Fig. 2.

Overview of image processing for the validation analysis.

A) For the internal validation, we used the lesion segmentation tool (LST) for cross-sectional lesion segmentation separately for each time point. These lesion segmentations were manually corrected. Lesion changes were created by difference images which were also manually corrected (manually corrected lesion change label, LCL). Manually corrected cross-sectional lesion segmentations served as starting point for the segmentation of white matter lesion changes by LST (including the joint lesion map). For validation analysis, manually corrected LCLs were compared to LCLs derived from LST.

B) For the external validation, MIAC AG (Medical Image Analysis Center Basel, Switzerland) delivered coregistered cross-sectional lesion segmentations of both time points and segmentations of white matter lesion changes (LCL). MIAC cross-sectional lesion segmentation served as starting point for the longitudinal lesion segmentation by LST including the joint lesion map. The latter was complemented by MIAC LCLs. For validation analysis, MIAC LCLs were compared to LCLs derived from LST.