Integration of Longitudinal and Transverse Radiomics from Ultrasound Images with Clinical Factors for HER-2 Status Prediction in Invasive Breast Cancer Patients.

Objective: This study developed a nomogram combining longitudinal and transverse ultrasound radiomics with clinical factors to identify human epidermal growth factor receptor 2 (HER2) status in invasive breast cancer (BC).

Materials and methods: We analyzed 537 invasive BC patients from two hospitals: 436 in the training cohort (Hospital A) and 101 in the test cohort (Hospital B). From longitudinal and transverse ultrasound planes, 788 radiomics features were extracted, with dimensionality reduced using least absolute shrinkage and selection operator regression. A radiomics nomogram integrating clinical predictors and radiomics scores (Rad-scores) was constructed.

Results: Fifteen and sixteen features from longitudinal and transverse ultrasound planes, respectively, were selected to generate Rad-scores, which differed significantly between HER2-positive and HER2-negative groups in both cohorts (p < 0.05). The combined radiomics model outperformed individual models with AUCs of 0.783 and 0.762 in the training and external test cohorts, respectively. Tumor size was an independent clinical predictor. The nomogram, incorporating Rad-scores and tumor size, achieved AUCs of 0.790 (training cohort) and 0.774 (test cohort). Decision curve analysis demonstrated its potential clinical utility.

Conclusion: A biplanar ultrasound radiomics nomogram effectively predicts HER2 status in invasive BC, potentially reducing the need for biopsies and supporting personalized treatment strategies.