. 2017 Sep 8;16:643–658. doi: 10.1016/j.nicl.2017.09.003

Table 2.

Table showing the results of each method over the whole dataset. Optimal parameter combinations indicated by *. Statistical differences between the closest competitor (optimised LPA) and the proposed method at a 5% significance level are in bold. For comparison, correlation between ground truth volumes and Fazekas scores is 0.829.

Method	Parameters	DSC	ASSD	HD	Prec.	Recall	ICC	Faz. corr.
LGA	κ = 0.3	0.382	5.77	48.6	0.925	0.265	0.693	0.782
LGA	t = 0.5	0.382	5.77	48.6	0.925	0.265	0.693	0.782
LPA	t = 0.5	0.536	2.60	37.3	0.926	0.416	0.874	0.846
LesionTOADS		0.497	2.74	34.3	0.667	0.498	0.488	0.358
LGA	κ = 0.11*	0.473	4.54	39.9	0.698	0.403	0.836	0.767
LGA	t = 0.01*	0.473	4.54	39.9	0.698	0.403	0.836	0.767
LPA	t = 0.15*	0.683	1.62	33.3	0.759	0.681	0.952	0.805
Proposed	w⁽²⁾ = 8*	0.703	1.23	38.6	0.763	0.695	0.985	0.862
Proposed	σ_γ = 2.5	0.703	1.23	38.6	0.763	0.695	0.985	0.862