Automated search and identification of Mycobacterium tuberculosis in digital images of micropreparations to improve the quality of bacterioscopic diagnosis of TB

Abstract

The aim of investigation was the design a computer system for Mycobacterium tuberculosis search in digital images of microscopic preparations. Materials and methods. To develop the algorithm of the selecting objects, we used biological material (sputum) from patients with pulmonary tuberculosis with the already established and proven disease. Preparation stained by Ziehl-Nelsen method. To identify objects used mathematical models of logistic regression, decision tree and artificial neural network. On the basis of the color characteristics of each pixel and differences there were developed the algorithm of the determination of the objects on digital image of the microscopic preparation. 114 morphometric parameters were measured. Results indicate that the percentage of correct identification of objects with artificial neural network has been reached and the maximum 96.7%.Thus, the developed algorithms allow the search and identification of objects in digital images microscopy diagnostic material, suggesting the possibility of using the system in public health institutions