APPLICATION OF SEMANTIC SEGMENTATION WITH U-NET ARCHITECTURE ON DYNAMIC ULTRASONOGRAPHY FOR TEMPOROMANDIBULAR JOINT DEFORMATION DETECTION

• Main Author: Sekar Prima, Aliya
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/85733
• Institution: Institut Teknologi Bandung

Ultrasound imaging is one of the medical imaging methods used by doctors as one of diagnostic procedures. Ultrasound imaging commonly used to observe soft tissue anatomy of the body, such as abdomen, thyroid glands, breast, and cardiac. In developing countries such as Indonesia, ultrasound imaging is one of the most affordable medical imaging methods for patients compared to other sophisticated medical imaging modalities, such as MRI and CT scan. Thus, the development of ultrasound imaging usage in various medical field is encouraged by a lot of newest research in recent days. Dentistry is one of the medical fields where the usage of ultrasound imaging is being developed. Temporomandibular joint disorder or TMD is one of the most common disorder found in patients who are in need of treatment by dentists. Temporomandibular joint or TMJ is often regarded as one of the more complex anatomy of joint due to its shape and location. In the diagnostic procedure of TMD, medical imaging method is often used in order to obtain additional information for doctors to be able to give accurate diagnosis for patients. As the forementioned challenges of medical imaging methods exist, the use of ultrasound imaging in the diagnostic procedure of TMD currently is being developed. The challenges of the use of ultrasound imaging in TMJ anatomy are the lack of standard procedure available and the subjectivity of ultrasound imaging qualities due to it being dependent to the operators. Thus, an objective method of observing TMJ ultrasound imaging is needed in order to solve the challenges. In this research, the application of computer-aided diagnosis algorithm with U-Net based semantic segmentation method is developed and observed. The results obtained from the experiments in this research have shown the huge potential of semantic segmentation method to be applied in TMJ deformation detection task. One of the semantic segmentation models developed in this research has reached the accuracy of 92.7% in predicting discus articular of TMJ in dynamic ultrasound imaging. This shows that semantic segmentation method with U-Net architecture has great potential to be applied in the development of computer-aided diagnosis in TMD. Keywords: ultrasonography, semantic segmentation, temporomandibular joint, U-Net?