Skin Lesion Classification on HAM10000 (Classification and Calibration)
88.34AccuracyTwo-Stage ResNet50
Evaluation results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| Two-Stage ResNet50Backbone=ResNet502026.06 | 88.34 | — | — | — | — | 95.59 | 89.13 | 87.56 | |
| LoRA-EnsembleEnsemble members=162024.05 | 88 | 78.3 | 3.7 | 0.342 | 17.5 | — | — | — | |
| Explicit EnsembleEnsemble members=162024.05 | 85.8 | 74.6 | 10.5 | 0.536 | 21.8 | — | — | — | |
| Typical ResNet50 baselineBackbone=ResNet502026.06 | 85 | — | — | — | — | 88 | 83 | 80 | |
| Snapshot EnsembleEnsemble members=162024.05 | 84.9 | 73.7 | 5.8 | 0.431 | 21.7 | — | — | — | |
| Single Network2024.05 | 84.1 | 71.4 | 13.9 | 1.138 | 29.1 | — | — | — | |
| MC Dropout2024.05 | 83.7 | 71 | 9.9 | 0.631 | 27 | — | — | — | |
| Single Net w/ LoRA2024.05 | 83.2 | 70.7 | 8.5 | 0.569 | 25.6 | — | — | — | |
| Batch-EnsembleEnsemble members=162024.05 | 76.8 | 58.4 | 6.4 | 0.651 | 33.2 | — | — | — | |
| LRFormer2024.05 | 74.3 | 52.1 | 5.3 | 0.737 | 35.4 | — | — | — | |
| L22024.05 | 74.1 | 50.7 | 6.5 | 0.766 | 36 | — | — | — | |
| DermAgentType=Medical Agent2026.05 | 61.83 | — | — | — | — | — | — | — | |
| MedAgent-ProType=Medical Agent2026.05 | 57.63 | — | — | — | — | — | — | — | |
| Hulu-Med-7BType=Medical MLLM2026.05 | 52.65 | — | — | — | — | — | — | — | |
| HuatuoGPTType=Medical MLLM2026.05 | 51.4 | — | — | — | — | — | — | — | |
| Qwen3-VL-8BType=General MLLM2026.05 | 51.09 | — | — | — | — | — | — | — | |
| DermoGPT-RLType=Dermatology MLLM2026.05 | 50 | — | — | — | — | — | — | — | |
| GPT-4oType=General MLLM2026.05 | 48.91 | — | — | — | — | — | — | — | |
| SkinVL-PubMMType=Dermatology MLLM2026.05 | 45.17 | — | — | — | — | — | — | — | |
| LLaVA-Med-v1.5Type=Medical MLLM2026.05 | 44.24 | — | — | — | — | — | — | — | |
| GPT-5.2Type=General MLLM2026.05 | 35.98 | — | — | — | — | — | — | — | |
| MDAgentsType=Medical Agent2026.05 | 16.82 | — | — | — | — | — | — | — | |
| Xu et al.Features=GNN-based graph-level anomaly detection (GLAD), Best classifiers=One-class graph transformation learning (OCGTL)2026.04 | — | — | — | — | — | 91.4 | — | — |