BI-RADS-based malignancy classification on EMBED (test)
91.8AUCMIL-PF
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| MIL-PF# Trainable Params (M)=0.04, Level=Breast, Backbone=MedSigLIP, Aggregation=max2026.03 | 91.8 | 84.5 | 73.5 | |
| MIL-PF# Trainable Params (M)=0.05, Level=Breast, Backbone=DINOv2, Aggregation=attn2026.03 | 91.6 | 85 | 76.2 | |
| MIL-PF# Trainable Params (M)=0.04, Level=Breast, Backbone=MedSigLIP, Aggregation=attn2026.03 | 91.4 | 84.3 | 74.6 | |
| MIL-PF# Trainable Params (M)=0.05, Level=Breast, Backbone=DINOv2, Aggregation=max2026.03 | 90.5 | 83.3 | 70.3 | |
| SIL-L GMIC-ResNet18# Trainable Params (M)=22.49, Level=Image2026.03 | 87.5 | 80.2 | 56.6 | |
| ES-Attside# Trainable Params (M)=22.89, Level=Case2026.03 | 83.6 | 66.8 | 55.8 | |
| FPN-AbMIL (mean)# Trainable Params (M)=1.76, Level=Breast2026.03 | 83.5 | 76.5 | 40.3 | |
| FPN-SetTrans (mean)# Trainable Params (M)=1.76, Level=Breast2026.03 | 83.5 | 76.7 | 45.1 | |
| FPN-AbMIL (max)# Trainable Params (M)=5.38, Level=Breast2026.03 | 81.7 | 75.9 | 33.4 | |
| GMIC# Trainable Params (M)=14.11, Level=Image2026.03 | 81.6 | 73.6 | 38 | |
| FPN-SetTrans (max)# Trainable Params (M)=5.38, Level=Breast2026.03 | 81.3 | 72.8 | 38.8 | |
| FPN-AbMIL# Trainable Params (M)=1.76, Level=Image2026.03 | 80.2 | 74 | 36.7 | |
| FPN-SetTrans# Trainable Params (M)=5.38, Level=Image2026.03 | 79.4 | 72.8 | 36.4 |