leelisae
commented
4 months ago @jcohenadad - The example data I shared via Dropbox passed QC (C2-C4 is the region of my interest).
However, with the CanProCo dataset, I noticed frequent QC failures, primarily related to vertebral labeling and/or registration, and in rare instances, segmentation. I approached each QC failure case-by-case, experimenting to find the most effective solution.
For example, while I initially preferred manual initialization for vertebral labeling corrections, I often found this insufficient, requiring me to manually place the labels entered in the SC. For registration, I’d adjust parameters (based on your previous guidance too), but occasionally had to resort to segmentation-based registration.
I replaced the filenames of the original QC-failed outputs with corrected versions and re-ran the rest of the code. I now understand that there are more sophisticated methods available, but given my limited scripting knowledge and storage capacity (I was provided only a MacBook and external hard drive), this approach seemed practical at the time.
Currently, I have several upcoming deadlines. However, I’m open to modifying the analysis script in the future to be more sophisticated, if needed. Thank you!
jcohenadad
@leelisae while working on #91 I was wondering, after you run the processing on a subject and you do the QC, if you notice issues with the segmentation, where is the manually corrected segmentation or mask stored? The storage location of derivative is important as it should be considered by the analysis script, such that, if it exists, instead of re-running the segmentation the code would instead use the manually corrected segmentation/label (as done, e.g., in this other project: https://github.com/sct-pipeline/spine-park).