人工智能辅助培训用于口腔医学本科生邻面龋诊断：一项随机实验研究。
完整摘要 · 中文译文
背景：人工智能（AI）在提高牙科X线片上邻面龋检测方面已显示出潜力；然而，其在提升本科生独立影像判读能力方面的教育价值仍不清楚。因此，本研究旨在评估AI辅助放射学培训与传统放射学培训相比，对口腔医学本科生评估邻面龋扩展诊断能力的影响。 方法：本随机实验研究纳入82名口腔医学本科生。在完成一项使用20张传统咬翼片的前测后——学生需在每张X线片上评估三个邻面——学生被随机分配到传统放射学培训组或使用Second Opinion®软件的AI辅助培训组。两组均接受一小时的培训课程。一天后，使用20张传统咬翼片进行无AI辅助的后测，学生再次在每张X线片上评估三个邻面。诊断表现采用完全匹配总分（0–60分）和绝对平均误差（0–5分制）进行评估。同时记录前测和后测的完成时间（分钟）。采用混合设计方差分析评估组别、时间及交互效应（α = 0.05）。 结果：79名学生完成了两次测试，其中41人接受AI辅助培训，38人接受传统X线片培训。培训组之间以及各时间点（干预前和干预后，p > 0. 05）。然而，常规组从测试前到测试后的绝对平均误差显著增加（p < 0.05），表明其与正确答案的偏差更大。AI组测试后完成时间显著短于常规组，且两组从测试前到测试后的完成时间均显著缩短（p < 0.05）。 结论：与常规放射学训练相比，AI辅助训练并未显著提高学生在邻面龋检测中的诊断能力。然而，AI训练提高了诊断效率，表现为较低的绝对平均误差，并且在移除AI支持后未损害学生独立进行影像判读的能力。AI可作为本科口腔医学教育中的支持性辅助手段，特别是在提高病变深度评估的效率和一致性方面。
完整摘要 · 英文原文
BACKGROUND: Artificial intelligence (AI) has shown promise in improving the detection of proximal caries on dental radiographs; however, its educational value in enhancing undergraduate students' independent radiographic interpretation remains unclear. Therefore, this study aimed to evaluate the effect of AI-assisted radiology training on undergraduate dental students' diagnostic competence in assessing proximal caries extension when compared with conventional radiology training. METHODS: This randomized experimental study included 82 undergraduate dental students. After completing a pre-test using 20 conventional bitewing radiographs, from which students evaluated three proximal surfaces per radiograph, students were randomly allocated to either conventional radiology training or AI-assisted training using Second Opinion® software. Both groups received a one-hour training session. A post-test using 20 conventional bitewing radiographs without AI assistance was conducted one day later, with students again evaluating three proximal surfaces per radiograph. Diagnostic performance was assessed using total exact-match scores (0-60) and absolute average error (0-5 scale). Completion time in minutes was also recorded for pre- and post-tests. Mixed-design ANOVA was used to evaluate group, time, and interaction effects (α = 0.05). RESULTS: Seventy-nine students completed both tests, with 41 trained using AI assistance and 38 trained using conventional radiography. No significant differences were found between training groups or across time points (pre- and post-intervention, p > 0.05). However, absolute average errors increased significantly in the conventional group from pre- to post-test (p < 0.05), indicating a greater deviation from the correct answers. Post-test completion time was significantly shorter in the AI group compared with the conventional group, and both groups demonstrated significant reductions in completion time from pre- to post-test (p < 0.05). CONCLUSIONS: AI-assisted training did not significantly improve students' diagnostic competency in proximal caries detection compared with conventional radiology training. However, AI training improved diagnostic efficiency, as indicated by the lower absolute average errors, and did not impair students' independent radiographic interpretation when AI support was removed. AI may serve as a supportive adjunct in undergraduate dental education, particularly for improving efficiency and consistency in lesion depth estimation.