An adaptive filtering approach for segmentation of tuberculosis bacteria in Ziehl-Neelsen sputum stained images

Tuberculosis is a disease with one of the most leading causes of deaths in the world, however, its fatality index could be reduced if it is diagnosed and treated on time. The Ziehl-Neelsen stained sputum smear method is the most used for bacilli detection and for developing a proper diagnosis by the specialist. Nevertheless, these stained images do not always present an adequate contrast, then, the elaboration of a reliable diagnosis is a complex, time consuming and a difficult process. This research proposes an alternative method to perform automatic bacilli segmentation in Ziehl-Neelsen images using Adaptive Signal Processing techniques, like the Least Mean Squares and Reduced Rank with Eigendecomposition algorithms. The quantitative results achieved, in correlation and true positives detection, are encouraging and suggest the use of this approach as a feasible alternative, when compared with the classical segmentation techniques, for automatic bacilli segmentation in the Ziehl-Neelsen images.