Fig. 3.

The MIMoSA procedure is demonstrated and visualized in an example axial slice. 1. MIMoSA first selected candidate voxels defined as being the 85th quantile or above in intensity on the FLAIR images. 2. Features inside the candidate mask were then extracted (full brain features derived from FLAIR volumes are only shown for simplicity). 3. To obtain T1 lesion (T1L) masks and T2 lesion (T2L) masks, separate models were fit. 4. Training the MIMoSA models on a subset of subjects with manual segmentations yielded segmentation models which were then applied to test subjects not included in the training set.