Meeting Banner
Abstract #3960

Undersampled Spectroscopic Imaging with Model-Based Reconstruction

Itthi Chatnuntawech1, Berkin Bilgic1, Elfar Adalsteinsson2, 3

1EECS, Massachusetts Institute of Technology, Cambridge, MA, United States; 2EECS, MIT, Cambridge, MA, United States; 3Harvard-MIT Division of Health Sciences and Technology, Cambridge, MA, United States

In this work, a two-step model-based method that leads to an accurate reconstruction from undersampled spectroscopic imaging data is proposed. This method takes advantage of a fast water reference scan to estimate a subset of (non-linear) unknowns, leaving only a few, linear unknowns to be determined in the next step. Then, a regularized optimization problem with prior knowledge on the structure of the data is formulated to reconstruct the spectroscopic imaging data. This method reduces acquisition time by undersampling while preserving high reconstruction quality. The proposed method yields significantly lower root mean square error than that of the conventional method which finds the minimum-norm solution without regularization.

Abstract PDF Presentation Video

Keywords

acceleration accuracy accurate acquisition adopted adopting advantage affects aids allows alternatively amplitudes approximately artifacts audience block blue bottom brain chemical choline chosen consisting constants contains corner decay decaying depicts determine determined deviations diagram division done errors evaluated even experiment exponentials expressed extending fast finds formulated formulation frequency fully greatly green gridded ground hand health imposing improve improves incorporated incorporates independently inside inversion investigators knowledge known leads least limitation limitations linear lipid locations long magenta manually maps masked metabolite metabolites middle minutes mitigates model modeled modeling must noise norm occurred operator optimization outer parametric post preserving prior problem processing proposed pulses quick radiology reconstruct reconstructed reconstruction reconstructions recovery reduces regularization regularized residual resolution respect respectively retrospectively sampled sciences scientists separate significantly simultaneously solution solved solving space spatial spectra spectral spectroscopic spectrum square step structure suppressed suppression tackle takes target technology term thesis transform truth unknowns upper variation various vector volume water white