Medical image registration methods have been thoroughly investigated and broadly utilized in a variety of clinical applications. In dentistry, particularly in orthodontic treatment and dental implant surgery, accurate multimodal data registration plays a critical role in preoperative diagnosis and planning. However, due to the limited accuracy of existing registration methods, clinical practice still relies heavily on manually placed landmarks, which significantly restricts the widespread adoption of automatic multimodal medical image registration in preoperative settings. Traditional medical image registration algorithms often suffer from low computational efficiency and large registration errors, making it difficult to meet the clinical demands for high precision and reliability. To address these challenges, this paper integrates deep learning with multimodal fusion techniques and proposes an efficient automatic registration framework tailored for dental applications, specifically designed for Cone-Beam Computed Tomography (CBCT) and Intraoral Scans (IOS) data. The proposed method first employs segmentation networks to separately extract the anatomical structures from CBCT and IOS data, obtaining segmented models of the teeth. These segmented 3D models are then input into a graph convolutional neural network-based registration module, which constructs local graph structures to model the spatial relationships between points and their neighbors, thereby enabling high-precision point cloud registration. We conducted a comprehensive evaluation of the proposed method and compared it with several existing registration techniques. Experimental results on real patient dental datasets demonstrate that our method outperforms existing dental image registration approaches in both accuracy and computational efficiency, indicating strong potential for clinical application and broader adoption.