Deep Learning-Assisted CBCT Precisely Locating MB2 Validated By Microscope

Based on the gold standard of Micro-Root-Canal Treatment(MRCT), this study established a CBCT imaging deep learning training dataset for the maxillary molars MB2 canal. With clinical cases screening from 2021 to 2023, hereinto, total 78 cases (maxillary first/second molars accounting for 68/10, respectively) that MB2 were detected through CBCT and were successfully negotiated via microscope eventually, was included. CBCT images of the treated tooth were collected, and MB2 canal contours were manually annotated in their cross-sections, having 1,962 images labeled. They were randomly grouped into a training group (62 cases, 1,412 images), a validation group (8 cases, 305 images), and a test group (8 cases, 245 images), subsequently, the training was conducted in the YOLO11 artificial intelligence model. The precision (B) of the MB2 canal segmentation model was 0.883, and the recall (B) was 0.920. The mAP 50(B) value was 0.961. This study integrates preoperative CBCT three-dimensional imaging data with intraoperative microscopic validation results establishing a MB2 canal detection deep learning training system via multimodal fusion pathway. It efficiently solved the clinical challenge that the canal specificity identification was insufficient in the existed single-imaging methods. The intelligent analytical model developed in this study can accurately recognize MB2 canal anatomical features in maxillary molar CBCT images, providing reliable decision support for clinicians in preoperative root canal treatment planning.