards, Linda On Tue, Aug 25, 2020 at 1:33 PM Ariel Rokem <arokem@gmail.com> wrote:
Hi Linda,
On Mon, Aug 17, 2020 at 10:51 AM Linda Jasmine Hoffman < tuf72977@temple.edu> wrote:
Hi Ariel,
No worries at all! I completely understand you've been very busy, especially with Neurohackademy going on. I appreciate your response!
Our lab has tried to implement a QSIprep pipeline in the past, and it presented a number of issues for us that we were unable to resolve, particularly given the nascent nature of the software. In light of this, I would prefer to incorporate the denoising/degibbing procedures into our existing preprocessing protocol if at all possible. I understand that denoising/degibbing must occur before any motion correction or other preprocessing is performed, and I don't believe that performing topup first violates this rule. To clarify, our topup script is executed as follows:
*#FSL topup script*
for n in 5022 5216 5302 5391 do
topup
--imain=/data/projects/tbi/dti/${n}/*a2p_p2a_b0.nii.gz *
--datain=/data/projects/tbi/dti/acqp.txt
--config=b02b0_1.cnf
--out=/data/projects/tbi/dti/${n}/topup_output/topup_output
--iout=/data/projects/tbi/dti/${n}/topup_output/my_hifi_b0
--fout=/data/projects/tbi/dti/${n}/topup_output/displacement
done
Note that the only data that goes into the *topup* command are our concatenated anterior-to-posterior and posterior-to-anterior b0 fieldmaps (i.e. *a2p_p2a_b0.nii.gz*). I thought it would be best to do *topup* first since it...
1. ...does not affect our DWI volumes directly - it merely gives us further information concerning motion and the susceptibility-induced field to feed into *eddy* - our most critical preprocessing step. 2. ...yields a high-fidelity brain mask that I was unable to obtain through other means (mainly through unsuccessfully running *bet* on my 4D DWI volumes, and through acquiring a suboptimal brain mask using *dwi2mask* in MRtrix3).
That makes sense.
I wanted to be sure to include a brain mask in my denoising pipeline since I didn't want the inclusion of skull matter to affect how MRtrix3 estimated the noise structure of my data. Did I go wrong by failing to denoise/degibb my fieldmaps in addition to my DWI volumes?
I don't think the fieldmaps need to be denoised.
As for problematic noise voxels in my MD image, I have taken the following screenshots for your reference from a representative subject (i.e. 5216):
*Noise voxel #1 signal: ~0.05* [image: Screen Shot 2020-08-17 at 1.34.46 PM.png]
*Noise voxel #2 singal: ~0.01* [image: Screen Shot 2020-08-17 at 1.35.35 PM.png]
I have attached the free-water corrected scalars for this subject, as well as their MRtrix3 extracted bvals/bvecs to this email for your reference. I sincerely appreciate all of your continued time and consideration on this matter!
Any chance you could share the preprocessed data for the whole volume for
this individual? It's hard to say without looking at the values in these voxels. My hypothesis is that these are simply voxels where the signal is very low. In which case, I don't think there is any harm in masking them out.
I look forward to hearing from you soon!
Kind regards, Linda
On Mon, Aug 17, 2020 at 12:43 PM Ariel Rokem <arokem@gmail.com> wrote:
Hi Linda,
Sorry for the slowness here... It's been... challenging.
Two thoughts:
1. Sorry if I wasn't clear about this before: It is usually recommended that denoising and Gibbs ringing removal be done *before *other steps in preprocessing. To be on the safe side, I would recommend using https://qsiprep.readthedocs.io/en/latest/ for preprocessing. It implements the state of the art, and can be run as a docker/singularity container, which simplifies installation issues.
2. I am wondering what the signal is like in these voxels that still appear with very high MD values. Is there something unusual about their B0 signal? Or are the other data so low as to be indistinguishable from the noise floor? If you could find the coordinate of one of these voxels, and then us that to share with us the signal values in this voxel (as well as b-values and b-vectors) it would help diagnose this.
Cheers,
Ariel
On Mon, Aug 17, 2020 at 9:14 AM Linda Jasmine Hoffman < tuf72977@temple.edu> wrote:
Good morning DIPY experts,
I hope you have all been doing well! I just wanted to follow up with you again as per my latest update re: persisting ventricular noise post-denoising & FWC. Please let me know if you can shed any light on why this noise may still be an issue, even after implementing Ariel's denoising/degibbing suggestion.
I look forward to hearing from you soon!
Kind regards, Linda
On Mon, Aug 3, 2020 at 7:33 PM Linda Jasmine Hoffman < tuf72977@temple.edu> wrote:
Good evening DIPY experts,
I just wanted to follow up with you all as per my last email to see if you've had the opportunity to give my questions some consideration.
Please let me know! I look forward to hearing from you soon!
Kind regards, Linda
On Tue, Jul 28, 2020 at 10:36 PM Linda Jasmine Hoffman < tuf72977@temple.edu> wrote:
Good evening DIPY experts,
I have developed a denoising protocol for my HYDI data, and it has afforded me some success in eliminating a portion of the excess ventricular noise that I have been finding in my free-water-corrected (FWC) scalars. Below is an example from a representative subject (i.e. "Subject 1") for whom this course of actions seems to have worked quite well:
*Subject 1: Original MD map (no denoising of DWI data):* [image: 5022_md.png]
*Subject 1: New MD map (with denoising of DWI data):* [image: 5022_md_denoised.png]
However, I have a few concerns. First, my data is still not as clean as I would like it to be, given the persisting residual noise that is still present in the sagittal view. Second, the denoising protocol that I have implemented did not work consistently well for all subjects. Here is an example from a second representative subject (i.e. "Subject 2") to illustrate this issue:
*Subject 2: Original MD map (no denoising of DWI data):* [image: 5216_md.png]
*Subject 2: New MD map (with denoising of DWI data):* [image: 5216_md_denoised.png]
What is particularly concerning about this is that the resultant image for Subject 2 is still not as clean as what is presented on your DIPY free-water elimination page <https://dipy.org/documentation/1.0.0./examples_built/reconst_fwdti/>. It is worth noting that quality assurance measures have been taken for all of our data, and this subject did not exhibit inordinate imaging artifacts.
For your reference, my denoising pipeline utilized the *dwidenoise* and *mrdegibbs* functions in MRtrix3. I incorporated these steps into my processing protocol in the following order:
1. FSL - topup 2. MRtrix3 - dwidenoise 3. MRtrix3 - mrdegibbs 4. FSL - eddy
Note that I completed *topup* first since this step does not affect the raw, DICOM-to-NIfTI-converted DWI volumes in any way, and it is necessary for yielding a hifi brain mask. The scripts that I used for denoising/degibbing are delineated below:
*#dMRI noise level estimation and denoising using Marchenko-Pastur PCA:* for n in 5022 5216 5302 5391 do
dwidenoise
-mask /data/projects/tbi/denoise/${n}/topup_output/my_hifi_b0_Tcollapsed_brain_mask.nii.gz -noise /data/projects/tbi/denoise/${n}/dwidenoise/noise_hifi_map.nii /data/projects/tbi/denoise/${n}/6-cmrr_mb3hydi_ipat2_64ch/output.nii /data/projects/tbi/denoise/${n}/dwidenoise/denoised_hifi_vol.nii
done
*#Remove Gibbs Ringing Artifacts:* for n in 5022 5216 5302 5391 do
mrdegibbs
/data/projects/tbi/denoise/${n}/dwidenoise/denoised_hifi_vol.nii /data/projects/tbi/denoise/${n}/mrdegibbs/denoised_degibbs_hifi_vol.nii
done
What are your thoughts on the scripts I have implemented? Might I have done something incorrectly, or is there something further I should do to optimize this denoising pipeline? Is there anything I can do in addition to denoising to eliminate these undue levels of post-FWC ventricular noise in my scalars?
Finally, do you recommend denoising and degibbing DWI data as a canonical part of my pipeline? I ask because I know there is a tradeoff between SNR and spatial resolution following noise reduction procedures, so I'm curious to know what best-practices are in this regard. At the very least it seems like an important step if one intends to pursue FWE.
I sincerely appreciate all of your time and consideration on this matter.
