TABLE 2.
The prediction performance comparison for two input groups with 95% CI.
| R 2 | MSE | MAE | MAPE | EV score | |
| Five signal inputs + ResNet + wavelet scatter transform features | 97.17% (95.36–99.01) | 2.52 (1.42–2.63) | 1.14 (1.06–1.19) | 0.043 (0.029–0.048) | 97.11% (96.29–98.53) |
| Five signal inputs + ResNet + raw signal feature | 91.62% (89.78–93) | 12.6 (8.64–18.99) | 5.31 (3.86–9.45) | 0.116 (0.62–0.186) | 90.86% (80.35–98.85) |
| Three signal inputs + ResNet + wavelet scatter transform features | 90.78% (89.16–94.35) | 11.55 (10.22–13.5) | 2.42 (2.06–2.85) | 0.91 (0.76–1.13) | 90.87% (85.32–93.31) |
| Three signal inputs + ResNet + raw signal features | 85.02% (82.18–91) | 14.23 (10.34–16.9) | 9.31 (6.6–11.25) | 1.33 (0.72–1.65) | 83.22% (80.35–85.59) |
Comparison between two input groups shows that ABP and CVP signals improved the prediction result 6.39% regarding R2 scores. Comparison between wavelet scatter transform features and raw signals shows that the wavelet method yields better performance scores in all aspects. Five signal inputs include ABP, CVP, respiration, PPG, and ECG. Three signal inputs encompass respiration, PPG, and ECG. MSE, mean of square error; MAE, mean of absolute error; MAPE, mean of absolute percentage error; EV score, explained variance score.