Soft computing techniques for mycobacterium tuberculosis identification from zn stained microscopic sputum images

Abstract

Tuberculosis (TB) is a terrible, transferrable disease instigated by infection with Mycobacterium tuberculosis bacillus. As it was a threatening disease, killing millions of people worldwide, its rapid and accurate diagnosis was very essential everywhere, especially in developing countries. The common diagnosis of this infection was the microscopic examination of early morning sputum smears of affected persons. The manual detection of bacterial objects degrades the accuracy of decision making as well as it takes more time for examination. Thus this research presents an automatic system for detecting and counting the bacilli objects present in the Zeihl Neelsen stained microscopic images of sputum. It eases the job of lab technician with error-prone manual detection and helps in accurate decision making. newlineThe automatic bacilli identification and tuberculosis infection level grading were performed using some of the image processing techniques such as image segmentation, feature extraction, and classification. After Zeihl Neelsen staining, the sputum smear microscopic images have pink colored bacterial elements within the blue background. Thus segmentation was performed by using color space transformation from RGB to CIE Luv color space model. Otsu threshold was applied to segment the bacterial objects and morphological openings were used to remove the unwanted noise present in the image. After segmentation, the rod-shaped bacilli that exist both in single, as well as clusters, were classified as bacilli, non-bacilli and overlapping bacilli. The classification was performed based on features extracted from segmented bacilli objects. This research proposes three different classifiers for bacilli counting and grading of Tuberculosis infection. Fuzzy and Hyco-entropy based Decision Tree classifier was proposed which utilizes decision trees created based on features selected using Hyco-entropy values.