I’m a beginner in both QSIPrep and DWI analysis. I currently have a dataset that I’ve successfully converted into BIDS format. For now, all my structural and diffusion data are placed under a single session named ses-anat. The folder structure is as follows:
Here’s the tricky part: acq-a and acq-b were acquired approximately six weeks apart (each with both AP and PA directions collected during the same session). However, I’m not interested in longitudinal changes—I simply want to take advantage of all available data to improve the signal-to-noise ratio for a better overall reconstruction.
Currently, I’m running QSIPrep with the following command:
My main questions are:
1. Is this QSIPrep command appropriate for my use case, especially with the --separate-all-dwis and --distortion-group-merge average options?
2. Once QSIPrep finishes, how should I proceed with running QSIRecon to best utilize all the available diffusion data in a combined fashion?
3. Are there any best practices or recommendations for merging multi-session (but closely related) DWI data like this?
Any insights or suggestions would be greatly appreciated. Thank you in advance for your help!
It would help to know more about the data, e.g., number of volumes, to know if this would really help.
That being said,
Maybe others will have different opinions, but in this case, I would not combine them in preprocessing. These should go into separate session folders. You might be better off instead averaging whatever scalar maps you produce at the end with QSIRecon. Or in your analyses you can do a test-retest reliability analysis using your two datasets.
That is the wrong repo. QSIPrep now uses pennlinc/qsiprep. We also recommend using the most up to date version, which is 1.0.1 at this time.
I wouldn’t use these option personally. Instead I would not separate DWIs and use the concat merging. This will make it so your final DWI (or DWIs since I recommend you do session-wise processing) that recon is done on reflects the entire AP/PA series.
Many thanks for your reply—I’ve learned a great deal from it. I will share additional details about my dataset shortly. Thank you again for your patience and constructive feedback .
It would help to know more about the data
These two acquisitions use different parameters:
In acq-a: 74 volumes, 76 slices (from dcmmeta_shape); 2.0 mm isotropic voxels; TR/TE = 4.3 s / 65 ms; multiband MB=2 with in-plane GRAPPA R=2; EffectiveEchoSpacing = 0.330 ms; TotalReadoutTime = 41.91 ms.
In acq-b: 160 volumes, 84 slices; 1.798 mm isotropic voxels; TR/TE = 3.2 s / 81 ms; multiband MB=4; EffectiveEchoSpacing = 0.650 ms; TotalReadoutTime = 73.45 ms.
Additionally, the two runs use different b-value tables.
Maybe others will have different opinions, but in this case, I would not combine them in preprocessing. These should go into separate session folders. You might be better off instead averaging whatever scalar maps you produce at the end with QSIRecon. Or in your analyses you can do a test-retest reliability analysis using your two datasets.
Your suggestion is very helpful—I’ll analyze my data following your advice. However, my dataset is quite large (it includes many functional runs), so I’d like to ask whether using --bids-filter-file would achieve an effect similar to separating the sessions.
In that case, with your acquisitions being very different from one another, I would definitely not combine them, and unfortunately they probably cannot be used for test-retest reliability either.
I do not know enough about your analysis goals to tell you with acquisition is better, but it looks like the acq-b will have better spatial and angular resolution.
If that is the case, I would not specify --output-resolution 2, as you would be downsampling your data unnecessarily.
It could, but on principle these data should be divided into different session folders.
