. Author manuscript; available in PMC: 2016 Dec 1.

Published in final edited form as: Comput Med Imaging Graph. 2015 Aug 20;46(Pt 2):197–208. doi: 10.1016/j.compmedimag.2015.08.002

Table 3.

Comparison of epithelial tissue classification performances. Training sets ST-EP_1.3 were established using 60.152 windows from 19 images. Trained SVM classifiers were applied to 191 testing images to separate epitehelium from stroma. Algorithms were evaluated by Jaccard, area overlap (O) and Rand (Ri) indices to determine the concordance of the algorithm-based prediction with pathologist manual annotations. J and O indices were calculated separately for concordance of ST and EP areas, whereas Ri was calculated for ST and EP together. Respective columns contain mean value ± standard deviation.

	J _ST	J _EP	O _ST	O _EP	Ri
Chen et al. [31]	49.9 ± 19.1	48.9 ± 13.0	75.0 ± 13.5	63.2 ± 17.7	57.6 ± 7.4
SVM(ST-EP₁)	53.2 ± 18.8	55.9 ± 16.8	72.0 ± 14.2	72.8 ± 19.6	62.1 ± 8.2
SVM(ST-EP₂)	49.6 ± 19.2	58.9 ± 14.8	61.9 ± 15.9	81.4 ± 12.2	61.5 ± 7.7
SVM(ST-EP₃)	50.8 ± 18.2	59.5 ± 14.6	63.7 ± 14.1	81.5 ± 12.1	62.0 ± 7.5