Machine learning for classification of periodontal defects (vertical versus horizontal) using CBCT datasets.

Accurate classification of periodontal bone defects is essential for treatment planning; yet conventional radiographic methods have limitations in complex anatomical regions. Therefore, it is of interest to develop and validated machine learning models to automatically classify periodontal bone defects using cone-beam computed tomography data from 1,847 teeth in 312 patients treated between January 2021 and December 2023. Convolutional neural networks, random forest, support vector machines and gradient boosting classifiers were trained using radiomic features and raw image data, with performance evaluated through five-fold cross-validation against expert consensus. The convolutional neural network achieved the highest performance with 91.4% accuracy, 89.8% sensitivity for vertical defects, 92.6% specificity for horizontal defects and an area under the ROC curve of 0.946. Thus, we show that machine learning; particularly deep learning approaches can reliably classify periodontal defect morphology on CBCT images and support improved diagnostic consistency and clinical decision-making in periodontology.