VGG16 transfer learning achieves 90.8% accuracy in automated retinal disease detection from fundus images, significantly outperforming baseline CNN models.

Researchers developed a deep learning pipeline using VGG16 pretrained backbone for binary retinal disease classification from fundus photographs

Transfer learning model achieved 90.8% test accuracy with 0.90 weighted F1-score, substantially beating baseline CNN's 83.1% accuracy

Class weighting techniques were applied to address class imbalance in the dataset

Automated fundus image analysis could expand early retinal disease detection access to underserved populations globally