Associated factors and multimodal radiomics modeling for treatment response to neoadjuvant therapy in breast cancer

Background Neoadjuvant therapy has become an integral component of comprehensive breast cancer treatment.

Both breast ultrasound and contrast-enhanced magnetic resonance imaging (MRI) serve as valuable diagnostic techniques for

predicting pathological complete response (pCR). However, the predictive performance of either modality alone remains suboptimal.

The integration of these two imaging modalities with pathological characteristics demonstrates enhanced predictive efficacy for neoadjuvant therapy outcomes, exhibiting a promising synergistic effect. Objective To develop a multimodal imaging-based predictive model for efficient and accurate assessment of pathological complete response (pCR) rates following neoadjuvant therapy in breast cancer.Methods Clinical, ultrasonographic, and pathological data from breast cancer patients undergoing neoadjuvant therapy in the First Medical Center of Chinese PLA General Hospital from January 1 to December 31, 2022 were retrospectively collected, with manual lesion segmentation performed on breast MRI images. Patients were stratified into pathological complete response (pCR) and non-pCR groups based on postoperative histopathological evaluation. Based on immunohistochemical indicators, ultrasound, and MRI imaging feature parameters, a multimodal imaging predictive model was constructed using logistic regression analysis. The predictive performance of the model was evaluated in both training and validation cohorts through ROC analysis. Calibration curves were employed to verify the accuracy of model predictions, while decision curve analysis (DCA) was applied to assess the clinical practicality of the model. Finally, a clinically applicable nomogram was constructed to facilitate model interpretation and implementation.Results A total of 153 patients were included in this study, with 42 cases (27.45%) achieving pCR. Through LASSO regression combined with univariate and multivariate logistic regression analyses, 6 key predictive features significantly associated with pCR were identified: human epidermal growth factor receptor 2 (HER2) status, Ki-67 proliferation index, pretreatment time-signal intensity curve parameters, pretreatment apparent diffusion coefficient (ADC) values, baseline MRIderived maximum tumor volume, and the rate of change in maximum tumor area between early-treatment ultrasound and pretreatment measurements. A multimodal imaging predictive model was developed by integrating these features, yielding AUC values of 0.866 (95% CI: 0.797 - 0.935) and 0.822 (95% CI: 0.702 - 0.943) for the training and validation cohorts, respectively. These findings suggested that compared with unimodal technology-based models, the multimodal imaging predictive model demonstrated superior predictive performance. Conclusion The multimodal imaging-based predictive model has been clinically validated to provide accurate probabilistic estimation of pathological complete response (pCR) attainment in breast cancer patients undergoing neoadjuvant therapy, thereby offering evidence-based decision support for personalized treatment planning and therapeutic optimization.