Parameter optimization of motion artifact canceling PPG-based heart rate sensor by means of cross validation

Abstract

Photoplethysmography (PPG) is one of the simple and non-invasive heart rate (HR) sensing methods, but when applying it to a person during exercise, the output is contaminated with motion artifact (MA). Furthermore, when the pressure to stabilize the sensor on the skin surface is lower, extremely large values referred to as "outliers" are often observed in the sensed heart rate. To cancel the MA and reject the outliers, we have proposed an MA canceling PPG-based HR sensor, and have confirmed its effectivity for persons during vigorous exercises. However, the HR sensor contains several parameters to be adjusted to obtain better performance, although the number of experiments using subjects is limited due to its complexity. In this paper, we discuss a parameter optimization method for the MA canceling PPG-based HR sensor by means of cross validation. We apply the leave-one-out cross validation (LOOCV) to experimental data changing the values of the parameters, and then determine the ones which can minimize the root mean square error (RMSE). Finally, we show that the proposed HR sensor can achieve the RMSE of less than 7.1 beats per minute (bpm) for exercises of walking, running and jumping.