Fusing Pulse and ECG Data for Coronary Heart Disease and Complications Identification

Abstract

Abstract Objectives: This study aimed to explore the potential of combining wrist pulse with limb lead electrocardiogram (ECG) data to develop an identification model for coronary heart disease (CHD) and its associated comorbidities. Methods: We utilized a pulse-detecting device equipped with a pressure sensor and an ECG sensor to simultaneously collect wrist pulse and limb lead ECG signals from patients with coronary heart disease (CHD) and various comorbidities, including hypertension and diabetes. Time-domain analysis was applied to extract features such as time-domain parameters and pulse rate variability from the wrist pulse signals, as well as time-domain parameters and heart rate variability from the limb lead ECG signals. We implemented the random forest (RF) machine learning algorithm, to establish disease identification models based on these features, and evaluated their performance. Results: The results indicated that the disease identification model which incorporated features from both pulse and ECG signals, exhibited improvements of 1.99%, 3.13%, 3.78% and 3.32% in terms of accuracy, average precision, average recall and F1 value, respectively, when compared to the model based solely on pulse features. Furthermore, when compared to the ECG-based model, the results were improved by 3.99%, 3.13%, 3.78% and 3.32% respectively. Conclusions: The fusion of information from multiple sources enhances the reliability of decision-making of the system. This approach presents a novel method for managing cardiovascular diseases and offers insights into the application and promotion of wearable pulse-detecting products.