Arrhythmia Classification on MIT-BIH
59.3Macro F1Rich-specialization model
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Rich-specialization modelArchitecture=DeepArrhythmia, Variant=Rich-specialization2026.05 | 59.3 | 95 | |
| Signal + image + peaks + featuresCategory=Matched non-agentic fusion baseline2026.05 | 59.3 | 92.8 | |
| Threshold-induced routed modelArchitecture=DeepArrhythmia, Variant=Threshold-induced routed2026.05 | 58.8 | 95.1 | |
| Signal + image + peaks + features + morphologyCategory=Matched non-agentic fusion baseline2026.05 | 56.7 | 93.7 | |
| Simple routing baselineArchitecture=DeepArrhythmia, Variant=Simple routing2026.05 | 56.6 | 94.3 | |
| Minimal-specialization modelArchitecture=DeepArrhythmia, Variant=Minimal-specialization2026.05 | 53.4 | 91.7 | |
| SVMCategory=Machine learning model, Seeds=Mean over 32026.05 | 52.3 | 93.4 | |
| Signal + image + peaksCategory=Matched non-agentic fusion baseline2026.05 | 45.6 | 85.6 | |
| TCNCategory=Deep learning models, Seeds=Mean over 32026.05 | 44.4 | 70.4 | |
| xECGCategory=Deep learning models, Seeds=Mean over 32026.05 | 44 | 89.3 | |
| 2D CNNCategory=Deep learning models, Seeds=Mean over 32026.05 | 42.3 | 83.9 | |
| Qwen3.5-4BCategory=Open-source VLMs, Decoding Temperature=02026.05 | 41.4 | 78.5 | |
| Gemini 3.1Category=Closed-source VLMs, Decoding Temperature=02026.05 | 41 | 87.8 | |
| Gemma-3-27BCategory=Open-source VLMs, Decoding Temperature=02026.05 | 37.8 | 88 | |
| STFT CNNCategory=Deep learning models, Seeds=Mean over 32026.05 | 36.6 | 85.3 | |
| TransformerCategory=Deep learning models, Seeds=Mean over 32026.05 | 36.3 | 90.7 | |
| 1D ResNetCategory=Deep learning models, Seeds=Mean over 32026.05 | 36.2 | 76 | |
| GPT-5.4Category=Closed-source VLMs, Decoding Temperature=02026.05 | 35.2 | 83.5 | |
| PULSECategory=ECG VLMs, Decoding Temperature=02026.05 | 21.1 | 64.6 | |
| GEMCategory=ECG VLMs, Decoding Temperature=02026.05 | 20.5 | 45.9 | |
| ECG-R1Category=ECG VLMs, Decoding Temperature=02026.05 | 19.9 | 44.6 |