摘要
目的基于人工智能建立鉴别CT表现为混合磨玻璃结节(mGGN)的腺体前驱病变(PGL)与微浸润腺癌(MIA)的有效模型。方法回顾性分析温州医科大学附属第一医院2017年1月至2023年6月经手术病理证实且CT表现为mGGN的180例肺腺癌患者的临床和CT影像资料,包括PGL患者66例和MIA患者114例。采用完全随机法以8∶2的比例将患者分为训练集(n=144)和测试集(n=36)。使用AI软件(联影科研平台uRP)全自动提取CT图像中病灶的定量参数及影像组学特征;通过降维纳入组学的最明显相关特征,建立5种机器学习分类器,包括逻辑回归(LR)、支持向量机(SVM)、随机森林(RF)、高斯过程(GP)及决策树(DT),以训练集曲线下面积(AUC)最高的分类器作为最佳影像组学模型,并将其结果输出为影像组学评分(Rad-score)。将两组患者的临床信息、CT形态特征及定量数据纳入多因素logistic回归分析,筛选有效鉴别PGL和MIA的独立影响因素,并建立临床模型。最终基于Rad-score和临床危险因素构建综合预测模型。采用受试者工作特征(ROC)曲线的AUC、灵敏度、特异度和准确度评价三种模型的诊断性能。结果通过LASSO降维得到11个鉴别PGL与MIA的影像组学特征。在5种机器学习分类器中,GP具有最佳的诊断效能,其在训练集和测试集的AUC分别为0.865、0.762。单因素、多因素logistic回归分析进行临床特征筛选,使用mGGN的平均CT值、长短径平均值和实性部分长径构建的临床模型,得到训练集和测试集的AUC分别为0.870和0.794。综合预测模型表现出更优的诊断效能,在训练集中的AUC、灵敏度、特异度、准确度分别为0.948、81.1%、91.2%、87.5%;在测试集中的AUC、灵敏度、特异度、准确度分别为0.883、76.9%、91.3%、86.1%。结论基于人工智能对肺结节定量及组学特征分析建立的综合预测模型能够较好地鉴别CT表现为mGGN的PGL与MIA,可用于指导临床治疗决策。
Objective To establish an effective model for distinguishing glandular prodromal lesions(PGL)mixed with ground-glass nodules(mGGN)from minimally invasive adenocarcinoma(MIA)on CT based on artificial intelligence.Methods A retrospective analysis was conducted on the clinical and CT image data of 180 patients with lung adenocarcinoma confirmed by surgical pathology and with CT manifestations of mGGN in the First Affiliated Hospital of Wenzhou Medical University from January 2017 to June 2023,including 66 patients with PGL and 114 patients with MIA.Patients were divided into the training set(n=144)and the test set(n=36)in an 8∶2 ratio using a completely random method.The quantitative parameters and radiomics features of the lesions in CT images were automatically extracted using artificial intelligence software(United Imaging Research Platform uRP).By incorporating the most obvious correlation features of omics through dimensionality reduction,five machine learning classifiers were established,including logistic regression(LR),support vector machine(SVM),Random forest(RF),Gaussian process(GP),and Decision Tree(DT).The classifier with the training set highest area under the curve(AUC)was selected as the best radiomics model,and output the result as radiomics score(Rad-score).The clinical information,CT morphological characteristics and quantitative data of the two groups were included in the multivariate logistic regression analysis to screen the independent influencing factors for effectively differentiating PGL and MIA,and a clinical model was established.Finally,a comprehensive prediction model was constructed based on Rad-score and clinical risk factors.The diagnostic performance of the three models was evaluated by using the AUC,sensitivity,specificity and accuracy of receiver operating characteristic(ROC)curve.Results Eleven radiomics features for distinguishing PGL from MIA were obtained through LASSO dimensionality reduction.Among the five machine learning classifiers,GP has the best diagnostic performance,with AUC of 0.865 in the training set and 0.762 in the test set,respectively.Univariate and multivariate logistic regression analyses were used for clinical feature screening.The clinical model was constructed by using the average CT value,average long and short diameter,and solid partial long diameter of mGGN,and the AUCs of the training set and the test set were 0.870 and 0.794,respectively.The comprehensive prediction model demonstrated superior diagnostic performance,with AUC,sensitivity,specificity,and accuracy in the training set being 0.948,81.1%,91.2% and 87.5% respectively,while 0.883,76.9%,91.3% and 86.1% respectively in the test set.Conclusion The comprehensive prediction model established based on the quantitative and omics feature analysis of pulmonary nodules by artificial intelligence can well distinguish mGGN mixed with PGL from MIA on CT,and can be used to guide clinical treatment decisions.
作者
陈永华
陈坚
林了一
陈聪
刘瑾瑾
孙厚长
杨运俊
傅钢泽
CHEN Yonghua;CHEN Jian;LIN Liaoyi;CHEN Cong;LIU Jinjin;SUN Houzhang;YANG Yunjun;FU Gangze(Department of Radiology,the First Affiliated Hospital of Wenzhou Medical University,Wenzhou,Zhejiang 325000,China;Department of Radiology,Taishun County Hospital of Traditional Chinese Medicine,Wenzhou,Zhejiang 325500,China)
出处
《重庆医学》
2025年第8期1848-1853,共6页
Chongqing Medical Journal
基金
浙江省温州市基础性公益科研项目(Y2023537)。
关键词
肺腺癌
腺体前驱病变
微浸润腺癌
人工智能
影像组学
lung adenocarcinoma
proglandular disease
microinvasive adenocarcinoma
artificial intelligence
radiomics
