ENHANCING THE EFFECTIVENESS OF TUBERCULOSIS DIAGNOSIS MODELS BASED ON VISUALIZATION TECHNIQUES

Abstract

The robust development of deep learning models has solved many practical problems, including supporting disease diagnosis through images. Although much progress has been achieved, explaining the decisions made by deep learning models remains a significant challenge. In some cases, these models use information outside the diagnostic area. This paper proposes a solution through visualization in diagnosing tuberculosis from chest X-ray images. It highlights the regions within the images that the deep learning model utilizes, aiming to detect discrepancies within the training dataset images. This enables data normalization and the application of techniques to improve the model's accuracy. The proposal has been implemented in trials and has shown effectiveness with deep learning models for chest X-ray images, aiding in the diagnosis of tuberculosis. It not only improves the reliability but also increases the accuracy of the model by adjusting and normalizing the training data based on real findings from the visualization process.