Predictive value of a mammography radiomics nomogram model for HER-2 expression status in mass-type breast cancer

GU Hongyu; LI Yonggang; SHEN Rui; DENG Xiaoyi; ZHOU Xiaofei

doi:10.19300/j.2025.L21940

Abstract：Objective To evaluate the utility of a mammography-based radiomics nomogram model in predicting human epidermal growth factor receptor 2(HER-2)expression status in breast cancer.Methods A total of 288 female breast cancer patients who underwent mammography and had surgical pathology confirmation were retrospectively collected from two hospitals.Patients from one hospital(n=215)were randomly divided into a training cohort(n=150)and a validation cohort(n=65)at a 7∶3 ratio;patients from the other hospital(n=73)served as an external test cohort.According to postoperative pathology,all patients were categorized into three groups:HER-2-positive(n=106),HER-2-low(n=128),and HER-2-zero(n=54).Radiomics features were extracted from mammograms,dimension-reduced and screened to construct a radiomics score(Radscore).Laboratory results,imaging T stage,maximum tumor diameter,shape,margin,and density were analyzed to identify clinical and mammographic indicators with statistically significant differences among three groups.Multivariate logistic regression was used to determine independent predictors of different HER-2 expression states,upon which a clinical model and a mammographic-feature model were built;these were then combined with the Radscore to develop nomogram models.Receiver operating characteristic(ROC)curves were used to assess predictive performance.The DeLong test compared differences in AUCs among three models.Decision curve analysis(DCA)evaluated clinical utility.Results For HER-2-positive type,a total of 8 optimal radiomic features were selected to establish Radscore 1.For HER-2-low expression type,6 optimal radiomic features were selected to establish Radscore 2.Tumor shape and CA153 were independent predictive factors for HER-2-positive,and they were combined with Radscore 1 to construct Nomogram Model 1.Maximum tumor diameter was an independent predictive factor for HER-2-low expression,and it was combined with Radscore 2 to construct Nomogram Model 2.Nomogram Model 1 achieved the highest AUCs in both the validation and external test cohorts for predicting HER2-positive(DeLong tests,both P＜0.001).For predicting HER2-low expression,Nomogram Model 2 showed higher AUC,sensitivity,and specificity than the mammographic-feature Model 2 in both the validation and external test cohorts(DeLong tests,both P＜0.001).DCA indicated greater net benefit for both Nomogram Models 1 and 2.Conclusion The radiomics-based nomogram effectively predicts HER-2 expression status in breast cancer,providing important clinical value for guiding HER-2 targeted treatment strategies.