Prediction of pathological complete response to neoadjuvant chemotherapy for invasive breast cancers based on longitudinal ultrasound and superb microvascular imaging: a single-center retrospective study

Purpose To examine whether dynamic alterations in conventional ultrasound (US) and superb microvascular imaging (SMI) can act as predictors of pathological complete response (pCR) following neoadjuvant chemotherapy (NAC) in breast cancer (BC). Methods This single-center, retrospective study included women with invasive BC who underwent NAC between January 2022 and December 2024. The features of conventional US and SMI characteristics of BC were analyzed before NAC and the change (Δ) after two cycles. Multivariate logistic regression analysis (Forward, Wald, α = 0.05) was used to screen factors independently associated with pCR. Area under the receiver operating characteristic curve (AUC) analysis was performed to confirm the predictive effectiveness and evaluate the internal validity through bootstrap resampling. A nomogram was created to graphically represent the predictive power of the various factors for pCR. Results Before NAC, the pCR group exhibited significantly higher negative rates for the estrogen receptor (ER) and progesterone receptor (PR). ( P < 0.001 and P = 0.005, respectively) and significantly higher positive rates of human epidermal growth factor receptor 2 (HER2) and echogenic rinds ( P < 0.001 and P = 0.029, respectively). Additionally, they exhibited significantly shorter largest diameters (LD) and shortest diameters (SD) ( P = 0.001 and P = 0.003). After two cycles of NAC, patients who achieved pCR exhibited a significantly higher proportion of monochrome superb microvascular imaging (mSMI) that had not expanded, as well as disappearance of the echogenic rind ( P < 0.001 and P = 0.002). Regarding the rate of change in LD, SD, and vascular index (VI), patients in the pCR group showed significantly higher values than those in the non-pCR group (all P < 0.001). The multivariate logistic regression model identified ΔVI (%), ΔSD (%), and SD to have the strongest association with pCR. The overall multivariate model demonstrated the best AUC (0.963), which was significantly higher than that of any single factor. Bootstrap resampling, calibration plots, and decision curve analysis (DCA) all demonstrated strong performance in both discrimination and calibration. Conclusion The baseline status of US and SMI, as well as the longitudinal changes, demonstrated good predictive performance for pCR in BC following NAC.