Accuracy Assessment Of Annotation Techniques For Gender Determination On Orthopantomogram

To develop an unique and accurate AI algorithm for the identification of specific landmarks on an OPG for gender determination and to compare the accuracy by using two different annotation techniques. After obtaining ethical clearance from the institution dataset of 574 annotated JPEG OPG images from dentate mandibles of adults (≥18 years) was used. Two annotation methods were applied: Object Detection with Bounding Boxes: Utilized YOLO algorithm for real-time object detection. Key Point Detection: Focused on identifying specific anatomical landmarks. Images were split into training, validation, and test sets, with augmentation and preprocessing applied. Fourteen YOLO models were trained and evaluated using mean Average Precision (mAP) and recall metrics. In object detection tasks, YOLO-NAS (medium) achieved a mean Average Precision (mAP) of 51.55%. Roboflow 3.0 Object Detection achieved an mAP of 57.9%, precision of 85.4%, and recall of 56.5%. YOLOv8 achieved an mAP of 63.6%, precision of 85.1%, and recall of 72%. Integrating AI-based landmark annotation techniques into OPG analysis holds promise for enhancing gender determination accuracy, with future improvements achievable through dataset expansion, advanced preprocessing, and model optimization.