Matthew R. Orton1, David J. Collins1, Dow-Mu Koh2, Michael Germuska1, Martin O. Leach1
1CR-UK and EPSRC Cancer Imaging Centre, Institute of Cancer Research, Sutton, Surrey, United Kingdom; 2Department of Radiology, Royal Marsden NHS Foundation Trust, Sutton, Surrey, United Kingdom
Many models have been proposed for describing diffusion-weighted data, but as the environment of the diffusion process is known to be very complex in biological systems, choosing an appropriate model is difficult. We present a Bayesian methodology for estimating the posterior probability (uncertainty) of a given selection of diffusion models, applied to clinical DWI data. This is of interest to indicate statistical model uncertainty, and therefore uncertainty in the interpretation of the data. By penalising over complicated models, this methodology provides diffusion metrics that are more stable, and therefore more sensitive to a wider range of treatment effects.