ECG Classification on PTB-XL (ACC, F1, AUROC)
96.78AUROCMSAIC-Net
Evaluation Results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| MSAIC-Net2026.06 | 96.78 | — | 85.76 | — | — | — | — | 93.43 | |
| CNN-LSTM (Alamatsaz et al.)2026.06 | 96.77 | — | 85.23 | — | — | — | — | 93.26 | |
| Resnet (Khan et al.)2026.06 | 96.35 | — | 83.95 | — | — | — | — | 92.71 | |
| ECG-NATprotocol=linear evaluation2026.05 | 96 | 80 | — | — | — | — | 79.9 | — | |
| Transformer (Ikram et al.)2026.06 | 95.56 | — | 79.56 | — | — | — | — | 90.62 | |
| CoRe-ECG2026.04 | 94.2 | 81.4 | 69.5 | — | — | — | — | — | |
| LSTM (Zhang et al.)2026.06 | 94.06 | — | 76.18 | — | — | — | — | 87.41 | |
| ST-MEM2026.04 | 92.9 | 80.9 | 63.4 | — | — | — | — | — | |
| MERITEvaluation Protocol=Linear Probing2026.05 | 92.7 | 95.8 | 42.63 | — | — | — | — | — | |
| D-BETAEvaluation Protocol=Linear Probing2026.05 | 91.52 | 95.53 | 39.34 | — | — | — | — | — | |
| MLAE2026.04 | 91.5 | 80.2 | 62.5 | — | — | — | — | — | |
| MoCo V32026.04 | 91.3 | 79.9 | 64.4 | — | — | — | — | — | |
| STMEMEvaluation Protocol=Linear Probing2026.05 | 91.02 | 95.22 | 39.3 | — | — | — | — | — | |
| MTAE2026.04 | 91 | 78.9 | 61.3 | — | — | — | — | — | |
| ECG Lens ModelArchitecture=4 convolutional layers and 3 fully-connected layers, Dropout rate=0.4, Filter range=128 to 10242026.04 | 90 | 80 | 78 | 80 | 76 | — | — | — | |
| Supervised2026.04 | 89.6 | 71.4 | 60.2 | — | — | — | — | — | |
| MERLEvaluation Protocol=Linear Probing2026.05 | 88.97 | 93.88 | 35.8 | — | — | — | — | — | |
| BMIRC2026.04 | 88.3 | 75.2 | 62.1 | — | — | — | — | — | |
| ECG-ChatEvaluation Protocol=Linear Probing2026.05 | 88.01 | 94.42 | 36.4 | — | — | — | — | — | |
| CMSC2026.04 | 87.7 | 72.4 | 51 | — | — | — | — | — | |
| LSTM ModelArchitecture=Multiple LSTM layers followed by dense layers2026.04 | 87 | 73 | 72 | 78 | 71 | — | — | — | |
| CNN ModelInput shape=(n,1000,12), Architecture=1D convolutional layers, max-pooling, flattening, and dense layers2026.04 | 85 | 71 | 69 | 73 | 66 | — | — | — | |
| ECG-FMEvaluation Protocol=Linear Probing2026.05 | 83.87 | 92.26 | 31.31 | — | — | — | — | — | |
| ST-MEMprotocol=linear evaluation2026.05 | 83.8 | 72.6 | — | — | — | — | 50.8 | — | |
| QoQEvaluation Protocol=Linear Probing2026.05 | 83.76 | 92.4 | 28.55 | — | — | — | — | — | |
| ASTCLprotocol=linear evaluation2026.05 | 82.3 | — | — | — | — | — | — | — | |
| MTAEprotocol=linear evaluation2026.05 | 80.7 | 68.3 | — | — | — | — | 43.7 | — | |
| CMSCprotocol=linear evaluation2026.05 | 79.7 | 68.1 | — | — | — | — | 44.1 | — | |
| MLAEprotocol=linear evaluation2026.05 | 77.9 | 64.9 | — | — | — | — | 38.2 | — | |
| ESIEvaluation Protocol=Linear Probing2026.05 | 74.81 | 84.19 | 18.22 | — | — | — | — | — | |
| MOCOV3protocol=linear evaluation2026.05 | 73.9 | 55.2 | — | — | — | — | 14.2 | — | |
| ProtoSSL HEEDBEvaluation Protocol=Tuned, Label Overlap with HEEDB=Yes2026.05 | — | — | — | — | — | 91.1 | — | — | |
| ProtoSSL HEEDBEvaluation Protocol=Probed, Label Overlap with HEEDB=Yes2026.05 | — | — | — | — | — | 88.1 | — | — | |
| SupProto DirectEvaluation Protocol=Trained from scratch, Label Overlap with HEEDB=Yes2026.05 | — | — | — | — | — | 90 | — | — | |
| SupProto HEEDBEvaluation Protocol=Tuned, Label Overlap with HEEDB=Yes2026.05 | — | — | — | — | — | 89.9 | — | — | |
| SupProto HEEDBEvaluation Protocol=Probed, Label Overlap with HEEDB=Yes2026.05 | — | — | — | — | — | 83.6 | — | — |