Figure 3: Performance of all four methods in a representative dataset for R = 12.

A) Optimal reconstructions of simulated data at $\theta ={10}^{\circ }$ and $\mu =0.10$ for CD+r. The CD+r reconstruction is compared to the results from L1, LR, and CD, where the white numbers overlayed on the images reflect the NRMSE and SSIM, as labeled, computed across all flip angles. The greyscale bar represents fractional error between the ground truth (GT) and each reconstruction. L1 yields a higher NRMSE of 0.17 and a lower SSIM of 0.87, compared to NRMSE values of 0.13 for the remaining methods. This higher error in the L1 reconstruction is evident in the magnified anatomical images, where there is loss of edge information in the grey matter region adjacent to the ventricle. In the absence of VFA model error, CD+r yielded the highest SSIM value among the reconstructions. (B) Analogous to the information in (A) for the raw data analysis. The results from the raw data analysis are consistent with the results in panel A, where all NRMSEs and SSIM values are within 0.1 of their simulation-based counterparts. CD+r performed similarly to CD when model error was no longer accounted for.