A Novel Approach for Statistical Estimation of HARDI Diffusion Parameters from Rician and Non-Central Chi Magnitude Images

Introduction: The statistics of noisy MRI magnitude and root sum-of-squares (SoS) images are accurately modeled by the Rician and non-central chi (NCC) probability distributions, respectively [1]. In diffusion MRI (dMRI), the noise bias introduced by these distributions leads to distortion of estimated quantitative diffusion parameters [2]. One approach to addressing this noise bias is to denoise the data prior to parameter estimation [3,4]. Another strategy is to apply maximum likelihood (ML) estimation methods to quantify diffusion, leveraging the known characteristics of the Rician/NCC distribution [5]. Due to computational complexity, the ML strategy has been primarily used in dMRI for simple DTI fitting with Rician noise, while use of ML estimation with more advanced dMRI models and/or NCC noise is much less common. However, recent work has demonstrated an efficient algorithmic majorize-minimize (MM) framework for solving ML estimation problems involving Rician and NCC distributions [6]. In this work, we adapt the MM framework from [6] to the context of estimating high angular resolution diffusion imaging (HARDI) parameters from Rician and NCC data.