A novel method for assessing cardiac function in patients with coronary heart disease based on wrist pulse analysis

Abstract

Background: The timely assessment of B-type natriuretic peptide (BNP) marking chronic heart failure risk in patients with coronary heart disease (CHD) helps to reduce patients' mortality.

Objective: To evaluate the potential of wrist pulse signals for use in the cardiac monitoring of patients with CHD.

Methods: A total of 419 patients with CHD were assigned to Group 1 (BNP < 95 pg/mL, n = 249), 2 (95 < BNP < 221 pg/mL, n = 85), and 3 (BNP > 221 pg/mL, n = 85) according to BNP levels. Wrist pulse signals were measured noninvasively. Both the time-domain method and multiscale entropy (MSE) method were used to extract pulse features. Decision tree (DT) and random forest (RF) algorithms were employed to construct models for classifying three groups, and the models' performance metrics were compared.

Results: The pulse features of the three groups differed significantly, suggesting different pathological states of the cardiovascular system in patients with CHD. Moreover, the RF models outperformed the DT models in performance metrics. Furthermore, the optimal RF model was that based on a dataset comprising both time-domain and MSE features, achieving accuracy, average precision, average recall, and average F1-score of 90.900%, 91.048%, 90.900%, and 90.897%, respectively.

Conclusions: The wrist pulse detection technology employed in this study is useful for assessing the cardiac function of patients with CHD.

Keywords: B-type natriuretic peptide; Chronic heart failure; Coronary heart disease; Machine learning algorithm; Multiscale entropy method; Time-domain method; Wrist pulse signal.