ICD-10 Coding on MIMIC-IV top-50 codes (test)
71.2Macro F1 ScoreSHIFAMIND
Evaluation results
| Method | Links | ||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| SHIFAMIND2026.05 | 71.2 | 73.5 | 94.2 | 95.9 | 64.7 | 51.7 | 32.7 | 71.1 | 84.5 | 94.7 | — | — | — | — | |
| LAAT2026.05 | 71.1 | 74.6 | 94.7 | 96.1 | 65.4 | 52.1 | 32.9 | 71.8 | 85 | 95 | — | — | — | — | |
| SFTBackbone=Qwen3-4B, Paradigm=LLM (Post-training)2026.06 | 70.3 | 73 | — | — | — | — | — | — | — | — | 74.3 | 71.7 | 71.8 | 69.6 | |
| PHIBackbone=Qwen3-4B, Paradigm=LLM (Post-training)2026.06 | 70.1 | 73.4 | — | — | — | — | — | — | — | — | 73.7 | 73.2 | 70.8 | 71.1 | |
| GRPOBackbone=Qwen3-4B, Paradigm=LLM (Post-training)2026.06 | 69.8 | 73.6 | — | — | — | — | — | — | — | — | 71.5 | 75.9 | 68.8 | 73.1 | |
| PHIBackbone=Qwen2.5-1.5B, Paradigm=LLM (Post-training)2026.06 | 69.3 | 72.3 | — | — | — | — | — | — | — | — | 73.8 | 70.8 | 71.1 | 68.6 | |
| SFTBackbone=Qwen2.5-1.5B, Paradigm=LLM (Post-training)2026.06 | 69.2 | 71.9 | — | — | — | — | — | — | — | — | 74.3 | 69.6 | 71.7 | 67.6 | |
| GRPOBackbone=Qwen2.5-1.5B, Paradigm=LLM (Post-training)2026.06 | 68.8 | 72.6 | — | — | — | — | — | — | — | — | 71.1 | 74.1 | 68.5 | 71 | |
| KEPT2026.05 | 68.7 | 72.7 | 94.1 | 95.7 | 64.3 | 50.8 | 32.3 | 70.6 | 83.3 | 93.9 | — | — | — | — | |
| PLM-ICDBackbone=RoBERTa, Paradigm=PLM Baselines2026.06 | 68.3 | 73.5 | — | — | — | — | — | — | — | — | 73.3 | 73.6 | 68.9 | 69.8 | |
| CAML2026.05 | 67.4 | 71.6 | 93.9 | 95.4 | 63.3 | 50.8 | 32.5 | 69.8 | 83.3 | 94.2 | — | — | — | — | |
| LAATBackbone=Bi-LSTM, Paradigm=PLM Baselines2026.06 | 65.9 | 71.8 | — | — | — | — | — | — | — | — | 71.9 | 71.6 | 66.9 | 67.2 | |
| GRUBackbone=Bi-GRU, Paradigm=PLM Baselines2026.06 | 65.4 | 71.1 | — | — | — | — | — | — | — | — | 71.9 | 70.3 | 66.9 | 65.5 | |
| PLM-ICD2026.05 | 65 | 69.9 | 92.9 | 94.7 | 62.3 | 49.5 | 31.7 | 68.6 | 81.4 | 92.5 | — | — | — | — | |
| GKI-ICD2026.05 | 64.9 | 70 | 93 | 94.9 | 62.6 | 49.6 | 31.8 | 68.9 | 81.6 | 92.8 | — | — | — | — | |
| MultiResCNNBackbone=ResNet, Paradigm=PLM Baselines2026.06 | 64.8 | 71.1 | — | — | — | — | — | — | — | — | 71.7 | 70.5 | 67.2 | 65.1 | |
| CNNBackbone=CNN, Paradigm=PLM Baselines2026.06 | 64.4 | 70.7 | — | — | — | — | — | — | — | — | 71.6 | 69.8 | 66 | 65.9 | |
| CAMLBackbone=CNN, Paradigm=PLM Baselines2026.06 | 63.4 | 69.4 | — | — | — | — | — | — | — | — | 70.3 | 68.6 | 64.3 | 64.6 | |
| Vanilla CBM2026.05 | 16.4 | 37.2 | 69.9 | 77.9 | 37.4 | 31.7 | 23.5 | 39.6 | 52.8 | 70.8 | — | — | — | — | |
| Few-shot + CoTBackbone=Qwen3-4B, Paradigm=LLM (Prompting)2026.06 | 8.5 | 11.7 | — | — | — | — | — | — | — | — | 7.3 | 28.6 | 6.9 | 27.4 | |
| Few-shotBackbone=Qwen3-4B, Paradigm=LLM (Prompting)2026.06 | 6.2 | 7.9 | — | — | — | — | — | — | — | — | 5.2 | 17 | 5.9 | 27.3 | |
| Zero-shot + CoTBackbone=Qwen3-4B, Paradigm=LLM (Prompting)2026.06 | 3.2 | 3.5 | — | — | — | — | — | — | — | — | 1.8 | 45.3 | 1.9 | 29.8 | |
| Few-shotBackbone=Qwen2.5-1.5B, Paradigm=LLM (Prompting)2026.06 | 1.1 | 2 | — | — | — | — | — | — | — | — | 1.1 | 13.5 | 0.8 | 6 | |
| Zero-shotBackbone=Qwen3-4B, Paradigm=LLM (Prompting)2026.06 | 0.8 | 0.9 | — | — | — | — | — | — | — | — | 0.4 | 34.3 | 0.4 | 24.4 | |
| Few-shot + CoTBackbone=Qwen2.5-1.5B, Paradigm=LLM (Prompting)2026.06 | 0.6 | 1 | — | — | — | — | — | — | — | — | 0.5 | 15.2 | 0.3 | 4.5 | |
| Zero-shot + CoTBackbone=Qwen2.5-1.5B, Paradigm=LLM (Prompting)2026.06 | 0.3 | 0.4 | — | — | — | — | — | — | — | — | 0.2 | 13.3 | 0.2 | 18 | |
| Zero-shotBackbone=Qwen2.5-1.5B, Paradigm=LLM (Prompting)2026.06 | 0.2 | 0.3 | — | — | — | — | — | — | — | — | 0.1 | 15 | 0.1 | 17 |