Automatic segmentation and identification of the Mycobacterium tuberculosis in sputum images

Abstract

Tuberculosis is a very serious disease whose control is based on early diagnosis. A method frequently employed in its diagnosis consists of the sputum analysis in order to detect the Mycobacterium tuberculosis. The sputum examination demands a great amount of time and a good training of the specialist is required to avoid to commit a great numbers of errors. Image processing techniques can be helpful in examinations. Thus, this paper presents a new technique, it tries to improve the precision and diminish the time used in the analysis sputum samples. This techniques uses the linguistic knowledge about the characteristics of the bacilli, using the color information for segmentation and a classification tree for bacilli identification to establishing if a sample is positive or negative.