Information Theoretic Metrics for Automatic Quality Assessment of Processed PPG Signals

Most wireless health monitoring devices are designed to continuously acquire, process, store and transmit photoplethysmogram (PPG) signal(s) that may undergo different kinds of waveform distortions at the processing stages of denoising, compression and transmission to be performed at the wireless biosensing node. Therefore, in this paper, we present and evaluate the performance of the information-theoretic metrics, including the mutual information (MI) and Kullback-Leibler divergence (KLD) by using various kinds of compression and reconstructed signals obtained based on the compressed sensing, discrete cosine transform, predictive coding, and discrete wavelet transform. Evaluation results on the 3-point and 2-point scale ratings of subjective quality evaluation test showed that the MI metric provides promising results in terms of higher Pearson correlation coefficient with subjective quality scores and prediction accuracy in correctly predicting the quality group as compared to that of the KLD metric.