Table 3.

Cases with no lesion mask identified, or lesions misclassified. Cases with no lesion mask identified yielded an empty file (containing only 0 value), and cases in which the lesion was misclassified contained lesioned voxels, but had no voxels overlapping with the expert segmentation. For comparisons between algorithms, we removed cases with no lesions identified (24 ALI + 23 LINDA – 8 overlapping = 39), and removed the 10 cases in which all three algorithms misclassified the lesion

	No lesion identified	Lesion misclassified	Total cases
ALI	24	28	52
lesion_gnb	0	39	39
LINDA	23	45	68
Overlapping cases	8 (ALI + LINDA only)	10 across all three approaches	32 across three approaches

Abbreviations: ALI, automated lesion identification; lesionGnb, Gaussian naïve Bayes lesion detection; LHS, left hemisphere stroke; LINDA, lesion identification with neighborhood data analysis.