医学影像学杂志

目的 探讨基于表观弥散系数(ADC)、动态对比增强磁共振(DCE-MRI)定量参数直方图特征联合临床指标预测前列腺癌患者内分泌治疗2年内进展至去势抵抗性前列腺癌(CRPC)的价值。方法 选取172例前列腺癌患者，按照7∶3分为训练集120例和内部测试集52例；另选取65例前列腺癌患者作为外部验证集。根据内分泌治疗2年内是否进展为CRPC将患者分为CRPC组和非CRPC组，172例患者中非CRPC组98例、CRPC组74例，65例患者中非CRPC组43例、CRPC组22例。通过Siemens Syngo. via工作站获得定量参数K^trans、K_ep、V_e伪彩图，在3D Slicer软件中参考轴位T_2WI序列在ADC、K^trans、K_ep、V_e伪彩图上逐层勾画全前列腺腺体感兴趣区，并提取一阶直方图特征，经LASSO降维筛选出9个特征，计算组学评分(Rad-score)；采用单因素及后向多因素Logistic回归分析发生CRPC的独立预测因素，在训练集、内部测试集、外部验证集中构建临床模型、Rad-score模型及联合模型，通过ROC曲线、校准曲线和决策曲线分析评价模型的效能，并通过净重新分类指数(NRI)评估联合模型与单独临床模型、Rad-score模型的增益价值，之后基于多因素分析的独立预测因素绘制列线图。结果 单因素Logistic回归分析显示，训练集、内部测试集和外部验证集中前列腺特异性抗原(PSA)、前列腺特异性抗原密度(PSAD)、T分期、骨转移、前列腺癌包膜外侵犯评分(ESUR-EPE)、Rad-score差异均有统计学意义(P均<0.001)；后向多因素Logistic回归分析结果显示，PSA(OR=0.979,95%CI:0.965～0.994,P=0.006)、PSAD(OR=0.261,95%CI:0.163～0.996,P<0.019)、ESUR-EPE(OR=2.1,95%CI:1.39～3.173,P<0.001)、Rad-score(OR=12.839,95%CI:5.822～28.315,P<0.001)是前列腺癌内分泌治疗2年内进展至CRPC的独立预测因素。临床模型在训练集、内部测试集及外部验证集的ROC曲线下面积(AUC)分别为0.860、0.799、0.768;Radscore模型在训练集、内部测试集及外部验证集的AUC为0.895、0.893、0.852；联合模型在训练集、内部测试集及外部验证集的AUC为0.945、0.927、0.879；联合模型与临床模型、直方图模型比较NRI均在0.37以上，决策曲线和校准曲线显示联合模型具有良好的稳定性和临床应用价值。结论 基于ADC、DCE-MRI定量参数的一阶直方图特征构建的Radscore是CRPC独立预测因素，临床模型、Rad-score模型联合在预测前列腺癌内分泌治疗2年内进展为CRPC中具有较好的价值。

Objective To explore the value of combining histogram features of apparent diffusion coefficient(ADC), dynamic contrast-enhanced magnetic resonance imaging(DCE-MRI) quantitative parameters with clinical indicators in predicting the occurrence of castration-resistant prostate cancer(CRPC) in prostate cancer patients within 2 years. Methods A total of 172 prostate cancer patients were selected, divided into a training set of 120 cases and an internal test set of 52 cases in a 7∶3 ratio; additionally, 65 prostate cancer patients were chosen as an external validation set. According to whether they developed CRPC within two years, the patients were classified into the CRPC group and the non-CRPC group. 172 patients were 98 cases in the non-CRPC group and 74 cases in the CRPC group. 65 patients were 43 cases in the non-CRPC group and 22 cases in the CRPC group. Quantitative parameters K^trans, K_ep, and V_e pseudocolors were obtained using the Siemens Syngo.via workstation. Using the 3D Slicer software, regions of interest(ROI) for the entire prostate gland were outlined slice by slice in the ADC, K^trans, K_ep, and V_e pseudocolors with reference to the axial T₂ WI sequence, and first-order histogram features were extracted. Through LASSO dimensionality reduction, nine features were selected to calculate the radiomics score(Rad-score). Univariate and backward multivariate logistic regression analyses were performed to identify independent predictors of CRPC. Clinical, Rad-score, and combined models were constructed in the training set, internal test set, and external validation set. The performance of the models was evaluated using receiver operating characteristic(ROC) curves, calibration curves, and decision curves, and the net reclassification improvement(NRI) was used to assess the gain value of the combined model over the clinical and Rad-score models alone. Finally, nomograms were plotted based on the independent predictors identified in the multivariate analysis. Results Univariate logistic regression analysis showed that the differences in PSA, PSAD, T staging, bone metastasis, ESUREPE, and Rad-score in the training set, internal test set, and external validation set were statistically significant(P<0.001). The results of the backward multivariate logistic regression analysis indicated that PSA(OR=0.979, 95% CI: 0.965-0.994, P=0.006), PSAD(OR=0.261, 95% CI: 0.163-0.996, P<0.019), ESUR-EPE(OR=2.1, 95% CI: 1.39-3.173, P<0.001), and Rad-score(OR=12.839, 95% CI: 5.822-28.315, P<0.001) were independent predictors of CRPC. The area under the ROC curve(AUC) for the clinical model in the training set, internal test set, and external validation set was 0.860, 0.799, and 0.768, respectively. The AUC for the Rad-score model in the training set, internal test set, and external validation set was 0.895, 0.893, and 0.852, respectively. The AUC for the combined model in the training set, internal test set, and external validation set was 0.945, 0.927, and 0.879, respectively. The NRI values for the combined model compared to the clinical model and histogram model were all above 0.37. The decision curve and calibration curve demonstrated that the combined model had good stability and clinical applicability. Conclusion The Rad-score constructed based on the first-order histogram features of the quantitative parameters of ADC AND DCE-MRI is an independent predictor of CRPC, and the combined clinical and Rad-score model has a good value in predicting CRPC, which provides a new basis for clinical treatment decisions.