“Squashing peanuts and smashing pumpkins”: How noise distorts diffusion‐weighted MR data

Abstract

New diffusion-weighted imaging (DWI) methods, including high-b, q-space, and high angular resolution MRI methods, attempt to extract information about non-Gaussian diffusion in tissue that is not provided by low-b-value (b Ϸ 1000 s mm -2 ) diffusion or diffusion tensor magnetic resonance imaging (DT-MRI). Additionally, DWI data with higher spatial resolution are being acquired to resolve fine anatomic structures, such as white matter fasciculi. Increasing diffusion-weighting or decreasing voxel size can reduce the signal-to-noise ratio so that some DWI signals are close to the background noise level. Here we report several new artifacts that can be explained by considering how background noise affects the peanut-shaped angular apparent diffusion coefficient (ADC) profile. These include an orientationally dependent deviation from Gaussian behavior of the ADC profile, an underestimation of indices of diffusion anisotropy, and a correlation between estimates of mean diffusivity and diffusion anisotropy. We also discuss how noise can cause increased gray/white matter DWI contrast at higher b values and an apparent elevation of diffusion anisotropy in acute ischemia. Importantly, all of these artifacts are negligible in the b-value range typically used in DT-MRI of brain (b Ϸ 1000 s mm INTRODUCTIONSensitizing the MR signal to the diffusion of water molecules provides useful information about tissue characteristics in vivo (1), ranging from changes in acute ischemia (2) through characterization of anisotropic structures (e.g., 3,4) to reconstructions of white matter pathways throughout the brain (e.g., 5). Such information has been obtained mostly using diffusion-weighted (DW) image data with moderate resolution (e.g., 2.5 mm ϫ 2.5 mm ϫ 2.5 mm) and moderate diffusion weighting (b Ϸ 1000 s mm -2 ) so that the diffusion displacement profile in each voxel can be adequately described by a Gaussian function (6). For example, the diffusion characteristics of a voxel containing a single fiber population can be characterized by a tensor whose associated ellipsoid is prolate. In voxels containing multiple fiber populations (e.g., fibers crossing), the diffusion characteristics observed at low b values can still be described by a single tensor, but the diffusion ellipsoid may be less prolate and may become spherical or even oblate (7). In such cases, the tensor model does not adequately reflect the underlying tissue microstructure, which can be problematic for applications such as tractography (5).Several groups (7-17) have proposed methods for elucidating complex tissue microstructure by studying the nonGaussian diffusion behavior that only becomes apparent at higher b values (6,8). For example, Frank's spherical harmonic approach (9) collects DW data at b ϭ 3000 s mm -2 , whereas diffusion spectrum imaging (DSI) (11) (closely related to q-space imaging) (12) attempts to infer the diffusion displacement probability profile directly from diffusion-weighted data collected at very high b values (up to 20,000 of each DW image).It...

Hybrid diffusion imaging

“…In DTI, low DW signals cause an underestimation of the signal attenuation, thus the diffusivities will be underestimated (Jones and Basser 2004). This was likely the case for our diffusion measurements in the outer (b = 6000 and 9375 s/mm 2 ) shells ( Figure 3 and Table 5,6).…”

Construction of a temperature‐controlled diffusion phantom for quality control of diffusion measurements

“…Figure 2a illustrates the location of the 3D ROIs on one slice of the phantom, overlaid onto a directionally color-coded FA map. To rule out the possibility that the temperature-dependent course of the ADCs and FA values was affected by a low SNR in the datasets (14), the SNR within the phantom was analyzed at every temperature. The SNR was calculated according to the National Electrical Manufacturers Association (NEMA) standard method (15).…”

“…The region with the smallest SNR exhibits an SNR of 29.5 in the 15°C dataset and an SNR of 28.6 in the 40°C dataset. Therefore, the SNR can be ruled out as a major contributor to the temperaturedependent course of the diffusion parameters (14). Not surprisingly, the SNR of the diffusion-weighted images drops (depending on the diffusion direction) with increasing temperature due to the longer diffusion pathways caused by the higher thermal energy of the particles.…”

Structural network analysis of brain development in young preterm neonates

“…We seed streamlines at all voxels with FA greater than 0.1 since this is the value of the noise floor [29]. By choosing this FA threshold, we obtain all possible tracts that can be extracted from the DTI.…”