Automatic classification of alveolar bone defects via 3D CBCT for dental implant preoperative evaluation

Abstract

Accurate assessment of alveolar bone defects is critical to the success of dental implant surgery, as it directly influences preoperative planning and surgical outcomes. However, the clinically adopted Terheyden classification method relies heavily on dentists' subjective judgment of Cone-Beam Computed Tomography (CBCT) images, leading to poor diagnostic consistency and a high misclassification rate. Meanwhile, existing studies are mostly limited to two-dimensional image analysis, failing to fully exploit the spatial geometric information of three-dimensional CBCT and thus unable to accurately characterize defect morphology. To address the above issues, this study proposes an automatic classification method for alveolar bone defects based on three-dimensional CBCT data. The Implant Signed Distance Function (SDF) is introduced to achieve implicit geometric characterization of the spatial relationship between implants and alveolar bone. A 3D Res2Net network is adopted for multi-scale feature extraction, taking into account both local details and global anatomical structures. An ordinal classification loss function based on a distance cost matrix is designed to reduce the probability of high-risk cross-level misclassification. Experimental results show that the proposed model achieves an accuracy of 94.76% and a macro-average F1 score of 94.78% on the test set. This method mitigates the limitations of the traditional subjective classification method, improves the objectivity and accuracy of grading, and provides a reliable auxiliary tool for preoperative planning of dental implant surgery.