EUVIMEDEUVIMED entdecken
Open Access

Improving Radiology Report Error Detection Using a Multipass Large Language Model: Framework Development and Validation

Songsoo Kim iD, Seungtae Lee iD, See Young Lee iD, Joonho Kim iD, Keechan Kan iD, Hyunji Lee iD, Dukyong Yoon iD

JMIR Medical Informatics · 2026

DOI: 10.2196/87368 ↗

Kurz gefasst

Fachartikel in der Zeitschrift JMIR Medical Informatics, 2026. Verfasst von Songsoo Kim und 6 weiteren Autor:innen. Beteiligte Einrichtung: Department of Radiology, Seoul National University Hospital, Seoul, Republic of Korea. Hauptthemen laut MEDLINE: Large Language Models, Radiology, Diagnostic Errors, Radiology Information Systems. Die Arbeit wurde bisher 1-mal zitiert (Crossref) und nennt 33 Literaturangaben. Der Volltext ist frei in PubMed Central verfügbar.

AbstractSchlagwörterBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Background

Large language model (LLM) proofreaders for radiology reports generate many false positives (FPs) due to the low prevalence of errors.

Objective

This study aimed to determine whether an optimized LLM framework could improve both precision and cost-efficiency without compromising error detection capability.

Methods

In this retrospective study, 1000 radiology reports (radiography, ultrasonography, computed tomography, and magnetic resonance imaging; 250 each) were sampled from the Medical Information Mart for Intensive Care III database. Two public chest radiography corpora (CheXpert and Open-i) served as external test sets. Three LLM frameworks were evaluated: single-prompt detector (framework 1); report extractor plus single-prompt detector (framework 2); and extractor, detector, and FP verifier (framework 3). Precision for each framework was assessed using positive predictive value (PPV) and detected errors per 1000 reports. Overall efficiency was estimated using model inference costs and reviewer labor costs.

Results

PPV increased from 0.063 (95% CI 0.036‐0.101) in framework 1 to 0.079 (95% CI 0.049‐0.118) in framework 2 and 0.159 (95% CI 0.090‐0.252) in framework 3 ( P

Schlagwörter

MeSH-Schlagwörter (MEDLINE)

Large Language ModelsRadiologyDiagnostic ErrorsRadiology Information SystemsHumansRetrospective StudiesReproducibility of Results

Stichwörter der Autor:innen

error detectionhuman-in-the-looplarge language modelsquality assuranceradiology report

Autor:innen und Institutionen

  • Department of Radiology, Seoul National University Hospital, Seoul, Republic of Korea.
  • Department of Radiology, Gangnam Severance Hospital, Seoul, Republic of Korea.
  • Department of Internal Medicine, Gangnam Severance Hospital, Seoul, Republic of Korea.
  • Department of Neurology, Severance Hospital, Seoul, Republic of Korea.
  • Department of Surgery, Samsung Medical Center, Seoul, Republic of Korea.
  • Department of Obstetrics and Gynecology, Kangbuk Samsung Hospital, Seoul, Republic of Korea.

ORCID-Profile: Songsoo Kim, Seungtae Lee, See Young Lee, Joonho Kim, Keechan Kan, Hyunji Lee, Dukyong Yoon

Zitieren

APA

Kim, S., Lee, S., Lee, S. Y., Kim, J., Kan, K., Lee, H., & Yoon, D. (2026). Improving Radiology Report Error Detection Using a Multipass Large Language Model: Framework Development and Validation. JMIR Medical Informatics. https://doi.org/10.2196/87368

Vancouver

Kim S, Lee S, Lee SY, Kim J, Kan K, Lee H, et al. Improving Radiology Report Error Detection Using a Multipass Large Language Model: Framework Development and Validation. JMIR Medical Informatics. 2026. doi:10.2196/87368

BibTeX
@article{Kim2026,
  author = {Kim, Songsoo and Lee, Seungtae and Lee, See Young and Kim, Joonho and Kan, Keechan and Lee, Hyunji and Yoon, Dukyong},
  title = {Improving Radiology Report Error Detection Using a Multipass Large Language Model: Framework Development and Validation},
  journal = {JMIR Medical Informatics},
  year = {2026},
  doi = {10.2196/87368}
}
Kostenlos starten →