人工智能识别阴道镜下子宫颈红区在子宫颈癌前病变诊断中的价值  

Value of artificial intelligence-assisted identification of cervical red area under colposcope in the diagnosis of cervical precancerous lesions

作　　者：冯慧[1] 赵撼宇 赵健[1] 王业全 FENG Hui;ZHAO Han-yu;ZHAO Jian;WANG Ye-quan(Department of Obstetrics and Gynecology,Peking University First Hospital,Beijing 100034,China;不详)

机构地区：[1]北京大学第一医院妇产科,北京100034 [2]北京智源人工智能研究院,北京100000

出　　处：《中国实用妇科与产科杂志》2025年第3期357-360,共4页Chinese Journal of Practical Gynecology and Obstetrics

基　　金：国家重点研发计划(2020AA0105200)。

摘　　要：目的探讨人工智能辅助阴道镜识别子宫颈红区在诊断子宫颈癌前病变中的价值。方法收集2020年8月至2023年8月北京大学第一医院妇产科子宫颈诊疗中心的阴道镜数据3974例。采用ViT、DeiT3、BEiTv2三种深度学习模型,构建阴道镜下子宫颈红区预测模型,其中训练集与验证集以阴道镜拟诊标注作为学习标签,测试集以病理结果为金标准,以活检率(BR)及组织病理学符合率(CR)为评估指标。结果ViT模型活检率为40.9%、组织病理学符合率77.8%;DeiT3模型则分别为44.1%、77.6%;BEiTv2模型分别为35.8%、82.8%。结论人工智能识别阴道镜下子宫颈红区的ViT、DeiT3、BEiTv2三种深度学习模型均有较高的子宫颈癌前病变诊断效力。Objective To explore the value of artificial intelligence(AI)-assisted colposcopy for recognition of cervical red area in the diagnosis of cervical precancerous lesions.Methods Colposcopy data of 3974 cases were collected from the Cervical Diagnosis and Treatment Center of the Gynecology and Obstetrics Department of Peking University First Hospital from August 2020 to August 2023.Three deep learning models of ViT,DeiT3 and BEiTv2 were used to construct the prediction model of the cervical red area under colposcopy.The training set and validation set took the colposcopic diagnosis as the learning label,the test set took the pathological results as the gold standard,and the biopsy rate(BR)and histopathology coincidence rate(CR)were used as the evaluation indicators.Results The BR and CR of ViT model were 40.9%and 77.8%,respectively.The BR and CR of DeiT3 model were 44.1%and 77.6%,respectively.The BR and CR of BEiTv2 model were 35.8%and 82.8%,respectively.Conclusion The three deep learning models of AI for identifying the cervical red area under colposcope in this study all have high diagnostic efficacy of cervical precancerous lesions.

关 键 词：子宫颈癌前病变 人工智能 阴道镜 子宫颈红区 

分 类 号：R711.74[医药卫生—妇产科学]