Fig. 4.
Quantitative results of the whole brain segmentation on the testing data. SLANT27 shows the smallest variation among the other baselines. UNesT achieves the overall best performance. Compared with SLANT27, UNesT further reduces the variation with improved median and quartiles of the DSC. * indicates statistically significant (p < 0.05) by Wilcoxon signed-rank test. Detailed quantitative performance comparison of 130 classes is shown in Fig A.17 in the supplementary material.