Table 6 Performance comparison with previous studies.
From: Deep learning approaches for assessing pediatric sleep apnea severity through SpO2 signals
Studies | Signals | Datasets | Train set (Dataset subjects/Train subjects) | Validation set (Dataset subjects/Validation subjects) | Test set (Dataset subjects/Test subjects) | Model | Four-class accuracy (%) | Four-class kappa |
|---|---|---|---|---|---|---|---|---|
Hornero et al. (2017)15 | SpO2 | UofC + 12 other pediatric sleep laboratories | UofC (4191/589) | - | UofC + 12 other pediatric sleep laboratories (4191/3602) | MLP | 54.7 | - |
Jiménez-García et al. (2020)17 | SpO2 + AF | UofC | UofC (974/584) | - | UofC (974/390) | AdaBoost | 57.95 | 0.39 |
Vaquerizo-Villar et al. (2020)27 | SpO2 | CHAT | CHAT Baseline and Follow up (746/400) | CHAT Baseline and Follow up (746/100) | CHAT Baseline and Follow up (746/246) | CNN | 67.15 | 0.31 |
Vaquerizo-Villar et al. (2021)28 | SpO2 | CHAT + BUH + UofC | CHAT Baseline and Follow up (859/859) | CHAT Nonrandomized (779/467) + BUH (578/347) + UofC (974/382) | CHAT Nonrandomized (779/312) | CNN | 72.8 | 0.51 |
BUH (578/231) | 60.2 | 0.42 | ||||||
UofC (974/392) | 61 | 0.42 | ||||||
Jiménez-García et al. (2022)29 | SpO2 + AF | CHAT + UofC | CHAT (1638/1006) | CHAT (1638/326) + UofC (974/584) | CHAT (1638/306) | CNN | 72.55 | 0.60 |
UofC (974/390) | 61.79 | 0.44 | ||||||
García-Vicente et al. (2023)30 | ECG | CHAT | CHAT (1638/988) | CHAT (1638/223) | CHAT (1638/299) | CNN | 57.86 | 0.37 |
García-Vicente et al. (2023)31 | SpO2 + AF | CHAT + UofC | CHAT (1638/1006) | CHAT (1638/326) + UofC (974/584) | CHAT (1638/306) | CNN-BiGRU | 74.51 | 0.62 |
UofC (974/390) | 62.31 | 0.44 | ||||||
This study | SpO2 | CHAT | CHAT Baseline and Follow up (844/400) | CHAT Baseline and Follow up (844/84) | CHAT Baseline and Follow up (844/253) | ResNet | 72.9 | 0.57 |
CNN-BiGRU-Attention | 75.95 | 0.63 |
