Hospital of Stomatology, Guanghua School of Stomatology, Sun Yat-Sen University, Guangzhou, China.
Guangdong Provincial Key Laboratory of Stomatology, Guangzhou, China.
BMC Oral Health. 2024 Sep 27;24(1):1132. doi: 10.1186/s12903-024-04922-2.
This study aims to verify the effectiveness of a deep neural network (DNN) in automatically identifying pulp calcification on cone beam computed tomography (CBCT) images.
This study retrospectively analysed 150 CBCT images. Pulp calcification was identified and manually annotated by three endodontists with 10 years of experience. A DNN model based on the U-Net architecture was constructed to identify pulp calcification, and ten rounds of fourfold cross-validation were conducted. The model performance was evaluated using sensitivity, specificity, accuracy, and the area under the receiver operating characteristic curve (AUC).
The model achieved a sensitivity of 75.91 ± 2.84% in automatically identifying pulp calcification, with a specificity of 68.88 ± 2.35%, an accuracy of 72.78 ± 2.13%, and an AUC of 73.68 ± 3.09%. According to the ranking for diagnostic tests, the proposed method achieved a "good" grade for sensitivity, accuracy, and AUC and a "fair" grade for specificity.
The results indicate that the proposed method shows promise for identifying pulp calcification on CBCT images. Future research aims to expand the dataset and refine the model, thereby enhancing its clinical applicability. The integration of artificial intelligence into diagnostic and treatment systems is anticipated to increase the efficiency of diagnosing pulp calcification and assist dentists in assessing the difficulty of root canal treatment cases preoperatively.
Registration was performed on the Chinese Clinical Trial Registry ( https://www.chictr.org.cn/ ) (Registration number: ChiCTR2300077078, 27/10/2023) and National Medical Research Registry Information System ( https://www.medicalresearch.org.cn/ , 30/10/2023) (Number: MR-44-23-039664).
本研究旨在验证深度神经网络（DNN）在自动识别锥形束计算机断层扫描（CBCT）图像上牙髓钙化方面的有效性。
本研究回顾性分析了 150 张 CBCT 图像。由 3 名具有 10 年经验的牙髓病专家对牙髓钙化进行识别和手动标注。构建了一个基于 U-Net 架构的 DNN 模型来识别牙髓钙化，并进行了十轮四折交叉验证。使用灵敏度、特异性、准确性和受试者工作特征曲线下面积（AUC）来评估模型性能。
该模型在自动识别牙髓钙化方面的灵敏度为 75.91±2.84%，特异性为 68.88±2.35%，准确性为 72.78±2.13%，AUC 为 73.68±3.09%。根据诊断测试的排名，所提出的方法在灵敏度、准确性和 AUC 方面达到了“良好”等级，在特异性方面达到了“一般”等级。
研究结果表明，该方法在 CBCT 图像上识别牙髓钙化具有一定的应用前景。未来的研究旨在扩大数据集并改进模型，从而提高其临床适用性。将人工智能集成到诊断和治疗系统中，有望提高牙髓钙化的诊断效率，并帮助牙医在术前评估根管治疗病例的难度。
本研究在中国临床试验注册中心（https://www.chictr.org.cn/）（注册号：ChiCTR2300077078，2023 年 10 月 27 日）和国家医学研究登记信息系统（https://www.medicalresearch.org.cn/，2023 年 10 月 30 日）（编号：MR-44-23-039664）进行了注册。
