基于双参数MRI影像组学机器学习鉴别前列腺癌Gleason评分分级分组中的应用  

Application of Radiomics Machine Learning in Differentiating Gleason Grade Group for Prostate Cancer Based on Biparametric Magnetic Resonance Imaging

作　　者：孙于越 张濬韬 苏月婷 田为中[3] SUN Yu-yue;ZHANG Jun-tao;SU Yue-ting;TIAN Wei-zhong(Graduate School of Dalian Medical University,Dalian 116044,Liaoning Province,China;GE Healthcare,GE Healthcare PDX AA team,Shanghai 210000,China;Department of Imaging,Taizhou People's Hospital,Taizhou 225300,Jiangsu Province,China)

机构地区：[1]大连医科大学研究生院,辽宁大连116044 [2]通用电气医疗通用电器药业高级应用团队,上海210000 [3]泰州市人民医院影像科,江苏泰州225300

出　　处：《中国CT和MRI杂志》2023年第8期118-120,共3页Chinese Journal of CT and MRI

基　　金：泰州市科技支撑计划(社会发展)项目(TS201906)。

摘　　要：目的探讨基于双参数MRI影像组学机器学习模型对前列腺癌(PCa)Gleason评分分级分组中的价值。方法回顾性收集经病理证实为PCa的患者138例,其中GG>2患者83例,GG≤2患者55例。所有病人术前均进行MRI检查。按7∶3将患随机分为训练集和测试集,分别用于影像组学模型的机器学习和验证,采用RF、SVM和XGboost构建3组模型(ADC、T_(2)WI、ADC+T_(2)WI),采用受试者操作特征(ROC)曲线评估析各模型鉴别GG≤2与GG>2PCa的诊断效能。结果测试集中应用SVM算法的T_(2)WI+ADC模型诊断效能最高,AUC为0.896。其次为T_(2)WI+ADC中的RF模型,AUC为0.871。在各特征集中,RF和SVM算法的模型的AUC均高于XGboost算法。结论基于双参数MRI影像组学机器学习模型可较好地鉴别GG≤2与GG>2PCa。Objective To investigate the value of biparametic MRI radiomics machine learning model in Gleason grade group and grouping of prostate cancer(PCa).Methods 138 patients with PCa confirmed by pathology were retrospectively collected,including 83 patients with GG>2 and 55 patients with GG≤2.All patients underwent preoperative biparametric(T_(2)WI+ADC)MRI examination.Patients were randomly divided into training set and test set at 7:3 for machine learning and validation of imaging radiomics models,respectively.RF,SVM and XGboost were used to construct three groups of models(ADC,T_(2)WI and ADC+T_(2)WI).Receiver operating characteristic(ROC)curve was used to evaluate the diagnostic efficiency of each model in distinguishing GG≤2 from GG>2PCa.Results In the test set,the T_(2)WI+ADC model using SVM algorithm had the highest diagnostic efficiency,with an AU C of 0.896.Secondly,the AU C of RF model in T_(2)WI+ADC was 0.871.In each feature set,the AU C of RF and SVM algorithm is higher than XGboost algorithm.Conclusion The radiomics machine learning model based on biparametric MRI can distinguish GG≤2 from GG>2PCa.

关 键 词：前列腺癌 Gleason评分分级分组 影像组学 机器学习 

分 类 号：R697.3[医药卫生—泌尿科学]