Table 1 Comparison between TIA-Net and models in three benchmark sets
From: Automatic glaucoma detection based on transfer induced attention network
Database | Methods | Acc (%) | Se (%) | Sp (%) | AUC |
|---|---|---|---|---|---|
Ours | HOS-LR | 69.9 | 91.1 | 55.6 | 0.719 |
Wavelet-LR | 68.9 | 69.5 | 58.9 | 0.715 | |
Gabor-LR | 70.5 | 86.7 | 62.2 | 0.776 | |
HWG | 72.1 | 93.2 | 61.9 | 0.802 | |
CNN | 80.2 | 91.4 | 77.0 | 0.869 | |
VGG | 80.7 | 87.7 | 79.1 | 0.871 | |
GoogLeNet | 79.8 | 80.7 | 73.8 | 0.870 | |
ResNet | 81.2 | 83.6 | 73.9 | 0.872 | |
Chen [11] | 80.9 | 89.1 | 77.8 | 0.875 | |
Shibata [29] | 81.7 | 87.5 | 80.2 | 0.879 | |
NMD+CNN | 84.1 | 84.7 | 83.4 | 0.911 | |
SOD+CNN | 83.7 | 84.2 | 80.6 | 0.903 | |
NMD+Attention | 84.5 | 84.4 | 84.9 | 0.911 | |
TIA-Net (SOD+Attention) | 85.7 | 84.9 | 86.9 | 0.929 | |
ORIGA | HOS-LR | 63.5 | 90.3 | 32.2 | 0.632 |
Wavelet-LR | 65.9 | 59.1 | 66.8 | 0.648 | |
Gabor-LR | 67.2 | 49.0 | 77.2 | 0.682 | |
HWG | 68.8 | 71.7 | 55.0 | 0.693 | |
CNN | 70.4 | 70.7 | 74.8 | 0.791 | |
VGG | 70.1 | 69.8 | 71.0 | 0.800 | |
GoogLeNet | 71.8 | 69.8 | 73.5 | 0.805 | |
ResNet | 71.5 | 71.3 | 71.7 | 0.803 | |
Chen [11] | 70.8 | 69.2 | 71.0 | 0.794 | |
Shibata [29] | 73.3 | 73.2 | 76.7 | 0.809 | |
NMD+CNN | 74.5 | 68.7 | 80.7 | 0.815 | |
SOD+CNN | 73.9 | 80.9 | 72.2 | 0.813 | |
NMD+Attention | 74.9 | 71.2 | 77.7 | 0.817 | |
TIA-Net (SOD+Attention) | 76.6 | 75.3 | 77.2 | 0.835 |