Efficient Tuberculosis Diagnosis Using Deep Learning based Object Detection Method

Abstract

Tuberculosis is a lung disease that poses a threat to life leading to causes of death so early and precise identification is important. The purpose of this work is to progress the field of tuberculosis (TB) detection by implementing deep learning techniques. This entails developing a model that can handle a variety of instances, including differences in ill lungs and the presence of tuberculosis bacteria. The work offers a complete methodology for tuberculosis detection based on deep learning. A diversified chest X-ray image dataset is collected and pre-processed. The proposed model, a combination of ResNet50 and Fast R-CNN (region-based convolutional neural network), creates a powerful synergy that provides a reliable method for precisely detecting and localizing objects within images. By utilizing ResNet50 as a backbone, Fast R-CNN benefits from the rich feature maps generated by the deep network. It facilitates precise region proposal generation and improves the overall accuracy of object localization. Alternative object detection algorithms, such as RetinaNet and SSD, are compared with the proposed model. We achieved a notable accuracy of 96.7% by using the Fast R-CNN approach with a ResNet50 backbone. This accomplishment highlights the effectiveness of our selected strategy by outperforming other algorithms.