An Integrated Artificial Intelligence Model for Assisting Streamline Orthodontic Analysis in 3D Cephalometry using CBCT Cone Beam Computerized Tomography

Skeletal assessment through lateral cephalometry plays a important role in orthodontics, offering a consistent and effective method for diagnostic classification. While most existing studies focus on two-dimensional image analysis, three-dimensional (3D) imaging-achieved through cone-beam computed tomography (CBCT)-offers a more detailed evaluation of craniofacial hard and soft tissues compared to traditional 2D radiographs. This study aims to evaluate and enhance the traditional 2D tracing method to meet the efficiency of 3D analysis by developing a system that focuses on the most important landmark needed to streamline cephalometric evaluations. A hybrid computerized system was created to automate manual tasks in 2D lateral cephalometric and CBCT analysis. Tested on a dataset of cephalometric radiographs from different Caucasian patients with informed consent, based on European Orthodontics standards. The findings highlight the effectiveness of combining handcrafted feature extraction with panoptic segmentation that enhance the role of Yolo-UNet. Moreover, Grad-CAM (Gradient-weighted Class Activation Mapping) provides visual insights into the decision-making process of deep learning models by generating heatmaps. These heatmaps highlight the specific regions in an image that the model focuses on when making predictions, thereby identifying the most relevant areas or regions of interest (ROI) and enhancing interpretability. The experimental results achieved an accuracy exceeding 98.9%. The research providing a clear image for 2D and 3D evaluation demonstrating the ROI necessary for orthodontics measurements and analysis process relies on artificial Intelligent techniques.