【摘要】 目的探讨磁共振T2WI的影像组学在前列腺癌预测中的价值,以期为临床应用提供依据。方法收集2015年6月-2019年8月经病理证实的122例前列腺癌及前列腺增生患者,应用MaZda软件对每例患者T2WI图像进行预处理并提取纹理特征参数,采用Fisher系数、分类错误概率与平均相关系数（Classification error probability and average correlation coefficients, POE+ACC）及互信息（Mutual information, MI）各筛选10个最具鉴别意义的纹理参数。原始数据分析（Raw data analysis, RDA）、主成分分析（Principal component analysis, PCA）、线性判别分析（Linear discriminant analysis, LDA）及非线性判别分析（Nonlinear discriminant analysis, NDA）计算每种降维方法的准确率、敏感度等,并分析各纹理参数后处理方式间的差异。结果在Fisher、POE+ACC及MI降维方法中,Horzl_GLevNonU和WavEnLL_s^-1均被筛选出。RDA、PCA、LDA及NDA四种判别方式中,Fisher/NDA法预测前列腺癌和前列腺增生的灵敏度最高（93.52%）,特异度、准确率最高为MI/NDA法、POE+ACC/NDA法（分别为95.05%、93.68%）。Fisher降维中判别方式PCA、LDA与NDA之间差异均具有统计学意义（P<0.05）,POE+ACC降维中RDA、PCA、LDA与NDA之间差异均具有统计学意义（P<0.05）,MI降维中RDA、PCA、LDA与NDA之间差异具有统计学意义（P<0.05）。结论基于磁共振T2WI的影像组学分析在前列腺癌的诊断、预测中具有一定价值,不同降维方法及判别方式能够影响影像组学的预测结果。更多还原

【Abstract】 Objective To investigate the value of radiomics based on T2 WI in the diagnosis of prostate cancer, and thus to provide the basis for clinical application. Methods A total of 122 patients with pathological-confirmed prostate cancer and prostatic hyperplasia were retrospectively collected from June 2015 to August 2019. The T2 WI images of each patient were pre-processed by MaZda software, and texture feature parameters were extracted. Fisher coefficient, classification error probability and average correlation coefficients（POE+ACC） and mutual information（MI） were used to screen 10 most significant texture parameters. Raw data analysis（RDA）, principal component analysis（PCA）, linear discriminant analysis（LDA）, nonlinear discriminant analysis（NDA） were used to calculate the accuracy and sensitivity of each dimensionality reduction method, and analyzed the differences among the post-processing methods of texture parameters. Results Horzl_GlevNonU and WavEnll_s^-1 were all screened out in three dimensionality reduction methods. Fisher/NDA has the highest sensitivity（93.52%）, and the highest specificity（95.05%） and accuracy（93.68%） were MI/NDA and POE+ACC/NDA in predicting prostate cancer and hyperplasia, respectively. There were significant differences between PCA, LDA and NDA in Fisher dimensionality reduction（P<0.05）. There were significant differences between RDA, PCA, LDA and NDA in POE+ACC dimensionality reduction（P<0.05）. There were significant differences between RDA, PCA, LDA and NDA in MI dimensionality reduction（P<0.05）.Conclusions The radiomics based on T2 WI has a certain value in the diagnosis and prediction of prostate cancer. Different dimensionality reduction and discrimination can influence the prediction results.更多还原