Addressing a systematic vibration artifact in diffusion-weighted MRI

Abstract

We have identified and studied a pronounced artifact in diffusion-weighted MRI on a clinical system. The artifact results from vibrations of the patient table due to low-frequency mechanical resonances of the system which are stimulated by the low-frequency gradient switching associated with the diffusion-weighting. The artifact manifests as localized signal-loss in images acquired with partial Fourier coverage when there is a strong component of the diffusion-gradient vector in the left-right direction. This signal loss is caused by local phase ramps in the image domain which shift the apparent k-space center for a particular voxel outside the covered region. The local signal loss masquerades as signal attenuation due to diffusion, severely disrupting the quantitative measures associated with diffusion-tensor imaging (DTI). We suggest a way to improve the interpretation of affected DTI data by including a co-regressor which accounts for the empirical response of regions affected by the artifact. We also demonstrate that the artifact may be avoided by acquiring full k-space data, and that subsequent increases in TE can be avoided by employing parallel acceleration.

Figure 1

Axial, coronal, and sagittal cross‐sections of a diffusion‐weighted volume demonstrating a severe example of the vibration artifact located mainly in the mesial parietal lobe. Diffusion‐gradient direction was [x, y, z] = [0.98, 0.02, −0.20].

Figure 2

Measured signal (arbitrary units) versus magnitude of x‐component of diffusion‐gradient direction during two repeats of a 60‐direction acquisition. Plots are shown for (a) an ROI in the splenium where a white‐matter tract runs in the left–right direction and (b) an ROI in gray matter affected by the vibration artifact. The images to the left show the location of the ROIs superimposed on the mean diffusion‐weighted image across all directions.

Figure 3

(a) Axial, coronal, and sagittal cross‐sections of directionally encoded color FA maps, color‐coded for the direction of the first eigenvector of the fitted diffusion tensor (red—left–right, green—anterior–posterior, blue—superior–inferior) for a dataset affected by the vibration artifact. (b) The same dataset using a tensor fit with a co‐regressor to account for the artifact. (c,d) Parietal fiber bundles connecting the two hemispheres using (c) a standard diffusion‐tensor fit and (d) using a co‐regressor to account for the artifact.

Figure 4

Example of a clinical scan (cytotoxic edema due to acute ischemic stroke from cardiac embolism) where an isotropic decrease in the apparent diffusion is made ambiguous by the vibration artifact which affects image (b). Images are shown with (a) no diffusion‐weighting as well as b = 1,000 s/mm² in the (b) left/right, (c) superior/inferior, and (d) anterior/posterior directions. In Routine, uniform diffusion hyperintensities in three orthogonal axes are required to confirm isotropic decreases in the apparent diffusion resulting from cytotoxic edema of an acute ischemic stroke.

Figure 5

Top row: (a) single axial slice of a diffusion‐weighted image affected by the vibration artifact (b = 1,000 s/mm² diffusion‐weighting applied in left–right direction), along with (b) the phase image (generated without k‐space apodization filter to improve phase information in low signal areas), and (c) the corresponding k‐space. Bottom row: A repeat of the same acquisition as above, but with TR = 18.6 s (instead of the standard 9.3 s).

Figure 6

(a) Ratio of measured data to predicted data [based on tensors fit to data for abs(r_x,i) < 0.6] within the artifact‐affected region from a single subject. Also shown is the Tukey filter with parameters determined by a least‐squares fit. (b) FA calculated by standard model versus FA calculated by model including artifact co‐regressor. Data shown for ∼200,000 voxels within entire brain. Solid black line indicates line of equality.

Figure 7

Maximum‐intensity projections of the mean distribution of the artifact co‐regressor across 22 subjects in (a) axial, (b) coronal, and (c) sagittal views. (d) The volume of brain affected by the artifact (k > 0.5) in each of the 22 subjects.

Figure 8

(a) Measured signal (arbitrary units) versus magnitude of x‐component of diffusion‐gradient direction using a 30‐direction acquisition within a gray‐matter ROI. Crosses show data using standard imaging parameters (3/4 k‐space, no acceleration) and circles show data using full k‐space and a GRAPPA acceleration factor of 2. Calculated FA: 0.45 versus 0.14, calculated ADC: 1.17 versus 1.34 mm²/s. (b,c) Axial, coronal, and sagittal views of the fitted artifact‐related parameter k for (b) the 3/4 k‐space data and (c) the full k‐space data.