Table 1:

WMH Challenge Results. Sources→Target: Trained on Sources and tested on Target among Amsterdam, Singapore, and Utrecht. avg: Average of the target results. gain: avg gain over the respective DeepAll.

Model	A+S→U	U+S→A	A+U→S	avg	gain
DSC ↑ (Higher is better)
DeepAll ([4] Setup)	0.430	0.674	0.682	0.595	-
BigAug [10]	0.534	0.691	0.711	0.645	0.050
UDA [4]	0.529	0.737	0.782	0.683	0.087
DeepAll (for Ours)	0.183	0.619	0.781	0.528	-
DANN	0.315	0.674	0.773	0.587	0.060
Mixup	0.619	0.691	0.835	0.715	0.187
MixDANN	0.694	0.700	0.839	0.744	0.217
H95 ↓ (Lower is better)
DeepAll (Our Setup)	11.46	11.51	9.22	10.73	-
BigAug	9.49	9.77	8.25	9.17	−1.56
UDA (Full Set)	10.01	7.53	7.51	8.35	−2.38
DeepAll	42.69	18.05	4.56	21.77	-
DANN	38.56	15.48	5.15	19.73	−2.03
Mixup	24.08	13.21	5.70	14.33	−7.44
MixDANN	20.57	12.75	3.10	12.14	−9.63
AVD ↓ (Lower is better)
DeepAll	54.84	37.60	45.95	46.13	-
BigAug	47.46	30.64	35.41	37.84	−8.29
UDA (Full Set)	54.95	30.97	22.14	36.02	−10.11
DeepAll	384.19	43.28	23.26	150.24	-
DANN	134.03	26.65	24.09	61.59	−88.66
Mixup	42.09	33.47	13.41	29.66	−120.59
MixDANN	23.40	26.48	12.81	20.89	−129.35
Lesion Recall ↑ (Higher is better)
DeepAll	0.634	0.692	0.641	0.656	-
BigAug	0.643	0.709	0.691	0.681	0.025
UDA (Full Set)	0.652	0.841	0.754	0.749	0.093
DeepAll	0.309	0.623	0.705	0.546	-
DANN	0.349	0.630	0.740	0.573	0.028
Mixup	0.556	0.700	0.790	0.682	0.136
MixDANN	0.604	0.685	0.797	0.695	0.150
Lesion F1 ↑ (Higher is better)
DeepAll	0.561	0.673	0.592	0.609	-
BigAug	0.577	0.704	0.651	0.644	0.035
UDA (Full Set)	0.546	0.739	0.649	0.645	0.036
DeepAll	0.288	0.554	0.697	0.513	-
DANN	0.309	0.610	0.708	0.542	0.029
Mixup	0.515	0.642	0.724	0.627	0.114
MixDANN	0.602	0.651	0.728	0.660	0.147