A computer algorithm for detection of tuberculosis bacilli in Ziehl Nellsen sputum smear images based on the adjustment of RGB primary component tones and geometric eccentricity

The present study proposes a method of automatic detection of tuberculosis (TB) bacilli from digital images of Ziehl Neelsen sputum smear baciloscopy. The method is based on an algorithm that aims to automate the interpretation of optical microscopic images of sputum smears. According to the World Health Organization (WHO), a specialist can not analyze and process more than 20 samples per day (in order to not affect the analysis sensitivity and commit errors in diagnosis). Therefore, an automated tool as the proposed here, is an important contribution to the current efforts to fight tuberculosis. The algorithm is based on geometric eccentricity of ellipses and improvement of RGB component tones. Correspondence functions adjusted to sample preparation conditions were applied in order to improve the RGB primary component tones of the image. This allows to obtain an adequate segmentation of interest objects. For the recognition of each object as bacillus, the geometric descriptor of eccentricity of the ellipse was applied. The algorithm was validated with 66 independent sputum samples from TB patients. A sensitivity of 88.75% and a specificity of 95.5% was obtained for the diluted pellet method for sample preparation.