Automated analysis on enhancing the diagnostic relevance of tuberculosis images using image processing and artificial intelligence

Abstract

Tuberculosis is a communicable disease for which an early diagnosis is essential to control the disease The microscopy based TB screening is the conventional method employed for TB identification and provides significant benefit to large number of TB burdened communities across the globe Manual screening using microscope is tedious and requires highly trained experts Besides huge variability in sensitivity manual newlinescreening for the identification of disease causing agent is a labor intensive task Further it is time consuming and depends on patient s level of infection and requires large number of images to be analyzed in one slide Hence there is a need to automate the diagnostic process to improve the sensitivity and accuracy of the test The sputum smear positive and negative images recorded under standard image acquisition protocol are considered for this work The non uniform illumination in microscopic digital TB images due to light source optics and camera noise degrades the visual perception of these newlineimages In this work pre processing step to correct the non uniform illumination using retrospective techniques such as Surface Fitting Method Multiple Regression Method and Bidirectional Empirical Mode Decomposition has been attempted The most appropriate illumination correction method is evaluated by calculating the error and newlinestatistical measures Multifractal analysis that describes both local and global pixel distribution in an image is performed to further validate the methods newline newline