The Use of Artificial Intelligence in Diagnosing Retinal Diseases

Abstract

Background: Early and accurate detection of retinal diseases is critical to
prevent vision loss and ensure effective treatment. Traditional diagnostic
methods, including fundus photography and optical coherence tomography
(OCT), rely on the expertise of ophthalmologists, which may not always be
readily available. Artificial intelligence (AI) offers a promising approach to
enhance diagnostic accuracy and accessibility. This study evaluates the
diagnostic performance of AI-based systems in identifying retinal diseases
compared to conventional diagnostic methods.
Methods: A retrospective study was conducted involving 200 participants
(aged 18–80 years) with suspected retinal diseases. Diagnostic metrics,
including true positives (TP), true negatives (TN), false positives (FP), and
false negatives (FN), were analyzed to calculate sensitivity, specificity, positive
predictive value (PPV), negative predictive value (NPV), and overall accuracy.
AI performance was compared against diagnoses made by board-certified
ophthalmologists using fundus photography and OCT as the gold standards.
Results: AI systems achieved 94.8% sensitivity, 90.3% specificity, 92.1% PPV,
93.4% NPV, and 92.7% accuracy in detecting retinal diseases. Notably, AI
exhibited superior performance in identifying diabetic retinopathy (DR) and
age-related macular degeneration (AMD) in their early stages. Comparatively,
ophthalmologists demonstrated 92.1% sensitivity, 88.7% specificity, 90.2%
PPV, 91.0% NPV, and 90.8% accuracy.
Conclusion: AI-based diagnostic systems show comparable or superior
accuracy to conventional methods in diagnosing retinal diseases, particularly in
early detection scenarios. AI tools can complement ophthalmologists,
enhancing efficiency and accessibility, especially in resource-limited settings.
Future research should focus on refining AI algorithms and integrating them
into clinical workflows.