Curvas Principais para a triagem de pacientes com Tuberculose via análise de imagens de Raio X

Abstract

Tuberculosis (TB), if not diagnosed early, can be fatal or can cause serious consequences. Therefore, it’s important that the tools used in screening and diagnosing this disease continue to be modernized. In this article, two computational intelligence methods for detecting tuberculosis via X-ray image analysis are proposed and compared. One method uses the Principal Curves (PC) technique in a specialist approach (approach I). The second method uses a Principal Curve to model the class of TB lung images in a structure similar to a Siamese Network (approach II). For this, we used a database available in the literature divided into the following classes of patients: TB, other diseases and healthy. The accuracy values obtained by the methods are 0,89 (approach I) and 0,84 (approach II). these results are competitive compared with other methods in the literature, especially considering their low computational complexity.