Table 3.

Evaluation metrics based on estimated lesion volumes for the two MS datasets. The best value for each metric is shown in bold. Last row shows results after excluding worst performing algorithm.

	Correlation with true lesion volume	Correlation with number of lesions	MAE [ml]
Dataset I
SAMSEG	0.91	0.67	2.81
LST	0.88	0.55	3.51
nicMSlesions	0.89	0.43	5.44
U-Net	0.51	0.64	5.46
TrUE-Net	0.92	0.28	9.5
Majority voting	0.92	0.83	2.59
STAPLE	0.93	0.80	4.08
Dataset II
SAMSEG	0.96	0.78	6.86
LST	0.95	0.24	4.90
nicMSlesions	0.93	0.75	7.57
U-Net	0.56	0.69	14.0
TrUE-Net	0.91	-0.09	8.24
Majority voting	0.97	0.94	5.99
STAPLE	0.98	0.87	2.05