Mailman 3 December 2020 - DIPY

IVIM fit using GPU
by Marek Dostál Dec. 21, 2020

Dec. 21, 2020

Dear Dipy users, I am fitting brain IVIM data using IvimModelVP and I would like to make it faster. I can use multi-CPU core via Parallel (whole brain fit still takes 135 min) but I cannot implement it into GPU model via numba. Can you help me how can I implement IvimModelVP into GPU calculation and make it faster? Thank you Mark

1 0

Request for corrected HCP tractography atlases to perform RecoBundles
by Krahulec, D. Dec. 6, 2020

Dec. 6, 2020

Dear DIPY experts and Dr. Yeh, I am working on a doctoral project that involves tractography and intend to use RecoBundles (as described in https://www.ncbi.nlm.nih.gov/pubmed/28712994) with an HCP atlas in my streamline generation approach. Lately, I encountered several inconsistencies in various tractography atlases (coming from here: https://www.ncbi.nlm.nih.gov/pubmed/29758339). While implementing my code, I spotted that the DIPY-recommended HCP atlas (HCP842) lacks its reference template image. Here is the link to this atlas: https://figshare.com/articles/Advanced_Atlas_of_80_Bundles_in_MNI_space/737…. In its README file, it is described that the bundles are in the MNI ICBM 2009a space, yet this is not the case when I overlay the TRK files on the 'mni_icbm152_t1_tal_nlin_asym_09a.nii' MNI template image (see Figure 1 attached). As you can see, the affine matrices are inconsistent as well. In this forum entry (https://github.com/dipy/dipy/issues/2007), I later found out that Dr. Francois Rheault had recommended using the 'corrected version' of the HCP842 series, registered to the 'mni_icbm152_t1_tal_nlin_asym_09c.nii' MNI ICBM 2009c space, available here: https://drive.google.com/open?id=1pObyKLGJNPFZhMIncYuLn9utJDKb-Uq9. Again, when I visualized the TRK tracks on the T1 reference, the tracks ended up lying outside of the reference geometry (see Figure 2). Moreover, I learnt there is another tractography atlas (HCP1065, available here: http://brain.labsolver.org/diffusion-mri-templates/tractography), which is coincidentally most ideal for my project, but does not come in the TRK format. Instead, it is available in the TT.GZ (https://pitt.app.box.com/v/HCP1065-TRACT) or NII.GZ (https://pitt.app.box.com/v/HCP1065-NIFTI) formats, which cannot be loaded using DIPY, can it? If you have the TRK-formatted tracks of this atlas, can you please share them with me? It appears that tractography atlas misalignments are an open issue that might not have been resolved yet: https://github.com/dipy/dipy/issues/2166. Also, it seems to me that the code example on the official DIPY website (https://dipy.org/documentation/1.3.0./examples_built/bundle_extraction/) contains a couple of confusing / buggy steps (target header creation, reference images for input TRK bundles in function load_trk(), etc.). As such, the TRK bundles of the HCP842 atlas suffice to perform virtual dissection, however, to be able to overlay the bundle models in any subject space (not in MNI) to compare the dissected bundles with these models, I need the right reference image ensuring the headers / coordinate systems correspond. I would appreciate your help / provision of a corrected and consistent atlas along with its reference template image in the coming days. Thank you very much in advance. Best regards, Daniel Krahulec, MSc (Tech.) PhD candidate Eindhoven University of Technology Eindhoven, The Netherlands

3 6

Come play with us! fibr.dev, a citizen science game for dMRI QA
by Ariel Rokem Dec. 4, 2020

Dec. 4, 2020

Dear DIPY aficionados, Come try our new citizen science app at https://fibr.dev! ### What is this thing? We developed a citizen science web app to help measure the quality of brain MRI scans. To play, simply log in to https://fibr.dev and swipe left if the brain scan is “bad” or swipe right if the brain scan is “good” (we describe what “bad” and “good” mean in a short tutorial). ### Why we’re doing this? Innovations in medical imaging are helping researchers understand the developing human brain. There is growing hope that these technologies will lead to clinical tools that can help diagnose and treat diseases, mental illness, and learning disorders. But to achieve this dream, we need massive MRI datasets to capture the broad spectrum of brain development. So the Child Mind Institute has launched the Healthy Brain Network ( https://healthybrainnetwork.org/), a landmark mental health study that is collecting MRI images and other assessment data from 10,000 New York City area children. We have developed algorithms that analyze these images to measure the integrity of connections between different parts of the brain. But they aren’t perfect and sometimes do a bad job, especially if the child moved a lot in the scanner (kids have a hard time staying still sometimes). We need humans to visually inspect the quality of the output. Computers can struggle with this sort of thing but humans are great at it! That’s why we need you. We need you to help researchers learn about pediatric mental health and learning disorders. ### How you can help Log in to https://fibr.dev. Classify these images as “pass” or “fail” by swiping right or left. You’ll get points for each sample you swipe and you can unlock fun animal badges based on the number of points you have. Go unlock them all so you can brag about being a unicorn!

1 0