基于前列腺病变周围区域的MRI影像组学特征对临床显著性前列腺癌的诊断价值

The Value of Prostate Perilesional MRI Radiomics Features for Diagnosis of Clinically Significant Prostate Cancer

目的探究前列腺病变周围区域的MRI影像组学特征对诊断临床显著性前列腺癌(csPCa)的价值。方法回顾性分析132例(训练集99例、测试集33例)进行过前列腺MRI扫描且前列腺内病变类型经病理确诊患者的T2WI、表观扩散系数(ADC)、动态增强磁共振成像(DCE-MRI)的图像。运用3D slicer软件勾画病变周围5 mm感兴趣区(ROI)并提取影像组学特征,通过单变量分析、组内相关系数(ICC)、最小绝对值收敛和选择算子(LASSO)降维并运用LASSO建立预测模型,通过受试者工作特征(ROC)曲线评价模型诊断效能。运用测试集患者数据对模型进行验证。结果测试集中ROC曲线的曲线下面积(AUC)为0.861(95%置信区间:0.7361~0.9861),特异度及敏感度分别为1.0、0.667。结论基于前列腺病变周围区域的MRI影像组学特征对于诊断csPCa具有应用价值。

Objective To evaluate the value of prostate perilesional MRI radiomics features for diagnosis of clinically significant prostate cancer(csPCa).Methods Patients undergoing prostate MRI scanning and biopsy or prostatectomywere retrospectively analyzed.One hundred and thirty-two eligible patients(99 in training cohort,33 in test cohort)were included.Region of interest(ROI)of 5 mm around prostate MRI lesionswas segmented on T2 weighted imaging(T2 WI),apparent diffusion coefficient(ADC)map and dynamic contrast-enhanced magnetic resonance imaging(DCE-MRI)using 3 D slicer softwareandradiomics features were extracted.Dimensionality reduction of radiomics featuresin the training cohort was performed usingintraclass correlation coefficients(ICC),univariate analysis and least absolute shrinkage and selectionoperator(LASSO)thenradiomics prediction models were constructed.Diagnostic performance of the prediction model was evaluated through receiver operating characteristic(ROC)curve and quantified by area under curve(AUC).Independent validation was performed using test cohort data and ROC analysis.AUC was used to assess the discrimination ability of the prediction model.Results The AUCof ROC curve in test cohort was0.861(95%CI:0.7361-0.9861)(specifity:1,sensitivity:0.667).Conclusion Prostate perilesional MRI radiomics features were valuable to the diagnosis of csPCa.