Kind regards, Linda
On Thu, Jul 9, 2020 at 5:40 PM Linda Jasmine Hoffman < tuf72977@temple.edu> wrote:
> I haven't; I'll try that now. > > Thank you! > Linda > > On Thu, Jul 9, 2020 at 5:38 PM Ariel Rokem <arokem@gmail.com> wrote: > >> Hi Linda, >> >> Have you had a chance to try Gibbs ringing removal or and/or >> denoising on at least one subject? >> >> Cheers, >> >> Ariel >> >> >> >> On Thu, Jul 9, 2020 at 2:34 PM Linda Jasmine Hoffman < >> tuf72977@temple.edu> wrote: >> >>> Hi everyone, >>> >>> I just wanted to touch base with you to see if you've had the >>> opportunity to give my previous email some consideration. Please let me >>> know what my next steps should be re: denoising my DWI data to >>> eliminate excessive ventricular artifacts post-fwc. >>> >>> Thank you! >>> Linda >>> >>> On Thu, Jul 2, 2020 at 12:41 PM Linda Jasmine Hoffman < >>> tuf72977@temple.edu> wrote: >>> >>>> Hi Ariel, >>>> >>>> Our preprocessing pipeline includes the following steps for noise >>>> reduction in FSL: >>>> >>>> - topup - correct for the susceptibility induced field and >>>> movement >>>> - eddy - correct for eddy current distortions and movement >>>> >>>> We don't have a step in our pipeline to correct for Gibbs >>>> artifacts. Do you think this particular type of artifact is what's >>>> underpinning this issue with the FWC scalar maps? If so, I found a command >>>> in MRtrix3 (mrdegibbs) that eliminates ringing, but it will necessitate >>>> that I go back and redo a large amount of preprocessing. Do you know of an >>>> alternative route to mitigate this problem that may obviate my need to >>>> reprocess my data? >>>> >>>> Thank you so much for your help! >>>> Kind regards, >>>> Linda >>>> >>>> On Thu, Jul 2, 2020 at 12:45 AM Ariel Rokem <arokem@gmail.com> >>>> wrote: >>>> >>>>> Hi Linda, >>>>> >>>>> With your permission, I am adding the DIPY mailing list, so >>>>> others can weigh in and/or benefit from the discussion. >>>>> >>>>> My hunch is that the noise you are seeing in the ventricles is >>>>> due to artifacts/noise. Do you do any removal of Gibbs ringing artifacts or >>>>> any denoising of the data before analyzing it with fwdti? >>>>> >>>>> Cheers, >>>>> >>>>> Ariel >>>>> >>>>> >>>>> On Wed, Jul 1, 2020 at 12:22 PM Linda Jasmine Hoffman < >>>>> tuf72977@temple.edu> wrote: >>>>> >>>>>> Good afternoon DIPY experts, >>>>>> >>>>>> My name is Linda Hoffman, and I'm the lab manager for Dr. >>>>>> Ingrid Olson's Cognitive Neuroscience Lab at Temple University. I have >>>>>> been working on implementing a DIPY-based free-water elimination (FWE) >>>>>> pipeline that my labmate, Katie Jobson, adapted from your >>>>>> website >>>>>> <https://dipy.org/documentation/1.0.0./examples_built/reconst_fwdti/> >>>>>> in order to extract free-water corrected (FWC) scalar maps from a HYDI >>>>>> dataset that I'm analyzing. For your reference, I am ultimately >>>>>> planning to calculate FWC DTI metrics for the fornix and genu of the corpus >>>>>> callosum after performing probabilistic tractography. I have preprocessed >>>>>> my data using FSL version 6.0 and MRtrix3 on a linux machine. >>>>>> >>>>>> While I have successfully extracted FWC FA, MD, RD, and AD maps >>>>>> from my data using this pipeline, there still seems to be a >>>>>> disproportionate amount of noise in the ventricles, especially when >>>>>> comparing my output to your examples on the website linked above. This is >>>>>> the case even after eliminating voxels with a water volume fraction (WVF) >>>>>> exceeding 70%. In light of this, I was wondering if you may be able to >>>>>> address the following questions: >>>>>> >>>>>> - Is the amount of ventricular noise post-FWE in my scalar >>>>>> maps within a normal range? Will this preclude me from extracting valid >>>>>> FWC DTI metrics from the fornix and the genu? Here are some screenshots >>>>>> from a representative subject's scalar maps: >>>>>> >>>>>> *FA map with WVF elimination at a threshold of 70%* >>>>>> [image: fa_70.png] >>>>>> *MD map with WVF elimination at a threshold of 70%* >>>>>> [image: md_70.png] >>>>>> *RD map with WVF elimination at a threshold of 70%* >>>>>> [image: rd_70.png] >>>>>> *AD map with WVF elimination at a threshold of 70%* >>>>>> [image: ad_70.png] >>>>>> >>>>>> >>>>>> - If this noise is not within an acceptable range, how >>>>>> might I be able optimize our DIPY script so that I can perform a better >>>>>> FWE? I tried comparing the results from using a stricter WVF threshold of >>>>>> 60% as well as using no WVF thresholding to the above results. Using a >>>>>> stricter threshold did not completely eliminate the noise problem, but it >>>>>> did help a little bit. However, I'm not sure if there is a precedent for >>>>>> this level of thresholding in the literature, or if it is actually >>>>>> appropriate. Screenshots from a representative subject are listed below: >>>>>> >>>>>> *FA map with WVF elimination at a threshold of 60%* >>>>>> [image: fa_60.png] >>>>>> >>>>>> *MD map with WVF elimination at a threshold of 60%* >>>>>> [image: md_60.png] >>>>>> *FA map with No WVF elimination threshold* >>>>>> [image: fa_none.png] >>>>>> *MD map with No WVF elimination threshold* >>>>>> [image: md_none.png] >>>>>> >>>>>> I have attached a zip file with the following information for >>>>>> your reference: >>>>>> >>>>>> 1. Input data from a representative subject. This includes >>>>>> DWI volumes collected at b values between 0 to 2000. This is contained in >>>>>> the *subject_data *subfolder. >>>>>> 2. Scalar maps collected with a WVF thresholding rate of >>>>>> 70% (*F>.7*), 60% (*F>.6*), and with no thresholding ( >>>>>> *no_F_threshold*). >>>>>> 3. Three versions of the DIPY script I've been using - each >>>>>> one accounts for a different rate of WVF thresholding. >>>>>> These scripts are contained in the >>>>>> *dipy_fwe_script_versions* subfolder. >>>>>> >>>>>> I sincerely appreciate all of your time and consideration, and >>>>>> look forward to hearing from you soon! >>>>>> >>>>>> Kind regards, >>>>>> Linda >>>>>> >>>>>> dipyfwe.zip >>>>>> <https://drive.google.com/a/temple.edu/file/d/1yvLYeB-cUNqBoce-43ler0-zf_vG8b...> >>>>>> -- >>>>>> *Lab Manager* >>>>>> *Cognitive Neuroscience Lab* >>>>>> Temple University >>>>>> 1701 N. 13th St. >>>>>> Philadelphia, PA 19122 >>>>>> >>>>>> *Pronouns: * She/Her >>>>>> *Phone*: (215) 204-1708 >>>>>> *Email*: tuf72977@temple.edu >>>>>> >>>>> >>>> >>>> -- >>>> *Lab Manager* >>>> *Cognitive Neuroscience Lab* >>>> Temple University >>>> 1701 N. 13th St. >>>> Philadelphia, PA 19122 >>>> >>>> *Pronouns: * She/Her >>>> *Phone*: (215) 204-1708 >>>> *Email*: tuf72977@temple.edu >>>> >>> >>> >>> -- >>> *Lab Manager* >>> *Cognitive Neuroscience Lab* >>> Temple University >>> 1701 N. 13th St. >>> Philadelphia, PA 19122 >>> >>> *Pronouns: * She/Her >>> *Phone*: (215) 204-1708 >>> *Email*: tuf72977@temple.edu >>> >> > > -- > *Lab Manager* > *Cognitive Neuroscience Lab* > Temple University > 1701 N. 13th St. > Philadelphia, PA 19122 > > *Pronouns: * She/Her > *Phone*: (215) 204-1708 > *Email*: tuf72977@temple.edu >
-- *Lab Manager* *Cognitive Neuroscience Lab* Temple University 1701 N. 13th St. Philadelphia, PA 19122
*Pronouns: * She/Her *Phone*: (215) 204-1708 *Email*: tuf72977@temple.edu
-- *Lab Manager* *Cognitive Neuroscience Lab* Temple University 1701 N. 13th St. Philadelphia, PA 19122
*Pronouns: * She/Her *Phone*: (215) 204-1708 *Email*: tuf72977@temple.edu
-- *Lab Manager* *Cognitive Neuroscience Lab* Temple University 1701 N. 13th St. Philadelphia, PA 19122
*Pronouns: * She/Her *Phone*: (215) 204-1708 *Email*: tuf72977@temple.edu
-- *Lab Manager* *Cognitive Neuroscience Lab* Temple University 1701 N. 13th St. Philadelphia, PA 19122
*Pronouns: * She/Her *Phone*: (215) 204-1708 *Email*: tuf72977@temple.edu
-- *Lab Manager* *Cognitive Neuroscience Lab* Temple University 1701 N. 13th St. Philadelphia, PA 19122 *Pronouns: * She/Her *Phone*: (215) 204-1708 *Email*: tuf72977@temple.edu
Hi Ariel, I just wanted to follow up with you as per my last email. Have you had the opportunity to look further into my ventricular noise problem? Once again, thank you so much for your continued support on this matter. Best, Linda On Tue, Aug 25, 2020 at 2:25 PM Linda Jasmine Hoffman <tuf72977@temple.edu> wrote:
Good afternoon Ariel,
Thank you so much for your response; this is extremely helpful!
I just sent you preprocessed data for subject 5216 (from the example above) through Temple University's secure file transfer system (i.e. TUSafesend). Please let me know if you encounter any issues picking up the data, and if you have any questions about how the data are organized. For your reference, you can find the eddy-corrected DWI volumes in the *eddy_output* folder, and the relevant file is titled " *eddy_corrrected_data.nii.gz*". However, the data volumes that I fed into the DIPY FWC script are located in the *dwiextract_2000* folder, and the relevant file is titled "*data_250_1000_2000.nii.gz*"
The FWC scalar maps themselves are located in the *fwe* folder.
Finally, if your hypothesis is indeed correct, should I mask-out the ventricles by creating a native space ventricular map for each subject and then subtract the mask out of each of my scalars? Might this have negative consequences for the goodness-of-fit of DIPY's free-water corrected tensor model?
Thank you again for all of your continued support; it is deeply appreciated!
Kind regards, Linda
On Tue, Aug 25, 2020 at 1:33 PM Ariel Rokem <arokem@gmail.com> wrote:
Hi Linda,
On Mon, Aug 17, 2020 at 10:51 AM Linda Jasmine Hoffman < tuf72977@temple.edu> wrote:
Hi Ariel,
No worries at all! I completely understand you've been very busy, especially with Neurohackademy going on. I appreciate your response!
Our lab has tried to implement a QSIprep pipeline in the past, and it presented a number of issues for us that we were unable to resolve, particularly given the nascent nature of the software. In light of this, I would prefer to incorporate the denoising/degibbing procedures into our existing preprocessing protocol if at all possible. I understand that denoising/degibbing must occur before any motion correction or other preprocessing is performed, and I don't believe that performing topup first violates this rule. To clarify, our topup script is executed as follows:
*#FSL topup script*
for n in 5022 5216 5302 5391 do
topup
--imain=/data/projects/tbi/dti/${n}/*a2p_p2a_b0.nii.gz *
--datain=/data/projects/tbi/dti/acqp.txt
--config=b02b0_1.cnf
--out=/data/projects/tbi/dti/${n}/topup_output/topup_output
--iout=/data/projects/tbi/dti/${n}/topup_output/my_hifi_b0
--fout=/data/projects/tbi/dti/${n}/topup_output/displacement
done
Note that the only data that goes into the *topup* command are our concatenated anterior-to-posterior and posterior-to-anterior b0 fieldmaps (i.e. *a2p_p2a_b0.nii.gz*). I thought it would be best to do *topup* first since it...
