Multi-Structure Segmentation of Type B Aortic Dissection from CT Images with U-Net

nnU-Net 3D segmentations of Type B aortic dissection CT achieved DSCs of 93.05% for the lumen, 50.22% for the intimal flap, and 69.94% for thrombus.

• Why it matters: Accurate delineation of the lumen, intimal flap, and false-lumen thrombosis from CT is important for risk assessment and treatment planning in TBAD, where manual analysis is challenging.
• What they did: The study trained and compared four U-Net-based deep learning models—nnU-Net 2D/3D and TransUNet 2D/3D—using voxel-wise annotations from 46 CT scans of 20 patients.
• The result: Across all three structures, 3D models outperformed 2D, with the largest gains for the thin flap and thrombus; the flap remained hardest and boundary errors persisted in some scans.

AI summary of the abstract · may contain errors · not medical advice · read the paper.