Automatic Detection of Pulmonary Tuberculosis with MobileNet and Entropy-Based Feature Fusion Strategy

Abstract

The mycobacterium tuberculosis causes the chronic necrotizing disease known as tuberculosis. Complex pulmonary tuberculosis is a bacterial infection brought on by the easily airborne Mycobacterium TB. The primary infection is caused by mycobacterium TB, which gradually multiplies in the lungs after becoming infected. Frequent medical types of active lung TB symptoms and indicators include fever, coughing, weight loss, hemoptysis, and night sweats. The lack of human resources and radiological interpretation know-how weakens TB screening programs, especially in TB endemic countries. Additionally, poor judgment, a lack of proficiency, and diagnostic mistakes can directly result in patient damage or death. Making the right choice and automatically detecting CXR can serve as a reliable substitute for more complex and technically demanding methods. So that to address this problem, we created an AI model that automatically detect the existence of Pulmonary Tuberculosis disease on human beings. The model attempts to support physicians and patients for the diagnosis and treatment of the disease. We collected the necessary dataset from Jimma University medical center. We used image processing techniques that include image preprocessing, segmentation, feature extraction, and detection. We compared the newly proposed TBDMobileNetV2 model with other state-of-the-art models. TBDMobileNetV2 achieved 96.11% testing accuracy; AlexNet model achieved 83.39% testing accuracy; and VGGNet model achieved 73.33% testing accuracy. So that the proposed TBDMobileNetV2 model outperforms the other models f or the detection of Pulmonary Tuberculosis disease.