基于多模态超声图像特征的Logistic回归模型预测三阴性乳腺癌病灶肿瘤浸润淋巴细胞表达的临床价值

Clinical value of a Logistic regression model based on multimodality ultrasound image characteristics in predicting the expression of tumor-infiltrating lymphocytes in triple negative breast cancer

目的探讨基于多模态超声(包括二维超声、剪切波弹性成像、超声造影及自动乳腺全容积成像)图像特征的Logistic回归模型术前预测三阴性乳腺癌(TNBC)病灶肿瘤浸润淋巴细胞(TILs)表达的临床价值。方法选取经病理证实的TNBC女性患者99例,根据TILs表达水平将其分为TILs低表达组41例(TILs表达水平<20%)和TILs高表达组58例(TILs表达水平≥20%),应用二维超声获取病灶形态、方位、边缘、内部回声、后方回声、钙化等特征,剪切波弹性成像检测病灶剪切波速度(SWV),自动乳腺全容积成像获取病灶有无汇聚征、晕环征、导管改变等特征,超声造影获取起始增强时间、增强强度、增强方向、增强模式、局灶性充盈缺损、周围血管征、增强后病变范围等特征。比较两组多模态超声图像特征的差异;应用多因素Logistic回归分析筛选预测TNBC病灶TILs高表达的独立影响因素,并建立回归模型。绘制受试者工作特征(ROC)曲线分析回归模型预测TNBC病灶TILs高表达的诊断效能。结果TIls高表达组二维超声图像特征形态规则、边缘光整、后方回声增强、内部回声不均匀,以及超声造影图像特征高增强、局灶性充盈缺损占比均高于TIls低表达组,差异均有统计学意义(均P<0.05);两组自动乳腺全容积成像特征及SWV比较差异均无统计学意义。多因素Logistic回归分析显示,形态规则、边缘光整、后方回声增强、高增强及局灶性充盈缺损均为预测TNBC病灶TILs高表达的独立影响因素(OR=6.858、3.824、5.909、1.945、6.522,均P<0.05);建立的回归模型为:Logit(P)=-2.989+1.925×形态规则+1.341×边缘光整+1.776×后方回声增强+0.665×高增强+1.875×局灶性充盈缺损;其预测TNBC病灶TILs高表达的ROC曲线下面积为0.772。结论基于多模态超声图像特征的Logistic回归模型对术前预测TNBC病灶TILs表达有一定的临床价值。

Objective To investigate the clinical value of a Logistic regression model based on multimodality ultrasound[including two-dimensional ultrasound,shear wave elastography,contrast-enhanced ultrasound(CEUS)and automated breast volume scanning]image characteristics in predicting the expression of tumor-infiltrating lymphocytes(TILs)in triple-negative breast cancer(TNBC)preoperatively.Methods Ninety-nine female patients with TNBC confirmed by pathology were divided into the TILs low expression group(<20%,n=41)and the TILs high expression group(≥20%,n=58)according to the expression of TILs.The shape,orientation,margin,internal echo,posterior echo and calcifications were obtained by two-dimensional ultrasound,and the mean shear wave velocity(SWV)was obtained by shear wave elastography(SWE),the convergence sign,halo sign and catheter change were obtained by automated breast volume scanner(ABVS),and the initial enhancement time,enhancement intensity,enhancement direction,enhancement mode,perfusion defects,peripheral vascularity and extent of lesions after enhancement were obtained by CEUS.The differences in multimodality ultrasound image characteristics between the two groups were compared.Multivariate Logistic regression was applied to analyze the independent influencing factors for predicting the high expression of TILs in TNBC,and a regression model was established.Receiver operating characteristic(ROC)curve was drawn to analyze the regression model in predicting the high expression of TILs in TNBC.Results In the TILs high expression group,the proportion of regular shape,circumscribed margin,enhanced posterior echo and heterogeneous echo pattern in two-dimensional ultrasound image characteristics as well as hyperenhancement and perfusion defects in CEUS image characteristics were higher than those in the TILs low expression group,and the differences were statistically significant(all P<0.05).The differences were not statistically significant in SWV and ABVS image characteristics.Multivariate Logistic regression analysis showed that regular shape,circumscribed margin and enhanced posterior echo in twodimensional ultrasound image characteristics and hyperenhancement and perfusion defects in CEUS image characteristics were all independent influencing factors for predicting the high expression of TILs in TNBC(OR=6.858,3.824,5.909,1.945,6.522,all P<0.05).The established prediction model was:Logit(P)=-2.989+1.925×regular shape+1.341×circumscribed margin+1.776×enhanced posterior echo+0.665×hyperenhancement+1.875×perfusion defects,and the sensitivity,specificity,accuracy and AUC for predicting the high expression of TILs in TNBC were 68.3%,27.6%,76.0%and 0.772,respectively.Conclusion Logistic regression model based on multimodality ultrasound image characteristics has certain value in predicting TILs expression in TNBC preoperatively.