Table 6.

Performance of different network architectures applied to the NTUH_ROCF^a data set.

Metric	AlexNet	Sketch-a-Net	Our system
			Without data augmentation and dropout techniques	With data augmentation and dropout techniques
Sensitivity, mean (SD)	0.698 (0.039)	0.671 (0.047)	0.756 (0.033)	0.847 (0.017)
Specificity, mean (SD)	0.790 (0.046)	0.820 (0.054)	0.864 (0.017)	0.905 (0.009)
Accuracy, mean (SD)	0.744 (0.034)	0.746 (0.019)	0.810 (0.020)	0.876 (0.010)
AUROC,b mean (SD)	0.814 (0.021)	0.819 (0.009)	0.851 (0.020)	0.913 (0.004)
Total parameters (×106), n	46.73	8.38	0.40	0.56
Time required to complete 1-fold training, minutes	10	29	9	2

^aNTUH_ROCF: National Taiwan University Hospital_Rey-Osterrieth Complex Figure.

^bAUROC: area under the receiver operating characteristic curve.