Abstract:
Tuberculosis is one of the deadly diseases worldwide developing countries including Ethiopia. It is caused by Mycobacterium tuberculosis that influenced fundamentally on human body lung in form of pulmonary tuberculosis disease. Sputum smear microscopy is the most widely used diagnostic tools in developing countries. The main aim of this study is to develop automatic PTB bacilli detection from microscopic sputum smear images using image processing techniques. In this study, an algorithm based on image processing technique is developed for identification of pulmonary tuberculosis bacilli in digital image of stained sputum smear. The techniques was used in this study, Gaussian filter to remove noise, contrast enhanced to enhance the quality of image and K-mean cluster used to separate image into region in image segment process. In addition, support vector machine and k-nearest neighbor classifiers were used to identify bacilli, which classified the computed based on combined both morphological and color features from sputum smear images in two classes which are bacilli detect and non-bacilli detect. Total sample size of image dataset of 180 from stained sputum images of PTB bacilli infected were obtained from EPHI. The accuracy performance measured shows that SVM algorithm found to be 94.4% and for KNN algorithm it was 92.6%. The results and observations show that SVM is a more suitable more than KNN classifier for the classify PTB bacilli detect from pathologists reading predefined image obtained from sputum smear images database. The accuracy, sensitivity, specificity and F-measures improved the performance of the prototype results of SVM are 94.4%, 95%, 94% and 96% respectively. The future work will look into the problem related to diagnosis of PTB drug resistant cases in the proposed algorithm by a clinical setting.
