Dynamic functional imaging promises powerful tools for the visualization and elucidation of important disease-causing biological processes, where the pixels often represent a composite of multiple biomarkers independent of spatial resolution. This study exploits both blind source separation and imagery marker characteristics to develop a hybrid method for the separation of mixed yet correlated biomarker distributions in DCE-MRI. A compartment latent variable model is constructed upon which a novel convex analysis framework is proposed to provide a close-form algebraic solution to separating composite markers with non-negativity and well-grounded points. A unique non-negative clustered component analysis is further developed to explicitly consider both partial volume effect and noise contamination. Experimental results show promising and robust extraction of time activity curves and vascular marker images in agreement with biomedical expectations.