DETECÇÃO E CONTAGEM AUTOMÁTICA DE BACILOS DE TUBERCULOSE EM IMAGENS MICROSCÓPICAS COM YOLO11

Abstract

This work presents the application of computer vision and deep learning techniques to assist in the automatic detection of bacilli in microscopic images used in tuberculosis diagnosis. The identification of these microorganisms is traditionally performed manually by healthcare professionals, which can make the process time-consuming and subject to variations in interpretation. To investigate the feasibility of automating this task, a model based on the YOLO architecture for object detection in images was developed. During the development of the study, several experimental tests were conducted to evaluate the impact of factors such as the number of classes used during training, the use of the image division technique (tiling), and the conversion of images to different color spaces. The results showed that the approach using a single class achieved better performance in bacilli detection, reaching mAP@0.5 values close to 0.79. The use of the tiling technique also proved important for improving the detection of small objects present in microscopic images. Thus, the results indicate that deep learning-based models can contribute to the analysis of microscopic images used in tuberculosis diagnosis, assisting in the automatic identification of bacilli.