Table 3. Classification performances achieved with CROP 10 images on both hold-out training and hold-out test sets.
HOLD-OUT TRAINING | HOLD-OUT TEST | ||||||||
---|---|---|---|---|---|---|---|---|---|
CNN features | AUC | Acc | Sens | Spe | AUC | Acc | Sens | Spe | |
AlexNET+clinical | 4 | 0.73 | 0.61 | 0.63 | 0.60 | 0.79 | 0.82 | 0.80 | 0.83 |
ResNet152V2+clinical | 11 | 0.80 | 0.78 | 0.66 | 0.84 | 0.54 | 0.50 | 1.0 | 0.39 |
InceptionV3+clinical | 11 | 0.75 | 0.70 | 0.83 | 0.64 | 0.72 | 0.89 | 0.60 | 0.96 |