Study of a trustworthy AI collaborative model focused on data annotation skill training for ophthalmology students

Liu Tianhang; Zhang Yinglin; Liu Ziying; Niu Yanyan; Hu Lingxi; Wang Xingyue; Liu Jiang

doi:10.3760/cma.j.cn116021-20251229-02319

您当前的位置：

首页 >

文章列表页 >

Study of a trustworthy AI collaborative model focused on data annotation skill training for ophthalmology students

Digital Intelligence MedicalEducation | 更新时间：2026-07-27

- Study of a trustworthy AI collaborative model focused on data annotation skill training for ophthalmology students
- Chinese Journal of Medical Education Research Vol. 25, Issue 7, Pages: 769-778(2026)
- 作者机构：
  
  1.南方科技大学计算机科学与工程系，深圳 518055
  2.温州医科大学眼视光学院，温州 325027
- 作者简介：
  
  Liu Jiang, Email: liuj@sustech.edu.cn
- 基金信息：
  
  General Project of National Natural Science Foundation of China(82272086);National Key Research and Development Program of China(2024YFC2510800);Southern University of Science and Technology Higher Education Teaching Research and Reform Project(SJZLGC202430)
- DOI：10.3760/cma.j.cn116021-20251229-02319
  CLC： R-0;R-4;R-05
- Received：29 December 2025，
  
  Online First：01 July 2026，
  
  Published：20 July 2026
- 稿件说明：
移动端阅览
刘天航, 张颖麟, 刘子莹, 等. 聚焦眼科学生数据标注能力培养的可信人工智能协同模式研究[J]. 中华医学教育探索杂志, 2026, 25(7): 769-778.

Liu Tianhang, Zhang Yinglin, Liu Ziying, et al. Study of a trustworthy AI collaborative model focused on data annotation skill training for ophthalmology students[J]. Chinese Journal of Medical Education Research, 2026, 25(7): 769-778.
刘天航, 张颖麟, 刘子莹, 等. 聚焦眼科学生数据标注能力培养的可信人工智能协同模式研究[J]. 中华医学教育探索杂志, 2026, 25(7): 769-778. DOI： 10.3760/cma.j.cn116021-20251229-02319.

Liu Tianhang, Zhang Yinglin, Liu Ziying, et al. Study of a trustworthy AI collaborative model focused on data annotation skill training for ophthalmology students[J]. Chinese Journal of Medical Education Research, 2026, 25(7): 769-778. DOI： 10.3760/cma.j.cn116021-20251229-02319.

摘要

目的

构建可信人工智能（artificial intelligence，AI）与眼科学生的协同标注模式，评估与研究其在眼科学生数据标注能力培养中的应用价值。

方法

本研究共纳入30张视网膜眼底图像，按难易程度分为简单组和复杂病例组，每组各15张。招募15名眼科学生，在无AI辅助、AI辅助、AI辅助+不确定信息图3种标注场景下开展标注实验。采用平均标注时间、Dice系数（Dice similarity coefficient，DSC）及几何平均比作为评价指标，根据数据分布特征选用配对

检验或Wilcoxon符号秩检验，分析评估该协同模式对眼科学生标注表现的影响。实验结束后通过问卷收集学生对AI协同模式在效率提升、辅助价值、使用意愿及AI认知影响等方面的反馈。

结果

在AI辅助下，标注者的眼底血管标注准确性与效率均有提升，差异具有统计学意义（

0.05）。简单病例组平均标注耗时从1 154.79 s降至292.09 s（降幅75.13%，效应量0.867 0），DSC从66.72%升至73.98%（提升11.18%，效应量0.858 0）；复杂病例组平均标注耗时从716.90 s降至207.78 s（降幅67.30%，效应量0.859 6），DSC从62.62%升至74.41%（提升18.98%，效应量0.855 6）。同时，个体间差异显著缩小：简单病例组标注耗时人均标准差从410.65 s降至130.19 s，DSC人均标准差从3.25%降至0.53%；复杂病例组标注耗时人均标准差从292.35 s降至76.78 s，DSC人均标准差从3.18%降至0.78%。问卷调查显示，所有参与的学生均认可该模式能提升标注效率与准确性，且提升了对AI的临床应用认知与信任度。

结论

可信AI协同模式可提升眼科学生眼底血管标注的效率与质量，有效缩小个体标注能力差异，增强对人工智能的认知与信心。该模式有望作为眼科临床教育的补充工具，为眼科学生数据标注能力培养提供初步参考路径。

Abstract

Objective

To construct a collaborative annotation model between trustworthy artificial intelligence (AI) and ophthalmology students

and to evaluate and explore its application value in cultivating the data annotation skills of ophthalmology students.

