Baseline multiparametric MRI combined with clinical factors for predicting pathological complete response of breast cancer after Neoadjuvant chemotherapy.

Purpose

Early prediction of response to neoadjuvant chemotherapy (NAC) may facilitate treatment stratification in breast cancer. This study aimed to evaluate the association of baseline virtual magnetic resonance elastography (vMRE) and intravoxel incoherent motion diffusion-weighted imaging (IVIM-DWI) parameters with pathological complete response (pCR) and assess their incremental value beyond pretreatment clinicopathological factors.

Methods

Consecutive patients with histologically confirmed breast cancer who underwent baseline multiparametric MRI before NAC were retrospectively evaluated. vMRE parameters (shifted apparent diffusion coefficient, sADC; diffusion-derived shear modulus, μdiff) and IVIM-DWI parameters (ADC, D, D* and f) were analyzed. pCR was defined as ypT0/Tis ypN0. Variables associated with pCR in univariable analysis were entered into multivariable logistic regression. A clinical baseline model (tumor size, ER status, HER2 status, and Ki-67) was compared with an extended model incorporating μdiff and D and a parsimonious model based on ER status, μdiff, and D. Model performance was assessed using receiver operating characteristic (ROC) curve analysis, DeLong test, calibration analysis, decision curve analysis (DCA), and bootstrap internal validation.

Results

Of the 125 patients, 35 achieved pCR and 90 did not. The pCR group showed higher sADC and D values, lower μdiff values, and a lower proportion of ER positivity. ER status, μdiff, and D were independently associated with pCR. The clinical baseline model achieved an area under the curve (AUC) of 0.670, which improved to 0.928 after addition of μdiff and D (ΔAUC = 0.258; P < 0.001). The parsimonious model achieved an AUC of 0.921 (95% CI, 0.874-0.967), with 91.4% sensitivity and 84.4% specificity, and demonstrated comparable performance to the extended model (P = 0.403). Bootstrap internal validation of the parsimonious model yielded an optimism-corrected AUC of 0.914.

Conclusion

Baseline μdiff, D, and ER status were associated with pCR after NAC. Incorporating diffusion-derived MRI parameters improved discrimination beyond pretreatment clinical factors, while a parsimonious model retained comparable discrimination. Prospective multicenter external validation is warranted before clinical application.