目的 采用人工智能人脸识别技术(简称人脸识别技术)实时评估医学生课堂专注度,评价其在应用中的可靠性。方法 2021年10月,选取北京大学口腔医学院2018级五年制和八年制口腔医学专业80名学生为研究对象,收集其在牙体解剖学8个课时的课堂录像,采用人脸识别技术对录像资料转换的图片(每秒抽取1帧)进行人脸分析,用随机函数抽取全部图片数据的1/5作为检测数据集并输出其识别“专注”学生的人脸数量,同时由授课教师肉眼观察学生面部状态以进行“专注”学生的人工识别,计算人脸识别的准确率和查全率。通过Wilcoxon检验和组内相关系数(intraclass correlation coefficient,ICC)进行数据分析。结果 对21 600张图片的“专注”学生进行人脸识别,4 320 张图片作为检测数据集,人脸识别的准确率为90.4%(41 288/45 668),查全率为81.4%(41 228/50 648)。人脸识别(378 830人次)和人工识别(345 689人次)专注”学生的差异无统计学意义(P=0.109);二者呈中等程度相关(ICC=0.65,P=0.026)。结论 人脸识别技术能够有效判断医学生在课堂上的专注程度,可以为实时评估教学效果提供有效技术支持。

Objective To explore the feasibility of artificial intelligence face recognition technology (referred to as face recognition technology) in evaluating medical students' classroom concentration in real time. Methods In October 2021, 80 students of five-year and eight-year majors in stomatology enrolled in 2018 in Peking University School of Stomatology were selected as the research participants. Their classroom videos of 8 class hours in the dental anatomy course were collected, and face recognition technology was used. Perform face analysis on the pictures converted from video data (extract one frame per second) to evaluate the number of students who are “focused” in real time; a random function selects 1/5 of the entire picture data as the detection data, and judging whether the student is listening to the class attentively according to the naked eye observation of the student's facial state by the teachers, to carry out the manual identification of “focused” students, calculate the accuracy and recall rate of face recognition. Wilcoxon test was used to analyze the difference between face recognition technology and human recognition. The consistency of face recognition and evaluator recognition results was tested by intraclass correlation coefficient (ICC). Results Artificial intelligence recognized the face of “focused” students with 21 600 pictures, and 4 320 pictures were used as detection data sets for manual marking. The accuracy of face recognition was 90.4% (41 288/45 668), and the recall rate was 81.4% (41 228/50 648). There was no significant difference between face recognition (378 830 person times) and manual recognition (345 689 person times); They were moderately correlated (ICC=0.65, P=0.026). Conclusions Face recognition technology can effectively judge the concentration of students in the classroom and can provide an effective technical support for real-time evaluation of teaching effect.