Classification of alveolar bone density using 3-D deep convolutional neural network in the cone-beam CT images: A 6-month clinical study

Abstract Background Computer-based diagnoses are a crucial study in the medical image analyzing and machine learning technologies. The cone beam computed tomography (CBCT) modality provides three-dimensional bone models to extract an interactive treatment plan at relatively low radiation dose and cost. For the first time in this study, the evaluation of alveolar bone density was performed by a 3-D deep convolutional neural network (CNN) at the CBCT images. The trabecular pattern of the bone was recognized and classified. Method This study aimed to present a methodology which was implementing 3D voxel-wise feature evaluation within a convolutional neural network. We presented a three-dimensional CNN method that evaluated the alveolar bone density from CBCT volumetric data which could efficiently capture the trabecular pattern. In clinical trials, 207 surgery target areas of 83 patients have been selected. Clinical parameters were measured and evaluated during the surgery and a 6-month follow-up. These parameters were used to database labeling and evaluate the performance of the proposed technique. Results Our method achieved the average precision score of 84.63% and 95.20% in the hexagonal prism and the cylindrical voxel shapes respectively. Furthermore, the alveolar bone classification was performed in 76 ms. In comparison to the state-of-art approaches, the efficiency of the suggested algorithm was proved. Conclusion An automatic classification can improve the proficiency and certainty of the radiologic evaluation. The outcome of this research may help the dentists in the implant treatment from diagnosis to surgery.