Diagnosis of Pulmonary Tuberculosis by Posterior-Anterior Lung X-Ray

Abstract

Tuberculosis (TB) remains a pressing global health issue, with an estimated 10.6 million cases projected by 2021. In Pakistan, TB prevalence is notably high, comprising 61% of the WHO Eastern Mediterranean TB burden. TB, primarily caused by Mycobacterium bacteria, affects multiple organs, often presenting with subtle or asymptomatic symptoms. Despite the gravity of the disease, early detection methods are limited, typically relying on model lung segmentation techniques. This research aims to enhance TB detection using chest X-ray images through a novel, model-less segmentation approach. By extracting statistical, geometric, and Hog descriptor features from lung images, coupled with various classifiers such as Convolutional Neural Network (CNN), Support Vector Machine (SVM), and K-Nearest Neighbor (KNN), the study achieved promising results. The highest accuracy attained was 91.88% using self-extracted features and linear regression, while CNN demonstrated competitive performance with an accuracy of 89.58%. To bolster the findings, visualization techniques were employed, confirming CNN's superior ability to discern patterns from segmented lung areas, thereby contributing to higher detection accuracy. This innovative approach holds significant potential for expediting computer-assisted TB diagnosis, benefiting clinical practice and public health initiatives.