Evaluation of the mandibular canal by CBCT with a deep learning approach

Abstract

Background/Aim: The mandibular canal including the inferior alveolar nerve (IAN) is important in the extraction of the mandibular third molar tooth, which is one of the most frequently performed dentoalveolar surgical procedures in the mandible, and IAN paralysis is the biggest complication during this procedure. Today, deep learning, a subset of artificial intelligence, is in rapid development and has achieved significant success in the field of dentistry. Employing deep learning algorithms on CBCT images, a rare but invaluable resource, for precise mandibular canal identification heralds a significant leap forward in the success of mandibular third molar extractions, marking a promising evolution in dental practices. Material and Methods: The CBCT images of 300 patients were obtained. Labeling the mandibular canal was done and the data sets were divided into two parts: training (n=270) and test data (n=30) sets. Using the nnU-Netv2 architecture, training and validation data sets were applied to estimate and generate appropriate algorithm weight factors. The success of the model was checked with the test data set, and the obtained DICE score gave information about the success of the model. Results: DICE score indicates the overlap between labeled and predicted regions, expresses how effective the overlap area is in an entire combination. In our study, the DICE score found to accurately predict the mandibular canal was 0.768 and showed outstanding success. Conclusions: Segmentation and detection of the mandibular canal on CBCT images allows new approaches applied in dentistry and help practitioners with the diagnostic preoperative and postoperative process.

Keywords

• deep learning
• artificial intelligence
• mandibular canal
• cbct