Detection of Mid Mesial Canal in Dental CBCT Images Using Combined YOLO and 2D U-Net Model

Failures in root canal treatments often arise from undetected canals and a limited comprehension of the intricate structure of the root canal system. An illustrative instance is the Middle Mesial Canal (MMC), situated between the mesiobuccal and mesiolingual canals of the mesial root of the first mandibular molar. Neglecting to identify and address these canals in the molar tooth can lead to subsequent treatment complications, resulting in persistent infection. The primary objective of this study is to evaluate the potential of three-dimensional dental Cone Beam Computed Tomography (CBCT) scans in detecting the presence of mid mesial canals through a deep learning approach. The 3D dental CBCT images are segmented into two-dimensional cross sections and processed to ascertain the occurrence of MMC. To achieve this, a deep learning Convolutional Neural Network (CNN) algorithm, known as YOLO (You Only Look Once), an object detection algorithm, and a segmentation algorithm called 2D U-Net are employed. The results obtained from YOLO reveal a confidence score of 0.89, a precision of 0.945, a recall of 0.902, and a Mean Average Precision (MAP) of 0.988, while the 2D U-Net achieves an accuracy of 0.9772.