We have two diffusion MRI acquisitions per subject, the two runs are collected in the same session. Reason for collecting two is because our clinician wants a standard clinical protocol, so we just add another research protocol on top of that. (At least that’s my understanding)
Clinical routine: fixed b = 1000 (30 directions)
Research routine: fixed b = 2000 (55 directions)
These two sequences were acquired with different TR and TE parameters. Our main goal is to construct a tract statistics dataset for group-level comparisons between a patient cohort and healthy controls.
When running QSIPrep, we observed that the default workflow concatenates the two dwi data and bval, bvec files across both acquisitions and performs preprocessing/distortion correction on the combined dataset simultaneously.
Question
Combining Acquisitions in QSIPrep: Is concatenated preprocessing expected/recommended when combining two diffusion runs with different TR/TE values?
Separate vs. Combined Strategy: If we opt to analyze these acquisitions separately:
What are the trade-offs between utilizing the research b=2000 with 55 directions or the clinical data of b=1000 with 30 directions.
Should one sequence be prioritized over the other, or is there a standard practice for running both pipelines in parallel and evaluating them side-by-side?
This is a bit complicated, I will try to answer the best I can.
To start, if you have any say in the acquisition parameters, it would be better to in the future combine these acquisitions into one at the scanner. You can always split up a nifti in to the low-b and high-b after preprocessing as needed if you want to compare the two imaging schemes (assuming you have enough b-0 volumes interspersed around the low and high b-valued images).
Most post-processing modeling choices do not depend on TR/TE, so concatenated images may work fine. So you can do some multishell options like NODDI and DKI. However, MAP-MRI depends on accurate big+little delta estimates, which could depend on TR/TE, so we do not recommend that workflow for data acquired like this. FOD estimation and tractography could still work too. But in all cases we recommend looking at the model outputs to make sure they look reasonable.
It is important in your case that data are denoised before combining. So make sure you are not using the`--denoise-after-combining flag. Also, it is important that each image (the low-b and high-b) have b0s, that are ideally similar across scans (otherwise you can get some weird scaling issues when combining the scans). I would also be sure to look at the final preprocessed image to make sure it looks reasonable. You may consider running each scan individually too so you can compare the individual vs. combined runs.
Multi-shell processing is generally better than single-shell. In addition to the increased angular resolution from having more directions in your image, your FODs will look sharper (leading to better tractography directional estimations), and you enable processing methods such as NODDI and DKI.
In general, as mentioned above, I would always recommend collecting multi-shell data. If you want to simulate clinical data for whatever reason, you can always just extract the single-shell low-b data from the overall acquisition.