Component-Adaptive and Lesion-Level Supervision for Improved Small Structure Segmentation in Brain MRI
The study introduced CATMIL, a unified objective function to improve segmentation of small brain lesions in MRI. It combines a component-adaptive term that reweights voxel contributions based on lesion size and a Multiple Instance Learning term that encourages the detection of individual lesions. This approach significantly boosts small lesion recall and reduces false negatives compared to standard methods.