Purpose

Magnetic resonance spectroscopic imaging (MRSI) of the brain provides insight into the distribution of metabolites, but is subject to comparatively low resolutions that limit the method’s usefulness. Approaches with higher resolution^1,2 were successfully introduced, but are in turn limited by increased measurement times. Another possibility is the application of super-resolution techniques to low resolution datasets to increase the information about metabolite distribution at high resolution. This work presents the validation of our patch-based super-resolution (PBSR) method on quantified N-acetyl-aspartate (NAA) and myoInositol (Ins) metabolite maps of multiple sclerosis (MS) patients at 7 T.

Methods

The PBSR method takes high resolution anatomical MR images (T₁-weighted and FLAIR) and low resolution MRSI images as input. The pipeline has three steps: (1) a preprocessing step, including (a) segmentation of MR images into grey matter, white matter (WM), cerebrospinal fluid and lesions³, (b) metabolite quantification of the acquired low resolution (LR) and high resolution (HR) MRSI data using the LCModel software, (2) a reconstruction step, aiming to find similar voxels (based on tissue segmentations and image intensities) in the neighbourhood of each central voxel, and (3) a mean correction step, imposing that the average of the L reconstructed values composing the low resolution voxel must be close to the corresponding original value of the low resolution image4. We validated the accuracy of our method on human datasets acquired in-vivo against three conventional interpolation techniques: nearest neighbours (NN), linear interpolation (LIN) and B-splines (BS). These datasets consisted of five MS patients (4 f, 1 m, age 32-46) that were measured with a Siemens 7 T Magnetom scanner and a 32-channel head coil (Nova Medical). Written informed consent and approval of the institutional review board were obtained. Using a CAIPRINHA-accelerated phase-encoded FID-MRSI sequence^1,5, a low resolution (TR = 600 ms, TE = 1.3 ms, 45° FA, 64×64 matrix, 3.4× 3.4× 8.0 mm³ voxel size, 6×acceleration, 5 min) and a high resolution (TR = 200 ms, TE = 1.3 ms, 29° FA, 100×100 matrix, 2.2× 2.2×8.0 mm³ voxel size, 4×acceleration, 6 min) single slice protocol were acquired, in addition to MPRAGE and FLAIR scans and a pre-saturation turbo-FLASH-based B₁⁺ map. Data processing utilised a MATLAB-based in-house routine⁶ that included MUSICAL⁷ coil combination, a spatial Hamming filter and the removal of lipid contamination by L2-regularisation⁸. After spectral fitted in the range of 1.8 to 4.2 ppm using LCModel, NAA and Ins maps were created. All four interpolation techniques were applied and their results compared.

Results

The acquired high resolution maps show the contrast between lesions and surrounding white matter which is not seen in the acquired LR maps. In the case of NAA (Fig.1), only PBSR achieves the same contrast, while for Ins (Fig. 2) the contrast is highest for LIN followed BS, NN and PBSR. In the resulting statistics (Fig. 3), PBSR has the highest averaged structural similarity index (aSSIM) and the best p-value and effect size for NAA for differentiating lesion voxels versus surrounding white matter, while for Ins this is the case for the acquired HR maps. In general, PBSR preserves tissue contrast and structural information best, matching well with the trend of the acquired HR MRSI.

Discussion

The results obtained in human in vivo datasets show that PBSR is a feasible and superior method for the up-sampling of quantified metabolite maps from MRSI. MRI-based weights allow to achieve a better tissue contrast than the other methods. This approach has a high potential for clinical neuroimaging applications at 7 T, as more metabolic information can be obtained from measurements with low resolution that can be conducted in a shorter measurement time. For future development, using absolutely quantified metabolite maps for PBSR may allow to show the methods accuracy even better.

Acknowledgements

This study was supported by TRANSACT (EU-FP7-PEOPLE-2012- ITN-316679); research support from the NIHR/WT Clinical Research Facility; the Austrian Science Fund (FWF): KLI-61; and the FFG Bridge Early Stage Grant #846505.