This repository was adapted from MASILab/tbi_ct_lesion_segmentation repository.

This repository can be used to take T1-weighted images where enlarged perivascular spaces (PVS) have been labelled, train a deep learning U-net, test it, and use it to segment unlabelled data.

PVS are a pathology that appear isointense with CSF on T1-weighted and T2-weighted images. They are a small vessel disease pathology that may be indicative of vascular damage and impaired clearance of molecular debris.

We recommend a labelled dataset of 50 images where 40 are used for training and 10 are used for testing.