基于MRI影像组学特征构建膀胱尿路上皮癌病理分级预测模型的价值

目的 探讨基于MRI影像组学特征构建膀胱尿路上皮癌病理分级预测模型的价值.方法 搜集经手术病理证实的100例膀胱尿路上皮癌患者,其中低级别尿路上皮癌(LGUC)28例和高级别尿路上皮癌(HGUC) 72例.通过随机分层抽样方法以7∶3的比例分为训练组及测试组.使用ITK-SNAP软件勾画T2WI、扩散加权成像(DWI)...

The Value of Prediction Model Based on MRI Radiomics in Evaluating the Histological Grade of Urothelial Carcinoma fo Bladder

Objective To explore the value of prediction model based on MRI radiomics in evaluating the histological grade of urothelial carcinoma of bladder. Methods A total of 100 patients with bladder urothelial carcinoma confirmed by postoperative pathology were retrospectively analyzed, including 28 cases of low-grade urothelial carcinoma(LGUC) and 72 cases of high-grade urothelial carcinoma(HGUC).All MRI data were divided into training and testing group by stratified sampling method with the ratio of 7∶3.The ITK-SNAP software was used to manually delineate the volume of interest(VOI) of tumor on T;WI,DWI and ADC maps, and then import A.K.software to extract radiomic features.The variance analysis, the minimal redundancy maximal relevance(mRMR) and the least absolute shrinkage and selection operator(LASSO) were used to select and reduce the dimension of the features. The Logistic regression algorithm was used to construct the predictive model and the receiver operating characteristic curve(ROC) was drawn to evaluate the performance of the model, and it was verified in the testing group. Results Four prediction models based on MRI radiomics were constructed: T;WI model, DWI model, ADC model and T;WI+DWI+ADC combined model.Among the prediction models based on single-sequence radiomics, the ADC model in training group and testing group has the highest AUC value in distinguishing LGUC from HGUC(AUC=0.825,0.818,respectively),which is higher than that of DWI model(AUC=0.794,0.750,respectively) and T;WI model(AUC=0.811,0.739,respectively).Compared with the prediction model based on single-sequence radiomics, the T;WI+DWI+ADC combined model had higher AUC value, and its sensitivity, specificity and AUC for distinguishing LGUC from HGUC were 100%,80.0% and 0.912 in the training group, and 77.3%,87.5% and 0.824 in the testing group. Conclusion The prediction model based on MRI radiomics has great application value in distinguishing LGUC and HGUC,and the T;WI+DWI+ADC combined mode shows higher performance than that based on single-sequence.