双参数MRI影像组学对前列腺癌Gleason分级的诊断价值  

Diagnostic value of biparametric MRI radiomics in Gleason classification of prostate cancer

作　　者：刘路路 徐峰 朱蒙蒙 岑朝敏 时金凤 王蕊 王前玉 LIU Lulu;XU Feng;ZHU Mengmeng;CEN Chaomin;SHI Jinfeng;WANG Rui;WANG Qianyu(Department of Radiology,Suqian First People’s Hospital,Suqian,Jiangsu Province 223800,China;Department of Pathology,Suqian First People’s Hospital,Suqian,Jiangsu Province 223800,China)

机构地区：[1]宿迁市第一人民医院影像科,江苏宿迁223800 [2]宿迁市第一人民医院病理科,江苏宿迁223800

出　　处：《实用放射学杂志》2024年第7期1121-1124,共4页Journal of Practical Radiology

基　　金：江苏省卫生健康委员会科研项目(M2022098);宿迁市科学技术局新型临床诊疗技术及公共卫生项目(SY202202)。

摘　　要：目的探讨双参数磁共振成像(bp-MRI)影像组学模型无创预测高危前列腺癌的价值。方法回顾性选取320例经病理证实的前列腺癌患者,所有患者均在病理前进行bp-MRI检查,包括T_(2)WI和扩散加权成像(DWI);从DWI中提取表观扩散系数(ADC)图。根据Gleason评分,将患者分为高危组(Gleason评分≥8)和中低危组(Gleason评分≤7)。使用3D Slicer软件,勾画出整个前列腺轮廓,利用Python软件计算参数,采用最小冗余最大相关性和序列后向消除算法提取和选择影像组学特征,并构建模型。分别构建3个影像组学(T_(2)WI、DWI、ADC)模型,并通过逻辑回归(LR)进行验证。采用受试者工作特征(ROC)曲线的曲线下面积(AUC)、特异度(SP)、敏感度(SE)和准确度(ACC)评价模型的性能。利用224例患者的临床数据和bp-MRI特征建立个体预测模型,并使用96例患者的数据进行验证。结果共提取1165个影像组学特征,通过特征筛选后分别筛选出2、4、6个影像组学特征构建预测高危前列腺癌的T_(2)WI模型、DWI模型及ADC模型。所有影像组学模型对鉴别中低危、高危组均有显著预测性能(P<0.05)。DWI模型的预测价值最高,在训练组的AUC、ACC、SE、SP分别为0.814、0.756、0.838、0.744,在验证组的AUC、ACC、SE、SP分别为0.840、0.756、0.848、0.784。结论基于bp-MRI的影像组学可以在术前较好地鉴别中低危、高危前列腺癌。Objective To explore the value of biparametric magnetic resonance imaging(bp-MRI)radiomics models in noninvasive prediction of high-risk prostate cancer.Methods A total of 320 patients with pathologically confirmed prostate cancer were retrospectively selected,and all patients underwent bp-MRI before pathology,including T_(2)WI and diffusion weighted imaging(DWI).Apparent diffusion coefficient(ADC)maps were extracted from DWI.All patients were divided into high-risk(Gleason score≥8)and medium-low risk(Gleason score≤7)groups based on the Gleason score.Using 3D Slicer software,the entire prostate gland was outlined.Python software was used to calculate parameters,and the minimum redundancy maximum correlation and sequence backward elimination algorithms were used to extract and select radiomics features and to build a model.Three radiomics(T_(2)WI,DWI,ADC)models were constructed and verified by logistic regression(LR).The performance of the model was evaluated by area under the curve(AUC)of receiver operating characteristic(ROC)curve,specificity(SP),sensitivity(SE),and accuracy(ACC).An individual prediction model was established via the clinical data of 224 patients and bp-MRI features,and validated via the data of 96 patients.Results A total of 1165 radiomics features were extracted.After feature screening,2,4 and 6 radiomics features were screened out to construct T_(2)WI model,DWI model and ADC model for predicting high-risk prostate cancer.All radiomics models had significant predictive performance in identifying medium-low risk and high-risk groups(P<0.05).The DWI model had the highest predictive value,and the AUC,ACC,SE,and SP in the training group were 0.814,0.756,0.838,and 0.744,respectively.The AUC,ACC,SE,and SP in the verification group were 0.840,0.756,0.848,and 0.784,respectively.Conclusion Radiomics based on bp-MRI can better identify medium-low risk and high-risk prostate cancer before surgery.

关 键 词：双参数磁共振成像 影像组学 前列腺癌 GLEASON评分 

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