Spatiospectral image processing workflow considerations for advanced MR spectroscopy of the brain

Abstract

Magnetic resonance spectroscopy (MRS) is one of the few non-invasive imaging modalities capable of making neurochemical and metabolic measurements in vivo. Traditionally, the clinical utility of MRS has been narrow. The most common use has been the "single-voxel spectroscopy" variant to discern the presence of a lactate peak in the spectra in one location in the brain, typically to evaluate for ischemia in neonates. Thus, the reduction of rich spectral data to a binary variable has not classically necessitated much signal processing. However, scanners have become more powerful and MRS sequences more advanced, increasing data complexity and adding 2 to 3 spatial dimensions in addition to the spectral one. The result is a spatially- and spectrally-variant MRS image ripe for image processing innovation. Despite this potential, the logistics for robustly accessing and manipulating MRS data across different scanners, data formats, and software standards remain unclear. Thus, as research into MRS advances, there is a clear need to better characterize its image processing considerations to facilitate innovation from scientists and engineers. Building on established neuroimaging standards, we describe a framework for manipulating these images that generalizes to the voxel, spectral, and metabolite level across space and multiple imaging sites while integrating with LCModel, a widely used quantitative MRS peak-fitting platform. In doing so, we provide examples to demonstrate the advantages of such a workflow in relation to recent publications and with new data. Overall, we hope our characterizations will lower the barrier of entry to MRS processing for neuroimaging researchers.

Keywords: LCModel; chemical shift imaging; image processing; magnetic resonance spectroscopy; single-voxel spectroscopy; spectral peak fitting.

Figure 1.

Overview of MRS approaches. (a) Clinically, SVS spectra are often visually reduced to a binary variable representing the presence or absence of a single metabolite peak. This approach does not take advantage of either spectral (or metabolite) or spatial information. (b) In research, SVS is used to quantify different metabolites in a single location, thus adding spectral information but lacking spatial information. (c) Up-and-coming CSI sequences are used in research to quantify spectral metabolite information at different locations, providing the opportunity for rich spatiospectral processing and inference. Lac = lactate.

Figure 2.

Existing and proposed workflows. (a) Existing fMRI and dMRI workflows rely on the NIFTI file standard to remove variability in DICOMs and facilitate spatiotemporal or spatial with gradient processing, respectively. (b) The fragmented data landscape for MRS has come to support spectral processing, but at the cost of increased logistical variability and minimal spatial processing support. (c) The proposed approach aligns MRS workflows with existing neuroimaging frameworks, replacing the temporal/gradient dimension with a spectral one, to facilitate spatiospectral processing and simplify entry into MRS analysis for those familiar with traditional MRI standards.

Figure 3.

Variability in MRS data between scanners. As seen in the plotted real components of the complex-valued spectra, Philips MRS DICOMs contain both water-suppressed and water-unsuppressed data (a) whereas Siemens DICOMs contain only suppressed (b). Further, the signal amplitude between suppressed spectra can vary widely between scanner manufacturers. a.u. = arbitrary units.

Figure 4.

The low resolution of CSI necessitates verification of spatial orientation. (a) Brain anatomy is largely left-right symmetric, especially when the resolution is on the order of cm with CSI, making it difficult to discern orientation accuracy. (b) Coarse asymmetric phantom studies can help ensure DICOM to NIFTI conversions are accurate. The MRS heat maps are produced by collapsing the spectral dimension across ppm with the log average real amplitude.

Figure 5.

Spatiospectral preprocessing and quality assurance of CSI in a representative sample. Plotting the interquartile spectra across the brain in gray and outliers in red, we can leverage the spatial information to identify outlying spectra and exclude those voxels from further analysis. a.u. = arbitrary units.

Figure 6.

Spatiospectral inference. Using the spatial information in CSI and T1w MRI segmentations (a) captured by the proposed workflow, voxel-wise weights for consolidating spectra for each tissue are computed (b), allowing tissue-weighted spectra to be computed in this representative sample. (c) These spectra can then be spectrally processed to obtain metabolite ratios. This technique revealed baseline differences (** p < 0.005, Wilcoxon sign-rank test) in NAA/Cr between WM and dGM in the T1D cohort. (d) In multifocal SVS, multiple SVS locations can be measured to obtain spatial coverage. A representative sample from the RA cohort is shown. (e) Using the proposed framework, multifocal SVS studies also can be spatiospectrally processed to identify statistically significant associations of WA and LWI with neurochemical markers in different cortical gyri (p-values uncorrected from linear models controlling for age and handedness). a.u. = arbitrary units.