Skip to main content
. 2020 Dec 7;8(12):e21790. doi: 10.2196/21790

Table 5.

Comparative performance analysis of the proposed ensemble-SDCNNa model with various state-of-the-art methods.

Comparative methods MCb SZc MC + SZ
F1 APd ARe ACCf AUCg F1 AP AR ACC AUC F1 AP AR ACC AUC
LBPh and SVMi,j [46] 0.537 0.58 0.5 0.58 0.675 0.76 0.76 0.76 0.76 0.83 0.729 0.729 0.729 0.729 0.763
HoGk and SVMi [47] 0.797 0.796 0.798 0.797 0.863 0.85 0.85 0.85 0.85 0.90 0.822 0.823 0.821 0.821 0.882
ShuffleNeti [43] 0.747 0.771 0.727 0.748 0.84 0.875 0.876 0.873 0.873 0.937 0.884 0.885 0.883 0.884 0.936
InceptionV3i [44] 0.739 0.773 0.711 0.74 0.828 0.882 0.883 0.881 0.881 0.942 0.887 0.89 0.884 0.885 0.944
MobileNetV2i [45] 0.762 0.769 0.755 0.769 0.833 0.876 0.878 0.875 0.875 0.941 0.886 0.888 0.883 0.884 0.946
Santosh et al [41] l 0.79 0.88 0.86 0.93
Hwang et al [17] 0.674 0.884 0.837 0.926
ResNet50i [29] 0.788 0.796 0.78 0.79 0.886 0.877 0.877 0.877 0.876 0.94 0.88 0.881 0.878 0.879 0.921
ResNet101i [29] 0.8 0.821 0.782 0.798 0.895 0.864 0.865 0.862 0.861 0.934 0.859 0.862 0.857 0.858 0.923
Alfadhli et al [14] 0.81 0.79 0.791 0.89
GoogLeNeti [20,21] 0.834 0.851 0.818 0.834 0.902 0.852 0.853 0.851 0.851 0.921 0.843 0.846 0.84 0.84 0.914
Lopes and Valiati [21] 0.826 0.926 0.847 0.904
Vajda et al [42] 0.783 0.87
Pasa et al [22] 0.79 0.811 0.844 0.9 0.862 0.925
Govindarajan and Swaminathan [15] 0.876 0.877 0.878 0.94
Proposed 0.929 0.937 0.921 0.928 0.965 0.908 0.909 0.908 0.908 0.948 0.9 0.902 0.898 0.899 0.95

aSDCNN: shallow–deep CNN.

bMC: Montgomery County.

cSZ: Shenzhen.

dAP: average precision.

eAR: average recall.

fACC: accuracy.

gAUC: area under the curve.

hLBP: local binary pattern.

iWe evaluated the performance of these models using our selected data sets and experimental protocol.

jSVM: support vector machine.

kHoG: histogram of oriented gradients.

l—: not available. These results were not reported in some existing studies.