Automated Tuberculosis screening using image processing tools

We present a preliminary work as a proof of concept on how image processing algorithms can be applied to detect and diagnose Tuberculosis in microscopic images of sputum samples stained with the Ziehl-Neelsen method. 300 images were acquired at the Hospital Nacional Dos de Mayo and processed using edge detection and mathematical morphology to extract objects of interest. Bacilli are discriminated from these objects applying a classifier based on the Mahalanobis distance and using shape characteristics as features. Results show a specificity value over 90% which is close to previously reported attempts on samples processed with Auramine.