基于ADC图的影像组学模型在前列腺癌Gleason危险度分级中的诊断价值  

Usefulness of Radiomics based on ADC Images in Gleason Risk Assessment of Prostate Cancer

作　　者：赵茹[1] 赵红[1] 宫希军[1] 虞红珍 张志强 王龙胜[1] ZHAO Ru;ZHAO Hong;GONG Xijun(Department of Radiology,The Second Hospital of Anhui Medical University,Hefei,Anhui Province 230601,P.R.China)

机构地区：[1]安徽医科大学第二附属医院放射科,安徽医科大学医学影像研究中心,合肥230601 [2]安徽医科大学第二附属医院病理科,安徽医科大学医学影像研究中心,合肥230601 [3]安徽医科大学第二附属医院泌尿外科,安徽医科大学医学影像研究中心,合肥230601

出　　处：《临床放射学杂志》2023年第10期1625-1629,共5页Journal of Clinical Radiology

基　　金：安徽省转化医学研究院科研基金项目(编号:2021zhyx-C65);安徽医科大学校科研基金项目(编号:2021xkj054)。

摘　　要：目的探讨基于ADC图的影像组学模型在鉴别前列腺癌Gleason危险度分级中的价值。方法回顾性分析2019年9月至2021年12月行前列腺MRI检查并行手术或穿刺病理学证实为前列腺癌的59例患者资料(28例Gleason评分≤3+4分,31例Gleason评分≥4+3分),采用ITK-SNAP软件对患者的ADC图像进行感兴趣区(ROI)勾画,共勾画病灶73个,采用Pyradiomics方法提取纹理特征Spearman去除相关性较高的特征,最小绝对收缩和选择算法(LASSO)回归进行特征筛选,最终采用迭代的方式建立支持向量机(SVM)分类模型,筛选最优诊断模型,并进行验证,结果用曲线下面积(AUC)表示。结果共提取94个特征,通过Spearman去除相关系数较高的72个特征后,经LASSO筛选出最佳特征10个,构建SVM最优模型,在训练集中其AUC为0.95(95%CI:0.90~1.00),经交叉验证后,在验证集中AUC为0.87(95%CI:0.72~1.00)。结论基于ADC图的影像组学模型在鉴别前列腺癌Gleason低危组和高危组中有一定的诊断价值。Objective To explore the usefulness of Radiomics based on ADC images in Gleason risk classification of prostate cancer.Methods 59 patients(28with Gleason score≤3+4 and 31 with Gleason score≥4+3)who underwent prostate MRI scan and were diagnosed as prostate cancer by surgical or biopsy pathology in our hospital from September 2019 to December 2021 were enrolled,retrospectively.ROI were delineated by using ITK-SNAP and 73 lesions of ROI were constructed.Texture features were extracted by Pyradiomics method,Spearman was used to remove features with high correlations,feature selection by using LASSO regression,and SVM classification model was established with iterative method.The optimal model was construct and verified,and the results were expressed with AUC.Results A total of 94 features were extracted.After Spearman selection,72 features with high correlations were removed.10 most predictive features were selected by LASSO and the SVM optimal model was constructed.In training set,the model showed a result with AUC 0.95(95%CI:0.90-1.00)while in validation set with AUC 0.87(95%CI:0.72-1.00).Conclusion Radiomics model based on ADC images have the potential in differentiating Gleason low-risk group from high-risk group.

关 键 词：前列腺癌 表观扩散系数 影像组学 

分 类 号：R737.25[医药卫生—肿瘤]