ABSTRACT
Coronary heart disease (CHD) is a leading cause of mortality globally and poses a significant threat to public health. Coronary angiography (CAG) is a gold standard for the clinical diagnosis of CHD, but its invasiveness restricts its widespread application. In this study, we utilized a pulse diagnostic device equipped with pressure and photoelectric sensors to synchronously and non-invasively capture wrist pressure pulse waves and fingertip photoplethysmography (FPPG) of patients undergoing CAG. The extracted features were utilized in constructing random forest-based models to assessing the severity of coronary artery lesions. Notably, Model 3, incorporating both wrist pulse and FPPG features, surpassed Model 1 (solely utilizing wrist pulse features) and Model 2 (solely utilizing FPPG features). Model3 achieved an Accuracy, Precision, Recall, and F1-score of 78.79%, 78.69%, 78.79%, and 78.70%, respectively. Compared to Model1 and Model2, Model 3 exhibited improvements by 4.55%, 5.25%, 4.55%, and 5.12%, and 6.06%, 6.58%, 6.06%, and 6.54% respectively. This fusion of wrist pulse and FPPG features in Model 3 highlights the advantages of multi-source information fusion for model optimization. Additionally, this research provides invaluable insights into the novel development of diagnostic devices imbued with TCM principles and their potential in managing cardiovascular diseases.
