MULTI-MODAL TUBERCULOSIS DISEASE DETECTION USING CONVOLUTIONAL NEURAL NETWORK

Abstract

Tuberculosis (TB) is a highly infectious disease caused by Mycobacterium tuberculosis that primarily affects the lungs but can also impact other parts of the body. However, accurately diagnosing TB poses a significant challenge, especially in developing countries where access to trained radiologists and expensive imaging methods is limited. Researchers are actively exploring the use of Computer-Aided Diagnosis (CAD) applications as a promising approach to enhance the diagnosis of TB. While many studies focus on analyzing pixel data from X-ray images, relying solely on this modality has limitations in effectively diagnosing the disease. This study proposes a multimodal fusion approach that integrates chest X-ray images and patient symptoms to improve TB detection. The researchers utilized a dataset collected from Wolkite University Specialized Hospital, consisting of 734 cases. Among these, 571 cases were classified as normal, while 163 cases were classified as abnormal. The proposed approach involved various techniques such as data preprocessing, augmentation, segmentation, and data balancing to ensure accurate results. Additionally, hyperparameter tuning was performed to optimize both unimodal and multimodal fusion models. The study employed both individual modality classification and multimodal fusion classification. For multimodal fusion, the researchers extracted features from individual classification models, allowing for a comprehensive analysis that combines the strengths of each modality. The proposed model is rigorously evaluated using 10-fold cross-validation and specific parameters. The results demonstrated that the multimodal fusion technique outperformed individual modalities, with the early fusion model achieving impressive performance metrics. It achieved a remarkable 99% accuracy, 99% sensitivity, and 98% specificity. This comprehensive approach significantly enhances TB detection accuracy, offering potential for real-world clinical applications. By improving diagnostic accuracy, healthcare professionals can devise more effective treatment plans. The use of multimodal fusion not only reduces the workload for radiologists but also ensures a more accurate and timely diagnosis.