Multi-Level Signal Quality Assessment Approach for Multi-Channel Non-Contact Smart Mattress Systems

Electrocardiogram (ECG) is crucial for sleep monitoring, requiring a convenient signal acquisition system and precise signal quality assessment methods in clinical applications. However, traditional ECG acquisition techniques face challenges in capturing signals during sleep. Wet electrodes of traditional contact-based ECG are uncomfortable, and the complex electrode arrangement of the standard 12-lead ECG is user-unfriendly. Moreover, traditional binary annotation strategies and signal quality recognition methods lack comprehensive and accurate analyses. To address this challenge, we designed two distinct multi-channel ECG acquisition systems using capacitive-coupled and flexible dry electrodes, integrating them into two different non-contact smart mattresses. For the two-channel acquisition system, we conducted a short-term simulated sleep experiment involving four constrained sleeping postures. For the eight-channel acquisition system, we conducted an overnight real-time sleep experiment under unconstrained sleeping postures. To ensure precise signal quality assessment, we proposed tailored annotation strategies and recognition methods for each sleep monitoring experiment. Using the random forest approach with 26 signal features, we built a four-level classification model for the simulated sleep experiment, and a regression model with discrete numerical ratings for the real-time sleep experiment. The result of the classification model achieved an average accuracy of 81.0% and an average acceptable overlap accuracy of 99.1%. The result of the regression model achieved an average acceptable prediction interval coverage probability under 2-Off deviation of 90.2%. Compared with existing methods, the proposed model demonstrated superior performance. This study offers great potential for more convenient and effective sleep monitoring with refined acquisition systems and signal quality assessment methods.