Main features of AF prediction studies based on ECGs in sinus rhythm
| Input data | Model | N ECGs | N AF ECGs | Time to onset of AF | AUC (%) |
|---|---|---|---|---|---|
| 12 lead ECG [18] | CNN | 80,125 | 13,824 | < 31 days | 87 |
| 12 lead ECG [19] | DNN | 1,632,487 | 9,377 | < 1 year | 85 |
| 12 lead ECG [20] | RNN | 1,057 | 1,355 | < 1 day | 79 |
| 12 lead ECG [21] | CNN | 45,770 | 2,171 | < 5 years | 82 |
| 12 lead ECG [22] | CNN | 907,858 | 28,117 | > 31 days | 86 |
| 12 lead ECG [13] | ResNet | 669,782 | 22,695 | < 5 years | 78 |
| 2 lead ECG [23] | CNN | 29,884 | 3,307 | < 15 days; > 1 day | 79 |
| 1 lead ECG [24] | CNN | 169,013 | 13,553 | NA | 62 |
| 1 lead ECG [24] | CNN | 166,447 | 15,052 | NA | 62 |
| 1 lead ECG [24] | CNN | 143,503 | 10,951 | NA | 80 |
AF: atrial fibrillation; ECGs: electrocardiograms; N ECGs: number of ECGs; AUC: area under the receiver operating characteristic curve; CNN: convolutional neural network; DNN: deep neural network; RNN: recurrent neural network; ResNet: residual network; NA: not available