REMOVED: Human-AI collaboration for mandibular canal tracing accuracy on cone-beam computed tomography: A multi-evaluator study

To evaluate the accuracy, efficiency, and reliability of manual, artificial intelligence (AI)-driven, and AI-assisted tracing methods for localizing the mandibular canal (MC) using cone-beam computed tomography (CBCT) across diverse clinical scenarios and evaluators. Ground truth was established using a dry skull. Ten CBCT scans, including anatomy, bone density, and dentition status variations, were assessed by six evaluators with varying experience levels using manual, AI-driven, and AI-assisted methods. Tracings were assessed using Dice Similarity Coefficient (DSC) and 95% Hausdorff Distance (HD95, mm). Task completion time was recorded. Accuracy, boundary precision, reliability, efficiency, and inter-evaluator agreement were analysed. A total of 312 MCs were evaluated. Manual methods showed excellent intra- and inter-evaluator reliability (DSC: 0.99, HD95: 1.11). AI-assisted methods showed good inter-evaluator (DSC: 0.91, HD95: 1.23) and near-perfect intra-evaluator reliability (DSC: 0.81, HD95: 2.2). Manual tracing achieved the highest accuracy (DSC: 0.98, 99.1%) but was the slowest (6.22 minutes). AI-driven tools were the fastest (13 seconds–3 minutes), with strong accuracy (DSC: 0.87, 99.8%), but their performance declined in complex cases. AI-assisted methods balanced speed (4.19 minutes) and accuracy (DSC: 0.83, 99.9%) with better boundary precision than manual ( p = 3×10 -4 ), and comparable to the AI-driven method ( p = 1.0). Inter-evaluator agreement was low for manual (ICC: 0.30), but good for AI-assisted and AI-driven methods (ICC: 0.72). Manual tracing is the most accurate method for the task, and AI-driven methods offer speed. AI-assisted tracing optimizes accuracy and efficiency, improves boundary precision, and reduces examiner variability. Artificial Intelligence-assisted tracing methods, with expert/clinician intervention, for mandibular canal tracing offer a practical balance between precision, efficiency, and time savings, reducing variability and improving reliability across users even in complex anatomical cases, where performance remained consistent despite occasional challenges. • This study evaluated manual, AI-driven, and AI-assisted methods for mandibular canal tracing in CBCT, using accuracy, efficiency, and reliability metrics across diverse cases. • AI-assisted tracing method offered a balanced solution, improving boundary precision, speed, and consistency, supporting broader clinical decision-making and diagnostics.