Acad Radiol. 2026 Aug 18:S1076-6332(26)00571-4. doi: 10.1016/j.acra.2026.07.062. Online ahead of print.
ABSTRACT
RATIONALE AND OBJECTIVES: To develop and validate a radiomics-based clinical model for the preoperative prediction of pathological complete response (pCR) to neoadjuvant chemotherapy (NACT) in human epidermal growth factor receptor 2 (HER2) -positive breast cancer patients.
MATERIALS AND METHODS: This retrospective study included HER2 -positive breast cancer patients who underwent preoperative NACT. Patients were recruited from Center 1 during the period January 2015 to December 2023, and from Center 2 during January 2018 to December 2024. When assessing pCR, surgical specimens served as the gold standard. The radiomics signature (RS) was first refined through redundancy and dimensionality reduction to mitigate overfitting. The retained features were then used to build a Least Absolute Shrinkage and Selection Operator (LASSO) regression model for further feature selection and RS optimization. Subsequently, an integrated model incorporating independent clinical risk factors and the RS was developed. Model performance was compared using the DeLong test on area under the curve (AUC) values.
RESULTS: A total of 659 patients were included, consisting of a training cohort (n=283) and an internal validation cohort (IVC) (n=121) from Center 1, and an independent external validation cohort (EVC) (n=255) from Center 2. The clinical model and RS model showed acceptable performance in predicting pCR, with AUC values of 0.699 and 0.817 respectively observed in the training cohort (TC). The integrated model demonstrated significantly higher predictive accuracy (AUC = 0.861) compared to both the clinical and RS models across all cohorts. The statistical significance of this difference was confirmed using DeLong’s test.
CONCLUSION: The clinical-radiomics combined model demonstrates improved and stable predictive performance for early response to NACT compared with radiomics-only and clinical-only models in both internal and external validation cohorts.
PMID:42613278 | DOI:10.1016/j.acra.2026.07.062
