Radiation-Free Cephalometric Landmark Detection Using Facial Profile Images

This study introduced a novel approach to cephalometric landmark detection using facial profile images instead of X-rays, addressing concerns about patient exposure to radiation. The model initially estimates landmark coordinates from facial profile features, leveraging high-resolution representation learning. Subsequently, spatial relationships refine the coordinates, followed by input into fully connected networks for enhanced accuracy. Experiments with 2000 facial profile images from female patients demonstrated promising results, suggesting performance on par with or potentially surpassing existing cephalogram-based methods. The proposed two-stage learning method estimates precise landmark position, indicating that X-rays may not be necessary for cephalometric landmark detection.