Fig. 6

Deep learning architecture overview. Mask R-CNN was used to detect CMBs slice-wise and to obtain a semantic segmentation mask of CMBs. Next, semantic segmentation masks of CMBs were spliced with the originally corresponding SWS to input the MIL network for classifying CSVD SWS-wise. CMBs cerebral microbleed, SWS susceptibility-weighted MR sequences, MIL multi-instance learning, CSVD cerebral small vessel disease