Table 1.
Errors (mmHg) on SBP and DBP prediction for different setups with MIMIC database
| Neural network (training dataset) | SBP | DBP | SBP | DBP |
|---|---|---|---|---|
| MAE | RMSE | |||
| Direct SBP/DBP prediction | ||||
| ResNet (PPG) | 9.556 | 4.217 | 13.572 | 6.012 |
| ResNet (PPG + ECG) | 4.667 | 2.445 | 6.227 | 3.042 |
| ResNet + LSTM (PPG) | 7.122 | 3.534 | 11.214 | 5.029 |
| ResNet + LSTM (PPG + ECG) | 4.118 | 2.228 | 5.682 | 2.986 |
| Entire BP prediction | ||||
| Fully connected (PPG) | 36.559 | 10.602 | 45.013 | 13.417 |
| Fully connected (PPG + ECG) | 29.753 | 12.759 | 39.330 | 15.198 |
| LSTM (PPG) | 12.118 | 5.018 | 17.875 | 6.890 |
| LSTM (PPG + ECG) | 7.603 | 3.688 | 11.846 | 5.320 |
| WaveNet (PPG) | 18.539 | 8.154 | 26.638 | 11.441 |
| WaveNet(PPG + ECG) | 14.501 | 7.224 | 22.922 | 10.477 |
| WaveNet + LSTM (PPG) | 14.353 | 6.311 | 21.323 | 9.150 |
| WaveNet + LSTM (PPG + ECG) | 8.812 | 3.471 | 12.967 | 4.864 |
| ResNet + LSTM (PPG) | 8.660 | 3.843 | 13.439 | 5.718 |
| ResNet + LSTM (PPG + ECG) | 4.507 | 2.209 | 6.414 | 3.101 |