Optimized Reconstruction Parameters for Noise Modeling in Multi-Task Bayesian Compressed Sensing for Sparse 2D Spectroscopy

Trina Kok¹, Elfar Adalsteinsson¹, ²

Metabolite spectra could be simulated and included as prior spectral information in the reconstruction of under-sampled 2D spectra via Multi-Task (MT) Bayesian CS. We previously showed that MT Bayesian CS successfully reconstructed peaks of Glu and Gln even with imperfect simulated metabolite spectra as priors. Spectroscopy data are intrinsically low SNR and here we extend previous work by incorporating noise modeling parameters for MT Bayesian CS and demonstrate improved reconstruction performance for under-sampled 2D spectra in CTPRESS compared to reconstruction without explicit noise modeling.