Hyperpolarized MRI with 13C-labelled compounds is an emerging clinical technique allowing in vivo metabolic processes to be characterized non-invasively. Accurate quantification of 13C data, both for clinical and research purposes, typically relies on the use of region-of-interest analysis to detect and compare regions of altered metabolism. However, it is not clear how this should be determined from the five-dimensional data produced and most standard methodologies are unable to exploit the multidimensional nature of the data. Here we propose a solution to the novel problem of 13C image segmentation using a hybrid Markov random field model with continuous fuzzy logic. The algorithm fully utilizes the multi-dimensional data format in order to classify each voxel into one of six distinct classes based on its metabolic characteristics. Bayesian priors fully incorporate spatial, temporal and ratiometric contextual information whilst image contrast from multiple spectral dimensions are considered concurrently by using an analogy from color image segmentation. Performance of the algorithm is demonstrated on in silico data, where the superiority of the approach over a reference thresholding method is consistently observed. Application to in vivo animal data from a pre-clinical subcutaneous tumor model illustrates the ability of the MRF algorithm to successfully detect tumor location whilst avoiding image artifacts. This work has the potential to assist the analysis of human hyperpolarized 13C data in the future.