AI-Based Working Length Determination in Curved Root Canals

Overview

This study tests whether artificial intelligence (AI) can accurately measure the length of curved root canals from dental x-rays. Curved root canals are hard to measure correctly, and wrong measurements can lower the success of root canal treatment. Adults over 18 years old who need root canal treatment can take part. Researchers will use x-rays taken during the patient's normal treatment. No extra x-rays, procedures, or visits are needed. An AI program will be trained to measure canal length automatically, and its measurements will be compared to measurements made by a human expert. The results may show whether AI can measure root canal length faster and more consistently, which could help dentists plan treatment more accurately in the future.

Working length determination is a critical step in root canal treatment, and inaccurate measurement can lead to under- or over-instrumentation, post-operative pain, and reduced treatment success. This is particularly challenging in curved root canals, where the apical foramen often deviates from the anatomical or radiographic apex, and where small endodontic files are difficult to visualize on periapical radiographs. Conventional methods for working length determination, including tactile sensation, electronic apex locators, and radiographic interpretation, are subject to observer variability and technical limitations, especially as canal curvature increases. Recent advances in artificial intelligence, particularly deep learning-based image analysis, have shown promise in improving the objectivity and consistency of measurements derived from dental radiographs. This study aims to develop and evaluate deep learning models (including convolutional neural network architectures such as GoogleNet Inception V3, U-Net, Mask R-CNN, and YOLO-based networks) for the automatic detection and measurement of curved root canal length on periapical radiographs, and to compare the performance of these models with measurements made by a human observer. Periapical radiographs will be obtained using the paralleling technique during the working length confirmation stage of routine root canal treatment, as part of the patient's standard clinical care. No additional radiographic exposure, procedure, or clinical visit will be performed solely for research purposes. Radiographic images will be anonymized and labeled using canal length segmentation, with pixel-based measurements calibrated to millimeters based on DICOM metadata. The dataset will be divided into training, validation, and test subsets. Model performance will be evaluated using standard classification and segmentation metrics, including sensitivity, precision, F1 score, Intersection over Union (IoU), and Receiver Operating Characteristic (ROC) curve analysis with area under the curve (AUC).