Table 33 Comparative analysis of clinical translation via cross-institutional and LOSO validation.
Model | Cross-institutional accuracy (%) | LOSO accuracy (%) | AUC-ROC (%) | Dice coefficient (%) | Remarks |
|---|---|---|---|---|---|
Aniso-ResCapHGBO-Net (Proposed) | 98.91 | 97.80 | 99.80 | 96.10 | High generalizability, preserves privacy, robust to institutional variance |
CNN | 93.24 | 91.05 | 94.80 | 88.30 | Moderate generalization; overfits local features |
dResU-Net | 95.63 | 93.80 | 96.50 | 91.50 | Strong segmentation, moderate cross-site stability |
CapsuleNet | 94.88 | 92.30 | 95.70 | 89.90 | Good spatial sensitivity; lower precision across sites |
SVM | 89.41 | 87.05 | 90.60 | 80.40 | Lower adaptability to heterogeneity |
KNN | 86.22 | 84.35 | 88.20 | 78.50 | Limited scalability; less robust in cross-domain applications |
