基于病变内部及周围MRI影像组学特征预测临床显著性前列腺癌

目的探究基于病变内部及病变周围MRI影像组学模型预测临床显著性前列腺癌的价值。材料与方法140例(训练集112例,测试集28例)进行过前列腺磁共振扫描的患者纳入研究。手动勾画T2加权成像(T2 weighted imaging,T2WI)、ADC图像病变内部区域(intralesional volume,ILV)及病变周围区域(perilesional volume)并分别提取影像组学特征,构建最小绝对值收敛和选择算子(least absolute shrinkage and selection operator,LASSO)回归预测模型。运用ROC曲线分析、临床决策曲线分析对模型进行评估。结果训练集的AUC、准确率分别为:0.93(95%置信区间:0.88~0.98,特异度:0.87,敏感度:0.89)、0.84(95%置信区间:0.76~0.90)。测试集的AUC、准确率分别为0.92(95%置信区间:0.81~1.00,特异度:0.95,敏感度:0.68)、0.89(95%置信区间:0.72~0.98)。临床决策曲线分析中诊断阈值位于0.01~0.83或0.87~0.98时,运用预测模型患者具有良好的净受益。结论基于病变内部及病变周围MRI影像组学特征对于临床显著性前列腺癌的预测具有应用价值。

Predicting clinically significant prostate cancer based on perilesional and intralesional MRI radiomics features

Objective:To explore value of perilesional volume(ILV)and intralesional volume(PLV)MRI radiomics features for the diagnosis of clinically significant prostate cancer(csPCa).Materials and Methods:One hundred and forty patients(train set:112,testing set:28)who underwent prostate MRI examination were included in that retrospective study.ILV and PLV were manually segmented on T2WI(T2 weighted imaging),ADC map and radiomics features were extracted.Radiomics features were selected via univariate analysis and least absolute shrinkage and selection operator(LASSO)combined with 10-fold cross validation.Prediction model was built based on LASSO regression and evaluated by ROC curve analysis,decision curve analysis(DCA).Results:AUC and accuracy of model in train set were 0.93(95%CI:0.88—0.98,specificity:0.87,sensitivity:0.89),0.84(95%CI:0.76—0.90),respectively.AUC and accuracy of model in test set were 0.92(95%CI:0.81—1,specificity:0.95,sensitivity:0.69),0.89(95%CI:0.72—0.98),respectively.Decision curve showed that if the cut-off point is between 0.01 and 0.83 or between 0.87 and 0.98,using that model has more net benefit than either the“positive-all”model or the“negative-all”model.Conclusions:ILV and PLV based MRI radiomics features is valuable for diagnosis of csPCa.