双参数MRI影像组学结合机器学习预测前列腺癌神经侵犯的价值  

The value of dual parameter MRI radiomics combined with machine learning in predicting neural invasion in prostate cancer

作　　者：王海林 李杰[2] 郑毅 陈炜越 卢陈英[1] 周永进[1] 王海永 纪建松[1] WANG Hailin;LI Jie;ZHENG Yi;CHEN Weiyue;LU Chenying;ZHOU Yongjin;WANG Haiyong;JI Jiansong(Zhejiang Key Laboratory of Imaging and Interventional Medicine,the Fifth Affiliated Hospital of Wenzhou Medical University,Lishui Central Hospital,Lishui 323000,China;Department of Urology,the Fifth Affiliated Hospital of Wenzhou Medical University,Lishui Central Hospital,Lishui 323000,China;Suzhou Institute of Biomedical Engineering and Technology,Chinese Academy of Sciences,Suzhou 215163,China)

机构地区：[1]温州医科大学附属第五医院(丽水市中心医院)全省影像与介入医学重点实验室,浙江丽水323000 [2]温州医科大学附属第五医院(丽水市中心医院)泌尿外科,浙江丽水323000 [3]中国科学院苏州生物医学工程技术研究所,江苏苏州215163

出　　处：《温州医科大学学报》2024年第12期1004-1011,共8页Journal of Wenzhou Medical University

基　　金：浙江省医药卫生科技计划项目(2023RC114)。

摘　　要：目的:探讨双参数MRI影像组学结合机器学习在术前预测前列腺癌(PCa)神经侵犯(PNI)的应用价值。方法:回顾性分析2016年4月至2023年5月期间在温州医科大学附属第五医院经病理证实的PCa患者672例。以7:3的比例随机分为训练集和验证集。提取T2加权成像(T2WI)和表观弥散系数(ADC)序列中肿瘤的影像组学特征,通过降维获得最佳特征,并建立了6种机器学习分类器。选择验证集中AUC最高的分类器作为最佳影像组学模型,将其结果输出为影像组学评分(Rad-score)。使用多因素Logistic回归筛选出预测PCa患者PNI的独立危险因素,并建立临床模型。进一步基于Rad-score和临床危险因素建立临床-影像组学融合模型,并绘制列线图。结果:通过降维得到19个最优影像组学特征用于构建机器学习分类器,其中12个来自T2WI序列,7个来自ADC序列。在验证集中,极端梯度提升决策树的预测性能最高,AUC为0.871。临床T分期和PI-RADS评分是预测PCa患者发生PNI的独立危险因素,进一步结合Rad-score建立列线图。结果显示,该列线图可表现出较高的诊断性能,在训练集和验证集中的AUC分别为0.945、0.896。结论:双参数MRI的影像组学对PCa患者PNI具有潜在的预测价值,进一步结合影像学特征建立的列线图能够更好地提升性能。Objective:To explore the application value of dual parameter MRI radiomics combined with machine learning in preoperative neural invasion(PNI)in prostate cancer(PCa).Methods:Retrospective analysis was made of 672 PCa patients confirmed by pathology from April 2016 to May 2023,who were divided into a training set and a validation set using random sampling method at a ratio of 7:3.Radiomic features of tumors were extracted from T2-weighted images(T2WI)and apparent diffusion coefficient(ADC)sequences,and the best features were obtained through dimensionality reduction.Six machine learning classifiers were established.The classifier with the highest area under the curve(AUC)in the validation set was selected as the optimal radiomic model,and output its results as the image omics score(Rad-score).Multivariate logistic regression was used to screen out independent risk factors for predicting PNI in PCa patients,and to establish a clinical model.Based on Rad-score and clinical risk factors,a Clinical-Radiomics fusion model was established and a nomogram was drawn.Results:By dimensionality reduction,19 radiomic features were obtained for constructing machine learning classifiers,including 12 from T2WI sequences and 7 from ADC sequences.In the validation set,the extreme gradient enhancement decision tree had the highest predictive performance,with the AUC of 0.871.Clinical T-stage and PI-RADS score were independent risk factors for predicting PNI in PCa patients.Further combined with Rad score for nomogram establishment,the results showed that the nomogram demonstrated high diagnostic performance,with AUCs of 0.945 and 0.896 in the training and validation sets,respectively.Conclusion:The radiomics of dual parameter MRI has potential predictive value for PNI in PCa patients,which can further improve the performance if combined with radiology features to establish nomogram.

关 键 词：前列腺癌 神经侵犯 影像组学 磁共振成像 

分 类 号：R445[医药卫生—影像医学与核医学]