Sodium MRI has the potential to probe the tumour microenvironment and provide quantitative biomarkers of treatment response in tumours, based on its sensitivity to cell viability and tissue microstructure^1,2. For example, total sodium concentration (TSC) has been shown to be elevated in many tumours^1,3, and is sensitive to chemotherapy in preclinical experiments⁴. If TSC is to have clinical utility as a biomarker in oncology, values should be repeatable within a given subject and comparable across different scanners and sites. This work presents preliminary results on the repeatability and reproducibility of TSC values in healthy volunteers, providing an initial step in the technical validation of sodium MRI-derived biomarkers⁵.Sodium brain MRI was performed on two male volunteers (aged 35 and 27) scanned twice at two sites, A and B. One site used a GE 3 T MR750, and the other a Philips 3 T Achieva, and both used a ¹H/²³Na dual-tuned head coil (RAPID Biomedical GmbH, Rimpar, Germany). Scan-rescan repeatability was assessed after a short time outside of the scanner of between 20 and 80 minutes. Inter-site/inter-scanner reproducibility was assessed by performing the above protocol at sites A and B, up to 3 days apart. The scan protocol consisted of a ¹H T₁-weighted scan, followed by a ²³Na 3D Cones⁶ acquisition: 4x4x4 mm³, 60x60x60 matrix, TR=100 ms, TE=0.5 ms, FA=90°, readout=10 ms, 3 averages, 11 minutes, G_max=30 mT/m. 6552 (2184 x 3 averages) readouts were performed at each site, with an additional 3 dummy acquisitions at site A. 3D Cones images were reconstructed using k-space density compensation weights^7,8 and the Image Reconstruction Toolbox⁹. Two effective matrix sizes, m = 250 and 350, were investigated for the weights calculation, both using an oversampling factor of 4. The same calibration phantoms (consisting of 4% agar, NaCl concentrations: 40, 80 mM) were used for all scans, and ROI mean signals in the phantoms were used to derive TSC maps, after applying a power image noise correction¹⁰. Manually-defined bilateral ROIs were drawn in cerebrospinal fluid (CSF), grey matter (GM), white matter (WM) and vitreous humour (VH), pooling all voxels for a given region. TSC repeatability was assessed using the coefficient of variation (CoV) of ROI median values for scans 1 and 2; mean repeatability across the 4 ROIs is reported for each subject at each site. Reproducibility was assessed using the CoV of ROI median values obtained for scan 1 at each site, again reporting mean values across the 4 ROIs.Figure 1 shows example images (subject 1), from each scan at the two sites, for reconstructions with m = 250 and 350. For a given m, images from site B were noticeably smoother than those from site A, a trend observed in both subjects. These results suggest that the weights calculation should be adjusted for each scanner, to prevent excessive smoothing. As images from site A using m=350 and from site B using m=250 were visually most similar, these datasets were used in the subsequent analysis. TSC maps are shown in Figure 2 (subject 2), and Figure 3 shows boxplots for each ROI in each subject, for both scans at both sites. General trends were consistent across all scans, with VH TSC higher than CSF, and similar values observed in GM and WM; mean±SD TSC for all scans were 92±11 mM (CSF), 24±6 mM (GM), 25±5 mM (WM) and 115±14 mM (VH). Precision may be improved by considering more ROIs or using GM/WM/CSF segmentation; note also that CSF and VH values are likely to be underestimated due to partial T₁ recovery, and may suffer from pulsation and movement artefacts. Scan-rescan repeatability CoVs were consistently higher at site A than site B, 11% vs 2% (subject 1), and 7% vs 2% (subject 2), suggesting TSC measurements are more repeatable at site B than site A. It is likely that these findings are affected by differing levels of image artefact and the k-space weights used, with the generally smoother images from site B yielding higher repeatability. Inter-site/inter-scanner reproducibility CoVs for scan 1 were 4% and 9% for subjects 1 and 2, respectively.

Consistent TSC estimates were observed across repeated scans at different sites with different scanners. k-space weights are an important consideration when comparing multi-site non-Cartesian data; this will be investigated in future work, along with phantom and analytical point spread function analysis accounting for site-specific SNR. These developments will aid the standardisation of the reconstruction and analysis of sodium images acquired at different sites, in order to robustly evaluate image quality and quantification.