Multipoint 5D flow cardiovascular magnetic resonance - accelerated cardiac- and respiratory-motion resolved mapping of mean and turbulent velocities

Figures

• 

  Fig. 1. 5D Flow data acquisition. a Data are sampled using a Cartesian pseudo-radial tiny Golden angle sampling pattern [33]. b Respiratory motion state detection is derived from profile d(kx, ky = 0, kz = 0), which is repeatedly acquired and processed using a combination of principal component analysis, lowpass filtering, and coil-clustering [36]. Each spoke is composed of 21 k-space profiles, one of them always through the k-space center . With Tr = 3.3. ms, respiratory motion is sampled at ca. 14 Hz. c Acquired data are sorted into four discrete respiratory motion states such that the acceleration factor for each motion state is similar

• 

  Fig. 2. Data driven respiratory motion detection. a Respiratory motion is extracted from the profiles through the k-space center d(kx, ky = 0, kz = 0). The temporal evolution of readouts through the k-space center shows a modulation with respiratory motion and cardiac motion. b The 2nd principal component provides the main variation of the signal. c A lowpass filter limits the spectral components to the expected frequency range of respiratory motion. d) Coil clustering is used to combine the signals from different channels and provides a final estimate of respiratory motion

• 

  Fig. 3. Image reconstruction using locally low rank approach followed by Bayesian multipoint unfolding. a) A locally low-rank approach is employed for each velocity encoding separately. The locally low-rank model divides the image into 3-dimensional patches. Each patch is reordered into local Casorati matrices for which a low rank is enforced by penalizing the nuclear norm. Compared to a global Casorati matrix, the values of the singular values decrease more rapidly. b) Following reconstruction of the individual velocity encodings, for each Cartesian direction the different velocity encodings kv are combined using a Bayesian multipoint approach. A Bayesian probability model [4] provides posterior probabilities for mean velocity v and intra-voxel standard deviation σ given the measured signal S. v and σ are chosen such that the posterior probability is maximized, providing maps of turbulent kinetic energy (TKE) and mean velocities

• 

  Fig. 4. Analysis of navigator-based respiratory gating in standard 4D Flow CMR. a) For an exemplary 4D Flow scan the diaphragm position measured by the pencil beam navigator over time is shown together with accepted data (red) and rejected data (yellow). When the navigator signal is within the acceptance window, data of all cardiac phases are accepted until the next navigator position is obtained. As a result, the effective acceptance window of the data varies as a function of heart phase and becomes wider towards diastole as shown in b)

• 

  Fig. 5. Comparison of 5D Flow LLR reconstructions in expiration relative to reconstructions using end-expiratory data only (EXP LLR) and without gating (NG LLR). 5D Flow LLR shows reduced aliasing artifacts in the magnitude, velocity and TKE maps. For both EXP LLR and NG LLR, residual motion artifacts are present in the magnitude images and increased noise is observed in the TKE maps

• 

  Fig. 6. Results in systole comparing 5D Flow LLR with 5D Flow TV and standard 4D Flow. For 5D Flow TV and the 4D Flow reference residual aliasing is observed in the magnitude images. Moreover, the TKE maps obtained with 5D Flow TV and the 4D Flow reference show more noise compared to 5D Flow LLR. Exemplary slices in the ascending aorta (AA) and descending aorta (DA) show qualitatively similar results for through-plane velocities in the ascending aorta whereas in the descending aorta artifacts can be observed in the in-plane velocity components for 5D Flow TV

• 

  Fig. 7. Comparison of TKE maps in systole for the 4D Flow reference versus 5D Flow LLR. In the 4D Flow reference data increased noise compared to 5D Flow LLR is seen. Moreover, TKE maps from 4D Flow show high values in the descending aorta whereas TKE values are lower in the descending aorta for most cases with 5D Flow LLR

• 

  Fig. 8. a) Exemplary magnitude and velocity magnitude images for different respiratory bins reconstructed with 5D LLR. Image quality reduces from end expiration to end inspiration and artifacts can be observed in the descending aorta in the magnitude and velocity images reconstructed in end inspiration. b): The respiratory curve shows that the expiratory bins have a narrower range of motion than inspiratory bins

• 

  Fig. 9. Quantitative analysis of results. Scan times, peak through-plane velocities and peak flow for 5D Flow LLR reconstructions are compared to the 4D Flow reference. Duration of the 5D Flow LLR acquisition is on average ca. 4.5 times lower than for 4D Ref. Bland-Altman analysis shows a mean difference of 3.1% and − 2.4% and limits of agreement of ±8.6% and ±13.6%