Meeting Banner
Abstract #1934

Comparison of Novel ICA-Based Approach to Existing Diffusion MRI Multi-Fiber Reconstruction Methods

Bryce Wilkins1, Namgyun Lee1, Kyungmin Nam1, Darryl Hwang1, Manbir Singh1

1Radiology and Biomedical Engineering, University of Southern California, Los Angeles, CA, United States

A novel Independent Component Analysis (ICA) based approach to resolving multiple fiber directions per voxel in diffusion-weighted MRI analysis is quantitatively compared against four alternatives: Generalized q-Sampling Imaging, Constrained Spherical Deconvolution, analytical Q-Ball Imaging, and Higher-Order Tensors. We investigate the performance of the various methods when processing limited sample data, as is likely to be acquired in clinical studies due to constrained scan time. Results of two phantom datasets and a human study are presented, revealing consistently higher metric scores for the ICA-based approach, especially in the case of limited gradient sampling.

Abstract PDF Poster PDF

Keywords

acquisition already alternative alternatives analytical analyzed anterior applied approximately authors ball better biomedical blue bottom brain chosen clinical code codes common complete complex component computation consideration constrained correctly covering created crossing deconvolution derived diffusion distributed elderly emphasized engineering environments error estimating evaluated evaluation exact examines existing extent fiber fibers filtered five fraction full generalized generates geometry gradient greater green ground half hours house human idealized illustrate important in vivo include included independent indicates individual intra kissing known lastly lead leads length likely limited listed many metric metrics minutes norm novel objectively often orientations part particularly pathway pathways performance phantom phantoms physical process processed processing prohibitive projects propagate propagating quantitative radiology reaching reconstruction remains representative representing resolve respectively reveals sample samples sampling scanning scores seed selecting sets several situations southern specifically spherical starting steer studies studio subject successfully target tensor tracts trans truth unclear uniformly uniquely versatility white wrong