Table 2.

Agreement measures for Dice index, Jacarrd index, positive predictive value, true positive rate, false positive rate, and volume difference between lesions segmented by the automated methods and a neuroradiologist for 25 subjects with mild WML loads.

Automated segmentation methods	Dice index (B = 1, W = 0)	Jaccard index (B = 1, W = 0)	Positive predictive value (B = 1, W = 0)	True positive rate (B = 1, W = 0)	False positive rate (B = 0, W = 1)	Volume difference (B = 0, W = ∞)
Proposed method	0.2648 ± 0.134	0.1598 ± 0.10	0.3383 ± 0.218	0.2555 ± 0.143	0.3656 ± 0.189	0.5056 ± 0.435
TMOD	0.2397 ± 0.123 NS	0.1420 ± 0.09 NS	0.2180 ± 0.139 **	0.3351 ± 0.150 *	0.5594 ± 0.191 ***	1.4350 ± 2.005 **
SLS	0.1471 ± 0.131 ***	0.0851 ± 0.09 **	0.1680 ± 0.167 ***	0.1609 ± 0.148 **	0.4748 ± 0.196 *	0.7978 ± 0.653 *
LPA	0.1275 ± 0.080 ***	0.0700 ± 0.05 ***	0.1331 ± 0.122 ***	0.1913 ± 0.157 *	0.5417 ± 0.229 **	1.4973 ± 1.673 **

Dice index, Jaccard index, positive predictive value, true positive rate, false positive rate, and volume difference are presented as mean ± standard deviation. B indicates Best, and W is Worst. NS is not statistically significantly different compared with the proposed method (P > 0.05); * is significant at P ≤ 0.05; ** is significant at P ≤ 0.01, and *** is significant at P ≤ 0.001. All paired t-tests were Bonferroni corrected for four comparisons.