双参数纹理分析结合机器学习在高级别前列腺癌中的诊断价值  

Diagnosis of Bi-Parametric Texture Analysis Combined with Machine Learning for High-Grade Prostate Cancer

作　　者：肖建明[1] 牛翔科 王娜[1] 陈志凡[1] 王宗勇[1] 彭涛[1] XIAO Jianming;NIU Xiangke;WANG Na;CHEN Zhifan;WANG Zongyong;PENG Tao(Department of Radiology,Affiliated Hospital of Chengdu University,Chengdu 610081,China)

机构地区：[1]成都大学附属医院放射科,四川成都610081

出　　处：《中国医学影像学杂志》2021年第2期177-180,共4页Chinese Journal of Medical Imaging

基　　金：2018年四川省卫生和计划生育委员会科研课题(18PJ150);2017年四川省卫生和计划生育委员会科研课题(17PJ430);2018年成都市卫生和计划生育委员会科研课题(2018055)。

摘　　要：目的对比MR双参数与多参数成像纹理及定量分析结合机器学习对高级别前列腺癌的诊断价值。资料与方法回顾性分析疑似前列腺癌并行前列腺多参数MRI和经直肠超声引导穿刺活检取得病理结果的194例患者,采用Omni-Kinetics软件分别在T2WI、表观扩散系数(ADC)、T1加权动态磁敏感增强(T1WI_DSC)序列勾画病灶所在全部层面兴趣区,提取病变区纹理及定量特征数据后采用多因素Logistic回归分析。应用受试者工作特征(ROC)曲线评价双参数(T2WI+ADC)与多参数(T2WI+ADC+T1WI_DSC)诊断高级别前列腺癌的差异。结果双参数诊断高级别前列腺癌的敏感度为82.61%,特异度为86.11%,准确度为84.75%;多参数诊断高级别前列腺癌的敏感度为86.97%,特异度为86.11%,准确度为86.44%。两验证组ROC曲线下面积差异无统计学意义(0.918比0.946,P=0.077)。结论MR双参数纹理分析结合机器学习诊断高级别前列腺癌有较高的准确性。Purpose To compare bi-parametric and multi-parametric MRI texture and quantitative analysis based on machine learning in detecting high-grade prostate cancer(HGPCa).Materials and Methods 194 patients suspected as prostate cancer underwent multiparameter MRI,and histology of all lesions was confirmed by transrectal ultrasound-guided biopsy.T2WI,apparent diffusion coefficients(ADC)and T1WI dynamic susceptibility contrast-enhanced(T1WI_DSC)including texture features and quantitative parameters were generated by the Omni-Kinetics software,the texture and quantitative features were extracted based on all lesions.Multi-Logistic regression models based on bi-parametric(T2WI+ADC)and multi-parametric(T2WI+ADC+T1WI_DSC)was used to predict HGPCa respectively.The area under the receiver operating characteristic(ROC)curve was used to evaluate each predictor.Results The diagnostic accuracy of bi-parametric MRI for detecting HGPCa showed a sensitivity of 82.61%,specificity of 86.11%,accuracy of 84.75%.The diagnostic accuracy of multi-parametric MRI for detecting HGPCa showed a sensitivity of 86.97%,specificity of 86.11%,accuracy of 86.44%.No significant difference of area under ROC curve was observed between validation sets(0.918 vs.0.946,P=0.077).Conclusion Biparametric MR texture and quantitative analysis based on machine learning show a good diagnostic performance for the detection of HGPCa.

关 键 词：前列腺肿瘤 磁共振成像 扩散加权成像 表观扩散系数 纹理分析 最少绝对收缩和选择算子 

分 类 号：R445.2[医药卫生—影像医学与核医学] R737.25[医药卫生—诊断学]