Synopsis
An eight-coil rotating radiofrequency coil array (RRFCA) has been proposed1 to boost the SNR and improve imaging acceleration performance. However, existing acceleration methods for RRFCA are either based on SENSE,2 requiring rotational sensitivity maps,3 or have restrictions on rotation speed4 which undermines the performance of the RRFCA. Therefore, in this work, a novel GRAPPA-based5,6 algorithm, radial-GRAPPA-RRFCA, is proposed to address these issues and reliably reconstruct images from radial k-space. The algorithm has been validated with phantom images and human brain images.
Methods
The RF coil was simulated using Sim4Life (ZMT, Zurich, Switzerland) and the GRAPPA-RRFCA algorithm was developed using Matlab (Mathworks, Natick, MA). The algorithm first corrects the physical rotation of the RRFCA by implementing a realignment procedure. Then, while in the reference frame of a particular coil, the radial k-spaces are approximated as a collection of near-Cartesian grids. Interpolation methods are used to estimate weights then reconstruct the unacquired data that has resulted from the inherently under sampled RRFCA. Finally, a Hankel transform method is implemented producing coil-by-coil images directly from the now fully sampled radial k-spaces. An outline of the algorithm is depicted in Figure 1.
Results
As shown in Figure 2, the proposed method returns accurate image reconstructions of the phantom and brain. For the brain image, for example, the respective SSIM and RMSE of the chosen axial slice are 0.9912 and 0.0401.
Conclusion
High-quality images are reconstructed for the RRFCA without using sensitivity maps or conventional gridding operations for the radial k-space. In the future, the SNR improvement and imaging acceleration capability facilitated by this method will be quantitatively analysed.
Figure 1: An outline of the radial-GRAPPA-RRFCA algorithm. The three-step process to reconstruct under sampled radial k-space data resulting from the RRFCA.
Figure 2: Validation of GRAPPA-RRFCA using the numerical Shepp-Logan Phantom and an axial human brain image. Local SSIM maps are also shown. A Stationary RRFCA corresponds to a fully sampled conventional parallel imaging array. Rotating the RRFCA to four positions inherently under samples each coil by a factor of four.