Artificial intelligence for detection and segmentation of accessory root canals in endodontic imaging: A systematic review.

Abstract

Accessory root canal anatomy, including second mesiobuccal (MB2) and middle mesial canals (MMC), is difficult to identify and may compromise endodontic outcomes when missed, supporting artificial intelligence (AI)-based diagnostic tools. This systematic review (CRD420261415943) included studies evaluating AI models for accessory canal or root canal orifice identification, emphasizing MB2, MMC, and canal orifice detection. Study characteristics, imaging modalities, AI architectures, reference standards, outcomes, and validation strategies were extracted. Risk of bias was assessed using QUADAS-2 and certainty of evidence using GRADE. Seven studies were included. Most evaluated MB2 detection or segmentation on cone-beam computed tomography (CBCT); one assessed root canal orifice detection using dental operating microscope images, and one evaluated MMC detection. AI approaches included U-Net, YOLO, CNN-based pipelines, Inception V3-based machine learning, ChatGPT-4o, and nnU-Net. For MB2 detection/classification, sensitivity or recall ranged from 80.0% to 100%, specificity from 99.4% to 100% when reported, accuracy from 84.9% to 97.8%, and area under the curve from 0.57 to approximately 0.90. Segmentation/localization outcomes were heterogeneous. AI shows promise as an adjunct for accessory canal detection, particularly MB2 on CBCT, but evidence remains limited.