1. ...does not affect our DWI volumes directly - it merely gives us further information concerning motion and the susceptibility-induced field to feed into *eddy* - our most critical preprocessing step. 2. ...yields a high-fidelity brain mask that I was unable to obtain through other means (mainly through unsuccessfully running *bet* on my 4D DWI volumes, and through acquiring a suboptimal brain mask using *dwi2mask* in MRtrix3).
That makes sense.
I wanted to be sure to include a brain mask in my denoising pipeline since I didn't want the inclusion of skull matter to affect how MRtrix3 estimated the noise structure of my data. Did I go wrong by failing to denoise/degibb my fieldmaps in addition to my DWI volumes?
I don't think the fieldmaps need to be denoised.
As for problematic noise voxels in my MD image, I have taken the following screenshots for your reference from a representative subject (i.e. 5216):
*Noise voxel #1 signal: ~0.05* [image: Screen Shot 2020-08-17 at 1.34.46 PM.png]
*Noise voxel #2 singal: ~0.01* [image: Screen Shot 2020-08-17 at 1.35.35 PM.png]
I have attached the free-water corrected scalars for this subject, as well as their MRtrix3 extracted bvals/bvecs to this email for your reference. I sincerely appreciate all of your continued time and consideration on this matter!
Any chance you could share the preprocessed data for the whole volume
for this individual? It's hard to say without looking at the values in these voxels. My hypothesis is that these are simply voxels where the signal is very low. In which case, I don't think there is any harm in masking them out.
I look forward to hearing from you soon!
Kind regards, Linda
On Mon, Aug 17, 2020 at 12:43 PM Ariel Rokem <arokem@gmail.com> wrote:
Hi Linda,
Sorry for the slowness here... It's been... challenging.
Two thoughts:
1. Sorry if I wasn't clear about this before: It is usually recommended that denoising and Gibbs ringing removal be done *before *other steps in preprocessing. To be on the safe side, I would recommend using https://qsiprep.readthedocs.io/en/latest/ for preprocessing. It implements the state of the art, and can be run as a docker/singularity container, which simplifies installation issues.
2. I am wondering what the signal is like in these voxels that still appear with very high MD values. Is there something unusual about their B0 signal? Or are the other data so low as to be indistinguishable from the noise floor? If you could find the coordinate of one of these voxels, and then us that to share with us the signal values in this voxel (as well as b-values and b-vectors) it would help diagnose this.
Cheers,
Ariel
On Mon, Aug 17, 2020 at 9:14 AM Linda Jasmine Hoffman < tuf72977@temple.edu> wrote:
Good morning DIPY experts,
I hope you have all been doing well! I just wanted to follow up with you again as per my latest update re: persisting ventricular noise post-denoising & FWC. Please let me know if you can shed any light on why this noise may still be an issue, even after implementing Ariel's denoising/degibbing suggestion.
I look forward to hearing from you soon!
Kind regards, Linda
On Mon, Aug 3, 2020 at 7:33 PM Linda Jasmine Hoffman < tuf72977@temple.edu> wrote:
Good evening DIPY experts,
I just wanted to follow up with you all as per my last email to see if you've had the opportunity to give my questions some consideration.
Please let me know! I look forward to hearing from you soon!
Kind regards, Linda
On Tue, Jul 28, 2020 at 10:36 PM Linda Jasmine Hoffman < tuf72977@temple.edu> wrote:
> Good evening DIPY experts, > > I have developed a denoising protocol for my HYDI data, and it has > afforded me some success in eliminating a portion of the excess ventricular > noise that I have been finding in my free-water-corrected (FWC) scalars. > Below is an example from a representative subject (i.e. "Subject 1") > for whom this course of actions seems to have worked quite well: > > *Subject 1: Original MD map (no denoising of DWI data):* > [image: 5022_md.png] > > *Subject 1: New MD map (with denoising of DWI data):* > [image: 5022_md_denoised.png] > > However, I have a few concerns. First, my data is still not as > clean as I would like it to be, given the persisting residual noise that is > still present in the sagittal view. Second, the denoising protocol that I > have implemented did not work consistently well for all subjects. Here is > an example from a second representative subject (i.e. "Subject 2") > to illustrate this issue: > > *Subject 2: Original MD map (no denoising of DWI data):* > [image: 5216_md.png] > > *Subject 2: New MD map (with denoising of DWI data):* > [image: 5216_md_denoised.png] > > What is particularly concerning about this is that the resultant > image for Subject 2 is still not as clean as what is presented on > your DIPY free-water elimination page > <https://dipy.org/documentation/1.0.0./examples_built/reconst_fwdti/>. > It is worth noting that quality assurance measures have been taken for all > of our data, and this subject did not exhibit inordinate imaging artifacts. > > For your reference, my denoising pipeline utilized the *dwidenoise* > and *mrdegibbs* functions in MRtrix3. I incorporated these steps > into my processing protocol in the following order: > > 1. FSL - topup > 2. MRtrix3 - dwidenoise > 3. MRtrix3 - mrdegibbs > 4. FSL - eddy > > Note that I completed *topup* first since this step does not affect > the raw, DICOM-to-NIfTI-converted DWI volumes in any way, and it is > necessary for yielding a hifi brain mask. The scripts that I used for > denoising/degibbing are delineated below: > > > > *#dMRI noise level estimation and denoising using Marchenko-Pastur > PCA:* > for n in 5022 5216 5302 5391 > do > > dwidenoise > > -mask > /data/projects/tbi/denoise/${n}/topup_output/my_hifi_b0_Tcollapsed_brain_mask.nii.gz > -noise /data/projects/tbi/denoise/${n}/dwidenoise/noise_hifi_map.nii > /data/projects/tbi/denoise/${n}/6-cmrr_mb3hydi_ipat2_64ch/output.nii > /data/projects/tbi/denoise/${n}/dwidenoise/denoised_hifi_vol.nii > > done > > > *#Remove Gibbs Ringing Artifacts:* > for n in 5022 5216 5302 5391 > do > > mrdegibbs > > /data/projects/tbi/denoise/${n}/dwidenoise/denoised_hifi_vol.nii > /data/projects/tbi/denoise/${n}/mrdegibbs/denoised_degibbs_hifi_vol.nii > > done > > What are your thoughts on the scripts I have implemented? Might I > have done something incorrectly, or is there something further I should do > to optimize this denoising pipeline? Is there anything I can do in > addition to denoising to eliminate these undue levels of post-FWC > ventricular noise in my scalars? > > Finally, do you recommend denoising and degibbing DWI data as a > canonical part of my pipeline? I ask because I know there is a tradeoff > between SNR and spatial resolution following noise reduction procedures, so > I'm curious to know what best-practices are in this regard. At the very > least it seems like an important step if one intends to pursue FWE. > > I sincerely appreciate all of your time and consideration on this > matter. > > Kind regards, > Linda > > On Thu, Jul 9, 2020 at 5:40 PM Linda Jasmine Hoffman < > tuf72977@temple.edu> wrote: > >> I haven't; I'll try that now. >> >> Thank you! >> Linda >> >> On Thu, Jul 9, 2020 at 5:38 PM Ariel Rokem <arokem@gmail.com> >> wrote: >> >>> Hi Linda, >>> >>> Have you had a chance to try Gibbs ringing removal or and/or >>> denoising on at least one subject? >>> >>> Cheers, >>> >>> Ariel >>> >>> >>> >>> On Thu, Jul 9, 2020 at 2:34 PM Linda Jasmine Hoffman < >>> tuf72977@temple.edu> wrote: >>> >>>> Hi everyone, >>>> >>>> I just wanted to touch base with you to see if you've had the >>>> opportunity to give my previous email some consideration. Please let me >>>> know what my next steps should be re: denoising my DWI data to >>>> eliminate excessive ventricular artifacts post-fwc. >>>> >>>> Thank you! >>>> Linda >>>> >>>> On Thu, Jul 2, 2020 at 12:41 PM Linda Jasmine Hoffman < >>>> tuf72977@temple.edu> wrote: >>>> >>>>> Hi Ariel, >>>>> >>>>> Our preprocessing pipeline includes the following steps for >>>>> noise reduction in FSL: >>>>> >>>>> - topup - correct for the susceptibility induced field and >>>>> movement >>>>> - eddy - correct for eddy current distortions and movement >>>>> >>>>> We don't have a step in our pipeline to correct for Gibbs >>>>> artifacts. Do you think this particular type of artifact is what's >>>>> underpinning this issue with the FWC scalar maps? If so, I found a command >>>>> in MRtrix3 (mrdegibbs) that eliminates ringing, but it will necessitate >>>>> that I go back and redo a large amount of preprocessing. Do you know of an >>>>> alternative route to mitigate this problem that may obviate my need to >>>>> reprocess my data? >>>>> >>>>> Thank you so much for your help! >>>>> Kind regards, >>>>> Linda >>>>> >>>>> On Thu, Jul 2, 2020 at 12:45 AM Ariel Rokem <arokem@gmail.com> >>>>> wrote: >>>>> >>>>>> Hi Linda, >>>>>> >>>>>> With your permission, I am adding the DIPY mailing list, so >>>>>> others can weigh in and/or benefit from the discussion. >>>>>> >>>>>> My hunch is that the noise you are seeing in the ventricles is >>>>>> due to artifacts/noise. Do you do any removal of Gibbs ringing artifacts or >>>>>> any denoising of the data before analyzing it with