基于术前多模态磁共振影像组学Radscore的膀胱癌肌层浸润性预测  

Prediction on muscle-invasive status of bladder cancer based on the preoperative multimodal MRI omics Radscore

作　　者：董琪 张静 王焕军[2] 张曦 刘洋 杜鹏 田强 卢虹冰 郭燕[2] 徐肖攀 DONG Qi;ZhANG Jing;WANG Huan-jun(Military Biomedical Engineering School,Air Force Medical University,Xi’an 710032,China)

机构地区：[1]空军军医大学军事生物医学工程学系,陕西西安710032 [2]中山大学第一附属医院放射科,广东广州510080 [3]空军军医大学唐都医院放射科,陕西西安710038

出　　处：《中国医学装备》2021年第6期17-21,共5页China Medical Equipment

基　　金：国家自然科学基金青年项目(81901698)“多模态MRI联合临床预测因子用于非肌层浸润膀胱癌的复发风险预测研究”。

摘　　要：目的:基于术前多模态磁共振成像(MRI)影像组学策略构建一个简明的量化模型,实现膀胱癌肌层浸润性的术前准确预测。方法:选取医院就诊的106例膀胱癌患者的影像数据,包括T2加权(T2W)、弥散加权(DW)及其表观弥散系数(ADC)图像数据,将其中来自中山大学附属医院的64例膀胱癌患者数据作为训练集,来自唐都医院的42例作为测试集。对所有患者的肿瘤区域进行勾勒,从获得的感兴趣区域(ROI)中提取5类影像学特征。采用特征回归剔除算法进行特征选择,获得最优特征子集。利用逻辑回归策略对所选特征进行回归分析,得到回归系数,构建Radscore模型,并据此对患者的膀胱癌肌层浸润性进行量化分析,观察膀胱癌肌层浸润性的预测准确率以及受试者工作特征(ROC)曲线下面积(AUC)。结果:在训练集和测试集中,选择的36个特征所构建的Radscore模型,对膀胱癌肌层浸润性的预测准确率分别为84.7%和80.9%,AUC分别为0.880和0.813。结论:基于术前多模态MRI影像组学Radscore的膀胱癌肌层浸润性预测的双中心研究,对医师制定个体化临床治疗策略具有重要意义。Objective:To construct a concise quantitative model based on the strategy of preoperative multimodal magnetic resonance imaging(MRI)omics so as to realize the preoperatively accurate prediction for muscle-invasive of bladder cancer.Methods:The image data of 106 patients with bladder cancer who admitted to hospital were selected,which included the image data of T2 weighting(T2W),diffusion weighting(DW),apparent diffusion coefficient(ADC).The data of 64 patients from the First Affiliated Hospital of Sun Yat-Sen University we used as the training set,and 42 patients from Tangdu hospital was used as the test set.The tumor region of all patients were outlined and 5 kinds of imaging characteristics were extracted from obtained region of interest(ROI).And the culling algorithm of characteristic regression was adopted to select characteristic and obtain the optimal characteristic subset.The strategy of logistic regression was used to implement regression analysis for the selected characteristics,and to obtain regression coefficient,and to construct Radscore model.And then,the muscle-invasive statuses of bladder cancer of patients were quantitatively analyzed according to above data.The accurate rate of prediction on muscle-invasive status of bladder cancer and the area under curve(AUC)of receiver operating characteristics(ROC)curve were observed.Results:In the training set and the test set,the accurate rates of the prediction of Radscore model,that were constructed from selected 36 characteristics,on muscle-invasive status of bladder cancer were 84.7%and 80.9%,and AUCs of the two sets were 0.880 and 0.813,respectively.Conclusion:The dual-center research of the prediction of muscle-invasive status of bladder cancer based on the preoperative multimodal MRI omics Radscore has important significance for formulating strategy of individually clinical treatment with physician.

关 键 词：膀胱肿瘤 肌层浸润性 多参数磁共振成像(MRI) 影像组学 支持向量机(SVM) 

分 类 号：R445.2[医药卫生—影像医学与核医学]