Background and purpose: The most clearly recognized benefit of neoadjuvant chemotherapy (NAC) is that it can increase the proportion of patients who can be treated with breast-conserving therapy (BCT). However

the shrinkage modes of the primary breast tumor after NAC have been confirmed as a predictor of BCT rate and prognosis. This study is to evaluate the accuracy of MRI predicting the shrinkage mode of the primary breast tumor after NAC with three-dimensional reconstruction technique. Methods: Sixty-one women with pathologically proven solitary invasive ductal carcinoma (ⅡA-ⅢC) were recruited. Breast specimens were prepared with PMSS

and residual tumors were microscopically outlined

scanned and registered by PHOTOSHOP software. The 3D model of residual tumors was reconstructed with 3D-DOCTOR software based on pathology and MRI imaging characteristics to evaluate the shrinkage mode. We devided the pathological shrinkage modes into surgical pCR (no residual tumors)

solitary lesions without surrounding lesions

multinodular lesions

solitary lesions with adjacent spotty lesions and diffuse lesions. Further

the clinical-pathological shrinkage modes were divided into 2 categories: distinct shrinkage mode (DSM

the longest diameter of the pathological residual tumors was less than 50% and ≤2 cm in comparison with the primary tumor before NAC) and non-distinct shrinkage mode (NDSM

the longest diameter of the pathological residual tumors was more than 50% and/or 2 cm in comparison with the primary tumor before NAC). Results: The surgical pCR

solitary lesions without surrounding lesions

multinodular lesions

solitary lesions with adjacent spotty lesions and diffuse lesions were observed in 23

17

5

9

7 and 18

3

13

20

7 patients by MRI and pathology

respectively (P=0.001). The accuracy

sensitivity and specificity of MRI for predicting pathological shrinkage modes were 86.2%

65.6% and 91.4%

respectively. The DSM was observed in 36 (59.0%) patients by pathology

and 38 (62.3%) patients by MRI. Two methods had a high consistency in clinical-pathological shrinkage modes (κ=0.863

P=0.000). The accuracy

sensitivity and specificity of MRI for predicting clinical-pathological shrinkage modes were 91.0%

64.0% and 94.8%

respectively. There was not a statistic difference in prediction between DSM and NDSM by MRI (P0.05). Receiver operating characteristic (ROC) curve analysis showed an AUC of 0.928 (P=0.000) for MRI to predict the clinical-pathological shrinkage mode. Conclusion: Three-dimensional MRI reconstruction after NAC could simulate and predict spatial location of residual tumors

and can be helpful in selecting patients who received BCT after NAC with tumor downstaging.