Machine Learning Approach Combined with Statistical Features in the Classification of Peripheral Pulse Morphology

Abstract

The aim of the attempt is to build a mechanism for objective evaluation of the autonomous nervous system (ANS) for disease diagnosis at an early stage. With the experience of data collection from various control subjects, BARC has identified eight different pulse morphologies. A Peripheral Pulse Analyser (PPA) measures peripheral blood flow. Blood flow was measured in control subjects (100) and patients (100). The morphology of a person's pulse changes throughout time. Pulse morphologies vary according to age, disease, and other parameters. More than 8500 signals from 200 humans were tested. Various pattern-matching and classification techniques are given in this research to detect the existence of specific pulse shapes in obtained PPA signals. Peaks of PPA blood flow patterns are detected, and features are extracted from the sample pattern. Various machine learning (ML) algorithms are used to identify various pulse shapes depending on the parameters of extracted features. We observed that in one PPA signal of the duration of 300 seconds, 3 to 4 defined pulse morphologies out of 8 are available. Every pulse morphology is different from the others. After training, the system was able to detect pulse shapes to assess the ANS of the subject with more than 94% to 97% accuracy. The proposed system will assist the doctor in making a decision quickly based on a few processed parameters rather than assessing several individual parameters at a crucial time. The output of the system is the assessment report of ANS. This is an attempt to replace traditional Ayurvedic pulse examination methos for disease detection.