Department of Conservative Dentistry, University Hospital Heidelberg, Heidelberg University, Heidelberg, Germany.
Deans Office of the Medical Faculty, Heidelberg University, Heidelberg, Germany.
Eur J Dent Educ. 2024 Nov;28(4):925-937. doi: 10.1111/eje.13028. Epub 2024 Jul 31.
Radiographic diagnostic competences are a primary focus of dental education. This study assessed two feedback methods to enhance learning outcomes and explored the feasibility of artificial intelligence (AI) to support education.
Fourth-year dental students had access to 16 virtual radiological example cases for 8 weeks. They were randomly assigned to either elaborated feedback (eF) or knowledge of results feedback (KOR) based on expert consensus. Students´ diagnostic competences were tested on bitewing/periapical radiographs for detection of caries, apical periodontitis, accuracy for all radiological findings and image quality. We additionally assessed the accuracy of an AI system (dentalXrai Pro 3.0), where applicable. Data were analysed descriptively and using ROC analysis (accuracy, sensitivity, specificity, AUC). Groups were compared with Welch's t-test.
Among 55 students, the eF group by large performed significantly better than the KOR group in detecting enamel caries (accuracy 0.840 ± 0.041, p = .196; sensitivity 0.638 ± 0.204, p = .037; specificity 0.859 ± 0.050, p = .410; ROC AUC 0.748 ± 0.094, p = .020), apical periodontitis (accuracy 0.813 ± 0.095, p = .011; sensitivity 0.476 ± 0.230, p = .003; specificity 0.914 ± 0.108, p = .292; ROC AUC 0.695 ± 0.123, p = .001) and in assessing the image quality of periapical images (p = .031). No significant differences were observed for the other outcomes. The AI showed almost perfect diagnostic performance (enamel caries: accuracy 0.964, sensitivity 0.857, specificity 0.074; dentin caries: accuracy 0.988, sensitivity 0.941, specificity 1.0; overall: accuracy 0.976, sensitivity 0.958, specificity 0.983).
Elaborated feedback can improve student's radiographic diagnostic competences, particularly in detecting enamel caries and apical periodontitis. Using an AI may constitute an alternative to expert labelling of radiographs.
放射诊断能力是牙科教育的主要重点。本研究评估了两种反馈方法，以提高学习成果，并探讨了人工智能（AI）支持教育的可行性。
第四年的牙科学生在 8 周内可使用 16 个虚拟放射学示例病例。根据专家共识，他们被随机分配接受详细反馈（eF）或结果知识反馈（KOR）。学生的诊断能力在检测龋病、根尖周炎、所有放射学发现的准确性和图像质量方面在牙尖片/根尖片上进行测试。我们还评估了人工智能系统（dentalXrai Pro 3.0）的准确性，在适用的情况下。数据使用描述性分析和 ROC 分析（准确性、敏感性、特异性、AUC）进行分析。使用 Welch 的 t 检验比较组间差异。
在 55 名学生中，eF 组在检测釉质龋（准确性 0.840±0.041，p=0.196；敏感性 0.638±0.204，p=0.037；特异性 0.859±0.050，p=0.410；ROC AUC 0.748±0.094，p=0.020）、根尖周炎（准确性 0.813±0.095，p=0.011；敏感性 0.476±0.230，p=0.003；特异性 0.914±0.108，p=0.292；ROC AUC 0.695±0.123，p=0.001）和评估根尖片图像质量方面的表现明显优于 KOR 组（p=0.031）。在其他结果方面未观察到显著差异。人工智能显示出几乎完美的诊断性能（釉质龋：准确性 0.964，敏感性 0.857，特异性 0.074；牙本质龋：准确性 0.988，敏感性 0.941，特异性 1.0；总体：准确性 0.976，敏感性 0.958，特异性 0.983）。
详细反馈可以提高学生的放射诊断能力，特别是在检测釉质龋和根尖周炎方面。使用人工智能可能是专家标记放射图像的替代方法。
