Improving cardiac cine MRI on 3T using 2D k-t accelerated auto-calibrating parallel imaging

Variable-Density k-t Sampling (VDkt): As shown in Figure 1, the entire k-t space is divided into sub-bands based on distance from central k-space and these sub-bands are sampled using time-shifted acquisition with linearly increasing acceleration from center to outer bands. Static Tissue Removal (STR): Signals of static tissue voxels can be estimated from the original undersampled k-t data [3]. Such static tissue signals can be subtracted from k-space such that the subsequent k-t reconstruction processes only dynamic tissue voxels within a smaller FOV and thus produces higher image quality [3]. The same method can be used to improve k-t accelerated 2D cine, especially at apical slices where dynamic signals originates from a very small heart transection. Short-axis 2D cine SSFP data was acquired on 3 volunteers on a GE 3T with a 32-channel cardiac coil. Imaging parameters were: 360 × 270 mm2 FOV, 224 × 168 matrix, 15 8 mm slices, 40° flip angle, 8 breathholds. Full k-space was collected and offline downsampled to simulate k-t sampling with various accelerations: A. 5x outer & 1x center (net: 3.6x), B. 6x outer & 2x center (net: 5.0x); C. 7x outer & 3x center (net: 6.2x). For comparison, breath-held (15 s) 3D cine was also performed with 40° flip angle and 9x center & 3x outer acceleration (net 9.6x).

As shown in Figure 2, VDkt (d) reduces aliasing artifacts vs. regular k-t sampling (e) with same acceleration and STR suppresses residual artifacts (d vs. k). At mid-ventricular slice, VDkt 5.0x (c) provides image quality similar to full k-space reference (a). At apical slice, higher acceleration of 6.2x (j) is obtainable without visible aliasing or blurring compared to full k-space image (g). In comparison, 3D cine (f, l) suffers from much lower contrast, especially near apex, and more severe banding artifacts at inferoposterior endocardium.