Tag: ophthalmology

A comparative analysis of leading Vision-Language Models (VLMs) – OpenAI's GPT-4o, GPT-4V, and Google's Gemini – reveals their potential and limitations in detecting and diagnosing inherited retinal diseases (IRDs) from fundus photographs. While GPT-4o and GPT-4V demonstrate strong feature extraction capabilities and high detection accuracy, Gemini struggles with misidentifying normal images. All models require further refinement for improved diagnostic accuracy and gene inference.

This article introduces MIRAGE, a groundbreaking multimodal foundation model designed for comprehensive retinal OCT image analysis. It addresses the limitations of existing models by integrating multiple imaging modalities and establishing a new benchmark for evaluating AI in ophthalmology.