目的 探索并验证基于华为云ModelArts平台构建的深度学习模型在宫颈液基细胞学(liquid-based cytology,LBC)非典型细胞诊断中的应用价值,并评估其对不同诊断经验医师的辅助效果.方法 回顾性分析2020年东莞市人民医院1 044例宫颈脱落细胞学标本,采用华为云ModelArts平台开发的人工智能(artifical intelligence,AI)辅助诊断系统与初级、中级、高级医师进行诊断比对,计算灵敏度、特异度、精确率、符合率、曲线下面积(area under the curve,AUC)等指标,评估AI系统的诊断效能及其对不同年资医师的辅助诊断效果.采用McNemar检验比较AI系统与人工诊断的差异.结果 在1 044例宫颈脱落细胞学标本中,AI系统在非典型细胞检出的灵敏度和特异度分别为98.96％和89.15％,均高于初级医师(81.95％和91.81％).AI系统的总体诊断精确率为93.68％,显著高于初级医师(87.26％,P＜0.001).AI辅助可显著提高初级医师的诊断性能,灵敏度从80.1％提升至96.5％,特异度从85.6％提升至92.3％.结论 本研究构建的AI辅助宫颈细胞学诊断系统性能优越,尤其能显著提高初级医师的诊断水平,具有良好的临床应用前景.
Application of Huawei Cloud ModelArts-driven AI-assisted diagnostic system in detecting atypical cervical cytology
Objective To explore and validate the application value of a deep learning model based on the Huawei Cloud ModelArts platform in the diagnosis of atypical cervical cells in liquid-based cytology(LBC)and to evaluate its assistive effect for pathologists with different diagnostic experiences.Methods We retrospectively analyzed 1 044 cervical cytology specimens from Dongguan People's Hospital in 2020.The artifical intelligence(AI)-assisted diagnostic system developed on the Huawei Cloud ModelArts platform was compared with junior,intermediate,and senior pathologists for diagnosis.Sensitivity,specificity,precision,recall,and area under the receiver operating characteristic curve(AUC)were calculated to assess the diagnostic performance of the Al system and its assistive effect for pathologists with different levels of experience.The McNemar test was used to compare the differences between the Al system and manual diagnosis.P＜0.05 was considered statistically significant.Results For the 1 044 cervical exfoliated cytology specimens,the sensitivity and specificity of the AI system in detecting atypical cells was 98.96％and 89.15％,both of which were higher than those of junior doctors(81.95％and 91.81％,respectively).The overall diagnostic accuracy of the Al system was 93.68％,which was significantly higher than that of junior doctors(87.26％,P＜0.001).Al assistance could significantly improve junior doctors'ability to detect atypical cells,with the sensitivity and specificity increasing from 80.1％to 96.5％and from 85.6％to 92.3％,respectively.Conclusion The AI-assisted cervical cytology diagnostic system developed in this study demonstrated superior performance,particularly in significantly improving the diagnostic level of junior pathologists,showing promising clinical application prospects.
万方数据
维普
