Background: ACRIN 6657, the imaging component of the I-SPY trial (CALGB 150007/150012), is a multi-center study testing the ability of MRI to provide in-vivo quantification of breast tumor response to neoadjuvant chemotherapy, for early prediction of response and stratification of risk-of-recurrence following treatment. We report results from preliminary analysis comparing MRI variables for correlation with pathologic response and disease progression. Methods: Women with ≥3 cm invasive breast cancer receiving an anthracycline-cyclophosphamide (AC) neoadjuvant chemotherapy regimen followed by a taxane (T) were enrolled between May 2002 and March 2006. Contrast-enhanced MRI was performed prior to start of treatment (baseline), following 1 cycle of AC chemotherapy (t2), between AC and T regimens (t3), and after all chemotherapy but prior to surgery (t4). MRI assessments included tumor longest diameter (MRLD), tumor volume (MRVol), and signal enhancement ratio (SER), a measure of contrast enhancement kinetics. Clinical size (cSize) and mammographic longest diameter (MGLD) were also recorded. Linear dimension was measured by the radiologist for MGLD and MRLD; MRVol was calculated by computer using SER thresholds. Pathologic residual disease size (pSize) and residual cancer burden (RCB) index were evaluated following surgery. Results: 237 patients were enrolled at 9 institutions. 216 patients with complete imaging formed the preliminary analysis set. At time of analysis 42 patients had progressed or died with mean time-to-progression of 21 months; 174 patients were progression-free with mean follow-up time of 42 months. At t4, MRVol was more strongly correlated with pSize than MRLD, SER or cSize (r=.61 vs .28, .24 and .43), while SER showed a stronger correlation with RCB than MRLD, MRVol or cSize (r = .45 vs .30, .31 and .37). MGLD at t4 did not show a significant correlation with either pSize or RCB. Early measurements of tumor size change from baseline by MRVol at t2, and MRVol, MRLD and SER at t3, all showed significant correlation with RCB. In univariate logistic regressions, all t4 measurements were found to be predictive of disease progression. Conclusion: Among clinical and imaging measurements of residual breast tumor size, MRI appears to most accurately reflect pathologic extent of disease following neoadjuvant treatment. Preliminary findings also suggest that tumor size and contrast kinetics measured by MRI may be useful early predictors of treatment response. ACRIN 6657 is continuing to collect follow-up data toward the primary aim testing MRI for stratification of post-treatment risk groups according to 3-year disease-free survival. This work is funded by NIH/ACRIN Grant U01 CA79778S2, CALGB Grants CA31946 and CA33601, and NCI SPORE Grant CA58207.