Automatic perivascular space segmentation on T2-weighted MR images
- Frangi filtering can effectively highlight the perivascular spaces in the white matter.
- However, since many false positives occur at the tissue interfaces, a predefined ROI mask and additional FP reduction step are often required.
- Frangi filtering (jupyter notebook)
- Input: 3D T2-weigthed MR image & Tissue segmentation from infant freesurfer
- Output: BG (Basal Ganglia) PVS segmentation
- BG PVS volume calculation (jupyter notebook)
- A deep learning is an effective way to reduce false positives.
- In our initial method, the DL model was designed in a way that classfies FP from the Frangi filtering result (HBM 2021).
- The new version was learned using the PVS and ROI segmentations, both ROI (white matter & deep gray matter regions) and PVS can be simultaneously segmented from T2 input.