APPLICATION OF CLASSIFICATION AND OBJECT DETECTION FOR TUBERCULOSIS DETECTION IN CHEST X-RAYS

• Main Author: Apriliyanti, Alya
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/86184
• Institution: Institut Teknologi Bandung

Tuberculosis, an infectious disease that generally attacks the lungs, has a significant impact on global health as the leading cause of death in the world after COVID-19 in 2022. One of the stages of examination for the diagnosis of tuberculosis is a chest x-ray examination. Several problems, such as irregular tuberculosis images on chest X-ray medical images, misinterpretation, and limited waiting time for examination, make increasing the efficiency of chest X-ray examinations, including in cases of tuberculosis, very much needed. AI (artificial intelligence) is one of the technologies that can solve various problems and increase efficiency in the health sector. Therefore, this final project will apply a classification and object detection model to detect tuberculosis cases in chest X-ray medical images. This detection system will receive chest medical images from X-rays. It will detect whether there is a case of tuberculosis in the lungs and then display the possible location of tuberculosis. The CNN architecture model used is EfficientNetB6 for image classification with new layers and YOLOv8x for object detection using the transfer learning method. Additional datasets from one of the hospitals in Bandung are used to train and test the model. The results of testing the model on the detection system show that the classification model can distinguish between medical images of chest X-rays in tuberculosis and regular with an accuracy value of around 95.86%. Meanwhile, the object detection model can detect the location of tuberculosis in the lungs with the limitation that the model can only detect large forms of tuberculosis (such as cloudy white spots or infiltration); it cannot detect forms of tuberculosis that are only in the form of nodules (small circles). The mAP value of the object detection model is 0.768.