Table 6.

Comparison of the proposed ensemble with state-of-the-art methods.

Method	SN(%)	SP(%)	GM(%)	BA(%)
LH10	0.6888	0.9376	0.7976	0.8132
VGG	0.6241	0.9319	0.7565	0.7780
VGG-P	0.6553	0.9703	0.7727	0.8128
GoogleNet-P	0.6930	0.9264	0.7943	0.8097
GoogleNet	0.6256	0.9341	0.7608	0.7798
VGG-EF	0.7762	0.9633	0.8647	0.8697
VGG-LT	0.8148	0.9678	0.8880	0.8913
Kim et al.33	0.3759	0.9027	0.5825	0.6393
Anthimopoulos et al.31	0.7687	0.9657	0.8567	0.8672
ECNN	0.9041	0.9818	0.9420	0.9430
BCNN2D	0.8615	0.9730	0.9152	0.9172
MSTAGE-CNN2D	0.8552	0.9740	0.9125	0.9146
MCONTEXT-CNN2D	0.8621	0.9766	0.9171	0.9194
BCNN2.5D	0.8426	0.9748	0.9051	0.9087
MSTAGE-CNN2.5D	0.8776	0.9754	0.9246	0.9265
BCNN3D	0.8271	0.9650	0.8917	0.8960
MSTAGE-CNN3D	0.8335	0.9727	0.8993	0.9031

The first group of methods (first seven rows) presents the results of the state-of-the-art and the proposed ensemble methods. The second group presents the results of the individual models.

SN: sensitivity; SP: specificity; GM: geometric mean; BA: balanced accuracy.