Deep learning based detection of critical anatomical structures in cone beam computed tomography images

This study aims to compare the performance of regional segmentation methods used for the detection of critical anatomical structures in cone-beam computed tomography (CBCT) images. In particular, the accurate detection of infraorbital foramen and maxillary sinus structures from radiographic images is the focus of the thesis. The maxillary sinus and infraorbital foramen are important anatomical structures in various clinical applications. Accurate and sensitive detection of these structures is crucial for proper diagnosis and treatment planning. In line with the thesis, different regional segmentation architectures with different structures were examined and compared for the detection of the maxillary sinus and infraorbital foramen. Each model was trained and tested for the task of maxillary sinus and infraorbital foramen detection. This process aims to evaluate the effect of different network architectures and learning strategies on performance. The original dataset used in the study includes slices from 3D tomography radiography images obtained from medical imaging devices. The detection performance of each model has been evaluated using metrics such as precision, accuracy, recall, and F1 score. The study aims to optimize image analysis processes in dentistry, maxillofacial surgery, and otolaryngology applications and to improve accuracy in treatment planning.