Automated <i>Mycobacterium tuberculosis</i> Detection in Multivariant Digitized Ziehl-Neelsen Staining Using Faster R-CNN Method

Tuberculosis (TB) is an infectious disease caused by Mycobacterium tuberculosis and remains a major public health concern in Indonesia. One of the most widely used diagnostic methods is the microscopic examination of sputum smears stained using the Ziehl-Neelsen technique. However, manual identification of TB bacteria presents significant challenges, particularly due to staining thickness variations that lead to inconsistent color intensities, making visual detection difficult and often subjective. This study is aimed at developing an automated TB bacteria detection system using deep learning, specifically the Faster R-CNN algorithm with ResNet-50 layers. The system is implemented using the Python programming language and the TensorFlow Object Detection API. We incorporated data augmentation in the form of random rotation, random flipping, and color processing such as hue variation, saturation stretching, brightness stretching, and exposure stretching. Experimental results show that the proposed model achieves an accuracy of 88%, with a precision of 94%, recall of 93%, and an F1-score of 94%. The model outputs annotated images indicating the locations of TB bacteria, which can assist medical professionals in the diagnostic process. These findings demonstrate the potential of deep learning-based approaches in automating TB detection, particularly in healthcare settings with limited human resources.