Figure 5.

Here the methods are compared in terms of RMSE of FA, MD, MK, and angular error for each input scan. The baseline and SHORE methods use all available shells while all other methods are given on the first shell of a lower b-value. On average, the Patch and Slice disentanglement models perform better in white and gray matter for both datasets across metrics. Notably the improvement in MK indicates the estimation of the second shell is successful. Wilcoxon signed-rank test shows that all methods are statistically significant (p-value<0.01).