多参数MRI影像组学模型预测胰腺导管腺癌病理分化程度的价值  

Value of Multiparameter MRI Radiomics Model in Predicting the Pathological Diferentiation of Pancreatic Ductal Adenocarcinoma

作　　者：庄雨 陈杰[1] 李静燕 张京刚[1] 刘琪 潘靓[1] 江曼 张悦[2] 陈学敏[2] ZHUANG Yu;CHEN Jie;LI Jingyan(Department of Radiology,The Third Affiliated Hospital of Soochow University,Changzhou,Jiangsu Province 213003,P.R.China)

机构地区：[1]苏州大学附属第三医院医学影像科,213003 [2]苏州大学附属第三医院肝胆胰外科,213003 [3]呼伦贝尔市人民医院影像科,021000 [4]南京医科大学常州临床医学院,213004

出　　处：《临床放射学杂志》2023年第11期1768-1773,共6页Journal of Clinical Radiology

摘　　要：目的建立多参数MRI影像组学模型,评估其在预测胰腺导管腺癌(PDAC)病理分化程度中的价值。方法回顾性分析经术后病理证实的96例PDAC患者术前临床和MRI资料。按病理分化程度分为低分化组(50例)和中-高分化组(46例)。分析肿瘤部位、边界、最大横截面长径、强化是否均匀、是否伴囊变/坏死、胰管胆管扩张、胰腺萎缩、血管受累、淋巴结增大及肿瘤与正常胰腺实质信号差值等常规MRI表现。采用随机分层抽样按7∶3比例分为训练组和测试组。应用ITK-SNAP软件、A.K.软件、Spearman相关性分析、最小绝对收缩和选择算法(LASSO)、十折交叉验证法分别在脂肪抑制T1WI、脂肪抑制T2WI和动态增强T1WI(DCE-T1WI)3个序列对肿瘤进行二维分割、特征提取和筛选。采用Logistic回归构建影像组学模型及联合临床资料、常规MRI表现的联合模型,使用受试者工作特征曲线下面积(AUC)评估模型预测效能,在测试组中进行验证。结果胰腺实质期肿瘤与正常胰腺组织信号差值在低分化组和中-高分化组间差异具有统计学意义(P<0.05),其他常规MRI表现和临床资料在两组间差异无统计学意义(P>0.05)。影像组学模型、联合模型的AUC值在训练组中分别为0.83、0.96,二者差异具有统计学意义(P<0.05),在测试组中分别为0.82、0.90。联合模型准确度、敏感度、特异度高于影像组学模型。结论多参数MRI影像组学模型在预测PDAC病理分化程度中具有良好的效能,联合模型预测效能更佳。Objective To establish a multiparameter MRI radiomics model and evaluate its value in predicting the path-ological differentiation of pancreatic ductal adenoearcinoma(PDAC).Methods The preoperative elinical and MRI data of 96 patients with PDAC confirmed by postoperative pathology were analyzed retrospectively.According to the degree of pathological differentiation,they were divided into low differentiation group(50 cases)and medium-high diferentiation group(46 cases).The conventional MRI findings such as tumor location,boundary,maximum cross-sectional length and diameter,whether the enhancement is uniform,whether there is cystic degeneration/necrosis,pancreatic duct and bile duct dilatation,pancreatic atrophy,vascular involvement,lymph node enlargement,and signal difference between tumor and normal pancreatic parenchyma were analyzed.Random stratified sampling was used to divide them into training group and test group according to the ratio of 7:3.ITK-SNAP software,A.K.software,Spearman correlation analysis,least absolute shrinkage and selection operator(lasso)and ten fold cross validation method were used to perform two-dimensional segmentation,feature extraction and screening of tumors in fat suppressed T,WI,fat suppressed T,WI and dynamic enhanced T,WI(DCE-T,WI)sequences respectively.Logistic regression was used to construct a radiomics models and a combined model combining clinical data and general imaging performance.The area under the working characteristic curve(AUC)of the subjects was used to evaluate the prediction efficiency of the model,which was verified in the test group.Results The signal difference of pancreatic parenchymal tumor and normal pancreatic tissue was statistically different between the poorly differentiated group and the medium-high differentiated group(P<0.05).No significant differences of other general imaging manifestations and clinical data was present between the two groups(P>0.05).The AUC values of the radiomics models and the combined model were respectively 0.83 and 0.96 in the

关 键 词：胰腺导管腺癌 影像组学 磁共振成像 病理分化程度 

分 类 号：R735.9[医药卫生—肿瘤] R445.2[医药卫生—临床医学]