Image Classification on Breast UltraSound (BUS) dataset
88.5AccuracyFull FT MoE
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| Full FT MoEBackbone=BiomedCLIP ViT-B/16, # Params (%)=760, Evaluation Protocol=Zero-shot2026.04 | 88.5 | — | — | — | — | — | — | |
| Hybrid-MT-ESTANLearning Paradigm=Multi-task learning, Backbone=SwinTransformer_V2_Base_256, Attention Module=AAA blocks2023.08 | 82.8 | 86.4 | 79.2 | 86 | 82.8 | 13.6 | 20.8 | |
| MT-ESTANLearning Paradigm=Multi-task learning2023.08 | 78.6 | 83.7 | 72.6 | 83 | 78.2 | 16.3 | 27.4 | |
| ChowderyLearning Paradigm=Multi-task learning2023.08 | 77.4 | 77.3 | 77.3 | 77 | 77.3 | 22.7 | 22.7 | |
| Swin TransformerLearning Paradigm=Single-task learning2023.08 | 77.4 | 72.6 | 82.5 | 74 | 77.6 | 27.4 | 17.5 | |
| MedQwenBackbone=BiomedCLIP ViT-B/16, # Params (%)=2.24, Evaluation Protocol=Zero-shot2026.04 | 77.23 | — | — | — | — | — | — | |
| Full FTBackbone=BiomedCLIP ViT-B/16, # Params (%)=100, Evaluation Protocol=Zero-shot2026.04 | 76.12 | — | — | — | — | — | — | |
| LoRA (rank32)Backbone=BiomedCLIP ViT-B/16, # Params (%)=5.98, Rank=32, Evaluation Protocol=Zero-shot2026.04 | 75.87 | — | — | — | — | — | — | |
| AdaMoLEBackbone=BiomedCLIP ViT-B/16, # Params (%)=2.33, Evaluation Protocol=Zero-shot2026.04 | 75.38 | — | — | — | — | — | — | |
| HydraLoRABackbone=BiomedCLIP ViT-B/16, # Params (%)=1.58, Evaluation Protocol=Zero-shot2026.04 | 75.08 | — | — | — | — | — | — | |
| LoRA (rank16)Backbone=BiomedCLIP ViT-B/16, # Params (%)=2.99, Rank=16, Evaluation Protocol=Zero-shot2026.04 | 74.98 | — | — | — | — | — | — | |
| MoELoRABackbone=BiomedCLIP ViT-B/16, # Params (%)=2.24, Evaluation Protocol=Zero-shot2026.04 | 74.89 | — | — | — | — | — | — | |
| EMT-NetLearning Paradigm=Multi-task learning2023.08 | 74.1 | 79.4 | 69.1 | 75 | 74.3 | 20.6 | 30.9 | |
| VGGA-ViTLearning Paradigm=Multi-task learning, Type=Hybrid CNN-Transformer2023.08 | 73.6 | 61.8 | 79.8 | 61 | 70.8 | 38.2 | 20.2 | |
| DenseNet121Learning Paradigm=Single-task learning2023.08 | 73 | 74 | 71 | 73 | 72.5 | 26 | 29 | |
| LoRABackbone=BiomedCLIP ViT-B/16, # Params (%)=1.49, Rank=8, Evaluation Protocol=Zero-shot2026.04 | 72.13 | — | — | — | — | — | — | |
| ViTLearning Paradigm=Single-task learning2023.08 | 72.1 | 74.1 | 69.3 | 73 | 71.7 | 25.9 | 30.7 | |
| MiLoRABackbone=BiomedCLIP ViT-B/16, # Params (%)=1.49, Rank=8, Evaluation Protocol=Zero-shot2026.04 | 71.92 | — | — | — | — | — | — | |
| PiSSABackbone=BiomedCLIP ViT-B/16, # Params (%)=1.49, Rank=8, Evaluation Protocol=Zero-shot2026.04 | 71.48 | — | — | — | — | — | — | |
| MobileNetLearning Paradigm=Single-task learning2023.08 | 71 | 82 | 61 | 74 | 71.5 | 18 | 39 | |
| SHA-MTLLearning Paradigm=Multi-task learning2023.08 | 69.6 | 48.1 | 90.8 | 58 | 69.5 | 51.9 | 9.2 |