Predicting neoadjuvant chemotherapy efficacy in triple-negative breast cancer using DCE-MRI and mammographic texture

This study aimed to develop and validate a predictive model integrating dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) kinetic features and digital breast tomosynthesis (DBT) texture features for predicting pathological complete response (pCR) following neoadjuvant chemotherapy (NAC) in patients with triple-negative breast cancer (TNBC). Additionally, its predictive performance was compared with that of single-modality models. A total of 360 patients with pathologically confirmed TNBC who received NAC between January 2022 and December 2025 were retrospectively and consecutively collected. Patients were randomly allocated to a training set ( n = 252) and a validation set ( n = 108) at a 7:3 ratio. Pre-NAC DCE-MRI kinetic and DBT texture features were extracted. After feature selection using least absolute shrinkage and selection operator (LASSO) regression, DCE-MRI and DBT radiomics scores (Rad-scores) were constructed. A nomogram prediction model was then developed using multivariable logistic regression incorporating clinicopathological and laboratory indicators. Model discrimination, calibration, and clinical net benefit were evaluated using the receiver operating characteristic curves, calibration curves, and decision curve analysis, respectively. These results were subsequently compared with those of a clinical-only model and single-modality combined models. The pCR rates in the training and validation sets were 41.67% and 40.74%, respectively. Multivariable logistic regression identified Ki-67 proliferation index, maximum tumor diameter, background parenchymal enhancement score, neutrophil-to-lymphocyte ratio, lactate dehydrogenase level, DCE-Rad score, and DBT-Rad score as independent predictors of pCR ( P < 0.05). The dual-modality nomogram achieved area under the curve (AUC) values of 0.784 (95% CI: 0.718–0.851) in the training set and 0.759 (95% CI: 0.644–0.873) in the validation set, outperforming the clinical-only model (0.727), the clinical plus DCE-Rad score model (0.761), and the clinical plus DBT-Rad score model (0.766). Calibration curves showed good agreement between predicted probabilities and actual outcomes. Furthermore, decision curve analysis confirmed the clinical net benefit of the proposed model. The multimodal radiomics fusion model based on DCE-MRI kinetic and DBT texture features effectively predicts pCR following NAC in TNBC patients. Its predictive performance is superior to that of single-modality models. Consequently, this model offers a non-invasive imaging tool for individualized pretreatment efficacy assessment. Nevertheless, further multicenter external validation is required to confirm its generalizability. For now, it should be viewed as a decision‑support aid that informs rather than dictates treatment choices, with specific actionable thresholds to be refined in future prospective studies.