Abstract
Large language models (LLMs) have shown promise for medical question answering (Q\&A), but their reliability in specialised clinical domains remains limited by subspecialty-specific knowledge demands, structured clinical reasoning, and evidence-grounded answer selection. We introduce EndoRAG, a domain-specific agentic retrieval-augmented generation (RAG) framework for endocrinology multiple-choice question (MCQ) answering. Collected source documents were ingested into separate endocrinology subspecialty vector databases. EndoRAG uses a graph-based agentic workflow integrating query interpretation, iterative evidence retrieval, agent-mediated option-wise reasoning, and verification to support evidence-grounded clinical Q\&A. We compared EndoRAG with LLM-only prompting and Single-pass RAG on 386 text-based questions from seven endocrinology MCQ datasets. Across four LLM backbones, EndoRAG achieved the best performance, reaching 88.71% macro- and 88.08% micro-average accuracy. These findings suggest that agentic RAG workflows, for endocrinology, can improve specialised medical MCQ answering, with performance improvements dependent on the underlying model. Upon acceptance, the code will be made available.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/MedAgent_013.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to Open Review
Open Review Page: https://openreview.net/forum?id=R95HesUxQj
BibTex
@InProceedings{PanMar_EndoRAG_MICCAISAT2026,
author = { Panagiotou, Maria AND Abdur Rahman, Lubnaa AND Papathanail, Ioannis AND Mougiakakou, Stavroula},
title = { { EndoRAG: An Agentic Retrieval-Augmented Generation Framework for Endocrinology Question Answering } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17263},
month = {pending},
page = {pending}
}