磁共振T_(2)WI纹理分析对前列腺癌雄激素剥夺治疗疗效评估的临床价值

MR T_(2)WI texture analysis for the evaluation of androgen deprivation therapy in prostate cancer

目的:探讨磁共振(MR)T_(2)WI纹理分析对前列腺癌雄激素剥夺治疗(ADT)后的临床指导及鉴别治疗后残存病灶与周围良性组织的价值。方法:回顾性分析2018年1月-2019年5月经本院穿刺病理证实为前列腺癌并进行7个月ADT治疗后的患者的病例资料。治疗后患者按照PSA水平及病理结果分组,所有患者行常规T_(1)WI、T_(2)WI和DWI扫描。采用ITK-SNAP软件在T_(2)WI上手动逐层勾画ROI采用GE公司的AK软件提取308个纹理特征。采用组内相关系数(ICC)评估特征可重复性,独立样本t检验或Mann-Whitney U检验筛选出在每组间差异有统计学意义的纹理特征,采用10倍交叉验证法及Lasso回归模型对特征进行进一步的筛选和建模,多因素逻辑回归模型最终构建三个机器学习模型,采用ROC曲线及决策曲线(DCA)评估模型诊断效能。结果:纳入ADT治疗后疗效差患者23例,疗效好患者20例。筛选出Correlationangle 135 offset4、Haralick Correlation All Directionoffset4SD、Elongation、Low Intensity LargeArea Emphasis这4个特征构建模型一,构建的模型一评估前列腺癌ADT治疗后是否存在病灶的ROC曲线下面积(AUC)为0.87,敏感度为0.739,特异度为0.75。筛选出2个纹理特征VoxelValue Sum、Long Run Emphasisangle45offset1构建模型二,鉴别癌灶与中央腺的AUC为0.91,敏感度为0.81,特异度为1。筛选出GLCMEntropy All Directionoffset7SD、Long Run Emphasisangle 135 offset4、Long Run High GreyLevel Emphasis All Directionoffset4SDNULLADC这3个特征构建模型三,鉴别癌灶与外周带的AUC为0.87,敏感度为0.952,特异度为0.67。结论:MR T_(2)WI纹理分析可以对前列腺癌ADT治疗后的不同疗效进行评估以指导下一步临床治疗,同时MR T_(2)WI纹理特征可以鉴别ADT治疗后残存病灶与周围良性组织。

Objective:The value of MR T_(2)WI texture analysis in guiding clinical study after androgen deprivation therapy for prostate cancer and differentiating residual lesions from surrounding benign tissues after treatment.Methods:The patients with prostate cancer confirmed by puncture pathology in our hospital from January 2017 to December 2018 were retrospectively analyzed after 7 months of ADT treatment.After treatment,patients were divided into groups according to PSA level and pathological results.All patients were scanned by routine T_(1)WI,T_(2)WI and DWI sequences.ITK-SNAP software was used to manually sketch ROI layer by layer on T_(2)WI.The AK software of GE company was used to extract 308 texture features.Intra-group correlation coefficient(ICC)was used to evaluate feature repeatability,and independent sample t-test or Mann-Whitney U test was used to screen out texture features with statistically significant differences between groups,and 10-fold cross-validation method and Lasso regression model were used to further screen and model the features.Finally,three machine learning models were constructed by multi-factor logical regression model,and the diagnostic efficiency of the model was evaluated by ROC curve and decision curve(DCA).Results:There were 23 patients with poor curative effect and 20 patients with good curative effect after ADT treatment.The four characteristics of Correlationangle135offset4,Haralick CorrelationAllDirectionoffset4SD,elongation and low intensity large area emphasis were selected to construct model one.The area under the ROC curve of the model one was 0.87.The sensitivity was 0.739 and the specificity was 0.75 to evaluate whether there were lesions in prostate cancer after ADT treatment.Two texture features voxel value sum and LongRun Emphasisangle45offset1 were selected to construct model 2.The area under the ROC curve was 0.91.The sensitivity was 0.81 and the specificity was 1.The three features of GLCM Entropy All irectionoffset7SD,Long un mphasisangle 135 offset4,Long Run High Grey Level Emphasis All Directionoffset4 SDNULLADC were selected to construct model 3.The area under the ROC curve,sensitivity and specificity for distinguishing cancer from peripheral zone were 0.87,0.952 and 0.67 respectively.Conclusion:MR T_(2)WI texture analysis can be used to evaluate the different efficacy of ADT treatment for prostate cancer to guide the next clinical treatment.MR T_(2)WI texture features can be used to differentiate residual lesions from surrounding benign tissues after ADT treatment.