Table 2.
Performance metrics of our ApneaFormer model on the OSASUD dataset
| Dataset | Model | Signal | Sens. | Spec. | Acc. | Prec. | F1 score |
|---|---|---|---|---|---|---|---|
| Stroke dataset (OSASUD) | ResNet43 | ECG | 16.8 | 82.4 | 71.6 | 15.6 | 0.162 |
| AIOSA43 | ECG | 62.8 | 79.6 | 76.9 | 37.7 | 0.471 | |
| Hybrid Transformer | ECG | 51.90 | 67.21 | 64.88 | 22.09 | 0.310 | |
| ApneaFormer (ours) | ECG | 68.05 | 82.43 | 80.24 | 40.96 | 0.511 | |
| ApneaFormer (ours) | ECG, RR interval | 63.41 | 83.01 | 80.03 | 40.06 | 0.491 | |
| ApneaFormer (ours) | ECG, Kurtosis | 66.53 | 85.71 | 82.80 | 45.50 | 0.540 | |
| ApneaFormer (ours) | ECG, Lyapunov Exp. | 66.96 | 82.90 | 80.47 | 41.26 | 0.511 | |
| ApneaFormer (ours) | ECG, Kurtosis, Lyapunov Exp. | 68.03 | 83.89 | 81.47 | 43.11 | 0.528 |
We evaluate the performance of our model using different feature combinations and compare it against other baseline models.