In this work, we evaluate the effect of a self-calibrated image-based non-linear background phase correction¹ on 2D Phase Contrast MRI data and validate it against gold-standard data after static phantom correction² in a pediatric and congenital heart disease population.

Patients: 109 patients ages 5 months to 60 years (mean age 15 years) with 227 PC sequences (100 aorta, 78 main pulmonary artery (MPA), 25 right pulmonary artery (RPA) and 24 left pulmonary artery (LPA)) from the New York-Presbyterian Morgan Stanley Children’s Hospital congenital CMR program between January 1, 2015 and June 30, 2015 were retrospectively identified. PC MRI free-breathing (respiratory triggered or multiple averages) data were acquired using two GE Signa HDx 1.5T scanners (Waukesha, WI) with commercially available RF coils as part of routine clinical CMR studies. Static phantom correction² was performed for each of the images.

Post-hoc correction: Nonlinear self-calibrated phase-contrast (SCPC)¹ postprocessing using 5 terms (constant + XYZ + concomitant field) was applied in all the PC series using Matlab 2013b. Static tissue detection incorporated an automated iterative removal of outliers and a higher weighting of velocities from the quiescent cardiac phase to reduce effects from flow artifacts at systole. In both cases, over-fitting was prevented. The algorithm incorporates automatic failure-mode detection when static tissue is not properly detectable or does not reflect the background phase within the FoV³. There are three main automatic failure-mode types: 1) Appearance of wrapping (aliasing in the FoV)³; 2) Less than a percentage value (set to 10% in this study) of the image FoV is detected as static tissue; and 3) Velocity to noise ratio (VNR) in the static tissue is less than a defined threshold (set to 0.02 in this study).

Analyses: Mean and standard-deviation velocity values in the static tissue of the flow images for all vessels were computed before and after non-linear correction. Additionally, velocities within a 4-cm-radius region at isocenter (near the great vessels) after non-linear correction and without correction were compared with ground-truth stationary phantom correction¹. Flow was evaluated using identical ROI contours for the same dataset without correction, after non-linear correction and evaluated against the static-phantom correction within the same ROI. Intraclass correlation (ICC) was calculated to compare static phantom and self-calibrated corrected data. Paired t-test of pairwise difference of the pulmonary-to-systemic flow ratio, Qp/Qs, between the uncorrected data and self-calibrated data versus the static-phantom corrected data (gold-standard) was computed in patients without known intracardiac shunts or significant valvular regurgitation.

Eleven PC series were discarded automatically by the algorithm due to insufficient detected static tissue (failure-mode type 2), and four more series were discarded due to incorrect static-phantom acquisition. For visualization purposes, Figure 1 shows the mean velocity maps in two acquisition planes (aorta and pulmonary acquisition planes) of a 1-year-old patient for the uncorrected data, after self-calibrated correction and after static phantom subtraction. The maps after both corrections appear visually very similar; however in areas marked with white arrows, the static tissue after self-calibrated correction appears more homogenous and closer to velocity values of zero than after static phantom correction. Figure 2 and Figure 3 show that non-linear correction reduced the background velocity mean error below ±6 mm/s with reduced variability. The net flow values obtained after self-calibrated correction showed strong agreement with phantom-corrected net flow measurements for all vessel types (ICC = 0.98): aorta (ICC = 0.95), MPA (ICC = 0.98), LPA (ICC = 0.99) and RPA (ICC = 0.99). Self-calibrated correction Qp/Qs showed statistically significant (p<0.04) lower mean error (|mean diff Qp/Qs| 0.069, SD diff Qp/Qs 0.10) than the Qp/Qs obtained from uncorrected data (|mean diff Qp/Qs| 0.1, SD diff Qp/Qs 0.14) with respect to the static-phantom corrected Qp/Qs in 39 patients without shunts or regurgitation.Self-calibrated non-linear background phase correction showed a very high level of agreement in great vessel net flow measurements and Qp/Qs ratios with phantom correction in a pediatric and follow-up congenital-heart-disease population. Although static-phantom correction is considered the gold-standard to correct for background phase correction in PC MRI, image-based correction algorithms, as the one evaluated here, have the potential to perform better as the background phase is evaluated under identical conditions, with the same coil loading, body shape, tissue susceptibilities, etc. These results are also encouraging for the future of accurate quantitation of 4D flow measurements as a static-phantom acquisition would prohibitively increase CMR study time.