Table 2.

Classification performance of FundusNet framework for referable vs non-referable DR compared to fully supervised baseline models.

Models	AUC (95% CI) cross-validated on EyePACS dataset	AUC (95% CI) on test dataset	P-value (vs FundusNet)	Sensitivity (95% CI)	P-value (vs FundusNet)	Specificity (95% CI)	P-value (vs FundusNet)
FundusNet	0.96 (0.938–0.972)	0.91 (0.898–0.930)	Ref	0.90 (0.895–0.917)	Ref	0.85 (0.830–0.862)	Ref
Baseline1 (ResNet50)	0.94 (0.919–0.953)	0.80 (0.783–0.820)	P < 0.001	0.81 (0.793–0.834)	P < 0.001	0.74 (0.731–0.758)	P < 0.005
Baseline2 (InceptionV3)	0.92 (0.905–0.961)	0.83 (0.801–0.853)	P < 0.001	0.84 (0.822–0.848)	P < 0.001	0.79(0.786–0.819)	P < 0.05

P value from measuring statistical significance using DeLong’s test for comparing pairwise AUCs.

DR diabetic retinopathy, CI confidence interval, AUC area under the ROC curve, Ref reference.