Table 1.

Total mean squared error (MSE) over 100 testing spectra for each acceleration factor.

Acceleration	D-UNet	${\ell }_1$
1.3x	0.039 ± 0.028	0.019 ± 0.0086
2x	0.017 ± 0.0061	0.068 ± 0.025
4x	0.033 ± 0.023	0.21 ± 0.096

A D-UNet model was trained to learn reconstruction for each sampling factor, and the results were compared to ${\ell }_1$-norm reconstruction as described in Eq. (3). Since the maximum signal is 1 for all spectra analyzed, the normalized MSE as a percentage for the D-UNet is 3.9%, 1.7%, and 3.3% for acceleration factors of 1.3x, 2x, and 4x, respectively. Similarly, the MSE as a percentage for the ${\ell }_1$-norm reconstruction is 1.9%, 6.8%, and 21% for acceleration factors of 1.3x, 2x, and 4x, respectively.