mschallenge-skill

Run the validate_mschallenge.py script with --input pointing to the dataset root and --output for the CSV report. It checks subject directories, timepoint folders, modality completeness for T1, T2, FLAIR, and PD, and ground truth lesion mask presence.

Load the binary mask with nibabel, count non-zero voxels, and multiply by the voxel volume from the header zooms to get volume in cubic millimeters. The analyze_lesions.py script automates this and also counts connected lesion components using scipy.ndimage.label.

Yes, pass the --no-ground-truth flag to the validation and QC scripts to skip lesion mask checks. This is intended for the 14 test subjects whose ground truth is hidden by the challenge organizers.

The scripts require numpy and nibabel for NIfTI loading and array operations, with scipy as an optional dependency for connected-component lesion counting. Without scipy, the analysis falls back to a simpler lesion presence count.

The longitudinal script compares binary masks voxel-wise between consecutive timepoints to classify new, resolved, and stable lesion voxels. Enlarged lesion detection is not implemented because it requires connected-component analysis across registered timepoints.