A variational network (VN) was implemented and trained on synthetic data to reconstruct multi-echo hyperpolarized 13C data acquired in the in vivo heart. VN reconstruction performance was studied using 2D and 3D synthetic data under low SNR conditions and for acceleration factors of 3 and 9, respectively. Relative to standard gradient descent based reconstruction, the network offers improved reconstruction accuracy and reduced signal leakage between metabolites while preserving information on lactate-to-bicarbonate ratios.
This abstract and the presentation materials are available to members only; a login is required.