建立基于MRI-T2WI的影像组学模型预测肝细胞癌Ki-67的表达水平

Development of MRI-TWI-Radiomics Model to Predict Ki-67 Expression Level in Hepatocellular Carcinoma

目的通过筛选出基于MRI-T2WI图像的最优影像组学标签,建立预测肝细胞癌(HCC)瘤体中细胞增殖抗原标记物(Ki-67)表达程度的预测模型。方法搜集行肝部MRI影像学检查并经术后病理证实为HCC、行Ki-67表达水平检测的108例患者。并将所有病例随机分组为训练集(80%)、测试集(20%)。经人工勾画出瘤体影像边界、提取感兴趣容积(VOI)的纹理特征参数。基于Radiomics cloud platform(慧影医疗)软件,先后通过移除低方差的特征、单变量特征选择、LASSO算法三次数据筛选分析,提取训练集最优影像组学特征。再利用所提取影像组学特征,经由逻辑式回归(LR)及支持向量机(SVM)两种分类器建立影像组学-Ki-67表达预测模型,通过建立训练集与测试集受试者工作特征(ROC)曲线评估预测模型的可行性。结果入组的108例患者按Ki-67阳性率≥50%分组,经由三次数据分析筛选出最优影像组学特征7个,并利用所得最优影像组学标签建立预测模型,以Ki-67≥50%分组时两种分类器所得测试集ROC曲线下面积(AUC)为0.793、0.871。结论HCC MRI-T2WI影像组学特征与瘤体Ki-67的表达具有相关性,可作为临床术前无创性预测HCC Ki-67表达的手段。

Objective To develop a radiomics model based onthe Optimal Radiomics Features which are produced from MRI-T2weighted imaging(T2WI)to predict expression level of Ki-67 in hepatocellular carcinoma(HCC).Methods A total of 108 patients with pathologically confirmed HCC who had abdominal MRI before operation and Ki-67 expression level test after resection were retrospectively collected.All cases were randomly dividedinto training set(80%)and test set(20%).The ROIs of HCC were drawn manually by a radiologist,and the texture feature parameters of volume of interest(VOI)were extracted.Based on the Radiomics cloud platform software(HuiY ing),we firstly performed data analysis three times by Removing Features with Low Variance,SelectK Best,and LASSO to extract the Optimal Radiomics Features form the training Set.Secondly using the OptimalR adiomicsfeatures,the Logistic Regression(LR)and support vector machine(SVM)classifiers used to establish a Radiomics-Ki-67 expression prediction model.The receiver operation characteristic curve(ROC)of the test set and the training set was used for evaluating the feasibility of the prediction model.Results108 patients were classified into two groups according to Ki-67 positive rates≥50%.Through three times data analysis 7 Optimal Radiomics Features were selected.Using the Optimal Radiomics features to develop a prediction model,we found that the area under the ROC curve(AUC)of the test set obtained by the two classifiers was 0.793 and 0.871,when grouped with Ki-67≥50%.Conclusion Radiomics features based on MRI-T2 WI images is correlated with the expression of tumor Ki-67 in hepatocellular carcinoma.Radiomics analysiscan be used as a non-invasive predictive method for the expression of Ki-67 in hepatocellular carcinoma.