The objective of this study is to establish deep learning (DL) algorithm to automatically classify multi-diseases from fundus photography and differentiate major vision-threatening conditions and other retinal abnormalities. The effectiveness and accuracy of the established algorithm will be evaluated in community derived dataset.

Retinal diseases seriously threaten vision and quality of life, but they often develop insidiously. To date, deep learning (DL) algorithms have shown high prospects in biomedical science, particularly in the diagnosis of ocular diseases, such as diabetic retinopathy, age-related macular degeneration, retinopathy of prematurity, glaucoma, and papilledema. However, there is still a lack of a single algorithm that can classify multi-diseases from fundus photography.

This cross-sectional study will establish a DL algorithm to automatically classify multi-diseases from fundus photography and differentiate major vision-threatening conditions and other retinal abnormalities. We will use the receiver operating characteristic (ROC) curve to examine the ability of recognition and classification of diseases. Taken the results of the expert panel as the gold standard, we will use the evaluation indexes, such as sensitivity, specificity, accuracy, positive predictive value, negative predictive value, etc, to compare the diagnostic capacity between the AI recognition system and human ophthalmologist.

Inclusion Criteria:

• fundus photography around 45° field which covers optic disc and macula
• complete patient identification information;

Exclusion Criteria:

• incomplete patient identification information

community derived dataset