Assessment of Photoplethysmography Signals using Customized Wavelet and Deep Learning Model

Photoplethysmography (PPG) is frequently influenced by interference, potentially resulting in inaccurate interpretation of physiological data. Hence, it is essential to conduct a quality assessment prior to extracting this information. This paper introduces a novel method for assessing PPG signal quality that combines multi-class features with multi-scale series information this approach overcomes the limitations of conventional machine learning methods, which tend to suffer from low accuracy, and deep learning methods that demand large training datasets. By extracting multi-class features, the reliance on sample size is reduced, while multi-scale series information is captured through a fusion of multi-scale wavelet transforms and deep learning techniques. The proposed method achieved the better accuracy of 98.3%, surpassing six other quality assessment methods across all metrics, including sensitivity, specificity, precision, and F1-score, based on 14,700 samples from seven experiments. This paper introduces a novel approach for quality assessment and information extraction in small PPG signal samples, showing potential for accurate monitoring and analysis of clinical and everyday PPG physiological data.