Superior Abdominal 4D Flow MRI Data Consistency with Adjusted Preprocessing Workflow and Noncontrast Acquisitions

Keller, E. J.; Collins, J. D.; Rigsby, C.; Carr, J. C.; Markl, M.; Schnell, S.

Acad Radiol. 2016 Dec 13; 24(3):350-358


RATIONALE AND OBJECTIVES: To assess the impact of an alternative preprocessing workflow on noncontrast- and contrast-enhanced abdominal four-dimensional flow magnetic resonance imaging (4D flow MRI) data consistency. MATERIALS AND METHODS: Twenty patients with cirrhosis and portal hypertension (5 women; 53 +/- 10 years old) underwent 4D flow MRI at 3.0T before and after administration of 0.03 mmol/kg of gadofosveset trisodium with velocity sensitivities of 100 and 50 cm/s for arterial and venous flow quantifications, respectively. 4D flow MRI data were preprocessed using the conventional workflow (workflow 1), applying noise filters prior to eddy current correction, and an alternative workflow (workflow 2), first correcting for eddy currents and using noise filtering only if needed for anti-aliasing. Vessel segmentation quality was ranked by independent reviewers and compared via Wilcoxon signed-rank tests. Flow quantification and conservation of mass at two portal and one arterial branch points were compared via paired t tests. RESULTS: Segmentation quality was significantly higher for workflow 2 (P < 0.05) with excellent interobserver agreement (kappa = 0.92). Workflow 2 resulted in larger flow values (P < 0.05) with improved conservation of mass (7.3 +/- 6.1% vs. 27.7 +/- 25.0%, P < 0.001 [portal]; 10.7 +/- 9.0% vs. 21.7 +/- 21.6%, P = 0.02 [arterial]). Peak velocities and abdominal aortic flow were similar (P > 0.05). Noncontrast acquisitions yielded significantly smaller portal flow values (P < 0.05) with improved conservation of mass (5.8 +/- 4.7% vs. 8.7 +/- 6.9%, P = 0.05 [portal]; 6.2 +/- 4.5% vs. 13.7 +/- 10.2%, P = 0.03 [arterial]). CONCLUSIONS: Superior abdominal 4D flow MRI data consistency was obtained by applying eddy current correction before any other data manipulation, using noise masking and velocity anti-aliasing cautiously, and using noncontrast acquisitions.

Read More on PubMed