Mycobacterium Tuberculosis Detection on Sputum Smear Microscopic Images Using Attention-Based Multi-Scale Convolutional Neural Network

Tuberculosis (TB) remains a significant global health issue caused by the bacterium Mycobacterium tuberculosis. In 2021, the World Health Organization (WHO) reported approximately 10.6 million TB cases and 1.6 million deaths due to the disease. Diagnosing TB involves various methods, including the Mantoux test, Interferon Gamma Release Assays, sputum smear microscopy, and chest X-rays. Among these, sputum smear microscopy is commonly used by laboratory technicians to identify TB, involving the Ziehl-Neelsen staining technique to detect red-colored bacteria in sputum samples. However, visual detection using the naked eye has limitations, such as time consumption and reliance on technician expertise, leading to potential inaccuracies and fatigue. This research proposes an automated detection system using deep learning to improve the efficiency and accuracy of TB diagnosis from sputum samples. The system aims to reduce the time required for diagnosis and assist technicians in making accurate decisions regarding TB infection. The proposed model employs Attention-based Multiscale CNN algorithms to automatically detect the presence of Mycobacterium tuberculosis in stained sputum samples. The model was tested and achieved an accuracy of 97.56%, demonstrating its effectiveness as a reliable tool for TB diagnosis. By implementing this system, laboratory technicians can benefit from more accurate and faster diagnosis, improving patient outcomes and addressing the challenges associated with manual detection. This research highlights the potential of deep learning technology in enhancing TB diagnostic processes.