fwdti? >>>>>> >>>>>> Cheers, >>>>>> >>>>>> Ariel >>>>>> >>>>>> >>>>>> On Wed, Jul 1, 2020 at 12:22 PM Linda Jasmine Hoffman < >>>>>> tuf72977@temple.edu> wrote: >>>>>> >>>>>>> Good afternoon DIPY experts, >>>>>>> >>>>>>> My name is Linda Hoffman, and I'm the lab manager for Dr. >>>>>>> Ingrid Olson's Cognitive Neuroscience Lab at Temple University. I have >>>>>>> been working on implementing a DIPY-based free-water elimination (FWE) >>>>>>> pipeline that my labmate, Katie Jobson, adapted from your >>>>>>> website >>>>>>> <https://dipy.org/documentation/1.0.0./examples_built/reconst_fwdti/> >>>>>>> in order to extract free-water corrected (FWC) scalar maps from a HYDI >>>>>>> dataset that I'm analyzing. For your reference, I am ultimately >>>>>>> planning to calculate FWC DTI metrics for the fornix and genu of the corpus >>>>>>> callosum after performing probabilistic tractography. I have preprocessed >>>>>>> my data using FSL version 6.0 and MRtrix3 on a linux machine. >>>>>>> >>>>>>> While I have successfully extracted FWC FA, MD, RD, and AD >>>>>>> maps from my data using this pipeline, there still seems to be a >>>>>>> disproportionate amount of noise in the ventricles, especially when >>>>>>> comparing my output to your examples on the website linked above. This is >>>>>>> the case even after eliminating voxels with a water volume fraction (WVF) >>>>>>> exceeding 70%. In light of this, I was wondering if you may be able to >>>>>>> address the following questions: >>>>>>> >>>>>>> - Is the amount of ventricular noise post-FWE in my scalar >>>>>>> maps within a normal range? Will this preclude me from extracting valid >>>>>>> FWC DTI metrics from the fornix and the genu? Here are some screenshots >>>>>>> from a representative subject's scalar maps: >>>>>>> >>>>>>> *FA map with WVF elimination at a threshold of 70%* >>>>>>> [image: fa_70.png] >>>>>>> *MD map with WVF elimination at a threshold of 70%* >>>>>>> [image: md_70.png] >>>>>>> *RD map with WVF elimination at a threshold of 70%* >>>>>>> [image: rd_70.png] >>>>>>> *AD map with WVF elimination at a threshold of 70%* >>>>>>> [image: ad_70.png] >>>>>>> >>>>>>> >>>>>>> - If this noise is not within an acceptable range, how >>>>>>> might I be able optimize our DIPY script so that I can perform a better >>>>>>> FWE? I tried comparing the results from using a stricter WVF threshold of >>>>>>> 60% as well as using no WVF thresholding to the above results. Using a >>>>>>> stricter threshold did not completely eliminate the noise problem, but it >>>>>>> did help a little bit. However, I'm not sure if there is a precedent for >>>>>>> this level of thresholding in the literature, or if it is actually >>>>>>> appropriate. Screenshots from a representative subject are listed below: >>>>>>> >>>>>>> *FA map with WVF elimination at a threshold of 60%* >>>>>>> [image: fa_60.png] >>>>>>> >>>>>>> *MD map with WVF elimination at a threshold of 60%* >>>>>>> [image: md_60.png] >>>>>>> *FA map with No WVF elimination threshold* >>>>>>> [image: fa_none.png] >>>>>>> *MD map with No WVF elimination threshold* >>>>>>> [image: md_none.png] >>>>>>> >>>>>>> I have attached a zip file with the following information for >>>>>>> your reference: >>>>>>> >>>>>>> 1. Input data from a representative subject. This >>>>>>> includes DWI volumes collected at b values between 0 to 2000. This is >>>>>>> contained in the *subject_data *subfolder. >>>>>>> 2. Scalar maps collected with a WVF thresholding rate of >>>>>>> 70% (*F>.7*), 60% (*F>.6*), and with no thresholding ( >>>>>>> *no_F_threshold*). >>>>>>> 3. Three versions of the DIPY script I've been using - >>>>>>> each one accounts for a different rate of WVF thresholding. >>>>>>> These scripts are contained in the >>>>>>> *dipy_fwe_script_versions* subfolder. >>>>>>> >>>>>>> I sincerely appreciate all of your time and consideration, and >>>>>>> look forward to hearing from you soon! >>>>>>> >>>>>>> Kind regards, >>>>>>> Linda >>>>>>> >>>>>>> dipyfwe.zip >>>>>>> <https://drive.google.com/a/temple.edu/file/d/1yvLYeB-cUNqBoce-43ler0-zf_vG8b...> >>>>>>> -- >>>>>>> *Lab Manager* >>>>>>> *Cognitive Neuroscience Lab* >>>>>>> Temple University >>>>>>> 1701 N. 13th St. >>>>>>> Philadelphia, PA 19122 >>>>>>> >>>>>>> *Pronouns: * She/Her >>>>>>> *Phone*: (215) 204-1708 >>>>>>> *Email*: tuf72977@temple.edu >>>>>>> >>>>>> >>>>> >>>>> -- >>>>> *Lab Manager* >>>>> *Cognitive Neuroscience Lab* >>>>> Temple University >>>>> 1701 N. 13th St. >>>>> Philadelphia, PA 19122 >>>>> >>>>> *Pronouns: * She/Her >>>>> *Phone*: (215) 204-1708 >>>>> *Email*: tuf72977@temple.edu >>>>> >>>> >>>> >>>> -- >>>> *Lab Manager* >>>> *Cognitive Neuroscience Lab* >>>> Temple University >>>> 1701 N. 13th St. >>>> Philadelphia, PA 19122 >>>> >>>> *Pronouns: * She/Her >>>> *Phone*: (215) 204-1708 >>>> *Email*: tuf72977@temple.edu >>>> >>> >> >> -- >> *Lab Manager* >> *Cognitive Neuroscience Lab* >> Temple University >> 1701 N. 13th St. >> Philadelphia, PA 19122 >> >> *Pronouns: * She/Her >> *Phone*: (215) 204-1708 >> *Email*: tuf72977@temple.edu >> > > > -- > *Lab Manager* > *Cognitive Neuroscience Lab* > Temple University > 1701 N. 13th St. > Philadelphia, PA 19122 > > *Pronouns: * She/Her > *Phone*: (215) 204-1708 > *Email*: tuf72977@temple.edu >
-- *Lab Manager* *Cognitive Neuroscience Lab* Temple University 1701 N. 13th St. Philadelphia, PA 19122
*Pronouns: * She/Her *Phone*: (215) 204-1708 *Email*: tuf72977@temple.edu
-- *Lab Manager* *Cognitive Neuroscience Lab* Temple University 1701 N. 13th St. Philadelphia, PA 19122
*Pronouns: * She/Her *Phone*: (215) 204-1708 *Email*: tuf72977@temple.edu
-- *Lab Manager* *Cognitive Neuroscience Lab* Temple University 1701 N. 13th St. Philadelphia, PA 19122
*Pronouns: * She/Her *Phone*: (215) 204-1708 *Email*: tuf72977@temple.edu
-- *Lab Manager* *Cognitive Neuroscience Lab* Temple University 1701 N. 13th St. Philadelphia, PA 19122
*Pronouns: * She/Her *Phone*: (215) 204-1708 *Email*: tuf72977@temple.edu
-- *Lab Manager* *Cognitive Neuroscience Lab* Temple University 1701 N. 13th St. Philadelphia, PA 19122 *Pronouns: * She/Her *Phone*: (215) 204-1708 *Email*: tuf72977@temple.edu
Hi Ariel, I just wanted to follow up with you to see if you've had the opportunity to give my previous email some consideration. Kind regards, Linda On Thu, Sep 10, 2020 at 1:00 PM Linda Jasmine Hoffman <tuf72977@temple.edu> wrote:
Hi Ariel,
I just wanted to follow up with you as per my last email. Have you had the opportunity to look further into my ventricular noise problem?
Once again, thank you so much for your continued support on this matter.
Best, Linda
On Tue, Aug 25, 2020 at 2:25 PM Linda Jasmine Hoffman <tuf72977@temple.edu> wrote:
Good afternoon Ariel,
Thank you so much for your response; this is extremely helpful!
I just sent you preprocessed data for subject 5216 (from the example above) through Temple University's secure file transfer system (i.e. TUSafesend). Please let me know if you encounter any issues picking up the data, and if you have any questions about how the data are organized. For your reference, you can find the eddy-corrected DWI volumes in the *eddy_output* folder, and the relevant file is titled " *eddy_corrrected_data.nii.gz*". However, the data volumes that I fed into the DIPY FWC script are located in the *dwiextract_2000* folder, and the relevant file is titled "*data_250_1000_2000.nii.gz*"
The FWC scalar maps themselves are located in the *fwe* folder.
Finally, if your hypothesis is indeed correct, should I mask-out the ventricles by creating a native space ventricular map for each subject and then subtract the mask out of each of my scalars? Might this have negative consequences for the goodness-of-fit of DIPY's free-water corrected tensor model?
Thank you again for all of your continued support; it is deeply appreciated!
Kind regards, Linda
On Tue, Aug 25, 2020 at 1:33 PM Ariel Rokem <arokem@gmail.com> wrote:
Hi Linda,
On Mon, Aug 17, 2020 at 10:51 AM Linda Jasmine Hoffman < tuf72977@temple.edu> wrote:
Hi Ariel,
No worries at all! I completely understand you've been very busy, especially with Neurohackademy going on. I appreciate your response!
Our lab has tried to implement a QSIprep pipeline in the past, and it presented a number of issues for us that we were unable to resolve, particularly given the nascent nature of the software. In light of this, I would prefer to incorporate the denoising/degibbing procedures into our existing preprocessing protocol if at all possible. I understand that denoising/degibbing must occur before any motion correction or other preprocessing is performed, and I don't believe that performing topup first violates this rule. To clarify, our topup script is executed as follows:
*#FSL topup script*
for n in 5022 5216 5302 5391 do
topup
--imain=/data/projects/tbi/dti/${n}/*a2p_p2a_b0.nii.gz *
--datain=/data/projects/tbi/dti/acqp.txt
--config=b02b0_1.cnf
--out=/data/projects/tbi/dti/${n}/topup_output/topup_output
--iout=/data/projects/tbi/dti/${n}/topup_output/my_hifi_b0
--fout=/data/projects/tbi/dti/${n}/topup_output/displacement
done
Note that the only data that goes into the *topup* command are our concatenated anterior-to-posterior and posterior-to-anterior b0 fieldmaps (i.e. *a2p_p2a_b0.nii.gz*). I thought it would be best to do *topup* first since it...
