Table 2.

Deep learning segmentation models for shoulder joint anatomy.

Name	Core advantages	Segmented anatomical structures	Imaging modality	Dice similarity coefficient (DSC)	References
nnUNet	Auto-adapts to datasets; high efficiency for bone segmentation	Humerus, glenoid cavity	MRI	Humerus: 0.95; Glenoid: 0.86	Mu et al. (2021) and Dai et al. (2024)
U-Net (basic version)	Strong soft tissue feature fusion; stable performance	Rotator cuff muscles (supraspinatus, infraspinatus)	MRI	Supraspinatus: 0.89; Infraspinatus: 0.91	Medina et al. (2021) and Alipour et al. (2024)
MSFFN (U-Net + AlexNet)	Multi-scale feature integration; optimized for sports-related injuries	Glenoid, humerus	MRI	Glenoid: 0.9265; Humerus: 0.9293	Dai et al. (2024)
U-Net-based rotator cuff special model	Automatic muscle selection; high generalization for Y-view images	Rotator cuff muscles (subscapularis, teres minor)	MRI (compatible with ultrasound)	≥0.93 (internal/external test sets)	Alipour et al. (2024)