Automated Cardiac Valve Tracking for Flow Quantification with Four-dimensional Flow MRI

Purpose To compare four-dimensional flow MRI with automated valve tracking to manual valve tracking in patients with acquired or congenital heart disease and healthy volunteers. Materials and Methods In this retrospective study, data were collected from 114 patients and 46 volunteers who underwent four-dimensional flow MRI at 1.5 T or 3.0 T from 2006 through 2017. Among the 114 patients, 33 had acquired and 81 had congenital heart disease (median age, 17 years; interquartile range IQR, 13-49 years), 51 (45%) were women, and 63 (55%) were men. Among the 46 volunteers (median age, 28 years; IQR, 22-36 years), there were 19 (41%) women and 27 (59%) men. Two orthogonal cine views of each valve were used for valve tracking. Wilcoxon signed-rank test was used to compare analysis times, net forward volumes (NFVs), and regurgitant fractions. Intra- and interobserver variability was tested by using intraclass correlation coefficients (ICCs). Results Analysis time was shorter for automated versus manual tracking (all patients, 14 minutes IQR, 12-15 minutes vs 25 minutes IQR, 20-25 minutes; P < .001). Although overall differences in NFV and regurgitant fraction were comparable between both methods, NFV variation over four valves was smaller for automated versus manual tracking (all patients, 4.9% IQR, 3.3%-6.7% vs 9.8% IQR, 5.1%-14.7%, respectively; P < .001). Regurgitation severity was discordant for seven pulmonary valves, 22 mitral valves, and 21 tricuspid valves. Intra- and interobserver agreement for automated tracking was excellent for NFV assessment (intra- and interobserver, ICC ≥ 0.99) and strong to excellent for regurgitant fraction assessment (intraobserver, ICC ≥ 0.94; interobserver, ICC ≥ 0.89). Conclusion Automated valve tracking reduces analysis time and improves reliability of valvular flow quantification with four-dimensional flow MRI in patients with acquired or congenital heart disease and in healthy volunteers. © RSNA, 2018 Online supplemental material is available for this article. See also the editorial by François in this issue.