Development of a novel wrist pulse system for early diagnosis of pathogenic bacterial infections using optimized feature selection with machine learning approaches

Abstract

Pelvic inflammatory disease (PID) and urinary tract infections (UTI) are two Pathogenic bacterial infections. PID affects the female reproductive system, whereas UTI affects the urine system. Females are more vulnerable to both forms of illnesses and challenging to detect simultaneously due to similar symptoms. Many Clinical procedures have previously been used to diagnose the diseases, but they are painful, costly, prone to radiation, and invasive. Therefore, this paper proposed the Ayurvedic and Traditional Chinese Medicine [TCM] based wrist pulse measurement to diagnose these bacterial infections non-invasively. An Artery tonometry-based test prototype was designed to measure the pulse signals at three doshas (Vata, Pitta, Kapha) locations on the radial artery under variable static force. The data is recorded at an intermediate force level to get maximum accuracy. Signals feature viz., time domain feature, autoregressive (AR) model features, approximation entropy, sample entropy, and multiscale entropy have been calculated. Further, the efficiency of the machine learning model is enhanced through optimization-based feature selection techniques. Three different optimization-based feature selection, viz., whale optimization (WOA), ant colony optimization (ACO), and proposed binary Greywolf optimization (BGWO), has been proposed to acquire the optimized features subsets for accurate classification of disease subjects. These subsets have been used by popular machine learning techniques to classify the PID and UTI. The results of the experiment demonstrate that as the static force changed, the wrist pulse signal first increased to an intermediate force level and then decreased. Second, the classification accuracy of the machine learning model with feature selection is more than conventional feature classification. Third, the wrist pulse analysis (WPA) shows the maximum accuracy in comparison to conventional clinical PID diagnosis, where Vata dosha has the highest classification accuracy (94.1%) when using BGWO-SVM, followed by Pitta dosha (88.2%) when using BGWO-BNN. Similarly, specificity has been used for UTI classification, where Vata dosha has the highest specificity of 100% with BGWO-SVM, followed by pitta dosha with 83.3% BGWO-BNN. Finally, three dosha classification findings demonstrate that Vata and Pitta doshas have been more significantly impacted, as they have the highest classification accuracy compared to conventional clinical diagnosis, which produces similar results consistent with ayurvedic literature. Therefore, WPA can provide a unique diagnostic method for early PID/UTI identification.