Bayati Mahsa, Alizadeh Savareh Behrouz, Ahmadinejad Hojjat, Mosavat Farzaneh
Post Graduate Student, Department of Oral and Maxillofacial Radiology, Faculty of Dentistry, Tehran University of Medical Sciences, Tehran, Iran.
PhD in Medical Informatic, Research and Development Manager, Department of Artificial Intelligence, Naaptech Co, Tehran, Iran.
Sci Rep. 2025 Feb 7;15(1):4641. doi: 10.1038/s41598-024-84737-x.
Dental caries is a very common chronic disease that may lead to pain, infection, and tooth loss if its diagnosis at an early stage remains undetected. Traditional methods of tactile-visual examination and bitewing radiography, are subject to intrinsic variability due to factors such as examiner experience and image quality. This variability can result in inconsistent diagnoses. Thus, the present study aimed to develop a deep learning-based AI model using the YOLOv8 algorithm for improving interproximal caries detection in bitewing radiographs. In this retrospective study on 552 radiographs, a total of 1,506 images annotated at Tehran University of Medical Science were processed. The YOLOv8 model was trained and the results were evaluated in terms of precision, recall, and the F1 score, whereby it resulted in a precision of 96.03% for enamel caries and 80.06% for dentin caries, thus showing an overall precision of 84.83%, a recall of 79.77%, and an F1 score of 82.22%. This proves its reliability in reducing false negatives and improving diagnostic accuracy. YOLOv8 enhances interproximal caries detection, offering a reliable tool for dental professionals to improve diagnostic accuracy and clinical outcomes.
龋齿是一种非常常见的慢性疾病，如果早期诊断未被发现，可能会导致疼痛、感染和牙齿脱落。传统的触觉视觉检查和咬合翼片放射摄影方法，由于检查者经验和图像质量等因素，存在内在的变异性。这种变异性可能导致诊断不一致。因此，本研究旨在使用YOLOv8算法开发一种基于深度学习的人工智能模型，以提高咬合翼片放射照片中邻面龋齿的检测率。在这项对552张放射照片的回顾性研究中，共处理了在德黑兰医科大学标注的1506张图像。对YOLOv8模型进行了训练，并根据精确率、召回率和F1分数对结果进行了评估，结果显示釉质龋的精确率为96.03%，牙本质龋的精确率为80.06%，总体精确率为84.83%，召回率为79.77%，F1分数为82.22%。这证明了其在减少假阴性和提高诊断准确性方面的可靠性。YOLOv8增强了邻面龋齿的检测能力，为牙科专业人员提供了一个可靠的工具，以提高诊断准确性和临床结果。
