An expectation-maximisation approach for simultaneous pixel classification and tracer kinetic modelling in dynamic contrast enhanced-magnetic resonance imaging

Traditionally, tracer kinetic modelling and pixel classification of DCE-MRI studies are accomplished separately, although they could greatly benefit from each other. In this article, we propose an expectation-maximisation scheme for simultaneous pixel classification and compartmental modelling of DCE-MRI studies. The key point in the proposed scheme is the estimation of the kinetic parameters (K trans and K ep) of the two-compartmental model. Typically, they are estimated via nonlinear least-squares fitting. In our scheme, by exploiting the iterative nature of the EM algorithm, we use instead a Taylor expansion of the modelling equation. We developed the theoretical framework for the particular case of two classes and evaluated the performances of the algorithm by means of simulations. Results indicate that the accuracy of the proposed method supersedes the traditional pixel-by-pixel scheme and approaches the theoretical lower bound imposed by the Cramer–Rao theorem. Preliminary results on real data were also reported.