Fig. 10.

Dice scores for grey matter (GM), compared with the lower and higher bound trained models, and the pseudo-labelling semi-supervised method for experiments with 1, 2, and 5 labeled training images. For GM segmentation, which is more challenging due to the narrow boundaries and shape of the cortex, our method outperformed the lower bound and pseudo models in experiments with 1 and 2 labeled images, but did not reach as high accuracy as the higher bound model.