Motion Correction in low SNR MRI using an approximate Rician log-likelihood

Ivor J A Simpson, Balazs Orzsik, Iris Asllani, Mara Cercignani

Keywords: Motion correction, Rician distribution, Low SNR

TL;DR: A generative model for motion correction of very noisy MRI data using a Rician log-likelihood function.

Abstract: Some MRI acquisitions, such as Sodium imaging, produce data with very low signal-to-noise ratio (SNR) and meaningful analysis may require several images to be averaged. As the data contains substantial noise, motion correction using standard registration tools may not be effective. This paper employs a simple generative model for the data, where the error is described as following a Rician distribution, which more accurately characterised the image noise. Maximum a posteriori inference is enabled by a differentiable approximation to the Rician log-likelihood function. We find that this approach substantially outperforms a Gaussian log-likelihood baseline on synthetic data that has been corrupted by Rician noise of varying degrees. We show results of our approach on real Sodium MRI data, and demonstrate that we can reduces the effects of substantial motion.