Jesper Leif
Roger Andersson1, Mark Jenkinson1
1fMRIB, Oxford University, Oxford,
Oxfordshire, United Kingdom
We
have developed a method for estimating and correcting distortions from
reverse-blip data with poor SNR. It is based on a forward model that allows
us to make predictions about the images and a Rician noise model that enables
us to calculate the probability of observed images. Bayesian inversion is
used to find the most probable distortion-free image and field. It performs
well even on data with very poor SNR.