1. ...does not affect our DWI volumes directly - it merely gives us further information concerning motion and the susceptibility-induced field to feed into *eddy* - our most critical preprocessing step. 2. ...yields a high-fidelity brain mask that I was unable to obtain through other means (mainly through unsuccessfully running *bet* on my 4D DWI volumes, and through acquiring a suboptimal brain mask using *dwi2mask* in MRtrix3).
That makes sense.
I wanted to be sure to include a brain mask in my denoising pipeline since I didn't want the inclusion of skull matter to affect how MRtrix3 estimated the noise structure of my data. Did I go wrong by failing to denoise/degibb my fieldmaps in addition to my DWI volumes?
I don't think the fieldmaps need to be denoised.
As for problematic noise voxels in my MD image, I have taken the following screenshots for your reference from a representative subject (i.e. 5216):
*Noise voxel #1 signal: ~0.05* [image: Screen Shot 2020-08-17 at 1.34.46 PM.png]
*Noise voxel #2 singal: ~0.01* [image: Screen Shot 2020-08-17 at 1.35.35 PM.png]
I have attached the free-water corrected scalars for this subject, as well as their MRtrix3 extracted bvals/bvecs to this email for your reference. I sincerely appreciate all of your continued time and consideration on this matter!
Any chance you could share the preprocessed data for the whole volume
for this individual? It's hard to say without looking at the values in these voxels. My hypothesis is that these are simply voxels where the signal is very low. In which case, I don't think there is any harm in masking them out.
I look forward to hearing from you soon!
Kind regards, Linda
On Mon, Aug 17, 2020 at 12:43 PM Ariel Rokem <arokem@gmail.com> wrote:
Hi Linda,
Sorry for the slowness here... It's been... challenging.
Two thoughts:
1. Sorry if I wasn't clear about this before: It is usually recommended that denoising and Gibbs ringing removal be done *before *other steps in preprocessing. To be on the safe side, I would recommend using https://qsiprep.readthedocs.io/en/latest/ for preprocessing. It implements the state of the art, and can be run as a docker/singularity container, which simplifies installation issues.
2. I am wondering what the signal is like in these voxels that still appear with very high MD values. Is there something unusual about their B0 signal? Or are the other data so low as to be indistinguishable from the noise floor? If you could find the coordinate of one of these voxels, and then us that to share with us the signal values in this voxel (as well as b-values and b-vectors) it would help diagnose this.
Cheers,
Ariel
On Mon, Aug 17, 2020 at 9:14 AM Linda Jasmine Hoffman < tuf72977@temple.edu> wrote:
Good morning DIPY experts,
I hope you have all been doing well! I just wanted to follow up with you again as per my latest update re: persisting ventricular noise post-denoising & FWC. Please let me know if you can shed any light on why this noise may still be an issue, even after implementing Ariel's denoising/degibbing suggestion.
I look forward to hearing from you soon!
Kind regards, Linda
On Mon, Aug 3, 2020 at 7:33 PM Linda Jasmine Hoffman < tuf72977@temple.edu> wrote:
> Good evening DIPY experts, > > I just wanted to follow up with you all as per my last email to see > if you've had the opportunity to give my questions some consideration. > > Please let me know! I look forward to hearing from you soon! > > Kind regards, > Linda > > On Tue, Jul 28, 2020 at 10:36 PM Linda Jasmine Hoffman < > tuf72977@temple.edu> wrote: > >> Good evening DIPY experts, >> >> I have developed a denoising protocol for my HYDI data, and it has >> afforded me some success in eliminating a portion of the excess ventricular >> noise that I have been finding in my free-water-corrected (FWC) scalars. >> Below is an example from a representative subject (i.e. "Subject 1") >> for whom this course of actions seems to have worked quite well: >> >> *Subject 1: Original MD map (no denoising of DWI data):* >> [image: 5022_md.png] >> >> *Subject 1: New MD map (with denoising of DWI data):* >> [image: 5022_md_denoised.png] >> >> However, I have a few concerns. First, my data is still not as >> clean as I would like it to be, given the persisting residual noise that is >> still present in the sagittal view. Second, the denoising protocol that I >> have implemented did not work consistently well for all subjects. Here is >> an example from a second representative subject (i.e. "Subject 2") >> to illustrate this issue: >> >> *Subject 2: Original MD map (no denoising of DWI data):* >> [image: 5216_md.png] >> >> *Subject 2: New MD map (with denoising of DWI data):* >> [image: 5216_md_denoised.png] >> >> What is particularly concerning about this is that the resultant >> image for Subject 2 is still not as clean as what is presented on >> your DIPY free-water elimination page >> <https://dipy.org/documentation/1.0.0./examples_built/reconst_fwdti/>. >> It is worth noting that quality assurance measures have been taken for all >> of our data, and this subject did not exhibit inordinate imaging artifacts. >> >> For your reference, my denoising pipeline utilized the *dwidenoise* >> and *mrdegibbs* functions in MRtrix3. I incorporated these steps >> into my processing protocol in the following order: >> >> 1. FSL - topup >> 2. MRtrix3 - dwidenoise >> 3. MRtrix3 - mrdegibbs >> 4. FSL - eddy >> >> Note that I completed *topup* first since this step does not >> affect the raw, DICOM-to-NIfTI-converted DWI volumes in any way, and it is >> necessary for yielding a hifi brain mask. The scripts that I used for >> denoising/degibbing are delineated below: >> >> >> >> *#dMRI noise level estimation and denoising using Marchenko-Pastur >> PCA:* >> for n in 5022 5216 5302 5391 >> do >> >> dwidenoise >> >> -mask >> /data/projects/tbi/denoise/${n}/topup_output/my_hifi_b0_Tcollapsed_brain_mask.nii.gz >> -noise >> /data/projects/tbi/denoise/${n}/dwidenoise/noise_hifi_map.nii >> /data/projects/tbi/denoise/${n}/6-cmrr_mb3hydi_ipat2_64ch/output.nii >> /data/projects/tbi/denoise/${n}/dwidenoise/denoised_hifi_vol.nii >> >> done >> >> >> *#Remove Gibbs Ringing Artifacts:* >> for n in 5022 5216 5302 5391 >> do >> >> mrdegibbs >> >> /data/projects/tbi/denoise/${n}/dwidenoise/denoised_hifi_vol.nii >> /data/projects/tbi/denoise/${n}/mrdegibbs/denoised_degibbs_hifi_vol.nii >> >> done >> >> What are your thoughts on the scripts I have implemented? Might I >> have done something incorrectly, or is there something further I should do >> to optimize this denoising pipeline? Is there anything I can do in >> addition to denoising to eliminate these undue levels of post-FWC >> ventricular noise in my scalars? >> >> Finally, do you recommend denoising and degibbing DWI data as a >> canonical part of my pipeline? I ask because I know there is a tradeoff >> between SNR and spatial resolution following noise reduction procedures, so >> I'm curious to know what best-practices are in this regard. At the very >> least it seems like an important step if one intends to pursue FWE. >> >> I sincerely appreciate all of your time and consideration on this >> matter. >> >> Kind regards, >> Linda >> >> On Thu, Jul 9, 2020 at 5:40 PM Linda Jasmine Hoffman < >> tuf72977@temple.edu> wrote: >> >>> I haven't; I'll try that now. >>> >>> Thank you! >>> Linda >>> >>> On Thu, Jul 9, 2020 at 5:38 PM Ariel Rokem <arokem@gmail.com> >>> wrote: >>> >>>> Hi Linda, >>>> >>>> Have you had a chance to try Gibbs ringing removal or and/or >>>> denoising on at least one subject? >>>> >>>> Cheers, >>>> >>>> Ariel >>>> >>>> >>>> >>>> On Thu, Jul 9, 2020 at 2:34 PM Linda Jasmine Hoffman < >>>> tuf72977@temple.edu> wrote: >>>> >>>>> Hi everyone, >>>>> >>>>> I just wanted to touch base with you to see if you've had the >>>>> opportunity to give my previous email some consideration. Please let me >>>>> know what my next steps should be re: denoising my DWI data to >>>>> eliminate excessive ventricular artifacts post-fwc. >>>>> >>>>> Thank you! >>>>> Linda >>>>> >>>>> On Thu, Jul 2, 2020 at 12:41 PM Linda Jasmine Hoffman < >>>>> tuf72977@temple.edu> wrote: >>>>> >>>>>> Hi Ariel, >>>>>> >>>>>> Our preprocessing pipeline includes the following steps for >>>>>> noise reduction in FSL: >>>>>> >>>>>> - topup - correct for the susceptibility induced field and >>>>>> movement >>>>>> - eddy - correct for eddy current distortions and movement >>>>>> >>>>>> We don't have a step in our pipeline to correct for Gibbs >>>>>> artifacts. Do you think this particular type of artifact is what's >>>>>> underpinning this issue with the FWC scalar maps? If so, I found a command >>>>>> in MRtrix3 (mrdegibbs) that eliminates ringing, but it will necessitate >>>>>> that I go back and redo a large amount of preprocessing. Do you know of an >>>>>> alternative route to mitigate this problem that may obviate my need to >>>>>> reprocess my data? >>>>>> >>>>>> Thank you so much for your help! >>>>>> Kind regards, >>>>>> Linda >>>>>> >>>>>> On Thu, Jul 2, 2020 at 12:45 AM Ariel Rokem <arokem@gmail.com> >>>>>> wrote: >>>>>> >>>>>>> Hi Linda, >>>>>>> >>>>>>> With your permission, I am adding the DIPY mailing list, so >>>>>>> others can weigh in and/or benefit from the discussion. >>>>>>> >>>>>>> My hunch