Sleep-Wake and Body Position Classification with Deep Learning using Pressure Sensor Mat Measurements

Polysomnography, the gold standard of sleep measurement, is a time-intensive, costly, and rather invasive procedure. Using under-bed pressure sensor arrays data simultaneously recorded with standard polysomnography, this study demonstrates that deep learning can be used to classify body position and differentiate sleep from wake. All measurements were performed in people with suspected sleep disorders referred for clinical assessments at a sleep laboratory. To perform the classification tasks, we used supervised learning and temporal convolution networks. Performance was assessed with leave-one-out cross validation on 84 participants for body position classification and 70 participants for sleep-wake classification. Results demonstrate that a pressure sensor array placed under the mattress with less than 100 sensors can outperform previous sleep-wake detection methods and is competitive with previous methods for body position classification. Our pressure sensor arrays differ from the mats used in previous work as they use significantly less sensors and are located under the mattress, making them less obtrusive. This tool has great potential as a cost-efficient mean of assessing sleep while reducing patient burden and the workload of specialized staff.