Vision-language Models for Structured Report Generation in Radiology: Towards Consistent and Reliable Chest X-ray Reporting
Miah MB, Bueß L, Maier A (2026)
Publication Type: Conference contribution
Publication year: 2026
Journal
Publisher: Springer Science and Business Media Deutschland GmbH
Pages Range: 84-89
Conference Proceedings Title: Informatik aktuell
Event location: Lübeck
ISBN: 9783658510992
DOI: 10.1007/978-3-658-51100-5_15
Abstract
Medical report generation (MRG) aims to automatically generate reports from medical images, reducing the workload on radiologists. Research in this field is progressing rapidly with large pretrained vision-language models (VLMs), but most are trained on general image-text data and fail to capture critical medical findings. Effective chest X-ray (CXR) report generation requires fine-tuning on high-quality datasets, but inconsistent reporting styles remain a key challenge. The structured radiology report generation (SRRG) approach addresses this by using large language models (LLMs) to standardize and generate consistent structured reports. In this study, we introduce SRRG-benchmark to systematically evaluate state-of-the-art LLMs for converting free-text CXR reports into structured formsuitable for training VLMs. We primarily focus on assessing the medical image interpretation capabilities of VLMs across both structured and conventional free-text report generation tasks. Our results demonstrate that structured reporting improves VLMs’ medical image interpretation performance compared to free-text fine-tuning, increasing MedGemma’s clinical accuracy (GREEN) from 0.50 to 0.53 and RadGraph F1 from 0.27 to 0.38, with similar gains for Qwen3-VL.
Authors with CRIS profile
Lukas Bueß
Lehrstuhl für Informatik 5 (Mustererkennung)
Andreas Maier
Lehrstuhl für Informatik 5 (Mustererkennung)
How to cite
APA:
Miah, M.B., Bueß, L., & Maier, A. (2026). Vision-language Models for Structured Report Generation in Radiology: Towards Consistent and Reliable Chest X-ray Reporting. In Heinz Handels, Katharina Breininger, Thomas Deserno, Andreas Maier, Klaus Maier-Hein, Christoph Palm, Thomas Tolxdorff (Eds.), Informatik aktuell (pp. 84-89). Lübeck, DE: Springer Science and Business Media Deutschland GmbH.
MLA:
Miah, Md Badhon, Lukas Bueß, and Andreas Maier. "Vision-language Models for Structured Report Generation in Radiology: Towards Consistent and Reliable Chest X-ray Reporting." Proceedings of the Bildverarbeitung für die Medizin Workshop, BVM 2026, Lübeck Ed. Heinz Handels, Katharina Breininger, Thomas Deserno, Andreas Maier, Klaus Maier-Hein, Christoph Palm, Thomas Tolxdorff, Springer Science and Business Media Deutschland GmbH, 2026. 84-89.
BibTeX: Download