CNNs for automatic glaucoma assessment using fundus images: an extensive validation

Table 4 Obtained results for HRF, Drishti-GS1, RIM, sjchoi86-HRF and the new public database ACRIMA, using Xception architecture represented in AUC, AUC’s confidence interval, Accuracy, sensitivity and specificity

Database	AUC	AUC’s 95% confidence interval	Accuracy	Sensitivity	Specificity	# images
HRF	0.8354	50.00–100.00%	0.8000	0.8333	0.7778	45
Drishti-GS1	0.8041	50.49–92.55%	0.7525	0.7419	0.7143	101
RIM-ONE	0.8575	77.53–91.12%	0.7121	0.7931	0.7990	455
sjchoi86-HRF	0.7739	66.44–86.85%	0.7082	0.7033	0.7030	401
ACRIMA	0.7678	68.41–81.81%	0.7021	0.6893	0.7020	705

ISSN: 1475-925X