Meeting Banner
Abstract #2117

Non-Negative Spherical Deconvolution for Fiber Orientation Distribution Estimation

Jian Cheng1, Dinggang Shen2, Pew-Thian Yap2

1Department of Radiology and Biomedical Research Imaging Center (BRIC) , The University of North Carolina at Chapel Hill, Chapel Hill, NC, United States; 2Department of Radiology and Biomedical Research Imaging Center (BRIC), The University of North Carolina at Chapel Hill, Chapel Hill, NC, United States

In diffusion MRI, Spherical Deconvolution (SD) was proposed to estimate the fiber Orientation Distribution Function (fODF) based on spherical deconvolution using a single-fiber response function. The peaks or the shape of fODFs can be used to infer local fiber directions. Constrained Spherical Deconvolution (CSD), which takes into consideration the non-negative of the fODF, is the most widely used method among SD variants. Although CSD is capable of accurately determining fiber directions, it is susceptible to false positive peaks especially in the regions with low anisotropy. This is a common drawback of all existing SD-based methods. Moreover, in practice the fODF estimated using CSD still has significant negative values. We propose a method called Non-Negative Spherical Deconvolution (NNSD) to solve the above two problems. Based on a Riemannian framework of ODFs and Square Root Parameterized Estimation for non-negative definite Ensemble Average Propagator, NNSD is formulated such that the non-negativity of the fODF is guaranteed with largely reduced false positive peaks.The synthetic data and real data experiments demonstrated the improvement of NNSD over CSD.

Keywords

accurately although among amount anisotropy authors axis background basis biomedical brain called capable chapel chose clean collinear colored common complex consideration constant constrained constraint contain continuous convolved cost crossing deconvolution defined definite degree density descent detect detected determining diagonal diffusion dimensions directionally distribution domain drawback eigenvalues element ensemble especially estimation evaluation evident existing fair false fiber field fixed formulated function generalized generated glyphs gradient guaranteed guarantees harmonic help hill human implemented indicated indicating infer initialization instead intensities intentionally isotropic kernel largely larger like limitations local mathematical matrix maximal measured medicine minimize minimizing mixture moreover naturally negative negativity noise normalization north novel opposed orientation orientations orthonormal overcome panels parameterized peaks perturbed positive practice prevalent problems propagator proposed radiology real recovered reduced reduces regularization removes represent respectively response root shape significantly signifying since smoothness software solve sphere spherical square squaring still susceptible symmetric synthetic takes tensors theoretical tracts unit unknown validated vector widely