Additional data files associated with "Distinct brain morphometry patterns revealed by deep learning improve prediction of post-stroke aphasia severity".

This .mat file contains most of the data (N=173/231) used in "Distinct brain morphometry patterns revealed by deep learning improve prediction of aphasia severity (Teghipco et al., under review). The data in the .mat file contain the most recent ARC database release (https://openneuro.org/datasets/ds004512/versions/2.0.0). Remaining data used in the manuscript will eventually be available as part of this public database repository.
Variables in DLAphasiaSeverityARCSubset.mat:
1) data: 4D matrix of x,y,z dimensions reflecting volumetric participant morphometry and lesion anatomy. The 4th dimension is participants.
2) subsOnly: participant IDs
3) wabClassi: severe(1) and nonsevere(0) aphasia status of participants
4) wabClass2i: more granular aphasia status of participants. Very severe(4), severe(3), moderate(2), mild(1)
Please see doi:10.18112/openneuro.ds004512.v2.0.0 on openneuro for raw and minimally processed ARC data. The core deep learning code we used with this data can be found here: https://github.com/alexteghipco/volDNN (this code is flexible to accommodate other deep learning problems, please see the parameters we used in our manuscript, which can be found here: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC10350198/).
Patient subtyping results are attached as the remaining two .mat files. Both files are save states for the brainSurfer GUI and can be loaded directly into brainSurfer to visualize the patient subcategories (clustering exemplars) exactly as we presented them in Figures 9 and 10 of our manuscript. See brainSurfer: https://github.com/alexteghipco/brainSurfer