Post

New research · Ophthalmology
NPJ digital medicine · 23h
AI / informaticsNPJ digital medicine · 2026

EyeRAG: graph retrieval-augmented generation for safe and accurate clinical dialogue in ophthalmology.

Kaikai Zhao, Daohuan Kang, Tao Yu … Kai Jin
Read paper
OphthalmologyAI / informatics

EyeRAG system reduces factual inaccuracies in ophthalmic clinical dialogue.

EyeRAG: graph retrieval-augmented generation for safe and accurate clinical dialogue in ophthalmology.

Kaikai Zhao et al. · NPJ digital medicine · 2026
Background

Large language models promise to transform ophthalmic clinical communication but face challenges from factual inaccuracies (hallucinations) and limited domain knowledge.

Purpose

This study introduces EyeRAG, a guideline-grounded GraphRAG system for ophthalmic dialogue that integrates OphthaKG, a domain-specific knowledge graph, constructed exclusively from clinical guidelines.

3.3%
Results
the percentage of factual inaccuracies in dialogue was reduced to this low level
n = 120 clinical scenarios
More results

Validated by LLM-as-a-judge and board-certified ophthalmologists on internal/external datasets, EyeRAG outperformed Vanilla LLMs and standard RAG.

LLM evaluations ranked EyeRAG highest (mean 1.61 ± 1.04 internal/1.72 ± 1.18 external).

“
Conclusion

It serves as a tool for patient interpretation rather than clinical decision-making, assisting in patient counselling and tele-ophthalmology by translating complex findings into accessible dialog under professional oversight.

Read paper
0 comments

No comments yet. Be the first.

Related papers

LatestFoundational
AI / Informatics
difference, 0
AI and human experts had the same median quality scores for eye care questions
Case-control
0 of 28
distinct structural changes were absent in all 28 healthy control eyes
Cohort Study
RR, 0.91
Black patients were less likely to start medication for their eye condition
AI / Informatics
100% diagnostic sensitivity
ROFI correctly found every case of eye disease
AI / Informatics
10.50-point increase
students trained with digital patients had higher history-taking assessment scores
AI / Informatics
0.877
The o1 large language model correctly answered 87.7% of ophthalmology questions.
Cohort Study
HR 4.85
chronic pain conditions mean a nearly five times higher risk of developing dry eye disease
Cohort Study
12 per 10,000
a higher rate of eye infections after dexamethasone implant injections