Fig. 2. ²³Na MRI combined with and DWI improves tumour vs non-tumour classification accuracy.

a Corresponding ¹H TurboRARE images and reconstructed ADC maps (diffusion tensor imaging, b values of 100, 300 and 700 s/mm²) were obtained from mice bearing MDA-MB-231 xenograft tumours at 2, 3 and 4 weeks post implantation. Tumour (red), phantom (blue, 50 mM NaCl) and non-tumour (green) regions of interest (ROIs) are shown. b ADC in tumour, phantom and non-tumour regions at each imaging timepoint (week 2, n = 4; week 3, n = 5; week 4, n = 6. mean ± SEM). c Tumour volume does not correlate with tumour ADC (pooled ROIs across all timepoints, n = 15). d Neither mean nor maximum ²³Na signal (²³Na 2D cartesian acquisition) from tumour ROIs correlate with tumour volume (n = 16). Both maximum (e) and mean (f) ²³Na signal negatively correlate with ADC across all regions (tumour, red; non-tumour, blue, all timepoints pooled, n = 30). g Principal component analysis biplot of maximum ²³Na signal and ADC (data from e) with loading vectors for maximum ²³Na signal and ADC with 95% concentration ellipses. Receiver-operating characteristics are shown (h) for linear discriminant models trained on maximum ²³Na values alone, ADC alone, or combination (training data from e, test data from DTX control cohort (Fig. 5, n = 32 regions from all timepoints). Classification sensitivity and specificity are presented in the inset. P-values for significant differences between groups calculated using an unpaired, two-tailed Student’s t test and for correlation using a Pearson r test.