Agricultural and Biological Research

Pulmonary tuberculosis bacilli detection in sputum smear microscopy images using image processing techniques

Diriba Abdeta, Chala Diriba and Worku Jimma*

Tuberculosis remains a global health threat, particularly in developing countries like Ethiopia, where Mycobacterium tuberculosis causes a significant impact, primarily affecting the lungs in the form of pulmonary tuberculosis disease. Sputum smear microscopy stands as the predominant diagnostic tool in such settings. This study aims to develop a K-Nearest Neighbor classifier model for the detection of pulmonary tuberculosis bacilli in microscopic sputum smear images. The study employed image processing techniques to identify pulmonary tuberculosis bacilli in digital images of stained sputum smears. K-Nearest Neighbor classifiers distinguish between two classes: Bacilli detection and non-bacilli detection. The image dataset, comprising 180 stained sputum images of pulmonary tuberculosis bacilli infections, was sourced from the Ethiopian Public Health Institute. The model's performance metrics, including accuracy, sensitivity, specificity and F-measure, demonstrate an impressive average accuracy of 92.6%. The developed model exhibits a sensitivity of 93%, specificity of 92% and an F-measure of 94.7%, highlighting its robust performance in pulmonary tuberculosis bacilli detection.