宫颈癌是威胁全球女性健康的一种恶性肿瘤，感染高危型人乳头瘤病毒是导致宫颈癌发病的直接原因。阴道镜检查是早期发现高级别上皮内瘤变的主要方法，采用适当的阴道镜检查方案，可以有效的预防宫颈癌的发生。然而，阴道镜的检查，非常依赖医生的经验，需要医生有大量操作和图像判读的临床实践经验。因此，为在我国实现世界卫生组织提出的消除宫颈癌的目标，就需要培养大批有经验的基层妇科医生。本文通过收集覆盖全国不同地区的多中心临床数据，通过组织临床专家结合临床信息和病理数据进行标注，构建阴道镜图像判读病例库。通过开发标注质控平台，实现了采用网页浏览器作为客户端的在线进行学习的阴道镜图像判读培训系统。该系统具有简单易用，电脑端和移动端均可使用的特点。通过招募有提升自身阴道镜图像判读水平意愿的医生进行培训，并在培训前后分别进行病例测试，验证了培训系统的效果。实验结果表明，通过使用该系统进行临床大样本病例数据的学习，受培训医生整体与3-10年临床经验医生的考试成绩相当。应用该培训系统，初步验证了短时间内通过大量临床数据训练，医生的临床诊断水平可得到提升的设想。该培训系统为提升临床阴道镜医生诊断水平、保证宫颈癌检查质量、更好地实现宫颈癌的早期防治，提供了新的临床学习渠道。
As a type of malignant tumor, cervical cancer, which directly caused by infection of high-risk human papillomavirus, threatens global female health. Colposcopy is the major method for early detection of high-grade epithelial neoplasia. Accurate colposcopy examination can effectively prevent the occurrence of cervical cancer. However, colposcopy hugely depends on the experience of doctors, which requires abundant experience of clinical practice and image interpretation. Therefore, large number of experienced community-level doctors are required in China, in order to achieve the goal of eliminating cervical cancer proposed by the World Health Organization.In this study, the multi-center clinical data of colposcopy with clinical information and pathological from different regions of China were collected and annotated by the clinical experts. A colposcopy image training system was developed on web browser for online learning by developing the annotation quality control platform. The system is convenient and easy to use on both computer and mobile devices.The doctors who were recruited to be trained by the developed colposcopy image training system, and the exam scores of the participated doctors were compared before and after training. The experimental result shows that the overall clinical assessment skill of doctors has been improved by using the system and learning from the big clinical data, the exam scores of these doctors could compare to the doctors who have 3-10 clinical experiences. In this study, the assumption that the clinical skill of colposcopy image annotation can be improved by training with a clinical big data in a short time was preliminary verified. The developed system provides a new learning platform to improve the diagnosis skill of doctors, which may further ensure the quality of colposcopy examination and realize the early prevention and intervention of cervical cancer.
