Assessing the efficacy of cancer treatments using in vivo imaging is shifting from qualitative techniques to quantitative imaging methods that characterize biologically relevant properties of tumor tissue. The use of modelfree or heuristic measures, such as the initial area under the gadolinium curve (IAUGC), or fully quantitative measures, such as the kinetic parameters from a compartmental model, are relatively well understood in the analysis of dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) (1-3). Analysis of an oncology imaging trial is usually achieved by applying statistical summaries, such as the mean or median, to the parameters of interest derived from tissue regions of interest (ROIs). That is, enhancing (tumor) voxels are identified from the DCE-MRI data for each scan across all subjects and those voxels are represented by a single parameter; for example, K trans from quantitative analysis and IAUGC 90 from a heuristic analysis. Hypothesis testing, either parametric or nonparametric, may then be applied to the derived statistics to assess the effects of treatment.Applying statistical summaries to the kinetic parameter maps from DCE-MRI however, discards a substantial amount of information contained in the contrast agent concentration time curves (CTCs) at each voxel, essentially abstracting thousands of observations in space and time to a single number per scan per subject. We believe that there is a wealth of potential information by retaining the collection of CTCs across all subjects and scans, at the same time acknowledging the fact that not all CTCs are the same and not all patients are the same.This article proposes a Bayesian hierarchical model to analyze all tumor CTCs across all patients and scans in a given study simultaneously based on the concept of a mixed-effects model. Mixed-effects models are well established in the statistical community and have found widespread applications in, for example, agriculture, economics, geophysics and the analysis of clinical trials (4,5). Previous examples of mixed-effects models in neuroimaging primarily exist for functional MRI studies (6-8). Mixedeffects models extend the concept of traditional linear or nonlinear models by combining both fixed effects and random effects in the same model. More generally, mixedeffects models are most often used to describe relationships between the measured response and explanatory variables in data that are grouped according to one or more factors. Fixed effects denote parameters that are associated with an entire population, and random effects denote parameters that are associated with random samples from a population. For example, the drug or radiation therapy given in a trial is a fixed effect, whereas patients are inherently random because they are sampled from the general population. Mixed-effect models can provide additional information by allowing such patient-specific treatment effects, and even when the treatment effect is assumed to be the same across patients a mixed model can improve the pr...