Development of AI-Based Field-Of-View Scanning Microscope for Automatic Detection of Tuberculosis

Prepared TB (Tuberculosis) slides by staining the pulmonary TB sputum slides using the Ziehl-Nielsen (ZN) procedure may have various qualities. A medical professional may also encounter challenges when he/she does an inspection of the slides under a conventional microscope. Additionally, extensive observation can lead to incorrect diagnosis due to fatigue. In this study, a mycobacterium tuberculosis (MTB) object identification system is constructed and integrated into a home-made FoV (Field-of-View) scanning microscope to overcome these issues. After capturing TB images by the microscope, a You-Only-Look-Once (YOLO) model is used for object detection. The test data is inputted into the model, then the system generates a bounding-box-labelled image and displays the number of the detected MTB. This number is then used for classifying severity level of the TB disease. Experimental results show that the system can effectively capture, detect, and classify actual TB images.