Author + information
Cardiovascular magnetic resonance imaging (CMR) is a non-invasive imaging technique of both function and structure of the cardiovascular system. The rapid imaging technique is essential since dynamic motion of heart is expected to be acquired. Traditionally, the compressed sensing (CS) technique is commonly applied to MRI, but the CS is not popular used in CMR due to its artifacts. In this study, we aimed to propose a two-level iterative reconstruction algorithm (TLIRA) that can perform reconstruction task while attenuate artifacts.
Variable sampling (VS) was used to undersample the k-space domain. Afterwards, we divided the under-sample k-space data into low-frequency and high-frequency bands. Two different thresholds were selected for low-frequency and high-frequency bands by grid-searching method. Finally, the reconstructed results over two bands were summarized.
A simulation experiment was carried out. We used a cardiac cine imaging and a cine-angiography. The acceleration factor was set to 10. We compared this proposed TLIRA with standard CS reconstruction. Peak signal-to-noise ratio (PSNR) was used to evaluate the performance of each method.
We find our TLIRA approach obtains better reconstruction results than standard CS reconstruction. Generally, the TLIRA achieved an improved PSNR of 2-3 dB over the cardiac cine imaging, and an improved PSNR of 1-2 dB over the cine-angiography.
Therefore, this proposed TLIRA approach is effective. It will be tested in the future with the combination of parallel imaging.