Detection of Mycobacterium Tuberculosis Using Real-time Detection Transformer

Tuberculosis is a severe disease caused by Mycobacterium tuberculosis (MT). In handling it, the sputum examination method was used because it is relatively cheap and easy. The importance of MT detection in medical personnel analysis. We propose detection of Mycobacterium tuberculosis (MT) with a 1-stage detection object using two models: Detection Transformer (DETR) and Real-time Detection Transformer (RT-DETR) with different backbone variations to determine the performance of both models and conduct experiments using other models, such as yolov5l and Faster-RCNN, to compare their evaluation results with those of the DETR and RT-DETR models. In this study, MT detection using DETR with the Backbone ResNet 101 variation obtained AP:50 = 0.681 and AP:50:95 = 0.276. The RT-DETR model with HGNetv2l backbone variations yielded AP:50 = 0.837 and AP:50:95 = 0.147. The Yolov5l model produced AP:50 = 0.75 and AP:50:95 = 0.326 values. while the Faster-RCNN model produced AP:50 = 0.696 and AP:50:95 = 0.30. The results of RT-DETR with the HGNetv2-l backbone demonstrated good performance in the detection of Mycobacterium tuberculosis.