A combined segmenting and non-segmenting approach to signal quality estimation for ambulatory photoplethysmography

Continuous cardiac monitoring of healthy and unhealthy patients can help us understand the progression of heart disease and enable early treatment. Optical pulse sensing is an excellent candidate for continuous mobile monitoring of cardiovascular health indicators, but optical pulse signals are susceptible to corruption from a number of noise sources, including motion artifact. Therefore, before higher-level health indicators can be reliably computed, corrupted data must be separated from valid data. This is an especially difficult task in the presence of artifact caused by ambulation (e.g. walking or jogging), which shares significant spectral energy with the true pulsatile signal. In this manuscript, we present a machine-learning-based system for automated estimation of signal quality of optical pulse signals that performs well in the presence of periodic artifact. We hypothesized that signal processing methods that identified individual heart beats (segmenting approaches) would be more error-prone than methods that did not (non-segmenting approaches) when applied to data contaminated by periodic artifact. We further hypothesized that a fusion of segmenting and non-segmenting approaches would outperform either approach alone. Therefore, we developed a novel non-segmenting approach to signal quality estimation that we then utilized in combination with a traditional segmenting approach. Using this system we were able to robustly detect differences in signal quality as labeled by expert human raters (Pearson’s r = 0.9263). We then validated our original hypotheses by demonstrating that our non-segmenting approach outperformed the segmenting approach in the presence of contaminated signal, and that the combined system outperformed either individually. Lastly, as an example, we demonstrated the utility of our signal quality estimation system in evaluating the trustworthiness of heart rate measurements derived from optical pulse signals.