DEVELOPMENT OF ARTIFICIAL INTELLIGENCE ONCOMPUTER AIDED DIAGNOSIS FOT DETECTION OF DENTAL DISORDERS BASED ON PANAROMIC X-RAY

• Main Author: Farras Kusuma, Rafi
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/85689
• Institution: Institut Teknologi Bandung

Dental disease is a significant global health problem. According to the 2017 Global Burden of Disease study, approximately 3.5 billion people worldwide experience dental disease, including untreated caries, severe periodontal damage, and severe tooth loss. Panoramic radiographs (PRs) are one of the frequently used methods for dental examinations due to their ability to show a comprehensive overview of the entire jaw structure in a single image. However, PRs also have limitations such as the absence of three-dimensionality and potential artefacts that may hinder accurate interpretation by clinicians. This study aims to develop an artificial intelligence (AI)-based Computer Aided Diagnosis (CAD) that utilises PRs to improve the accuracy and efficiency of dental abnormality detection. The applied methodology involves training two models, namely Mask R-CNN for Instance Segmentation (ISeg) which includes segmentation as well as bounding box (Bbox) determination and ResNet50 for classification. Mask R-CNN is trained using a dataset without disease labels while ResNet50 uses cropping results from a dataset labelled with diseases. The two models were then combined through transfer learning and further trained with disease-labelled datasets to improve the overall performance of the system. The results showed that the developed AI system achieved a score of 90.1% for Segmentation and Bbox accuracy. Meanwhile, the classification showed a Precision score of 90%, Recall 91.2%, Accuracy 91.2%, and F1 Score 90.5%. This system is proven to increase efficiency and objectivity in dental PRs image analysis and is able to process images faster than conventional methods. However, there are still challenges in accurate detection of abnormalities such as Deep Caries and Periapical Lesions, indicating the need for further development of the model to improve detection accuracy for these classes. Keywords: Panoramic Radhiographic, Artificial Intelligence, Dental Diseases, Mask R-CNN, Instance Segmentation, Classification, ResNet50