Multi-step retrieval and reasoning improves radiology question answering with large language models

Sebastian Wind, Jeta Sopa, Daniel Truhn, Mahshad Lotfinia, Tri-Thien Nguyen, Keno Bressem, Lisa Adams, Mirabela Rusu, Harald Köstler, Gerhard Wellein, Andreas Maier, Soroosh Tayebi Arasteh · 2025

Read Multi-step retrieval and reasoning improves radiology question answering with large language models by Sebastian Wind, Jeta Sopa, Daniel Truhn, Mahshad Lotfinia, Tri-Thien Nguyen, Keno Bressem, Lisa Adams, Mirabela Rusu, Harald Köstler, Gerhard Wellein, Andreas Maier, Soroosh Tayebi Arasteh (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/pmc-12749912.txt.

Description

Sebastian Wind and co-authors present, in twenty twenty-five, a method that helps large language models answer radiology questions more reliably. Radiology is the reading of medical scans. Standard retrieval systems fetch facts once, then answer. That single step struggles with complex reasoning. Their framework, called RaR, works in steps: it restates the question, fetches evidence, and builds an answer, then repeats. They tested twenty-five models of many sizes on expert radiology questions and on board exam questions.

The stepped method raised diagnostic accuracy over plain prompting and over single-step retrieval. Gains were largest for mid-sized and small models. Very large models barely changed. The method also cut made-up claims and gave useful evidence in close to half the cases.

Catalog record

CREATORSebastian Wind, Jeta Sopa, Daniel Truhn, Mahshad Lotfinia, Tri-Thien Nguyen, Keno Bressem, Lisa Adams, Mirabela Rusu, Harald Köstler, Gerhard Wellein, Andreas Maier, Soroosh Tayebi Arasteh
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE110 KB
VERSION1.0
LICENSECC BY 4.0
RIGHTSCC BY
SOURCEEurope PMC

Source: https://europepmc.org/articles/PMC12749912

Browse the full library · About the Tower of Babel