Stochastic Modeling for Photoplethysmography Compression.

Photoplethysmography (PPG) has been widely involved in health monitoring for clinical medicine and wearable devices. To make full use of PPG signals for diagnosis and health care, raw PPG waveforms have to be stored and transmitted in a storage and power-efficient way, which is data compression. In this study, we proposed a new approach for PPG compression using stochastic modeling. This new method models a single cardiac period of PPG waveform using two sets of Gaussian functions to fit the forward and backward waves of the PPG pulse, representing the signal with a few numbers of parameters that share high similarity inter cardiac periods. An adaptive quantization based on higher-order statistics of inter-cardiac- period parameters was then adopted to quantize continuous parameters into transmissive-friendly integers of different bits. Although further ASCII encoding was not applied in this research, comparison results on a wearable PPG dataset with 30 subjects show that the proposed approach can achieve a much higher compression ratio (up to 41 under 200 Samples/s for 18-bit data) than conventional delta modulation-based methods under clinical-acceptable recover quality, with percentage root-mean-square difference (PRD) lower than 9%. This algorithm may also have comparative results with state-of-the-art methods after introducing lossless encoding, which is hardly absent from the latter. This study indicates the high potential of using stochastic modeling in PPG compression, especially for reflective PPG collected by wearable devices where the amplitudes of signals can be significantly affected by respiration.Clinical Relevance-This research establishes a new approach of photoplethysmography compression, which contributes to remote and telehealth monitoring in wearable devices.