CLASSIFICATION OF TUBERCULOSIS SPUTUM SAMPLES IN MIDTI ACCORDING TO THE IUATLD SCALE USING YOLOV7-BASED METHODS

• Main Author: Aulia, Suci
• Format: Dissertations
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/83744
• Institution: Institut Teknologi Bandung

Microscopic detection of acid-fast bacilli (AFB) from Mycobacterium tuberkulosis (MTB) in Ziehl- Neelsen (ZN)-stained sputum samples is essential for detecting tuberculosis. Pathologists encounter many challenges that may result in incorrect diagnoses, such as the heterogeneous shape and irregular appearance of AFB-MTB, the intensity of ZN staining, and the fact that the process of scanning each Field of View (FoV) using a conventional microscope is very time- consuming for pathologists, which also has the risk of human error because it depends on the concentration of their eyes. To address this issue, we propose a method based on the You Only Look Once v7 (YOLOv7) algorithm to develop a computer-aided diagnosis (CAD) for tuberculosis (TB). CAD-TB classifies ZN-stained sputum smear samples into International Union Against Tuberculosis and Lung Disease (IUATLD) grades: negative, scanty, TB1+, TB2+, and TB3+. The initial step of CAD-TB is to detect AFB-MTB in each field of view. This detection is done digitally on images of sputum samples with ZN staining. After detecting and counting the number of AFB- MTB in each field of view, then mapped on the IUATLD scale into five classes. To support CAD, this study also created a new database, called the Microscopic Imaging Database of Tuberculosis Indonesia (MIDTI). MIDTI was acquired by WSI according to WHO guidelines and IUATLD grades used facilities from the Institut Teknologi Bandung (ITB), including an Olympus CX-31 microscope connected to a modified DSLR 700D camera . The test was conducted to classify all 40 sputum samples on the MIDTI (test data) into IUATLD grades using YOLOv7-RepVGG algorithm obtained precision 93.75%, sensitivity or recall 96.77%, F1-score 95.24%, and specificity from 77.78%, and accuracy level 92.50%. Each test result in this dissertation research has been validated by a team of pathologists at the microbiology laboratory of the Research Centre for Care and Control of Infectious Disease, Faculty of Medicine, Padjadjaran University (RC3ID FK UNPAD), Bandung. The opportunity for future research is, firstly, in terms of AFB-MTB detection methods, developing or optimizing the algorithm used to calculate with precision in detecting the number of AFB-MTB in colonies is necessary. The second opportunity is to create a new automatic scanner or integrate detection methods with existing automatic scanners in previous research.