确保牙科教育诚信:开发新型人工智能模型用于临床前牙髓治疗中一致且可追溯的图像分析
本研究开发并评估了一种基于孪生神经网络的人工智能模型,用于检测临床前牙髓根管治疗课程中放射影像不一致性,以提升教育诚信。模型在3390张放射影像上训练,测试集准确率89.31%、精确率76.82%、灵敏度84.82%、F1分数80.50%,最佳相似度阈值为0.48。结论认为该模型能有效检测放射影像不一致性,可作为维护学术诚信的客观工具。
要点
- AI模型在临床前根管治疗放射影像不一致性检测中准确率达89.31%。
- 最佳相似度阈值为0.48,交叉验证标准差为1.95%。
- 模型旨在提升评估公平性与可靠性并减轻教育者行政负担。
核心结论
- 基于孪生神经网络的AI模型能有效检测临床前根管治疗过程中的放射影像不一致性。
- 实施该模型可作为维护牙科教育学术诚信的客观工具,提升评估的公平性与可靠性,促进学生诚信文化并减轻教育者行政负担。
研究设计与样本
- 设计并评估一种利用孪生神经网络比较不同根管治疗过程放射影像的人工智能模型。
- 模型在3390张放射影像上训练,并应用数据增强以提高泛化能力。
- 数据集分为训练、验证和测试子集,采用交叉验证和超参数调优优化模型。
- 性能指标包括准确率、精确率、灵敏度(召回率)和F1分数。
主要发现
- AI模型准确率为89.31%,精确率为76.82%,灵敏度为84.82%,F1分数为80.50%。
- 最佳相似度阈值为0.48,此时观察到最大准确率。
- 混淆矩阵显示高正确分类率,交叉验证确认模型稳健性,各折标准差为1.95%。
关键数值与推荐
- 最佳相似度阈值为0.48。
- 模型准确率89.31%、精确率76.82%、灵敏度84.82%、F1分数80.50%。
- 交叉验证各折标准差为1.95%。
临床怎么用
- 原文未据此给出推荐。
局限性
- 原文未报告局限性。
以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。
原文
AIM: Academic integrity is crucial in dental education, especially during practical exams assessing competencies. Traditional oversight may not detect sophisticated academic dishonesty methods like radiograph substitution or tampering. This study aimed to develop and evaluate a novel artificial intelligence (AI) model utilizing a Siamese neural network to detect inconsistencies in radiographic images taken for root canal treatment (RCT) procedures in preclinical endodontic courses, thereby enhancing educational integrity. METHODODOLOGY: A Siamese neural network was designed to compare radiographs from different RCT procedures. The model was trained on 3390 radiographs, with data augmentation applied to improve generalizability. The dataset was split into training, validation, and testing subsets. Performance metrics included accuracy, precision, sensitivity (recall), and F1-score. Cross-validation and hyperparameter tuning optimized the model. RESULTS: Our AI model achieved an accuracy of 89.31%, a precision of 76.82%, a sensitivity of 84.82%, and an F1-score of 80.50%. The optimal similarity threshold was 0.48, where maximum accuracy was observed. The confusion matrix indicated a high rate of correct classifications, and cross-validation confirmed the model's robustness with a standard deviation of 1.95% across folds. CONCLUSIONS: The AI-driven Siamese neural network effectively detects radiographic inconsistencies in RCT preclinical procedures. Implementing this novel model will serve as an objective tool to uphold academic integrity in dental education, enhance the fairness and reliability of assessments, promote a culture of honesty amongst students, and reduce the administrative burden on educators.
引用
- Ibrahim M, Omidi M, Guentsch A, et al. Ensuring integrity in dental education: Developing a novel AI model for consistent and traceable image analysis in preclinical endodontic procedures. Int Endod J. 2025. PMID: 40536276. doi:10.1111/iej.14273
供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。