Intelligent Tuberculosis Detection System with Continuous Learning on X-ray Images

Tuberculosis (TB) has become a global health threat with millions of cases each year. Therefore, rapid and accurate detection is needed to control its spread. The application of artificial intelligence, especially Deep Learning (DL), has shown great potential in improving the accuracy of TB detection through DL-based X-ray image analysis. Although many studies have developed X-ray image classification models, very few have integrated them into web or mobile platforms. In addition, the models integrated into these platforms generally do not apply continuous learning methods so that model performance cannot be updated. Thus, it is necessary to build an intelligent system based on a web application that integrates the ResNet-101 model for TB detection in X-ray images. This system utilizes continuous learning methods, allowing the model to automatically update itself with new data, thereby improving detection performance over time. The results showed that before continuous learning, the model successfully classified all TB images correctly, but was only able to classify two normal images correctly, resulting in an accuracy of 62.5%. After manual continuous learning, the model showed an increase in accuracy to 71.4%, with better ability to recognize normal images, although there was a slight decrease in performance in detecting TB.