Automated signal quality assessment for wearable SCG and NIRS brain monitoring

Abstract

In remote health monitoring, signal quality assessment (SQA) is crucial for precise data analysis and trustworthy interpretation of physiological events. This thesis aims to develop reliable methods for SQA to evaluate the acceptability of signals from seismocardiography (SCG) measured from the chest and near-infrared spectroscopy (NIRS) measured from the brain. Several signal processing metrics were proposed and evaluated. For NIRS signals, these metrics included standard deviation (SD), relative standard deviation (rSD), ratio of power to standard deviation, and ratio of power to relative standard deviation. For SCG signals, the metrics included standard deviation, standard deviation of peak amplitudes, and the proportion of peak amplitude to noise amplitude.

The techniques demonstrated notable gains in accuracy and statistical significance for NIRS signals, proving reliability across different datasets, subjects, and protocols. The higher the Curve (AUC), the better the performance of the method, with lower p-values indicating higher statistical significance. For NIRS signals, the best performing method was the ratio of power to rSD, achieving a p-value of 2.84e-129 and an AUC of 0.99 for 5-second segments, and a p-value of 1.16e-65 and an AUC of 0.99 for 10-second segments. However, SCG signal processing presented challenges, with the SD metric yielding the best results, evidenced by an AUC of 0.98 and a p-value of 7.99e-33, indicating high performance and statistical significance.

This research underscores the importance of advanced signal processing methods in enhancing the reliability of remote health monitoring, ultimately improving patient outcomes.