影像组学联合临床指标预测PI-RADS V2.13分病变中有临床意义前列腺癌

Radiomics Combined with Clinical Indicators Predicts Clinically Significant Prostate Cancer in PI-RADS V2.1 Category 3 lesions

目的构建并验证临床-影像组学模型对前列腺影像报告和数据系统(PI-RADS)3分病变中有临床意义前列腺癌(csPCa)的预测效能。方法回顾性搜集170例PI-RADS 3分病变患者的MRI图像。按照7∶3的比例将患者随机分配至训练集和测试集。提取每个病灶轴位T_(2)WI、DWI和ADC图像的影像组学特征。首先采用单因素回归分析筛选出与PI-RADS 3分病变明显相关的临床特征。然后采用最小冗余最大相关(mRMR)和最小绝对收缩和选择算法(LASSO)对影像组学特征进行筛选和降维,并计算影像组学评分(Radscore)。最后将临床特征及Radscore纳入多因素逻辑回归分析,建立3个预测模型:临床模型、影像组学模型以及临床-影像组学联合模型,并以诺模图的形式将联合模型直观展示。Delong检验比较联合模型与临床模型和影像组学模型的诊断效能。结果最终选取7个影像组学特征用于构建影像组学模型。Radscore在csPCa患者和非csPCa患者之间有显著差异(训练集:P<0.001;测试集:P=0.0035)。多因素分析显示Radscore、年龄、前列腺特异性抗原密度(PSAD)和直肠指检(DRE)阳性可作为鉴别csPCa的独立预测因子。测试集联合模型预测csPCa的曲线下面积(AUC)为0.85(95%CI:0.74~0.96),高于临床模型(AUC=0.79,95%CI:0.66~0.92,P=0.034)和影像组学模型(AUC=0.74,95%CI:0.60~0.88,P<0.001)。决策曲线分析(DCA)表明临床-影像组学模型能使患者获得更高的净收益。结论临床-影像组学模型能有效识别PI-RADS 3分病变中的csPCa,从而避免不必要的活检,提高患者的生存质量。

Objective To determine the predictive performance of the integrated model based on clinical factors and radiomics features for the accurate identification of clinically significant prostate cancer(csPCa)among PI-RADS 3 lesions.Methods A retrospective study of 170 patients who underwent 3.0 T MRI with PI-RADS 3 lesions.Patients were randomly divided into the training set and the testing set at a ratio of 7∶3.Radiomics features were extracted from axial T_(2)WI,DWI and ADC images of each patient.First,univariate regression analysis was used to screen out clinical features significantly associated with PI-RADS 3 lesions.Then,the Minimum redundancy maximum relevance(mRMR)and LASSO feature selection methods were used to screen and reduce the dimension of the radiomics features,and the feature subset with the least prediction deviation and the most relevant to the target task was obtained to caculate the radiomics score(Radscore).Finally,the clinical features and Radscore were included in the multivariate logistic regression analysis,and three prediction models were established:clinical model,radiomics model and clinical-radiomics combined model.The combined model was visually displayed in the form of normogram.The performance of the integrated model was compared with radiomics model and clinical model on testing set.Results N=7 radiomics features were selected and used for radiomics model construction producing a radiomic score(Radscore).Radscore was significantly different between the csPCa and the non-csPCa patients(training set:P<0.001;testing set:P=0.0035).Multivariable Logistic regression analysis showed that age,PSAD,DRE could be used as independent predictors for csPCa identification.The clinical-radiomics model produced the AUC in the testing set was 0.85(95%CI:0.74-0.96),which was higher than clinical model(AUC=0.79,95%CI:0.66-0.92,P=0.0034)and the radiomics model(AUC=0.74,95%CI:0.60-0.88,P<0.001).The decision curve analysis implies that the clinical-radiomics model could be beneficial in identify csPCa among PI-RADS 3 lesions.Conclusion The clinical-radiomics model could effectively identify csPCa among biparametric PI-RADS 3 lesions,and thus could help avoid unnecessary biopsy and improve the life quality of patients.