Table 1.
System performance – agreement between SSA and PSG.
| Classifications | Accuracy | Cohen’s kappa | |||
|---|---|---|---|---|---|
| Mean ± SD | Median (95% CI) | Mean ± SD | Median (95% CI) | ||
| Real-time | WRN (3-Class) | 82.2 ± 6.4 | 82.6 (69.0–93.3) | 0.590 ± 0.122 | 0.598 (0.323–0.798) |
| WS (2-Class) | 90.3 ± 5.3 | 90.9 (75.0–98.4) | 0.619 ± 0.147 | 0.642 (0.296–0.890) | |
| Offline | WRN (3-Class) | 86.9 ± 4.8 | 87.3 (76.3–95.0) | 0.694 ± 0.113 | 0.700 (0.377–0.869) |
| WS (2-Class) | 91.7 ± 4.3 | 92.7 (81.1–98.0) | 0.676 ± 0.145 | 0.704 (0.341–0.922) | |
System performance was evaluated on the validation dataset using the real-time and offline estimations protocols. Accuracy and Cohen’s kappa coefficient were calculated between sleep sound analysis (SSA) and polysomnography (PSG) for each subject using epoch-by-epoch (30 sec) analysis. Data was analyzed for three classes of wake (W), rapid-eye-movement (R), non-rapid-eye-movement (N), and two classes of wake (W) vs. sleep (S). Values are presented as a mean percent agreement ± standard deviation (SD) and median 95% confidence interval (95% CI).