Analysis on pulse features of coronary heart disease patients with or without a history of ischemic stroke

Abstract

Objective

To evaluate the capability of wrist pulse analysis in distinguishing three physiological and pathological conditions: healthy individuals, coronary heart disease (CHD) patients without a history of ischemic stroke, and CHD patients with a history of ischemic stroke.

Methods

Study participants were recruited from Shuguang East Hospital, Yueyang Hospital of Integrated Traditional Chinese and Western Medicine, and Shanghai Municipal Hospital of Traditional Chinese Medicine, affiliated with Shanghai University of Traditional Chinese Medicine, from April 15 to September 15, 2021. They were categorized into three groups: healthy controls (Group 1), CHD patients without a history of ischemic stroke (Group 2), and CHD patients with a history of ischemic stroke (Group 3). The wrist pulse signals of the study participants were non-invasively collected using a pulse diagnosis instrument. The linear time-domain features and nonlinear time-series multiscale entropy (MSE) features of the pulse signals were extracted using time-domain analysis and the MSE methods, which were subsequently compared between groups. Based on these extracted features, a recognition model was developed using a random forest (RF) algorithm. The classification performance of the models was evaluated using metrics, including accuracy, precision, recall, and F1-score derived from confusion matrix as well as the area under the receiver operating characteristics (ROC) curve (AUC).

Results

A total of 189 participants were enrolled, with 63 in Group 1, 61 in Group 2, and 65 in Group 3. Compared with Group 1, Group 2 showed significant increases in pulse features H2/H1, H3/H1, W1, W2, and W2/T, and decreased in MSE₁ – MSE₇ (P < 0.05), while Group 3 showed significant increases in pulse features T5/T4, T, H1/T1, W1, W2, AS, and Ad, and decreased in MSE₁ – MSE₂₀ (P < 0.05). Compared with Group 2, Group 3 demonstrated notable increases in H1/T1 and As (P < 0.05). The RF model achieved precision of 80.00%, 61.54%, and 61.54%, recall of 74.29%, 60.00%, and 68.97%, F1-scores of 70.04%, 60.76%, and 65.04%, and AUC values of 0.92, 0.74, and 0.81 for Groups 1, 2, and 3, respectively. The overall accuracy was 67.69%, with micro-average AUC of 0.83 and macro-average AUC of 0.82.

Conclusion

Differences in pulse features reflect variations in arterial compliance, peripheral resistance, cardiac afterload, and pulse signal complexity among healthy individuals, CHD patients without a history of ischemic stroke, and those with such a history. The developed pulse-based recognition model holds the potential in distinguishing between these three groups, offering a novel diagnostic reference for clinical practice.