Kelvin J. Layton1, 2, Mark Morelande1, Peter M. Farrell1, Bill Moran1, Leigh A. Johnston1, 3
1Electrical and Electronic Engineering, University of Melbourne, Melbourne, Victoria, Australia; 2National ICT Australia, Melbourne, Victoria, Australia; 3Florey Neuroscience Institutes (Parkville), Melbourne, Victoria, Australia
The underlying distribution of relaxation times contributing to an MR signal can provide useful information about the structure of brain tissue. Recently, this distribution has been modelled by discrete T2 components, which are estimated using a gradient-based least-squares optimisation algorithm. In this work, we demonstrate that this algorithm highly dependent on initialisation and SNR levels, often producing inaccurate estimates of the T2 components. We propose a Bayesian algorithm that overcomes these limitations and provides reliable T2 estimates at clinically achievable SNR. The algorithm is demonstrated through simulated and experimental data.