Swin-UNet cGAN Improves Knee Cartilage Segmentation

Researchers at Yonsei University and collaborators (2026) develop a Swin-UNet conditional GAN to automatically segment femoral and tibial cartilage from knee MRIs, training and testing on 232 internal scans plus an external validation set. The model achieved the highest mean Dice and IoU scores and significantly better tibial boundary metrics (ASSD, HD95), indicating improved cartilage delineation that could support MRI-based patient-specific surgical planning.