Methods

A total of 30 retinal fundus images were included in this study and divided into a simple case group and a complex case group according to the difficulty level

with 15 images in each group. Fifteen ophthalmology students were recruited to conduct annotation experiments under three scenarios: no AI assistance

AI-assisted annotation

and AI-assisted annotation with uncertainty maps. The impact of this collaborative model on the annotation performance of ophthalmology students was analyzed and evaluated using average annotation time

Dice similarity coefficient (DSC)

and geometric mean ratio (GMR)

with statistical analysis performed by paired

-test or Wilcoxon signed-rank test (depending on data distribution characteristics). After the experiment

a questionnaire was used to collect students' feedback on the AI collaborative model in terms of efficiency improvement

auxiliary value

willingness to use

and impact on AI cognition.

Results

With AI assistance

the accuracy and efficiency of fundus vessel annotation by the annotators were significantly improved (

0.05). In the simple case group

the average annotation time decreased from 1 154.79 s to 292.09 s (a reduction of 75.13%

effect size 0.867 0)

and the DSC increased from 66.72% to 73.98% (an improvement of 11.18%

effect size 0.858 0). In the complex case group

the average annotation time decreased from 716.90 s to 207.78 s (a reduction of 67.30%

effect size 0.859 6)

and the DSC increased from 62

.62% to 74.41% (an improvement of 18.98%

effect size 0.855 6). Meanwhile

the inter-individual variability was significantly reduced: in the simple case group

the inter-individual standard deviation of annotation time decreased from 410.65 s to 130.19 s

and the inter-individual standard deviation of DSC decreased from 3.25% to 0.53%; in the complex case group

the inter-individual standard deviation of annotation time decreased from 292.35 s to 76.78 s

and the inter-individual standard deviation of DSC decreased from 3.18% to 0.78%. The questionnaire survey showed that all participating students recognized that this model could improve annotation efficiency and accuracy

and enhanced their cognition of and confidence in the clinical application of AI.

Conclusions

The trustworthy AI collaborative model can improve the efficiency and quality of fundus vessel annotation by ophthalmology students

effectively reduce the inter-individual differences in annotation skills

and enhance students' cognition of and confidence in artificial intelligence. This model is expected to serve as a supplementary tool for clinical ophthalmology education and provides a preliminary reference path for the cultivation of data annotation skills of ophthalmology students.

关键词

Keywords

references

胥瀚文，陈墨馨，梁小乙，等 . 基于面部照片的眼病智能诊断研究进展 [J ] . 上海交通大学学报（医学版）， 2025 ， 45 （ 9 ）： 1249 - 1255 . DOI: 10.3969/j.issn.1674-8115.2025.09.017 http://dx.doi.org/10.3969/j.issn.1674-8115.2025.09.017 .

林宇晖 . 基于人工智能的血常规分析技术在临床应用中的研究进展 [J ] . 中国冶金工业医学杂志， 2025 ， 42 （ 4 ）： 384 - 385 . DOI: 10.13586/j.cnki.yjyx1984.2025.04.066 http://dx.doi.org/10.13586/j.cnki.yjyx1984.2025.04.066 .

李海强，苏微微，李秀萍，等 . 机器人辅助甲状腺手术的研究进展 [J ] . 江苏医药， 2025 ， 51 （ 8 ）： 829 - 833 . DOI: 10.19460/j.cnki.0253-3685.2025.08.016 http://dx.doi.org/10.19460/j.cnki.0253-3685.2025.08.016 .

门茜儒，刘林 . 人工智能赋能医学教育现代化的应用挑战与现实路径 [J ] . 医学教育研究与实践， 2025 ， 33 （ 5 ）： 641 - 646 . DOI: 10.13555/j.cnki.c.m.e.2025.05.006 http://dx.doi.org/10.13555/j.cnki.c.m.e.2025.05.006 .

梁爽，池元伟，孟巍，等 . 住院医师人工智能素养现状及其影响因素分析：基于问卷调查的实证研究 [J ] . 牡丹江医科大学学报， 2025 ， 46 （ 3 ）： 140 - 144 . DOI: 10.13799/j.cnki.mdjyxyxb.2025.03.013 http://dx.doi.org/10.13799/j.cnki.mdjyxyxb.2025.03.013 .

