In the brain, sodium plays a key role in neuronal action potentials, mediates the transport of metabolic substrates through cell membranes and is involved in osmoregulation and pH regulation. Therefore sodium (²³Na) MRI shows great promise as a marker for cerebral metabolic dysfunction in studying stroke, cancer and neurodegenerative pathologies. However, the ²³Na MRI signal is heightened in cerebrospinal fluid (CSF)¹, such that measurements of tissue sodium can be biased by tissue atrophy. Another cerebral fluid compartment, which has not been considered previously, is cerebral blood vessels, occupying approximately 5% of the cerebral volume, with 140-150 mM sodium concentrations². The space occupied by cerebral blood vessels may be perturbed in pathology, such as cancer (angiogenesis) or stroke. In this study, we consider the ²³Na MRI signal from cerebral veins, to assess whether this is significantly different from GM and WM. Three ²³Na sequences were investigated: a density-weighted (DW) sequence, a sequence sensitive to the fast T₂f component (PACMAN)³ and an inversion recovery fluid attenuated sequence (SIRFLA)⁴.Five healthy participants undertook this study (19-30 yr; 2F/3M) on a Varian Inova 4.7 T system. For anatomical segmentation, whole brain MPRAGE (1 mm isotropic; TR/TI/TE = 508.5/300/4.5 ms) and T₂*-weighted FLASH (SWI) (0.8×0.8×4 mm³; TR/TE = 1540/15 ms) were acquired using a ¹H birdcage transmit head coil and 4-element receive array. A single-tuned ²³Na birdcage head coil was used to acquire whole head images (3.2×3.2×6.4 mm³; twisted projection). DW (TR/TE = 85/0.11 ms; flip angle = 30°) and PACMAN³ (TR/TE = 25/2.5 ms; flip angle = 110°) images were acquired on all five participants, whilst an additional SIRFLA⁴ sequence was acquired in two of the participants (TR/TI/TE = 150/37/0.22 ms; flip angle = 64°). Tissue segmentation was performed on the MPRAGE dataset (FSL BET & FAST), giving gray matter (GM), white matter (WM) and CSF partial volume estimates (PVEs) at higher spatial resolution than the sodium images. Veins were segmented using the SWI dataset (Matlab; 5 mm FWHM Gaussian low-pass filter and intensity threshold). Vein and tissue segmentations were co-registered and down-sampled to the sodium images (SPM). The respective sodium signal contributions of each compartment were resolved by fitting each voxel’s vein, CSF, GM and WM proportions to the sodium signal using a multiple linear regression. Signals were normalized to WM for comparison across subjects, since WM had less partial voluming with CSF and veins than GM, so smaller signal variance.The vein segmentation defined 2.8±0.3% of the brain as venous, example data in Figure 1a. The venous contribution to the sodium signal is significantly higher than for GM and WM (P<0.05, Bonferroni corrected) for both DW and PACMAN data (Figure 1b; Table 1). The enhanced venous ²³Na DW signal suggests higher sodium concentrations in cerebral veins than in GM and WM, while the PACMAN signal suggests a large T₂f increase in the venous tissue, similar to CSF. Statistical testing was not performed on the SIRFLA data, due to the low sample size (N = 2), but the venous signal appears less affected by the fluid attenuated sequence than CSF (Figure 1b; Table 1), consistent with an intermediate venous T₁ relaxation⁴ between tissue and CSF. Similar results were found using ROI-based analysis, where voxels with 10% or greater CSF were excluded. Venous DW and PACMAN sodium signal remained higher than GM and WM (P<0.05), suggesting that the heightened venous signal is not due to spatial overlap with neighboring CSF.

The heightened venous signal will contribute to a partial volume contamination of the sodium signal in tissues of interest, similar to that caused by CSF. Whilst large veins, occupying 2.8% of the brain volume, were investigated here, a typical tissue voxel will include approximately 5% microvasculature. These small blood vessels, not resolved in the SWI dataset, will also have a heightened sodium signal, scaling proportionally with volume occupied. The vascular sodium signal will also be an important factor for studies attempting to use sodium MRI as a method for direct detection of neuronal activity⁵, where focal blood volume increases will cause an increased signal. The DW data presented here has a shorter TR than typically used, but a much lower flip angle, yielding a small expected T₁ weighting of 5% in CSF at 4.7T (T₁ = 65 ms). This will lead to slight attenuation of the CSF and potentially venous signals. This abstract only considers the venous half of the cerebrovasculature, but future work will identify arteries using an angiogram, encompassing the full sodium signal contribution from cerebral blood vessels.