Table 3 The diagnostic ability of each model for AHI thresholds of 1, 5 AND 10 E/H.
From: Deep learning approaches for assessing pediatric sleep apnea severity through SpO2 signals
Model | Fold | AHI threshold | Se (%) | Sp (%) | Acc (%) | PPV (%) | NPV (%) | LR + | LR- |
|---|---|---|---|---|---|---|---|---|---|
ResNet | Fold-1 | 1 e/h | 93.04 | 56.71 | 83.46 | 85.71 | 74.50 | 2.14 | 0.12 |
5 e/h | 76.47 | 98.38 | 92.51 | 94.54 | 91.95 | 47.41 | 0.23 | ||
10 e/h | 64.00 | 99.12 | 95.66 | 88.88 | 96.18 | 73.27 | 0.36 | ||
Fold-2 | 1 e/h | 99.45 | 37.87 | 83.26 | 81.77 | 96.15 | 1.60 | 0.01 | |
5 e/h | 78.46 | 96.77 | 92.03 | 89.47 | 92.78 | 24.32 | 0.22 | ||
10 e/h | 80.00 | 98.67 | 96.81 | 86.95 | 97.80 | 60.26 | 0.20 | ||
Fold-3 | 1 e/h | 97.91 | 51.66 | 86.90 | 86.63 | 88.57 | 2.02 | 0.04 | |
5 e/h | 76.47 | 97.60 | 90.47 | 94.20 | 89.07 | 31.92 | 0.24 | ||
10 e/h | 75.75 | 98.17 | 95.23 | 86.20 | 96.41 | 41.47 | 0.24 | ||
CNN-BiGRU-Attention | Fold-1 | 1 e/h | 97.33 | 47.76 | 84.25 | 83.87 | 86.49 | 1.86 | 0.05 |
5 e/h | 82.35 | 97.85 | 93.70 | 93.33 | 93.81 | 38.29 | 0.18 | ||
10 e/h | 72.00 | 99.13 | 96.46 | 90.00 | 97.01 | 82.44 | 0.28 | ||
Fold-2 | 1 e/h | 95.14 | 72.73 | 89.24 | 90.72 | 84.21 | 3.48 | 0.06 | |
5 e/h | 76.92 | 96.24 | 91.24 | 87.72 | 92.27 | 20.43 | 0.23 | ||
10 e/h | 76.00 | 98.67 | 96.41 | 86.36 | 97.38 | 57.25 | 0.24 | ||
Fold-3 | 1 e/h | 96.35 | 63.33 | 86.11 | 88.49 | 84.44 | 2.62 | 0.05 | |
5 e/h | 74.11 | 97.60 | 89.68 | 94.02 | 88.10 | 30.94 | 0.26 | ||
10 e/h | 75.75 | 98.63 | 95.63 | 89.28 | 96.42 | 55.30 | 0.24 |
