Focus Session
SCMR 22nd Annual Scientific Sessions
Jonas Walheim, MSc
Doctoral Candidate
ETH Zurich
Hannes Dillinger, MSc
Doctoral Candidate
ETH Zurich
Richard Droste, MSc
Doctoral Candidate
University of Oxford
Sebastian Kozerke, PhD
Professor of Imaging Sciences and Biomedical Engineering
ETH Zurich and King's College London
Background:
4D Flow MRI is typically acquired with respiratory gating which leads to increased scan durations. An alternative approach is to use data from the entire respiratory motion cycle and resolve respiratory motion during image reconstruction [1]. This work presents a framework for imaging respiratory motion-resolved multipoint 4D Flow MRI (5D Flow) which makes scan time independent of respiratory motion and allows for accurate quantification of velocities and turbulent kinetic energy (TKE) [2] over a large dynamic range. In addition, flow can be assessed and compared for different respiratory motion states, yielding additional physiological information [3]–[5].
Methods:
The data acquisition is illustrated in Figure 1. 5D Flow was acquired with a pseudo-radial Cartesian sampling pattern [6], [7]. Respiratory motion was detected based on self-gating [8]. For reconstruction a locally low-rank approach [9]–[11] was applied to exploit correlations among heart phases and respiratory motion states. A Bayesian approach was employed to combine the data measured with 7 different velocity encodings (2x3 kv points + 1 kv = 0 reference) by maximizing the posterior probability of velocities and intra-voxel dephasing given the measured data [12].
Flow data in the aortic arch of 9 healthy volunteers was acquired on a 3T Philips Ingenia system (Philips Healthcare, Best, the Netherlands) using a Cartesian phase-contrast gradient-echo sequence with a spatial resolution of 2.5x2.5x2.5 mm3 and 25 cardiac phases. Velocities were encoded with 50 cm/s and 150 cm/s for each of the three directions. For comparison, a navigator-gated parallel imaging protocol [13] (4D Flow) was acquired with regular undersampling of R=2 in the in-plane phase encode direction, uniform VENC of 150 cm/s and a navigator window size of 5 mm.
Data analysis was performed in MATLAB. Peak velocities in planes along the aorta were assessed after applying a 3x3x3 median filter to the velocity field components and data obtained with 5D Flow and 4D Flow MRI were compared using a Bland-Altman analysis [14].
Results:
Exemplary results of magnitude and velocity magnitude images in inspiration and expiration obtained with 5D Flow are compared relative to the 4D Flow results in expiration in Figure 2. Figure 3 compares peak velocities assessed in cross-sections of the aorta in expiration in a Bland-Altman analysis. On average, peak velocities assessed with 5D Flow are higher than for 4D Flow (2.01 +/- 6.88%). TKE maps provided by 4D Flow show a high level of noise compared to 5D Flow. Net scan times for 5D Flow were 2.5 times lower compared to 4D Flow and showed less variation (7.17 ± 0.63 min vs. 17.79 ± 3.87 min).
Conclusion:
5D Flow MRI offers fixed scan durations independent of respiratory motion of the subject. Scan times can be reduced by a factor of 2.5 and the employed multipoint encoding provides velocity maps free of phase aliasing and allows assessing TKE.