Frank Riemer^{1}, Marco Reisert^{2}, Bhavana S Solanky^{3}, Claudia AM Wheeler-Kingshott^{3,4,5}, Golay Xavier^{3}, Renate Grüner^{1,6}, and Ivan I Maximov^{7}

^{1}MMIV, Haukeland University Hospital, Bergen, Norway, ^{2}University Medical Center, Freiburg, Germany, ^{3}UCL Queen Square Institute of Neurology, London, United Kingdom, ^{4}Brain Connectivity Centre, IRCCS Mondino Foundation, Pavia, Italy, ^{5}Department of Brain and Behavioural Sciences, University of Pavia, Pavia, Italy, ^{6}Department of Physics and Technology, University of Bergen, Bergen, Norway, ^{7}Western Norway University of Applied Sciences, Bergen, Norway

Due to quadrupolar interactions, 23Na exhibits a bi-exponential T2. Previous approaches to estimate T2* have relied on simple least squares approaches or treating it as an inverse problem. Here we present a fast method based on Bayesian estimation and illustrate it on an in vivo dataset.

S = v

where v

An existing dataset from a previously published study [2] was utilised as the

Tissue probability maps for white and grey matter (WM and GM respectively) were created in SPM (UCL, London, UK) and all processing including the Bayesian estimation were carried out in Matlab 9.5 (the MathWorks, Natick, MA).

Figure 1 shows the results of the Bayesian model training on the simulated data. Importantly, the predicted relaxation times for both fast and slow time components are almost the same for the uniform and Gaussian fraction distributions and demonstrated a very good prediction rate.

The algorithm produces a map for the volume fractions and relaxation times each. Table 1 shows summary measures extracted using tissue probability maps (>0.95) for total white and grey matter.

WM and GM were not significantly different from one another for v

Compared to previous studies, the T

Using a uniform as opposed to a Gaussian distribution did not affect estimated T

In future, we plan to develop a strategy to find the optimal training parameters with an iterative procedure and test it on a numerical T2* phantom in simulation to investigate the effects of training parameters, resolution and SNR on contrast in the maps for v

Characterisation of sodium T2* may be able to disentangle cellularity changes in cancer and neurological diseases. The presented method could be used to create fast T2* maps in in vitro models where the lengthy MR acquisition protocols required for multi-echo based T2* mapping is not an issue.

Woessner D, NMR relaxation of spin-3/2 nuclei: Effects of structure, order, and dynamics in aqueous heterogeneous systems. Concepts Magn Reson 2001;13(5):294-325.

Riemer F, Solanky BS, Wheeler-Kingshott CAM, Golay X. Bi-exponential 23Na T2* component analysis in the human brain. NMR Biomed. 2018;31(5):e3899.

Blunck Y, Josan S, Taqdees SW, et al. 3D-multi-echo radial imaging of 23Na (3D-MERINA) for time-efficient multi-parameter tissue compartment mapping. MRM. 2018;79(4).

Reisert M, Kellner E, Dhital B, Henning J, Kiselev VG. Disentangling micro from mesostructure by diffusion MRI: A Bayesian approach. Neuroimage 2017; 147:964-975.

Figure 1. The results of Bayesian model training in prediction of fast and slow relaxation model terms. Two initial distributions were used for the simulations: Signal fractions were uniformly (left) and Gaussian (right) distributed. Sub-captions a) and b) are the signal fractions for fast and slow terms, c) and d) are the relaxation times in ms for the fast and slow components respectively.

Table 1: Summary measures extracted using tissue probability maps (>0.95) for total white and grey matter.

DOI: https://doi.org/10.58530/2022/1286