A Deep Learning Approach to the Estimation of Bias and Variance in HARDI

• Activity Number: 84 - SPEED: A Mixture of Topics in Health, Computing, and Imaging
• Type: Contributed
• Date/Time: Sunday, July 29, 2018 : 4:00 PM to 4:45 PM
• Sponsor: Section on Statistics in Imaging
• Author(s): Allison Hainline* and Hakmook Kang and Bennett Landman
• Companies: Vanderbilt University and Vanderbilt University Medical Center and Vanderbilt University
• Keywords: diffusion MRI; HARDI; deep learning; neuroimaging; neural network

Abstract

The bias and variance of high angular resolution diffusion imaging (HARDI) methods have been shown to be estimable via simulation extrapolation (SIMEX) and the wild bootstrap, respectively. These methods, however, can be extremely computationally intensive, as metrics are determined on a voxel-wise basis. We propose an updated solution, utilizing a deep neural network to estimate both bias and variance, voxel-wise, without requiring extensive computation. Results are compared to the SIMEX and bootstrap estimates as well as the true values.