基于DCE-MRI定量参数图影像组学模型预测子宫内膜癌Ki-67表达

Value of Radiomics Based on DCE-MRI in Predicting of Ki-67 Expression in Endometrial Cancer

目的探讨基于动态增强磁共振成像(DCE-MRI)定量参数图影像组学模型对术前子宫内膜癌Ki-67表达水平的预测价值。方法回顾性搜集术前行DCE-MRI检查子宫内膜癌患者99例。根据Ki-67表达水平分为高表达组(42例)和低表达组(57例),按照样本量7∶3随机分为训练集(69例)和验证集(30例)。使用Omni-Ki-netics软件在DCE-MRI定量参数图上共提取201个影像组学特征,同时获取患者的独立特征。用Lasso回归对影像组学特征进行降维筛选和影像组学标签建立,采用K-S检验、t检验或卡方检验筛选有统计学差异的独立特征。Logistic回归用建立基于影像组学特征、独立特征和两者联合的诊断模型,采用受试者工作特征曲线下面积评估各模型预测效能,Delong检验用于比较各模型的预测效能。结果最后筛选出7个影像组学特征和5个独立特征来构建模型。在训练集和验证集,联合模型和影像组学模型的预测效能均显著优于独立特征模型(P值均<0.05)。结论基于DCE-MRI定量参数图构建的影像组学模型对术前预测子宫内膜癌Ki-67表达情况具有较高效能。

Objective To investigate the preoperative predictive value of radiomics based on quantitative parameter maps of dynamic contrast-enhanced magnetic resonance image(DCE-MRI)in the expression of Ki-67 in endometrial cancer.Methods 99 endometrial cancer patients undertaken DCE-MRI were collected retrospectively.According to the expression level of Ki-67,42 cases were divided into high expression group and 57 cases were divided into low expression group.According to the sample size of 7∶3,they were randomly divided into training set(69 cases)and validation set(30 cases).Omni-Kinetics software was used to extract total 201 radiomics from DCE-MRI quantitative parameter maps,and obtain the independent features of patients at the same time.Lasso regression was used to select radiomics and calculate radscore.K-S test,t test or chi square test were used to select the independent features with statistical differences.Logistic regression was used to establish diagnostic models based on radiomics,independent features and the combination of them.The area under the receiver operating characteristic(ROC)curve(AUC)was used to evaluate the predictive efficiency of each model,and the Delong test was used to compare the efficiency of each model.Results Finally,7 Radiomics and 5 independent features were selected to establish the models.In the training set and validation set,the predictive efficiency of the combined model and radiomics model were significantly better than the independent features model(P<0.05).Conclusion Radiomics based on DCE-MRI quantitative parameter maps have high efficiency in preoperative predicting the expression of Ki-67 in endometrial cancer.