Detection of tuberculosis bacilli from Ziehl Neelson stained sputum smear images

Abstract

Tuberculosis is a contagious illness caused by the Mycobacterium Tuberculosis, also known as Koch bacillus. Many developing countries follow the manual method for diagnosing TB, which causes false alarms in the detection of TB positive or negative. In order to reduce the intervention of human we have developed an effective algorithm for the detection of tuberculosis bacilli as an automated system. This paper proposes a color segmentation and classification approach for automatic detection of Mycobacterium Tuberculosis, which causes TB from the image of Ziehl-Nielsen stained sputum smear obtained from a bright microscope. Segment the bacilli called candidate bacilli using its characteristics from the image using Particle Swarm Optimization technique, depending on pixel intensities, each bacillus is segmented by extracting blue component of pixel values. The candidate bacilli are then grouped together using connected component analysis after using morphological operations. Detection of Tuberculosis bacilli from sputum smear by random forest technique is a prominent method used in diagnosing the tuberculosis by classifying the subject samples. The combination of particle swarm optimization and random forest classification provides better results and correct diagnosis in term of infection level. The experimental result shows that our approach is significantly better compared to the existing approaches.