Brendan Moloney ¹ Xin Li ^1* Michael Hirano ² Assim Saad Eddin ³ Jeong Y. Lim ⁴ Debosmita Biswas ² Anum Kazerouni ² Alina Tudorica ⁵ Isabella Li ² Mary L. Bryant ² Courtney Wille ⁶ Chelsea Pyle ⁵ Habib Rahbar ^2,7 Su Hsieh ³ Travis L. Rice-Stitt ⁸ Suzanne M. Dintzis ^7,9 Amani Bashir ^10,11 Evthokia Hobbs ¹² Alexandra Zimmer ¹² Jennifer M. Specht ^13,7 Sneha Phadke ^10,14 Nicole Fleege ^10,14 James H. Holmes ^10,3 Savannah Partridge ^2,7 Wei Huang ^1*

• ¹ Advanced Imaging Research Center, Oregon Health & Science University, Portland, Oregon, United States
• ² Department of Radiology, University of Washington, Seattle, California, United States
• ³ Department of Radiology, University of Iowa, Iowa City, Iowa, United States
• ⁴ Biostatistics Shared Resource, Knight Cancer Institute, Oregon Health & Science University,, Portland, United States
• ⁵ Department of Diagnostic Radiology, Oregon Health & Science University, Portland, United States
• ⁶ Institute for Clinical and Translational Science, University of Iowa, Iowa City, Iowa, United States
• ⁷ Fred Hutchinson Cancer Center, Seattle, Washington, United States
• ⁸ Department of Pathology, Oregon Health & Science University, Portland, United States
• ⁹ Department of Pathology, University of Washington, Seattle, California, United States
• ¹⁰ Holden Comprehensive Cancer Center, University of Iowa, Iowa City, Iowa, United States
• ¹¹ Department of Pathology, University of Iowa Hospitals and Clinics, Iowa City, United States
• ¹² Hematology and Medical Oncology Division, Knight Cancer Institute, Oregon Health & Science University, Portland, United States
• ¹³ Division of Hematology and Oncology, University of Washington,, Seattle, California, United States
• ¹⁴ Department of Internal Medicine, University of Iowa Hospitals and Clinics, Iowa City, Iowa, United States

Quantitative dynamic contrast-enhanced (DCE) MRI as a promising method for prediction of breast cancer response to neoadjuvant chemotherapy (NAC) has been demonstrated mostly in single-center and single-vendor platform studies. This preliminary study reports initial experience in implementing quantitative breast DCE-MRI in a multi-center (MC) and multi-vendor platform (MP) setting to predict NAC response. MRI data including B1 mapping, variable flip angle (VFA) measurements of native tissue R1 (R1,0), and DCE-MRI were acquired during NAC at three sites using 3T systems with Siemens, Philips, and GE platforms, respectively. High spatiotemporal resolution DCE-MRI was performed using similar vendor product sequences with k-space undersampling during acquisition and view-sharing during reconstruction. A breast phantom was used for quality assurance/quality control (QA/QC) across sites. The Tofts model (TM) and Shutter-Speed model (SSM) were used for pharmacokinetic (PK) analysis of the DCE data. Additionally, tumor region of interest (ROI)-and voxel-based analysis in combination with the use of VFAmeasured R1,0 or fixed, literature-reported R1,0 were investigated to determine the optimal analysis approach. Results from fifteen patients who have completed the study to date are reported. Voxelbased PK analysis using fixed R1,0 was deemed as the optimal approach which allowed inclusion of data from one vendor platform where VFA measurements produced ≥ 100% overestimation of R1,0. Semi-quantitative signal enhancement ratio (SER) and quantitative PK parameters outperformed tumor longest diameter (LD) in prediction of pathologic complete response (pCR) vs. non-pCR after the first NAC cycle, while K trans consistently provided more accurate predictions than both SER and LD after the first NAC cycle and at NAC midpoint. Both TM and SSM K trans and kep were excellent predictors of response at NAC midpoint with ROC AUC > 0.90, while the SSM parameters (AUC ≥ 0.80) performed better than the TM counterparts (AUC < 0.80) after the first NAC cycle. The initial experience of this ongoing study indicates the importance of QA/QC using a phantom and suggests that deploying voxel-based PK analysis using a fixed R1,0 may mitigate random errors from R1,0 measurements across platforms and potentially eliminate the need for B1 and VFA acquisitions in MC and MP trials.