基于MRI影像组学三阴性乳腺癌保乳术后肿瘤复发的预测模型构建及其应用价值  

Study on the Construction of A Model for Predicting Tumor Recurrence after Breast-Conserving Surgery for Triple-Negative Breast Cancer Based on MRI Imaging and Its Application Value

作　　者：李玉舟[1] 隋愿[1] 金红瑞[2] 孟令松 林光耀[1] 黄文起[1] LI Yu-zhou;SUI Yuan;JIN Hong-rui;MENG Ling-song;LIN Guang-yao;HUANG Wen-qi(Department of Magnetic Resonance,The First Affiliated Hospital of Zhengzhou University,Shangqiu 476000,Henan Province,China;Department of Radiology,the Third Affiliated Hospital of Zhengzhou University,Zhengzhou 450000,Henan Province,China)

机构地区：[1]商丘市第一人民医院医学影像中心,河南商丘476000 [2]郑州大学第一附属医院磁共振科,河南郑州450000 [3]郑州大学第三附属医院放射科,河南郑州450000

出　　处：《中国CT和MRI杂志》2023年第3期103-106,共4页Chinese Journal of CT and MRI

基　　金：2018年度河南省医学科技攻关计划(2018020940)。

摘　　要：目的探讨基于MRI影像组学三阴性乳腺癌保乳术后肿瘤复发的预测模型构建及其应用价值。方法回顾性收集2011年1月至2016年10月于三家医院收治并行保乳手术治疗的240例TNBC女性患者。将240例TNBC患者采用计算机产生随机数法以3:1的比例分为训练集(180例)和验证集(60例)。分析训练集和验证集患者的一般临床资料和一般影像学特征。根据训练集中TNBC患者保乳术后复发情况分为复发组(28例)和未复发组(152例)。比较2组一般临床资料。由2位高年资医师提取动态增强扫描早期时相图像的影像组学特征。利用训练集数据分别构建影像组学、临床因素的Logistic回归模型,并计算每个患者的影像组学得分(Radiomics score,Rad-score)和临床得分(Clinic-score)。利用广义线性回归模型进一步筛选变量建立联合预测模型,并计算联合得分(Combine-score)。结果临床得分(Clinic-score)=-0.816×年龄+1.406×(CD19+)+2.013×自然杀伤细胞+1.777×(K i-67)+1.300×淋巴结转移-1.790。影像组学评分(Rad-score)=-1.356×original_ngtdm_Busyness+0.926×wavelet.LHL_firstorder_Median+2.815×wavelet.LHH_ngtdm_Busyness-0.719×log.sigma.3.0.mm.3D_gldm_Dependence Variance-1.528。联合得分(Combine-score)=1.047×Rad-score-0.457×年龄+1.662×(CD19+)+2.093×自然杀伤细胞+0.918×(Ki-67)+1.289×淋巴结转移。Rad-score、Clinic-score及Combine-score在复发和未复发组间比较差异均有统计学意义(P<0.05)。训练集和验证集中临床预测模型分别与影像组学模型、联合预测模型的AUC比较差异均有统计学意义(P<0.05)。Hosmer-Lemeshow检验表明各模型在训练集和验证集中拟合均较好(P>0.05)。结论联合预测模型的鉴别能力优于单纯临床预测模型和影像组学模型,提示影像组学特征联合临床因素用于无创预测TNBC保乳术后肿瘤复发具有一定可行性。Objective To investigate the construction of a model for predicting tumor recurrence after breastconserving surgery for triple-negative breast cancer based on MRI imaging and its application value.Methods A total of 240 female patients with TNBC who were treated with breast conserving surgery in three hospitals from January 2011 to October 2016 were collected retrospectively.240 patients with TNBC were divided into training set(n=180)and verification set(n=60)according to the proportion of 3:1 by computer-generated random number method.The general clinical data and imaging features of patients in training set and verification set were analyzed.According to the recurrence after breastconserving surgery in the training group,TNBC patients were divided into recurrence group(n=28)and non-recurrence group(n=152).The general clinical data of the two groups were compared.The imaging features of the early phase images of dynamic contrast-enhanced scanning were extracted by two senior doctors.The Logistic regression models of imaging and clinical factors were constructed by using the training set data,Rad-score and Clinic-score of each patient were calculated.The generalized linear regression model was used to further screen variables to establish a joint prediction model,and Combine-score was calculated.Results Clinic-score=-0.816×age+1.406×(CD19+)+2.013×natural killer cells+1.777×(Ki-67)+1.300×lymph node metastasis-1.790.Rad-score=-1.356×original_ngtdm_Busyness+0.926×wavelet.LHL_firstorder_Median+2.815×wavelet.LHH_ngtdm_Busyness-0.719×log.sigma.3.0.mm.3D_gldm_Dependence Variance-1.528.Combine-score=1.047×Rad-score-0.457×age+1.662×(CD19+)+2.093×natural killer cells+0.918×(Ki-67)+1.289×lymph node metastasis.There were significant differences in Rad-score,Clinic-score and Combine-score between recurrent and nonrecurrent groups(P<0.05).The AUC of clinical prediction model in training set and verification set was significantly different from that in imaging group model and joint prediction model(P<0.05).Hosmer

关 键 词：MRI影像组学 三阴性乳腺癌 保乳术 肿瘤复发 预测模型 

分 类 号：R737.9[医药卫生—肿瘤]