Cloud Processing of Bed Pressure Sensor Data to Detect Sleep Apnea Events

Home-based monitoring of well-being through sensors that are either ambient or worn by the resident is a source of significant data about the resident. However, wide-scale use of such data creates a significant challenge in that the data has to be processed in a scalable way across many residents and residences. Cloud-based classification and machine learning tools, such as the IBM Watson cloud services, are now available as a potential deployment model. The assessment of sleep apnea is one example application that could benefit from residential assessment as a sleep lab assessment is costly and inconvenient. Home-based clinical information could allow physicians to triage patients for the costlier hospital-based testing. In this work, the performance of a sleep apnea algorithm is compared between the Matlab classification environment that was used for its development and IBM Watson cloud classification services. The work shows that similar but slightly different performance was achieved between the two systems with the IBM Watson tools having lower accuracy (mean 92.7% compared to 92.9%), higher precision (92.7% compared to 92.5%), lower recall (92.7% compared to 93.4%), and lower weighted F1 measure (92.7% compared to 93.0%). This slight difference demonstrates the portability of Matlab lab results into a cloud solution, thus facilitating scalability.