Auxiliary Assessment of Cardiovascular Health Using High-Dimensional Characteristics of Camera-Based iPPG Monitoring

Noninvasive, convenient, and reliable cardiovascular health assessment using pervasive healthcare applications can help in preventing cardiovascular diseases (CVDs). This study proposes a novel cardiovascular health state detection procedure based on the multiple imaging photoplethysmography (iPPG) characteristics collected by camera-based iPPG monitoring. In this study, we collect a dataset from 226 subjects, derive 20 iPPG characteristic variables, and construct a two-way classification to predict whether subjects are CVDs relevant patients with a history of coronary heart disease or hypertension. The cross-validation accuracy of full-feature random forest (RF) model is 0.860 in training set and weighted F1-score and area under curve (AUC) 0.900 and 0.7831 in test set, respectively. The contribution ratio of heart rate variability (HRV) and iPPG average waveform characteristics is 3:2. Sample equalization can improve the model performance and increase AUC to 0.975, with the contribution ratio of HRV and iPPG average waveform characteristics remaining consistent as 7:5. The proposed scheme emphasizes the advantages of using noncontact and high-dimensional features of camera-based iPPG healthcare approach for auxiliary assessment. The effective use of iPPG full-features and RF modeling has a strong application significance for the identification of the cardiovascular health of status and is hoped to further contribute to the prevention and intervention of CVDs.