Automation & Analysis of Chest X-Ray and Microscopy Image Detection for Tuberculosis

Abstract

TB (Tuberculosis) is a contagious disease which is caused by a bacterium named Mycobacterium Tuberculosis. Mycobacterium Tuberculosis invades the host cell when the immunity of the body gets weakened. TB is an air born infection meaning that the bacterium is transferred through atmospheric air from TB suspect to a healthy person by coughing and sneezing. This bacterium has the tendency to divide rapidly. Screening is done to confirm the presence of TB. There are different screening techniques available i.e. Chest X-ray, Microscopy, Gene Xpert and Culture etc. Medical image processing is a rapidly growing field of image processing that is used to automate different medical procedures. In this research we have designed two automated systems for the screening of TB patients. A sample of 50 sample images of microscopy slides and chest X-ray radiographs were taken. The proposed algorithms were implemented using MATLAB version 9.2. The comparative analysis of results has been done using PASW statistics version 18. The comparison between results from proposed algorithm and reference standard data results was done. The sensitivity of 98% was obtained for chest radiography algorithm and 92% was obtained for bacilli segmentation algorithm. The specificity of 96% was obtained for bacilli segmentation algorithm and 70% for chest radiography algorithm. The accuracy was calculated through Receiver operating curve. The area under the curve was found to be greater for bacilli segmentation algorithm. The proposed bacilli segmentation algorithm gave an accuracy of 94%, whereas the proposed chest radiography algorithm gave an accuracy of 92%. Chi square testing was performed on the two algorithms. The p-value was lesser than 0.05 that showed that the automated techniques are independent form the standard reference techniques. The computation time was lesser for proposed bacilli segmentation algorithm and the method was found to be more reliable for use. The accuracy of the two algorithms was found to be good being above 90, so we can use any of these algorithms for screening of TB patients. These will make the screening process robust and more reliable. Moreover, the proposed systems are expected to reduce laborious fatigue and human errors. The proposed algorithms will assist physicians, doctors and microbiologists in screening of TB patients. Further work could be done to detect other abnormalities of lungs i.e. lung cancer and heart diseases using the proposed chest radiography algorithm. Bacilli segmentation is done on the basis of color, so for future work one could also consider size and shape parameters of bacilli to make the system better