is that the noise you are seeing in the ventricles is >>>>>>> due to artifacts/noise. Do you do any removal of Gibbs ringing artifacts or >>>>>>> any denoising of the data before analyzing it with fwdti? >>>>>>> >>>>>>> Cheers, >>>>>>> >>>>>>> Ariel >>>>>>> >>>>>>> >>>>>>> On Wed, Jul 1, 2020 at 12:22 PM Linda Jasmine Hoffman < >>>>>>> tuf72977@temple.edu> wrote: >>>>>>> >>>>>>>> Good afternoon DIPY experts, >>>>>>>> >>>>>>>> My name is Linda Hoffman, and I'm the lab manager for Dr. >>>>>>>> Ingrid Olson's Cognitive Neuroscience Lab at Temple University. I have >>>>>>>> been working on implementing a DIPY-based free-water elimination (FWE) >>>>>>>> pipeline that my labmate, Katie Jobson, adapted from your >>>>>>>> website >>>>>>>> <https://dipy.org/documentation/1.0.0./examples_built/reconst_fwdti/> >>>>>>>> in order to extract free-water corrected (FWC) scalar maps from a HYDI >>>>>>>> dataset that I'm analyzing. For your reference, I am ultimately >>>>>>>> planning to calculate FWC DTI metrics for the fornix and genu of the corpus >>>>>>>> callosum after performing probabilistic tractography. I have preprocessed >>>>>>>> my data using FSL version 6.0 and MRtrix3 on a linux machine. >>>>>>>> >>>>>>>> While I have successfully extracted FWC FA, MD, RD, and AD >>>>>>>> maps from my data using this pipeline, there still seems to be a >>>>>>>> disproportionate amount of noise in the ventricles, especially when >>>>>>>> comparing my output to your examples on the website linked above. This is >>>>>>>> the case even after eliminating voxels with a water volume fraction (WVF) >>>>>>>> exceeding 70%. In light of this, I was wondering if you may be able to >>>>>>>> address the following questions: >>>>>>>> >>>>>>>> - Is the amount of ventricular noise post-FWE in my >>>>>>>> scalar maps within a normal range? Will this preclude me from extracting >>>>>>>> valid FWC DTI metrics from the fornix and the genu? Here are some >>>>>>>> screenshots from a representative subject's scalar maps: >>>>>>>> >>>>>>>> *FA map with WVF elimination at a threshold of 70%* >>>>>>>> [image: fa_70.png] >>>>>>>> *MD map with WVF elimination at a threshold of 70%* >>>>>>>> [image: md_70.png] >>>>>>>> *RD map with WVF elimination at a threshold of 70%* >>>>>>>> [image: rd_70.png] >>>>>>>> *AD map with WVF elimination at a threshold of 70%* >>>>>>>> [image: ad_70.png] >>>>>>>> >>>>>>>> >>>>>>>> - If this noise is not within an acceptable range, how >>>>>>>> might I be able optimize our DIPY script so that I can perform a better >>>>>>>> FWE? I tried comparing the results from using a stricter WVF threshold of >>>>>>>> 60% as well as using no WVF thresholding to the above results. Using a >>>>>>>> stricter threshold did not completely eliminate the noise problem, but it >>>>>>>> did help a little bit. However, I'm not sure if there is a precedent for >>>>>>>> this level of thresholding in the literature, or if it is actually >>>>>>>> appropriate. Screenshots from a representative subject are listed below: >>>>>>>> >>>>>>>> *FA map with WVF elimination at a threshold of 60%* >>>>>>>> [image: fa_60.png] >>>>>>>> >>>>>>>> *MD map with WVF elimination at a threshold of 60%* >>>>>>>> [image: md_60.png] >>>>>>>> *FA map with No WVF elimination threshold* >>>>>>>> [image: fa_none.png] >>>>>>>> *MD map with No WVF elimination threshold* >>>>>>>> [image: md_none.png] >>>>>>>> >>>>>>>> I have attached a zip file with the following information for >>>>>>>> your reference: >>>>>>>> >>>>>>>> 1. Input data from a representative subject. This >>>>>>>> includes DWI volumes collected at b values between 0 to 2000. This is >>>>>>>> contained in the *subject_data *subfolder. >>>>>>>> 2. Scalar maps collected with a WVF thresholding rate of >>>>>>>> 70% (*F>.7*), 60% (*F>.6*), and with no thresholding ( >>>>>>>> *no_F_threshold*). >>>>>>>> 3. Three versions of the DIPY script I've been using - >>>>>>>> each one accounts for a different rate of WVF thresholding. >>>>>>>> These scripts are contained in the >>>>>>>> *dipy_fwe_script_versions* subfolder. >>>>>>>> >>>>>>>> I sincerely appreciate all of your time and consideration, >>>>>>>> and look forward to hearing from you soon! >>>>>>>> >>>>>>>> Kind regards, >>>>>>>> Linda >>>>>>>> >>>>>>>> dipyfwe.zip >>>>>>>> <https://drive.google.com/a/temple.edu/file/d/1yvLYeB-cUNqBoce-43ler0-zf_vG8b...> >>>>>>>> -- >>>>>>>> *Lab Manager* >>>>>>>> *Cognitive Neuroscience Lab* >>>>>>>> Temple University >>>>>>>> 1701 N. 13th St. >>>>>>>> Philadelphia, PA 19122 >>>>>>>> >>>>>>>> *Pronouns: * She/Her >>>>>>>> *Phone*: (215) 204-1708 >>>>>>>> *Email*: tuf72977@temple.edu >>>>>>>> >>>>>>> >>>>>> >>>>>> -- >>>>>> *Lab Manager* >>>>>> *Cognitive Neuroscience Lab* >>>>>> Temple University >>>>>> 1701 N. 13th St. >>>>>> Philadelphia, PA 19122 >>>>>> >>>>>> *Pronouns: * She/Her >>>>>> *Phone*: (215) 204-1708 >>>>>> *Email*: tuf72977@temple.edu >>>>>> >>>>> >>>>> >>>>> -- >>>>> *Lab Manager* >>>>> *Cognitive Neuroscience Lab* >>>>> Temple University >>>>> 1701 N. 13th St. >>>>> Philadelphia, PA 19122 >>>>> >>>>> *Pronouns: * She/Her >>>>> *Phone*: (215) 204-1708 >>>>> *Email*: tuf72977@temple.edu >>>>> >>>> >>> >>> -- >>> *Lab Manager* >>> *Cognitive Neuroscience Lab* >>> Temple University >>> 1701 N. 13th St. >>> Philadelphia, PA 19122 >>> >>> *Pronouns: * She/Her >>> *Phone*: (215) 204-1708 >>> *Email*: tuf72977@temple.edu >>> >> >> >> -- >> *Lab Manager* >> *Cognitive Neuroscience Lab* >> Temple University >> 1701 N. 13th St. >> Philadelphia, PA 19122 >> >> *Pronouns: * She/Her >> *Phone*: (215) 204-1708 >> *Email*: tuf72977@temple.edu >> > > > -- > *Lab Manager* > *Cognitive Neuroscience Lab* > Temple University > 1701 N. 13th St. > Philadelphia, PA 19122 > > *Pronouns: * She/Her > *Phone*: (215) 204-1708 > *Email*: tuf72977@temple.edu >
-- *Lab Manager* *Cognitive Neuroscience Lab* Temple University 1701 N. 13th St. Philadelphia, PA 19122
*Pronouns: * She/Her *Phone*: (215) 204-1708 *Email*: tuf72977@temple.edu
-- *Lab Manager* *Cognitive Neuroscience Lab* Temple University 1701 N. 13th St. Philadelphia, PA 19122
*Pronouns: * She/Her *Phone*: (215) 204-1708 *Email*: tuf72977@temple.edu
-- *Lab Manager* *Cognitive Neuroscience Lab* Temple University 1701 N. 13th St. Philadelphia, PA 19122
*Pronouns: * She/Her *Phone*: (215) 204-1708 *Email*: tuf72977@temple.edu
-- *Lab Manager* *Cognitive Neuroscience Lab* Temple University 1701 N. 13th St. Philadelphia, PA 19122
*Pronouns: * She/Her *Phone*: (215) 204-1708 *Email*: tuf72977@temple.edu
-- *Lab Manager* *Cognitive Neuroscience Lab* Temple University 1701 N. 13th St. Philadelphia, PA 19122 *Pronouns: * She/Her *Phone*: (215) 204-1708 *Email*: tuf72977@temple.edu
Sorry - I am having some pretty busy days here, so no, I have not gotten back to a detailed exploration yet. I'm sorry. My intuition as it is now is that everything is fine and the anomalies you are seeing are due to the signal going into the noise floor, but have not verified yet. You could look at the signal yourself and see whether the b0 signal is lower than any of the DWI signals. I am guessing that is what is going on here. In which case, I probably wouldn't worry about it too much. Ariel On Tue, Sep 15, 2020 at 11:39 AM Linda Jasmine Hoffman <tuf72977@temple.edu> wrote:
Hi Ariel,
I just wanted to follow up with you to see if you've had the opportunity to give my previous email some consideration.
Kind regards, Linda
On Thu, Sep 10, 2020 at 1:00 PM Linda Jasmine Hoffman <tuf72977@temple.edu> wrote:
Hi Ariel,
I just wanted to follow up with you as per my last email. Have you had the opportunity to look further into my ventricular noise problem?
Once again, thank you so much for your continued support on this matter.
Best, Linda
On Tue, Aug 25, 2020 at 2:25 PM Linda Jasmine Hoffman < tuf72977@temple.edu> wrote:
Good afternoon Ariel,
Thank you so much for your response; this is extremely helpful!
I just sent you preprocessed data for subject 5216 (from the example above) through Temple University's secure file transfer system (i.e. TUSafesend). Please let me know if you encounter any issues picking up the data, and if you have any questions about how the data are organized. For your reference, you can find the eddy-corrected DWI volumes in the *eddy_output* folder, and the relevant file is titled " *eddy_corrrected_data.nii.gz*". However, the data volumes that I fed into the DIPY FWC script are located in the *dwiextract_2000* folder, and the relevant file is titled "*data_250_1000_2000.nii.gz*"
The FWC scalar maps themselves are located in the *fwe* folder.
Finally, if your hypothesis is indeed correct, should I mask-out the ventricles by creating a native space ventricular map for each subject and then subtract the mask out of each of my scalars? Might this have negative consequences for the goodness-of-fit of DIPY's free-water corrected tensor model?
Thank you again for all of your continued support; it is deeply appreciated!
Kind regards, Linda
On Tue, Aug 25, 2020 at 1:33 PM Ariel Rokem <arokem@gmail.com> wrote:
Hi Linda,
On Mon, Aug 17, 2020 at 10:51 AM Linda Jasmine Hoffman < tuf72977@temple.edu> wrote:
Hi Ariel,
No worries at all! I completely understand you've been very busy, especially with Neurohackademy going on. I appreciate your response!
