Table 4.
Comparison to other sleep monitor devices.
| Approaches | Sleep estimation accuracy (kappa) | Sleep parameters correlation (absolute error) | Validation | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Research/product | Sensor used | Class | Wake Sleep | Wake REM NREM | TST | SL | SE | WASO | RP | NP | RL | RC | Cohort | Reference |
| This study (SSA) | non-contact microphone | 3 | 91.7% (0.68) | 86.9% (0.69) | 0.95 (11.7) | 0.95 (6.8) | 0.82 (3.5) | 0.76 (14.8) | 0.89 (2.6) | 0.79 (4.1) | 0.73 (27.6) | 0.7 (0.4) | 100 | PSG |
| DeepSleepNet19 | EEG | 5 | 97.3% (0.86) | 93.8% (0.86) | — | — | — | — | — | — | — | — | 62* | PSG |
| Zhu et al.39 | EEG | 5 | 97.7% (0.95) | 92.8% (0.88) | — | — | (7) | — | — | — | — | — | 8 | PSG |
| Zeo40 | EEG | 4 | 92.6% (0.65) | 85.4% (0.69) | 0.95 | 0.42 | 0.95 | 0.9 | 0.51 | — | 0.02 | — | 26 | PSG |
| WatchPAT41–43 | accelerometer oximeter PAT Snore | 4 | 86.3% (0.56) | 77.2% (0.57) | 0.66–0.68 | — | — | — | 0.58 | — | — | 0.34 | 31 | PSG |
| Earlysense44 | Mattress pressure sensor | 4 | 90.5% (0.68) | 77.8% (0.56) | 0.87 | — | — | — | — | — | — | — | 85 | PSG |
| PulseOn45 (SmartWatch) | PPG, accelerometer | 3 | – | #81.4% | – | – | – | – | – | – | – | – | 108 | PSG |
| Tararaidze et al.46 | RIP | 3 | 88.4% (0.67) | 80.4% (0.65) | — | — | — | — | — | — | — | — | 29 | PSG |
| Tararaidze et al47. | Bio-motion sensor (RF) | 4 | 86.3% (0.57) | 75.9% (0.55) | — | — | — | — | — | — | — | — | 32 | PSG |
| Zaffaroni et al.48 | Bio-motion sensor (RF) | 4 | 90.6% (0.51) | 79.2% (0.53) | – | 0.41 (8.0) | 0.67 (4.8) | — | — | — | — | — | 20 | PSG |
| Philip et al.49 | Bio-motion sensor (RF) | 2 | 78% (0.38) | — | (50) | — | (12) | — | — | — | — | — | 113 | PSG |
| Sleep Hunter50 | Smartphone’s microphone lumination accelerometer | 3 | † | #73% (0.44) | — | † | † | † | — | — | — | — | 45 | Zeo |
| Actigraph52 | accelerometer | 2 | 86.3% (0.36) | — | — | — | — | 0.6 | — | — | — | — | 77 | PSG |
TST – total sleep time; SL – Sleep latency; SE – Sleep efficiency; WASO – Wake time after sleep onset; RP – Rapid-eye-movement percentage; NP – non-Rapid-eye-movement percentage; RL – REM latency; RC – number of REM cycles; SSA – Sleeping sound analysis; EEG – electroencephalogram; PPG-photoplethysmography; RIP – respiratory inductance plethysmography; PSG – Polysomnography. The approaches class column represents the maximum classes seperatable using the respective approach. Underline mark “—” indicates that this performance score was not originally included in the paper and was estimated by us using the complementary data available in the paper. † indicates that the wresearches considered Wake as REM; * indicates 31-fold cross-validations over 62 subjects; # indicates two-class, REM vs. NREM classification, in this case, our SSA yeilds 93% (0.76). It is important to mention that our sleep parameters correlations were calculated using concordance-correlation while other approaches did not report and might have used Pearson correlation values instead.