18F-FDG PET/CT影像组学预测胃癌患者HER-2表达状态研究

Study on the Prediction of HER-2 Expression Status in Gastric Cancer Patients using 18 F-FDG PET/CT Radiomics

目的 探讨联合18F-FDG PET/CT常规代谢参数和影像组学特征构建的模型在预测患者胃癌HER-2(human epidermal grouth factor receptor 2)表达状态中的应用价值。方法 回顾性分析兰州大学第二医院和甘肃省人民医院经病理证实为胃癌的110例患者,治疗前均行18F-FDG PET/CT检查。将兰州大学第二医院的病例以7∶3随机分为训练集(n=61)和内部验证集(n=26),以甘肃省人民医院数据(n=23)为外部验证集。在训练集中,采用LASSO回归和10折交叉算法筛选PET常规参数和影像组学特征,通过Logistic回归分别建立PET模型、组学模型及组合模型。应用受试者工作特征曲线(receiver operating characteristic, ROC)及DeLong检验评价和比较模型的预测效能,通过决策曲线分析模型的临床价值,校准曲线观察拟合效果。结果基于3、2、5个特征分别构建PET模型、组学模型和组合模型。ROC曲线显示3种模型均具有良好的预测效能,DeLong检验显示,训练集中,组合模型和组学模型差异有统计学意义(P<0.05),而PET模型和组学模型差异无统计学意义(P>0.05);内、外部验证集中,3种模型差异均无统计学意义(P>0.05)。校准曲线显示各模型均具有良好拟合效果,临床决策曲线显示组合模型较另两种模型具有较高的临床实用价值。结论18F-FDG PET/CT影像组学和临床特征联合对胃癌患者的HER-2表达状态具有良好的预测价值。

Objective To assess the predictive value of the model combined conventional metabolic parameters and radiomics features from 18 F-FDG PET/CT for predicting HER-2 expression status in gastric cancer.Methods A retrospective analysis included 110 gastric cancer patients from The Second Hospital Lanzhou University and The Gansu Provincial People′s Hospital.All patients underwent 18 F-FDG PET/CT before treatment.The Lanzhou cases were randomly divided into a 7∶3 training set(n=61)and an internal validation set(n=26),while data from Gansu Provincial People′s Hospital(n=23)served as an external validation set.LASSO regression and 10-fold cross-validation were employed to select PET parameters and radiomics features in the training set.Logistic regression was used to create PET,radiomics,and combined model.Evaluation included ROC curves and the DeLong test for model comparison.Clinical utility was assessed using decision curve analysis,and model consistency was observed through calibration curves.Results Models based on 3,2 and 5 selected features for the PET,radiomics,and combined model.ROC curves demonstrated strong predictive performance for all models.The DeLong test showed a significant difference between the combined model and radiomics model in the training set(P<0.05),with no statistical difference between the PET model and radiomics model(P>0.05).Internal and external validation sets showed no statistical differences among the three models(P>0.05).Calibration curves indicated good fitting effects for each model,and decision curve analysis revealed higher clinical utility for the combined model compared to the other two models.Conclusion The combined model provides a robust predictive tool for determining HER-2 expression status in gastric cancer patients.