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,LindaOn 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,LindaOn 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):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 1: New MD map (with denoising of DWI data):Subject 2: Original MD map (no denoising of DWI data):Subject 2: New MD map (with denoising of DWI data):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. 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:
- FSL - topup
- MRtrix3 - dwidenoise
- MRtrix3 - mrdegibbs
- 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
dodwidenoise-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.niidone#Remove Gibbs Ringing Artifacts:for n in 5022 5216 5302 5391domrdegibbs/data/projects/tbi/denoise/${n}/dwidenoise/denoised_hifi_vol.nii /data/projects/tbi/denoise/${n}/mrdegibbs/denoised_degibbs_hifi_vol.niiWhat 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?doneFinally, 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,LindaOn 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!LindaOn 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,ArielOn 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!LindaOn 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,LindaOn 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,ArielOn 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 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%MD map with WVF elimination at a threshold of 70%RD map with WVF elimination at a threshold of 70%AD map with WVF elimination at a threshold of 70%
- 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%I have attached a zip file with the following information for your reference:MD map with WVF elimination at a threshold of 60%FA map with No WVF elimination thresholdMD map with No WVF elimination thresholdI sincerely appreciate all of your time and consideration, and look forward to hearing from you soon!
- 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.
- Scalar maps collected with a WVF thresholding rate of 70% (F>.7), 60% (F>.6), and with no thresholding (no_F_threshold).
- 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.
Kind regards,Linda--Lab ManagerCognitive Neuroscience LabTemple University1701 N. 13th St.Philadelphia, PA 19122Pronouns: She/HerPhone: (215) 204-1708Email: tuf72977@temple.edu--Lab ManagerCognitive Neuroscience LabTemple University1701 N. 13th St.Philadelphia, PA 19122Pronouns: She/HerPhone: (215) 204-1708Email: tuf72977@temple.edu--Lab ManagerCognitive Neuroscience LabTemple University1701 N. 13th St.Philadelphia, PA 19122Pronouns: She/HerPhone: (215) 204-1708Email: tuf72977@temple.edu--Lab ManagerCognitive Neuroscience LabTemple University1701 N. 13th St.Philadelphia, PA 19122Pronouns: She/HerPhone: (215) 204-1708Email: tuf72977@temple.edu--Lab ManagerCognitive Neuroscience LabTemple University1701 N. 13th St.Philadelphia, PA 19122Pronouns: She/HerPhone: (215) 204-1708Email: tuf72977@temple.edu--Lab ManagerCognitive Neuroscience LabTemple University1701 N. 13th St.Philadelphia, PA 19122Pronouns: She/HerPhone: (215) 204-1708Email: tuf72977@temple.edu--Lab ManagerCognitive Neuroscience LabTemple University1701 N. 13th St.Philadelphia, PA 19122Pronouns: She/HerPhone: (215) 204-1708Email: tuf72977@temple.edu