Our lab has tried to implement a QSIprep pipeline in the past, and it presented a number of issues for us that we were unable to resolve, particularly given the nascent nature of the software. In light of this, I would prefer to incorporate the denoising/degibbing procedures into our existing preprocessing protocol if at all possible. I understand that denoising/degibbing must occur before any motion correction or other preprocessing is performed, and I don't believe that performing topup first violates this rule. To clarify, our topup script is executed as follows:
*#FSL topup script*
for n in 5022 5216 5302 5391 do
topup
--imain=/data/projects/tbi/dti/${n}/*a2p_p2a_b0.nii.gz *
--datain=/data/projects/tbi/dti/acqp.txt
--config=b02b0_1.cnf
--out=/data/projects/tbi/dti/${n}/topup_output/topup_output
--iout=/data/projects/tbi/dti/${n}/topup_output/my_hifi_b0
--fout=/data/projects/tbi/dti/${n}/topup_output/displacement
done
Note that the only data that goes into the *topup* command are our concatenated anterior-to-posterior and posterior-to-anterior b0 fieldmaps (i.e. *a2p_p2a_b0.nii.gz*). I thought it would be best to do *topup* first since it...
1. ...does not affect our DWI volumes directly - it merely gives us further information concerning motion and the susceptibility-induced field to feed into *eddy* - our most critical preprocessing step. 2. ...yields a high-fidelity brain mask that I was unable to obtain through other means (mainly through unsuccessfully running *bet* on my 4D DWI volumes, and through acquiring a suboptimal brain mask using *dwi2mask* in MRtrix3).
That makes sense.
I wanted to be sure to include a brain mask in my denoising pipeline since I didn't want the inclusion of skull matter to affect how MRtrix3 estimated the noise structure of my data. Did I go wrong by failing to denoise/degibb my fieldmaps in addition to my DWI volumes?
I don't think the fieldmaps need to be denoised.
As for problematic noise voxels in my MD image, I have taken the following screenshots for your reference from a representative subject (i.e. 5216):
*Noise voxel #1 signal: ~0.05* [image: Screen Shot 2020-08-17 at 1.34.46 PM.png]
*Noise voxel #2 singal: ~0.01* [image: Screen Shot 2020-08-17 at 1.35.35 PM.png]
I have attached the free-water corrected scalars for this subject, as well as their MRtrix3 extracted bvals/bvecs to this email for your reference. I sincerely appreciate all of your continued time and consideration on this matter!
Any chance you could share the preprocessed data for the whole volume
for this individual? It's hard to say without looking at the values in these voxels. My hypothesis is that these are simply voxels where the signal is very low. In which case, I don't think there is any harm in masking them out.
I look forward to hearing from you soon!
Kind regards, Linda
On Mon, Aug 17, 2020 at 12:43 PM Ariel Rokem <arokem@gmail.com> wrote:
Hi Linda,
Sorry for the slowness here... It's been... challenging.
Two thoughts:
1. Sorry if I wasn't clear about this before: It is usually recommended that denoising and Gibbs ringing removal be done *before *other steps in preprocessing. To be on the safe side, I would recommend using https://qsiprep.readthedocs.io/en/latest/ for preprocessing. It implements the state of the art, and can be run as a docker/singularity container, which simplifies installation issues.
2. I am wondering what the signal is like in these voxels that still appear with very high MD values. Is there something unusual about their B0 signal? Or are the other data so low as to be indistinguishable from the noise floor? If you could find the coordinate of one of these voxels, and then us that to share with us the signal values in this voxel (as well as b-values and b-vectors) it would help diagnose this.
Cheers,
Ariel
On Mon, Aug 17, 2020 at 9:14 AM Linda Jasmine Hoffman < tuf72977@temple.edu> wrote:
> Good morning DIPY experts, > > I hope you have all been doing well! I just wanted to follow up > with you again as per my latest update re: persisting ventricular noise > post-denoising & FWC. Please let me know if you can shed any light on why > this noise may still be an issue, even after implementing Ariel's > denoising/degibbing suggestion. > > I look forward to hearing from you soon! > > Kind regards, > Linda > > On Mon, Aug 3, 2020 at 7:33 PM Linda Jasmine Hoffman < > tuf72977@temple.edu> wrote: > >> Good evening DIPY experts, >> >> I just wanted to follow up with you all as per my last email to see >> if you've had the opportunity to give my questions some consideration. >> >> Please let me know! I look forward to hearing from you soon! >> >> Kind regards, >> Linda >> >> On Tue, Jul 28, 2020 at 10:36 PM Linda Jasmine Hoffman < >> tuf72977@temple.edu> wrote: >> >>> Good evening DIPY experts, >>> >>> I have developed a denoising protocol for my HYDI data, and it has >>> afforded me some success in eliminating a portion of the excess ventricular >>> noise that I have been finding in my free-water-corrected (FWC) scalars. >>> Below is an example from a representative subject (i.e. "Subject >>> 1") for whom this course of actions seems to have worked quite >>> well: >>> >>> *Subject 1: Original MD map (no denoising of DWI data):* >>> [image: 5022_md.png] >>> >>> *Subject 1: New MD map (with denoising of DWI data):* >>> [image: 5022_md_denoised.png] >>> >>> However, I have a few concerns. First, my data is still not as >>> clean as I would like it to be, given the persisting residual noise that is >>> still present in the sagittal view. Second, the denoising protocol that I >>> have implemented did not work consistently well for all subjects. Here is >>> an example from a second representative subject (i.e. "Subject 2") >>> to illustrate this issue: >>> >>> *Subject 2: Original MD map (no denoising of DWI data):* >>> [image: 5216_md.png] >>> >>> *Subject 2: New MD map (with denoising of DWI data):* >>> [image: 5216_md_denoised.png] >>> >>> What is particularly concerning about this is that the resultant >>> image for Subject 2 is still not as clean as what is presented on >>> your DIPY free-water elimination page >>> <https://dipy.org/documentation/1.0.0./examples_built/reconst_fwdti/>. >>> It is worth noting that quality assurance measures have been taken for all >>> of our data, and this subject did not exhibit inordinate imaging artifacts. >>> >>> For your reference, my denoising pipeline utilized the >>> *dwidenoise* and *mrdegibbs* functions in MRtrix3. I >>> incorporated these steps into my processing protocol in the following order: >>> >>> 1. FSL - topup >>> 2. MRtrix3 - dwidenoise >>> 3. MRtrix3 - mrdegibbs >>> 4. FSL - eddy >>> >>> Note that I completed *topup* first since this step does not >>> affect the raw, DICOM-to-NIfTI-converted DWI volumes in any way, and it is >>> necessary for yielding a hifi brain mask. The scripts that I used for >>> denoising/degibbing are delineated below: >>> >>> >>> >>> *#dMRI noise level estimation and denoising using Marchenko-Pastur >>> PCA:* >>> for n in 5022 5216 5302 5391 >>> do >>> >>> dwidenoise >>> >>> -mask >>> /data/projects/tbi/denoise/${n}/topup_output/my_hifi_b0_Tcollapsed_brain_mask.nii.gz >>> -noise >>> /data/projects/tbi/denoise/${n}/dwidenoise/noise_hifi_map.nii >>> /data/projects/tbi/denoise/${n}/6-cmrr_mb3hydi_ipat2_64ch/output.nii >>> /data/projects/tbi/denoise/${n}/dwidenoise/denoised_hifi_vol.nii >>> >>> done >>> >>> >>> *#Remove Gibbs Ringing Artifacts:* >>> for n in 5022 5216 5302 5391 >>> do >>> >>> mrdegibbs >>> >>> /data/projects/tbi/denoise/${n}/dwidenoise/denoised_hifi_vol.nii >>> /data/projects/tbi/denoise/${n}/mrdegibbs/denoised_degibbs_hifi_vol.nii >>> >>> done >>> >>> What are your thoughts on the scripts I have implemented? Might I >>> have done something incorrectly, or is there something further I should do >>> to optimize this denoising pipeline? Is there anything I can do in >>> addition to denoising to eliminate these undue levels of post-FWC >>> ventricular noise in my scalars? >>> >>> Finally, do you recommend denoising and degibbing DWI data as a >>> canonical part of my pipeline? I ask because I know there is a tradeoff >>> between SNR and spatial resolution following noise reduction procedures, so >>> I'm curious to know what best-practices are in this regard. At the very >>> least it seems like an important step if one intends to pursue FWE. >>> >>> I sincerely appreciate all of your time and consideration on this >>> matter. >>> >>> Kind regards, >>> Linda >>> >>> On Thu, Jul 9, 2020 at 5:40 PM Linda Jasmine Hoffman < >>> tuf72977@temple.edu> wrote: >>> >>>> I haven't; I'll try that now. >>>> >>>> Thank you! >>>> Linda >>>> >>>> On Thu, Jul 9, 2020 at 5:38 PM Ariel Rokem <arokem@gmail.com> >>>> wrote: >>>> >>>>> Hi Linda, >>>>> >>>>> Have you