Pulse waves unveiled: harnessing convolutional neural networks for accurate classification in pulse diagnosis

Abstract

Pulse diagnosis, an essential aspect of Sri Lankan traditional medicine and ancient civilizations like China, India, and Greece, involves skilled practitioners assessing pulse qualities through palpation for valuable health insights. However, the classification of pulse wave patterns presents challenges due to the subjective nature of palpation, complex wave patterns, and limited diagnostic tools. To address these difficulties, this study aims to compare finger pulse signals using a convolutional neural network (CNN) model to enhance pulse-based diagnosis and specifically target pulse wave patterns classification (PWPC) in patients with non-communicable diseases (NCDs) and healthy subjects. In this study, a binary classification was performed to distinguish between two healthy and unhealthy classes. Finger pulse signals from 50 patients with NCDs and 50 healthy control subjects were collected using a multipara patient monitor. The pulse signal images were pre-processed, including normalization, and then analyzed using CNN model, known for its effectiveness in image processing tasks. By isolating individual pulse cycles, the CNN model achieved an impressive identification accuracy of 92% in classifying pulse wave patterns. These findings highlight the efficacy of the proposed CNN model in accurately categorizing and understanding the distinctive pulse wave patterns associated with NCD patients and healthy individuals. The successful implementation of the CNN model in PWPC holds great promise for advancing pulse-based diagnosis. By enhancing diagnostic accuracy, healthcare practitioners can make more informed decisions and develop tailored treatment plans for individuals based on their specific pulse wave characteristics. Ultimately, this research contributes to the growing knowledge of pulse diagnosis and its potential to revolutionize healthcare practices.