Table I.

Dice scores on the test sets used in this study, for the two tasks of segmenting the whole lesion (top) and the Gadolinium enhancing structures (bottom). BRATS results are calculated on the whole brain, stroke results in the lesion area. Inter-rater represents the overlap over multiple segmentations of the corresponding task and datasets done by human raters [1]. Reported are mean with standard deviation and median with median absolute deviance.

Task: complete lesion (FLAIR)	mean (±std)	median (±MAD)
BRATS glioma – generative (raw)	.58 (±.22)	.67 (±.11)
BRATS glioma – gener. (postproc.)	.62 (±.21)	.72 (±.11)
BRATS glioma – gener.-discr. (region)	.69 (±.24)	.79 (±.06)
BRATS glioma – gener.-discr. (pixel)	.78 (±.13)	.83 (±.05)
INTER-RATER (4 raters)	.86 (±.06)	.87 (±.06)
Zurich stroke	.78 (±.11)	.79 (±.07)
INTER-RATER (2 raters)	.79 (±.11)	.80 (±.12)

Task: enhancing core (T1c)	mean (±std)	median (±MAD)
BRATS glioma – generative (raw)	.46 (±.26)	.60 (±.15)
BRATS glioma – gener. (postproc.)	.51 (±.27)	.64 (±.15)
BRATS glioma – gener.-discr. (region)	.53 (±.27)	.66 (±.14)
BRATS glioma – gener.-discr. (pixel)	.54 (±.29)	.66 (±.15)
INTER-RATER (4 raters)	.76 (±.10)	.78 (±.08)
Zurich stroke	.45 (±.33)	.64 (±.18)
INTER-RATER (2 raters)	.82 (±.08)	.83 (±.05)