Phase-Contrast (PC) cine MRI is an established method for the assessment of blood flow and tissue motion patterns in cardiovascular MRI. Prospective ECG-triggering bear several limitations due to incorrect trigger signals caused by respiratory motion, magnetohydrodynamic effects in high field MRI or changes of the heart rate. We present a highly accelerated self gated PC-MRI method based on continuously pseudo random sampled k-t-Sparse-Sense MRI.

Processing of Cardiac and Respiratory SG-Signal

A self-gating (SG) signal was acquired by performing an additional DC (i.e. without phase encoding) - projection along the frequency encoding direction consisting of 10 data points after each k-space line acquisition. Relevant information for cardiac and respiratory motion was extracted using first principle components of the SG Matrix. Thereupon the individual navigator signals for cardiac and respiratory motion were transformed using a complex wavelet transform at adapted scaling levels for respiratory and cardiac motion. The resulting phase-signal of the wavelet transform was then applied to directly sort k-space lines into different bins for cardiac and respiratory motion. Data of high respiratory motion were excluded from further data processing.

Imaging-Sequence

The MR-sequence was built on a 2D-FLASH sequence provided with additional 3D-motion-encoding gradients (centered four point encoding scheme) followed by 10-point DC-readout. The selection of k-space-lines was realized using a low-discrepancy random sequence generating Niederreiter-Numbers and was additionally weighted with a gauss-function to over-represent kspace-center selection¹. Additionally to the random sampling scheme the central kspace-lines (10% of the total k-space width) were acquired linearly for each time frame at the beginning of each measurement, to ensure fully sampled k-space-center.

Image Reconstruction

First coil maps were generated using the ESPIRiT-algorithm². The undersampled and reordered PC-k-t-data where subsequently reconstructed iteratevely with k-t-SPARSE-SENSE applying joint sparsity for the different flow encoding data-sets³. The single iteration steps were aimed to minimize the following objective function.

$$$D(x)=||F S_l x-y||_2+ \lambda ||Wx||_1$$$

Where S_l represent the coil sensitive-maps, F the fourier transform, y the measured k-space-data and W the applied sparsity transform. As a sparsity transform we applied a temporal fourier transform. The different iterations where performed using a soft threshold algorithm. The regularization parameter lambda was adapted in a decreasing manner for each regularization³.

Fig. 1.c shows the sampling pattern of a reordered 2D flow measurement for an undersampling factor of 6. The accompanying SG-signal is presented in Fig. 1.a. Flow visualization of the reconstructed MR-data and the underlying magnitude image can be seen in Fig. 1.d for left ventricular (LV) blood flow during early systole (imaging parameters: Matrix 256×174, FOV 300×174 mm², slice-thickness 4 mm, TE 5.6 ms, TR 10 ms). All data where reconstructed using Matlab. Performing 15 iteration steps resulted in a total reconstruction time of 2 minutes (8 cores I7).Total measurement time was 25-35 sec. In this paper we presented a highly accelerated self-gated PC-MRI-sequence based on free breathing random sampled data acquisition. Data acquired during respiratory motion as well as any other source of undesirable motion can be excluded from the post-processing enabling for free breathing during the whole measurement process. Moreover ECG-signal acquisition which is prone to distortions especially at higher field strength can be avoided. The high flexibility of data processing would also allow the correction of unstable heart rate during the measurement. The combination of these advantages yield high potential for the use in clinical applications.