Machine learning and neural net techniques are increasingly being applied to magnetic resonance studies, including NMR. Here, we present ISTANet - a novel neural network-based algorithm for NMR data processing. ISTANet's robust design, based on a working physical model, allows for both spectral reconstruction and linewidth deconvolution of non-uniformly sampled NMR experiments.
Initial results from ISTANet, as compared to both more traditional NUS techniques and other machine learning algorithms in the literature, are presented, for both synthetic and experimental data. The design and training strategies used for ISTANet, and measures for assessing its effectiveness are discussed, in addition to future challenges in desiging and comparing AI techniques for NMR data analysis.