Fig. 5.

Dice score after 20 interaction cycles for (a) whole tumor and (b) other tumor regions. Errors show standard error of the mean. Horizontal bars indicate that $p<0.001$ for the Wilcoxon signed-rank test of the two methods, with a dash indicating the method with poorer performance. This is essentially a cross section of Fig. 3 after 20 interactions. Again, in the larger regions (edema and whole tumor) MISCLASS, MISCLASS-B, and UNCERTAIN-MB perform comparably while in the other region MISCLASS-B and UNCERTAIN-MB dominate and UNCERTAIN and MISCLASS perform similarly. CERTAIN-MB performs poorly in all regions.