Volumetric Patch-Based Super-Resolution Reconstruction of Hyperpolarized 13C Cardiac MRI

The achievable spatial resolution of 13C metabolic images acquired with hyperpolarized 13C-pyruvate is worse than 1H images typically by an order of magnitude due to the rapidly decaying hyperpolarized signals and the low gyromagnetic ratio of 13C. This study is to develop and characterize a volumetric patch-based super-resolution reconstruction algorithm that enhances spatial resolution 13C cardiac MRI by utilizing structural information from 1H MRI. The reconstruction procedure comprises anatomical segmentation from high-resolution 1H MRI, calculation of a patch-based weight matrix, and iterative reconstruction of high-resolution multi-slice 13C MRI. The method was tested with a multi-compartmental digital phantom for optimizing the patch size and an anthropomorphic cardiac MR phantom for validating the performance. Finally, the method was applied to human cardiac 13C images, acquired with an injection of hyperpolarized [1-13C]pyruvate. The phantom studies demonstrated that high-resolution multi-slice 13C images, reconstructed from a single-slice low-resolution input 13C image, retained the signal intensity range. The reconstruction accuracy was asymptotically improved as the patch size increased whereas intra-segmental spatial fluctuations were preserved better with smaller patches. However, a structurally non-identified tissue region was not restored regardless of the patch size. The cardiac MR phantom and the human cardiac images demonstrated improved spatial resolutions in the reconstructed images ( $10\times 10 \times 30$ mm3/voxel to $2\times 2 \times 5$ mm3/voxel). The volumetric patch-based super-resolution method reconstructs multi-slice high-resolution of 13C images, enhancing the cardiac structure, while preserving the quantitative accuracy. The proposed method is applicable to other multi-modal images that suffer from limited spatial resolution.