Echo State Network-Based Estimation of Photoplethysmography Sensor-To-Skin Contact Force

Abstract

A photoplethysmographic signal, widely used in cardiovascular monitoring, is susceptible to the sensor's mounting conditions, including the contact force at the sensor-to-skin interface. We aimed to extract this concomitant parameter from a reflective photoplethysmographic signal to enable better observation of varying measurement conditions. Evaluation of a regressor based on an echo state network yields promising results when modeling the relationship between a reference force signal delivered from a force-sensitive resistor and the infrared and red photoplethysmographic signal components with an average normalized root mean square error of 0.101 (range of 0.051–0.150) for the considered test cases. The echo state network regressors using as few as 10 neurons show potential for deployment and online adaptation in resource-constrained hardware, e.g., microcontrollers.