Objective: To validate automatic sleep stage classification using deep neural networks on sleep assessed by radar technology in the commercially available sleep assistant Somnofy® against polysomnography (PSG). Methods: Seventy-one nights of overnight sleep in healthy individuals were assessed by both PSG and Somnofy at two different institutions. The Somnofy unit was placed in two different locations per room (nightstand and wall). The sleep algorithm was validated against PSG using a 25-fold cross validation technique, and performance was compared to the inter-rater reliability between the PSG sleep scored by two independent sleep specialists. Results: Epoch-by-epoch analyses showed a sensitivity (accuracy to detect sleep) and specificity (accuracy to detect wake) for Somnofy of 0.97 and 0.72 respectively, compared to 0.99 and 0.85 for the PSG scorers. The sleep stage differentiation for Somnofy was 0.75 for N1/N2, 0.74 for N3 and 0.78 for R, whilst PSG scorers ranged between 0.83 and 0.96. The intraclass correlation coefficient revealed excellent and good reliability for total sleep time and sleep efficiency, while sleep onset and R latency had poor agreement. Somnofy underestimated total wake time by 5 min and N1/N2 by 3 min. N3 was overestimated by 4 min and R by 3 min. Results were independent of institution and sensor location. Conclusion: Somnofy showed a high accuracy staging sleep in healthy individuals and has potential to assess sleep quality and quantity in a sample of healthy, mostly young adults. More research is needed to examine performance in children, older individuals and those with sleep disorders.

Background Respiratory rate (RR) is arguably the most important vital sign to detect clinical deterioration. Change in RR can also, for example, be associated with the onset of different diseases, opioid overdoses, intense workouts, or mood. However, unlike for most other vital parameters, an easy and accurate measuring method is lacking. Objective This study aims to validate the radar-based sleep monitor, Somnofy, for measuring RRs and investigate whether events affecting RR can be detected from personalized baselines calculated from nightly averages. Methods First, RRs from Somnofy for 37 healthy adults during full nights of sleep were extensively validated against respiratory inductance plethysmography. Then, the night-to-night consistency of a proposed filtered average RR was analyzed for 6 healthy participants in a pilot study in which they used Somnofy at home for 3 months. Results Somnofy measured RR 84% of the time, with mean absolute error of 0.18 (SD 0.05) respirations per minute, and Bland-Altman 95% limits of agreement adjusted for repeated measurements ranged from –0.99 to 0.85. The accuracy and coverage were substantially higher in deep and light sleep than in rapid eye movement sleep and wake. The results were independent of age, sex, and BMI, but dependent on supine sleeping position for some radar orientations. For nightly filtered averages, the 95% limits of agreement ranged from −0.07 to −0.04 respirations per minute. In the longitudinal part of the study, the nightly average was consistent from night to night, and all substantial deviations coincided with self-reported illnesses. Conclusions RRs from Somnofy were more accurate than those from any other alternative method suitable for longitudinal measurements. Moreover, the nightly averages were consistent from night to night. Thus, several factors affecting RR should be detectable as anomalies from personalized baselines, enabling a range of applications. More studies are necessary to investigate its potential in children and older adults or in a clinical setting.

The gold standard for assessing sleep apnea, polysomnography, is resource intensive and inconvenient. Thus, several simpler alternatives have been proposed. However, validations of these alternatives have focused primarily on estimating the apnea-hypopnea index (apnea events per hour of sleep), which means information, clearly important from a physiological point of view such as apnea type, apnea duration, and temporal distribution of events, is lost. The purpose of the present study was to investigate if this information could also be provided with the combination of radar technology and pulse oximetry by classifying sleep apnea events on a second-by-second basis. Fourteen patients referred to home sleep apnea testing by their medical doctor were enrolled in the study (6 controls and 8 patients with sleep apnea; 4 mild, 2 moderate, and 2 severe) and monitored by Somnofy (radar-based sleep monitor) in parallel with respiratory polygraphy. A neural network was trained on data from Somnofy and pulse oximetry against the polygraphy scorings using leave-one-subject-out cross-validation. Cohen’s kappa for second-by-second classifications of no event/event was 0.81, or almost perfect agreement. For classifying no event/hypopnea/apnea and no event/hypopnea/obstructive apnea/central apnea/mixed apnea, Cohen’s kappa was 0.43 (moderate agreement) and 0.36 (fair agreement), respectively. The Bland-Altman 95% limits of agreement for the respiratory event index (apnea events per hour of recording) were -8.25 and 7.47, and all participants were correctly classified in terms of sleep apnea severity. Furthermore, the results showed that the combination of radar and pulse oximetry could be more accurate than the two technologies separately. Overall, the results indicate that radar technology and pulse oximetry could reliably provide information on a second-by-second basis for no event/event which could be valuable for management of sleep apnea. To be clinically useful, a larger study is necessary to validate the algorithm on a general population.