Table 3.

Evaluation of model performance on test set based on augmentation and contrastive learning techniques compared to the state of the art.

Method	AUC (95% CI)
	100% data	10% data
SimCLR18	0.83 (0.80–0.85)	0.71 (0.66–0.73)
Lesion based CL (confidence threshold 0.7)25	0.86 (0.81–0.87)	0.75 (0.71–0.79)
Lesion based CL (confidence threshold 0.8)25	0.87 (0.84–0.89)	0.75 (0.74–0.78)
FundusNet (final proposed model)	0.91 (0.898–0.930)	0.81 (0.77–0.84)

Confidence threshold refers to the threshold set in²⁵ to reduce un confident prediction of patches from fundus images.

DR diabetic retinopathy, CI confidence interval, AUC area under the ROC curve, CL contrastive learning.