MM-Retinal-Reason Dataset and OphthaReason Model for Ophthalmic AI Reasoning

Researchers have introduced MM-Retinal-Reason, the first ophthalmic multimodal dataset designed to cover the full spectrum of perception and reasoning, from basic visual feature matching to complex clinical reasoning that integrates heterogeneous clinical information with multimodal imaging data. Building on this dataset, they developed the OphthaReason model to emulate realistic clinical thinking patterns. The work aims to bridge the gap in current medical multimodal large language models (MLLMs), which have focused primarily on shallow inference. The dataset includes both basic and complex reasoning tasks to enhance visual-centric fundamental reasoning capabilities. The research was published on arXiv under identifier 2508.16129.