CSF contamination-invariant statistics in diffusion-weighted MRI

Abstract

In diffusion-weighted magnetic resonance imaging (DW-MRI), the detection of microstructural change in brain white matter structures that border cerebrospinal fluid (CSF) is complicated by partial volume effects. The conventional diffusion tensor model is very sensitive to CSF contamination, which may leave subtle microstructural change undetected or may lead to detection of spurious microstructural change due to white matter atrophy. In this paper we present a novel method to detect microstructural change in CSF contaminated voxels of white matter structures imaged with conventional DW-MRI protocols. To the diffusion-weighted images (DWIs) a two-compartment tensor model, which has a tissue and a CSF compartment, is fitted by maximum likelihood estimation. However, this estimation problem has an infinite set of (almost) equally likely solution for DWIs acquired at a single b-value. We use statistical properties of this infinite set as statistics by fitting a trace-constrained bi-tensor model. We demonstrate on simulated diffusion-weighted data that our method is more sensitive to detect subtle changes in the microstructure of CSF contaminated voxels than the conventional single tensor model. In a small pilot study, using data of aging subjects to investigate the fornix with the proposed method, we found that the compartment fraction of the tissue compartment decreases significantly with age, whereas the anisotropy of the tissue compartment did not. Our method enables studying the microstructure of white matter regions that may contain CSF contamination.