俞立军 . WB公司数据标注业务外包质量提升对策研究 [D ] . 呼和浩特：内蒙古工业大学， 2024 .

刘江，聂秋实，王星月，等 . 生成式人工智能背景下的弱监督学习教学法探索 [J ] . 计算机教育， 2025 （ 1 ）： 198 - 203， 209 . DOI: 10.3969/j.issn.1672-5913.2025.01.043 http://dx.doi.org/10.3969/j.issn.1672-5913.2025.01.043 .

张绍衡，李晓丹，谢玥，等 . AI赋能的循证决策培训：内科规培智能化路径探索 [J ] . 中国医学教育技术， 2025 ， 39 （ 6 ）： 694 - 697， 704 . DOI: 10.13566/j.cnki.cmet.cn61-1317/g4.202506002 http://dx.doi.org/10.13566/j.cnki.cmet.cn61-1317/g4.202506002 .

徐赫，焦洋 . 人工智能在临床推理教学新模式中的研究及展望 [J ] . 中华医学教育探索杂志， 2025 ， 24 （ 10 ）： 1372 - 1378 . DOI: 10.3760/cma.j.cn116021-20250310-02066 http://dx.doi.org/10.3760/cma.j.cn116021-20250310-02066 .

谭思创，胡琪康，王彬，等 . 基于研究生视角的AI对医学作用初探 [J ] . 中国医学教育技术， 2025 ， 39 （ 6 ）： 790 - 795 . DOI: 10.13566/j.cnki.cmet.cn61-1317/g4.202506017 http://dx.doi.org/10.13566/j.cnki.cmet.cn61-1317/g4.202506017 .

刘江，章晓庆 . 面向非计算机专业的人工智能导论课程建设与探索 [J ] . 中国大学教学， 2022 （ 1 ）： 46 - 51 . DOI: 10.3969/j.issn.1005-0450.2022.01.009 http://dx.doi.org/10.3969/j.issn.1005-0450.2022.01.009 .

张传钊，陈月，王俊江，等 . ChatGPT在肝胆胰腺外科临床实习教学中的应用 [J ] . 医学教育研究与实践， 2025 ， 33 （ 4 ）： 571 - 577 . DOI: 10.13555/j.cnki.c.m.e.2025.04.017 http://dx.doi.org/10.13555/j.cnki.c.m.e.2025.04.017 .

Hoover A ， Kouznetsova V ， Goldbaum M . STARE: structured analysis of the retina dataset [D/OL ] . [ 2025-07-30 ] . https://cecas.clemson.edu/~ahoover/stare/ https://cecas.clemson.edu/~ahoover/stare/ .

Staal J , Abramoff MD , Niemeijer M , et al . Ridge-based vessel segmentation in color images of the retina [J ] . IEEE Trans Med Imag , 2004 , 23 ( 4 ): 501 - 509 . DOI: 10.1109/TMI.2004.825627 http://dx.doi.org/10.1109/TMI.2004.825627 .

Jin K , Huang X , Zhou J , et al . FIVES: a fundus image dataset for artificial intelligence based vessel segmentation [J ] . Sci Data , 2022 , 9 : 475 . DOI: 10.1038/s41597-022-01564-3 http://dx.doi.org/10.1038/s41597-022-01564-3 .

Zhang Y , Higashita R , Zeng L , et al . Online Bayesian approximation based uncertainty aware model for ophthalmic image segmentation [J ] . IEEE J Biomed Health Inform , 2026 , 30 ( 3 ): 2368 - 2380 . DOI: 10.1109/JBHI.2025.3593983 http://dx.doi.org/10.1109/JBHI.2025.3593983 .

Views

下载量

CSCD

Alert me when the article has been cited

提交

Tools

Publicity Resources

Artificial intelligence literacy and teaching application needs of clinical teachers: a survey from the perspectives of awareness and practical use

Practice and exploration of computer-based case simulations in pediatric clerkship teaching for eight-year clinical medicine program students

Latent class analysis of continuing medical education needs among medical staff in non-provincial capital municipal tertiary hospitals of Guangdong, China

Differences in the application of artificial intelligence tools in domestic and overseas scenarios among medical staff and their implications

Application of a guided notes-centered blended model in teaching the course of Digestive System

Related Author

Liu Tianhang

Zhang Yinglin

Liu Ziying

Niu Yanyan

Hu Lingxi

Wang Xingyue

Related Institution

No data

⁰