Abstract Introduction: We investigate the ability of multi-parametric, voxel-based characterizations of tumor heterogeneity from magnetic resonance imaging (MRI) to predict the response of breast cancer patients to neoadjuvant therapy (NAT). In particular, we use high-dimensional analysis of the longitudinal changes in the vascular and cellular characteristics provided by quantitative dynamic contrast enhanced MRI (DCE-MRI) to predict response. Methods: DCE-MRI data was acquired from 34 patients with stage II/III breast cancer before initiating NAT (t1) and after one cycle of NAT (t2). Pathological complete response (pCR) and non-pCR was defined at the time of surgery. Non-pCR patients were further subdivided into partial response (>30% decrease in tumor volume), progressive disease (PD, >20% increase in tumor volume), and stable disease according to the Response Evaluation in Solid Tumors (i.e., RECIST) criteria. For each tumor voxel, DCE-MRI data was modeled to extract the extravascular, extracellular volume fraction, ve, the plasma volume fraction, vp, and the volume transfer coefficients, Ktrans and kep, which correspond to the rate of wash-in and wash-out of contrast agent, respectively. Multi-parametric voxel-based maps of physiological parameters Ktrans, kep, ve, and vp were used to identify different tumor subpopulations. Dimension reduction was completed using t-distributed Stochastic Neighbor Embedding (t-SNE) and subpopulations were identified using DBSCAN clustering. Using the t-SNE generated low-dimensional maps, a subset subpopulations were analyzed based on cluster differences between groups at t2. Based on MRI parameter values, the selected voxel subpopulations fell into one of two categories: high vascularity-high cellularity (HV-HC: high vp , Ktrans , kep values, low ve values) and low vascularity-low cellularity (LV-LC: low vp , Ktrans , kep values, high ve values). The contribution of each cluster to overall tumor volume was analyzed for each patient and quantified as percent tumor volume. Results: No differences were observed between patient groups at t1. At t2, patients with pCR revealed increased percent tumor volumes containing LV-LC subpopulations, compared to PD patients (p = 0.03). Furthermore, PD patients demonstrated significant increase in percent tumor volumes with HV-HC subpopulations, compared to pCR patients (p=0.01). Compared to pCR patients, combined non-pCR patients showed decreased percent tumor volumes of LV-LC subpopulations, trending toward significance (p=0.06). Conclusion: The results indicate that analysis of high-dimensional parameter maps derived from quantitative MRI can be utilized to characterize intratumoral heterogeneity and identify subpopulations of tumor response within patients. These characterizations can potentially be used to define and model tumor habitats and ultimately predict patient treatment response. Citation Format: Anum Syed, Anna G. Sorace, Stephanie L. Barnes, Lori Arlinghaus, Xia Li, Thomas E. Yankeelov. Assessing heterogeneity in DCE-MRI data of breast cancer to predict treatment response [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2017; 2017 Apr 1-5; Washington, DC. Philadelphia (PA): AACR; Cancer Res 2017;77(13 Suppl):Abstract nr 3941. doi:10.1158/1538-7445.AM2017-3941