Real-Time Quality-Aware PPG Waveform Delineation and Parameter Extraction for Effective Unsupervised and IoT Health Monitoring Systems

Abstract

In this paper, we present a real-time quality-aware pulse waveform delineation and parameter extraction method for accurate and reliable measurements of pulse parameters from photoplethysmogram (PPG) signals. It consists of three major stages: the PPG signal quality assessment (PPG-SQA) using autocorrelation function (ACF) and number of threshold-crossings (NTC) features, the zero-frequency resonator (ZFR) based pulse onset and peak determination, and the pulse parameter extraction. The method is implemented on the Arduino Due with a 32-bit Atmel SAM3X8E ARM Cortex-M3 CPU, 512-kB flash memory, 96-kB SRAM, and 84-MHz clock speed. The method is evaluated on the recorded PPG signals and three standard PPG databases. The PPG-SQA algorithm achieves an average sensitivity (Se)=98.62%, specificity (Sp)=97.37%, and overall accuracy (OA)=98.09%. The algorithm achieves an average Se=99.88%, positive predictivity (Pp)=99.89, Se=99.82%, and Pp=99.95%, respectively with the delineation errors (mean ± standard deviation) of 8.45 ± 9.39 $ms$ and 0.23 ± 1.33 $ms$ for finding onsets and peaks, respectively. The statistical analysis demonstrates that the parameter measurement errors are minimum for most of the pulse cycles. Results show that our quality-aware PPG analysis scheme can achieve a false alarm rate reduction (FARR) of 97.36% which outperforms the other existing SQA algorithms. It can lead to save transmission and processing energy from 8.33% to 95.63% and 8.33% to 59.77% for a duration from 5 to 60 s, respectively. The method has great potential for low-energy IoT and unsupervised health monitoring devices.