Table 2 Breath-by-breath classification performance for each device in the training and test groups. Each breath was labeled as being either apnea/hypopnea or normal breathing based on polysomnographic judgment and was classified as positive or negative based on respiratory events (REs) detected as a reduction in respiratory amplitude or frequency, derived from acceleration and gyroscope signals.
From: Detection of sleep apnea using smartphone-embedded inertial measurement unit
Device | Group | Number of breathes | Classification performance | |||||||
|---|---|---|---|---|---|---|---|---|---|---|
TP | FP | FN | TN | Sensitivity | Specificity | PPV | NPV | F1 score | ||
Amue link | Training | 2915 | 564 | 1242 | 183,343 | 70.1% | 99.7% | 83.8% | 99.3% | 0.763 |
Test | 1809 | 459 | 524 | 98,715 | 77.5% | 99.5% | 79.8% | 99.5% | 0.786 | |
Xperia | Training | 3970 | 504 | 1769 | 205,783 | 69.2% | 99.8% | 88.7% | 99.1% | 0.777 |
Test | 2359 | 398 | 634 | 109,034 | 78.8% | 99.6% | 85.6% | 99.4% | 0.821 | |
iPhone | Training | 1764 | 565 | 924 | 146,349 | 65.6% | 99.6% | 75.7% | 99.4% | 0.703 |
Test | 1281 | 223 | 433 | 75,765 | 74.7% | 99.7% | 85.2% | 99.4% | 0.796 | |
