Impact of Wrap-Around Artifacts on Background Phase Correction and Stroke Volume Quantification in Phase-Contrast MRI

Background: Phase-contrast MRI (PC-MRI) is routinely used to quantify stroke volume (SV), shunt flow, and mitral regurgitant fraction (1). Guidelines state that static tissue background phase correction should be performed as a post-processing step to minimize background phase errors that corrupt velocity measurements (2). A mask of static tissue is either manually or automatically calculated, and the background phase is usually modeled as a second-order polynomial, extrapolated to the entire field of view, and subtracted from each frame (3).

It is recommended to exclude voxels with phase-encode wrap-around artifacts from the static tissue mask. However, the effect on flow parameters has not been studied.

The objective of this study was to quantify the error in stroke volume calculations when images have various degrees of wrap-around artifact in the static tissue background phase correction masks.

Methods: IRB approval was obtained; waiver of informed consent was granted for retrospective data use. Breath-hold 2D through-plane PC-MRI studies of the ascending aorta (n=24, age 62 [51–70], 91% male) and aortic valve (n=17, 69 [58–75], 94% male) were retrospectively identified. The images were cropped to remove any empty space and wrap already present. Wrap-around artifact was simulated by overlaying increasing fractions of the complex image edges onto the opposite edges in the phase-encode direction. Static tissue background correction was then performed. Static tissue was identified in each study as voxels whose velocity standard deviation over all time frames was below the 15th percentile (3). A second-order polynomial was fit to the phase of all identified static voxels using least-squares, then subtracted from the phase of each frame. Stroke volume (SV) was integrated over the corrected velocities during the entire cardiac cycle using manual segmentations of the aortic lumen or across the aortic valve. The SV bias was calculated relative to the original, non-wrapped, background-corrected reference. Example images are shown in Figures 1 and 2.

Results: With FOV reduced to 80%, median SV bias increased to −4.1 [−6.3 to −2.4] ml at the ascending aorta and 0.6 [−0.1 to 1.8] ml at the aortic valve. In general, bias and interquartile range increased as the FOV decreased. Studies where wrap-around artifact entered the SV measurement region of interest were excluded from the analysis.

Conclusion: Including wrap-around voxels in static tissue background correction results in SV bias, increased variance, and should be avoided. The median bias is small, unpredictable, and varies depending on the slice orientation.

Figure 1. Example of the ascending aorta with (A) magnitude and (B) velocity images for a range of FOV reductions with static voxels indicated in red. (C) Estimated background velocities for the same range of FOV reductions. The region of interest used for stroke volume calculation is indicated by a blue line. Note the apparent differences in the estimated background velocity fields as a function of FOV reduction.
Figure 2. Example for the aortic valve with (A) magnitude and (B) velocity images for a range of field-of-view (FOV) reductions with static voxels indicated in red. (C) Estimated background velocities for the same range of FOV reductions. The region of interest used for stroke volume calculation is indicated by a blue line. Note the apparent differences in the estimated background velocity fields as a function of FOV reduction.
Figure 3. Boxplots of stroke volume (SV) differences relative to baseline (no wrap) for decreasing field-of-view (FOV). The bias and variance generally increase with reductions in the FOV.