Characterizing breast cancer response to neoadjuvant therapy based on biophysical modeling and multiparametric magnetic resonance imaging data.

Personalized medicine efforts are focused on identifying biomarkers to guide individualizing neoadjuvant therapy regimens. In this work, we aim to validate a previously developed image data-driven mathematical modeling approach for dynamic characterization of breast cancer response to neoadjuvant therapy using a large, multi-site cohort. We retrospectively analyzed patients enrolled in the BMMR2 ACRIN 6698 subset at 10 institutions. Patients enrolled received four MRI examinations during neoadjuvant therapy with acquisitions at baseline (T ), 3-weeks/early-treatment (T ), 12-weeks/mid-treatment (T ), and completion of therapy prior to surgery (T ). A biophysical mathematical model of tumor growth is used extract metrics to characterize the dynamics of treatment response. Using predicted response at therapy conclusion and histogram summary metrics to quantify estimated tumor proliferation maps, we found univariate model-based metrics able to predict pathological response, with area under the receiver operating characteristic curve (AUC) ranging from 0.58 and 0.69 analyzing between T and T , and AUCs ranging from 0.72-0.76 analyzing between T and T . For hormone receptor (HR)-negative, human epidermal growth factor receptor 2 (HER2)-positive breast cancer patients our model-based metrics achieved an AUC of 0.9 analyzing between T and T and AUC of 1.0 analyzing between T and T . This data shows the significant promise in developing these imaging-based biophysical mathematical modeling methods of dynamic characterization into a clinical decision support tool for individualizing treatment regimens based on patient-specific response.