had a chance to try Gibbs ringing removal or and/or >>>>> denoising on at least one subject? >>>>> >>>>> Cheers, >>>>> >>>>> Ariel >>>>> >>>>> >>>>> >>>>> On Thu, Jul 9, 2020 at 2:34 PM Linda Jasmine Hoffman < >>>>> tuf72977@temple.edu> wrote: >>>>> >>>>>> Hi everyone, >>>>>> >>>>>> I just wanted to touch base with you to see if you've had the >>>>>> opportunity to give my previous email some consideration. Please let me >>>>>> know what my next steps should be re: denoising my DWI data to >>>>>> eliminate excessive ventricular artifacts post-fwc. >>>>>> >>>>>> Thank you! >>>>>> Linda >>>>>> >>>>>> On Thu, Jul 2, 2020 at 12:41 PM Linda Jasmine Hoffman < >>>>>> tuf72977@temple.edu> wrote: >>>>>> >>>>>>> Hi Ariel, >>>>>>> >>>>>>> Our preprocessing pipeline includes the following steps for >>>>>>> noise reduction in FSL: >>>>>>> >>>>>>> - topup - correct for the susceptibility induced field and >>>>>>> movement >>>>>>> - eddy - correct for eddy current distortions and movement >>>>>>> >>>>>>> We don't have a step in our pipeline to correct for Gibbs >>>>>>> artifacts. Do you think this particular type of artifact is what's >>>>>>> underpinning this issue with the FWC scalar maps? If so, I found a command >>>>>>> in MRtrix3 (mrdegibbs) that eliminates ringing, but it will necessitate >>>>>>> that I go back and redo a large amount of preprocessing. Do you know of an >>>>>>> alternative route to mitigate this problem that may obviate my need to >>>>>>> reprocess my data? >>>>>>> >>>>>>> Thank you so much for your help! >>>>>>> Kind regards, >>>>>>> Linda >>>>>>> >>>>>>> On Thu, Jul 2, 2020 at 12:45 AM Ariel Rokem <arokem@gmail.com> >>>>>>> wrote: >>>>>>> >>>>>>>> Hi Linda, >>>>>>>> >>>>>>>> With your permission, I am adding the DIPY mailing list, so >>>>>>>> others can weigh in and/or benefit from the discussion. >>>>>>>> >>>>>>>> My hunch is that the noise you are seeing in the ventricles >>>>>>>> is due to artifacts/noise. Do you do any removal of Gibbs ringing artifacts >>>>>>>> or any denoising of the data before analyzing it with fwdti? >>>>>>>> >>>>>>>> Cheers, >>>>>>>> >>>>>>>> Ariel >>>>>>>> >>>>>>>> >>>>>>>> On Wed, Jul 1, 2020 at 12:22 PM Linda Jasmine Hoffman < >>>>>>>> tuf72977@temple.edu> wrote: >>>>>>>> >>>>>>>>> Good afternoon DIPY experts, >>>>>>>>> >>>>>>>>> My name is Linda Hoffman, and I'm the lab manager for Dr. >>>>>>>>> Ingrid Olson's Cognitive Neuroscience Lab at Temple University. I have >>>>>>>>> been working on implementing a DIPY-based free-water elimination (FWE) >>>>>>>>> pipeline that my labmate, Katie Jobson, adapted from your >>>>>>>>> website >>>>>>>>> <https://dipy.org/documentation/1.0.0./examples_built/reconst_fwdti/> >>>>>>>>> in order to extract free-water corrected (FWC) scalar maps from a HYDI >>>>>>>>> dataset that I'm analyzing. For your reference, I am ultimately >>>>>>>>> planning to calculate FWC DTI metrics for the fornix and genu of the corpus >>>>>>>>> callosum after performing probabilistic tractography. I have preprocessed >>>>>>>>> my data using FSL version 6.0 and MRtrix3 on a linux machine. >>>>>>>>> >>>>>>>>> While I have successfully extracted FWC FA, MD, RD, and AD >>>>>>>>> maps from my data using this pipeline, there still seems to be a >>>>>>>>> disproportionate amount of noise in the ventricles, especially when >>>>>>>>> comparing my output to your examples on the website linked above. This is >>>>>>>>> the case even after eliminating voxels with a water volume fraction (WVF) >>>>>>>>> exceeding 70%. In light of this, I was wondering if you may be able to >>>>>>>>> address the following questions: >>>>>>>>> >>>>>>>>> - Is the amount of ventricular noise post-FWE in my >>>>>>>>> scalar maps within a normal range? Will this preclude me from extracting >>>>>>>>> valid FWC DTI metrics from the fornix and the genu? Here are some >>>>>>>>> screenshots from a representative subject's scalar maps: >>>>>>>>> >>>>>>>>> *FA map with WVF elimination at a threshold of 70%* >>>>>>>>> [image: fa_70.png] >>>>>>>>> *MD map with WVF elimination at a threshold of 70%* >>>>>>>>> [image: md_70.png] >>>>>>>>> *RD map with WVF elimination at a threshold of 70%* >>>>>>>>> [image: rd_70.png] >>>>>>>>> *AD map with WVF elimination at a threshold of 70%* >>>>>>>>> [image: ad_70.png] >>>>>>>>> >>>>>>>>> >>>>>>>>> - If this noise is not within an acceptable range, how >>>>>>>>> might I be able optimize our DIPY script so that I can perform a better >>>>>>>>> FWE? I tried comparing the results from using a stricter WVF threshold of >>>>>>>>> 60% as well as using no WVF thresholding to the above results. Using a >>>>>>>>> stricter threshold did not completely eliminate the noise problem, but it >>>>>>>>> did help a little bit. However, I'm not sure if there is a precedent for >>>>>>>>> this level of thresholding in the literature, or if it is actually >>>>>>>>> appropriate. Screenshots from a representative subject are listed below: >>>>>>>>> >>>>>>>>> *FA map with WVF elimination at a threshold of 60%* >>>>>>>>> [image: fa_60.png] >>>>>>>>> >>>>>>>>> *MD map with WVF elimination at a threshold of 60%* >>>>>>>>> [image: md_60.png] >>>>>>>>> *FA map with No WVF elimination threshold* >>>>>>>>> [image: fa_none.png] >>>>>>>>> *MD map with No WVF elimination threshold* >>>>>>>>> [image: md_none.png] >>>>>>>>> >>>>>>>>> I have attached a zip file with the following information >>>>>>>>> for your reference: >>>>>>>>> >>>>>>>>> 1. Input data from a representative subject. This >>>>>>>>> includes DWI volumes collected at b values between 0 to 2000. This is >>>>>>>>> contained in the *subject_data *subfolder. >>>>>>>>> 2. Scalar maps collected with a WVF thresholding rate of >>>>>>>>> 70% (*F>.7*), 60% (*F>.6*), and with no thresholding ( >>>>>>>>> *no_F_threshold*). >>>>>>>>> 3. Three versions of the DIPY script I've been using - >>>>>>>>> each one accounts for a different rate of WVF thresholding. >>>>>>>>> These scripts are contained in the >>>>>>>>> *dipy_fwe_script_versions* subfolder. >>>>>>>>> >>>>>>>>> I sincerely appreciate all of your time and consideration, >>>>>>>>> and look forward to hearing from you soon! >>>>>>>>> >>>>>>>>> Kind regards, >>>>>>>>> Linda >>>>>>>>> >>>>>>>>> dipyfwe.zip >>>>>>>>> <https://drive.google.com/a/temple.edu/file/d/1yvLYeB-cUNqBoce-43ler0-zf_vG8b...> >>>>>>>>> -- >>>>>>>>> *Lab Manager* >>>>>>>>> *Cognitive Neuroscience Lab* >>>>>>>>> Temple University >>>>>>>>> 1701 N. 13th St. >>>>>>>>> Philadelphia, PA 19122 >>>>>>>>> >>>>>>>>> *Pronouns: * She/Her >>>>>>>>> *Phone*: (215) 204-1708 >>>>>>>>> *Email*: tuf72977@temple.edu >>>>>>>>> >>>>>>>> >>>>>>> >>>>>>> -- >>>>>>> *Lab Manager* >>>>>>> *Cognitive Neuroscience Lab* >>>>>>> Temple University >>>>>>> 1701 N. 13th St. >>>>>>> Philadelphia, PA 19122 >>>>>>> >>>>>>> *Pronouns: * She/Her >>>>>>> *Phone*: (215) 204-1708 >>>>>>> *Email*: tuf72977@temple.edu >>>>>>> >>>>>> >>>>>> >>>>>> -- >>>>>> *Lab Manager* >>>>>> *Cognitive Neuroscience Lab* >>>>>> Temple University >>>>>> 1701 N. 13th St. >>>>>> Philadelphia, PA 19122 >>>>>> >>>>>> *Pronouns: * She/Her >>>>>> *Phone*: (215) 204-1708 >>>>>> *Email*: tuf72977@temple.edu >>>>>> >>>>> >>>> >>>> -- >>>> *Lab Manager* >>>> *Cognitive Neuroscience Lab* >>>> Temple University >>>> 1701 N. 13th St. >>>> Philadelphia, PA 19122 >>>> >>>> *Pronouns: * She/Her >>>> *Phone*: (215) 204-1708 >>>> *Email*: tuf72977@temple.edu >>>> >>> >>> >>> -- >>> *Lab Manager* >>> *Cognitive Neuroscience Lab* >>> Temple University >>> 1701 N. 13th St. >>> Philadelphia, PA 19122 >>> >>> *Pronouns: * She/Her >>> *Phone*: (215) 204-1708 >>> *Email*: tuf72977@temple.edu >>> >> >> >> -- >> *Lab Manager* >> *Cognitive Neuroscience Lab* >> Temple University >> 1701 N. 13th St. >> Philadelphia, PA 19122 >> >> *Pronouns: * She/Her >> *Phone*: (215) 204-1708 >> *Email*: tuf72977@temple.edu >> > > > -- > *Lab Manager* > *Cognitive Neuroscience Lab* > Temple University > 1701 N. 13th St. > Philadelphia, PA 19122 > > *Pronouns: * She/Her > *Phone*: (215) 204-1708 > *Email*: tuf72977@temple.edu >
-- *Lab Manager* *Cognitive Neuroscience Lab* Temple University 1701 N. 13th St. Philadelphia, PA 19122
*Pronouns: * She/Her *Phone*: (215) 204-1708 *Email*: tuf72977@temple.edu
-- *Lab Manager* *Cognitive Neuroscience Lab* Temple University 1701 N. 13th St. Philadelphia, PA 19122
*Pronouns: * She/Her *Phone*: (215) 204-1708 *Email*: tuf72977@temple.edu
-- *Lab Manager* *Cognitive Neuroscience Lab* Temple University 1701 N. 13th St. Philadelphia, PA 19122
*Pronouns: * She/Her *Phone*: (215) 204-1708 *Email*: tuf72977@temple.edu
-- *Lab Manager* *Cognitive Neuroscience Lab* Temple University 1701 N. 13th St. Philadelphia, PA 19122
*Pronouns: * She/Her *Phone*: (215) 204-1708 *Email*: tuf72977@temple.edu
participants (2)
-
Ariel Rokem -
Linda Jasmine Hoffman