Motivation: Magnetic resonance spectroscopic imaging (MRSI) enables the 3-dimensional visualization of metabolic concentrations in healthy subjects and patients. However, metabolic maps can be distorted due to artifacts originating from large water and lipid signals. Goal(s): The removal of nuisance signal in water unsuppressed 7T MRSI Approach: A Deep-Learning based nuisance identification neural network is trained to predict water and lipid signals, which are consequently subtracted from the original input. Results: Simulated data reveal a thorough removal of nuisance signals. Metabolic maps show an agreement with water suppressed MRSI from the same subject. Impact: WALINET+ is an initial attempt for a deep learning based removal of nuisance signals in water unsuppressed 7T MRSI, and has the potential of reducing acquisition times by several minutes.
Weiser et al. (Tue,) studied this question.