ALO: Addressing Class Imbalance in Radiology Report Generation through Anatomy-Level Oversampling

Lukas Buess, Robert Kurin, Adarsh Bhandary Panambur, Tomas Arias-Vergara, Andreas Maier

Keywords: Class imbalance, Structured report generation, Vision-language models

Abstract: Radiology report generation aims to connect visual understanding with clinical language, yet most methods rely on free-text supervision, which is highly variable and difficult to evaluate. Clinical datasets are also dominated by normal findings, causing models to underreport abnormalities. While recent works focus on architectural advances, we show that structured supervision and balanced sampling can yield substantial gains in clinical performance. We convert free-text reports into structured anatomy-level representations and introduce Anatomy-Level Oversampling (ALO), a data centered sampling strategy that balances normal and abnormal findings for each anatomical region. This structure provides consistent supervision and enables more informative evaluation. Across three public datasets, ALO improves sensitivity to pathological findings while remaining fully model agnostic. On internal validation, ALO increases F1-Score by 50\% and CRG by 5.8\%, and on external validation, it increases F1-Score by 45.1\% and CRG by 5\%. These results highlight the importance of structured data and balanced sampling for reliable report generation.

Primary Subject Area: Generative Models

Secondary Subject Area: Application: Radiology

Registration Requirement: Yes

Reproducibility: https://github.com/Kurin-FAU/ALO

Visa & Travel: No

Read CFP & Author Instructions: Yes

Originality Policy: Yes

Single-blind & Not Under Review Elsewhere: Yes

LLM Policy: Yes

Midl Latex Submission Checklist: Ensure no LaTeX errors during compilation., Replace NNN with your OpenReview submission ID., Includes \documentclass{midl}, \jmlryear{2026}, \jmlrworkshop, \jmlrvolume, \editors, and correct \bibliography command., Did not override options of the hyperref package., Did not use the times package., Use the correct spelling and format, avoid Unicode characters, and use LaTeX equivalents instead., Any math in the title and abstract must be enclosed within $...$., Did not override the bibliography style defined in midl.cls and did not use \begin{thebibliography} directly to insert references., Avoid using \scalebox; use \resizebox when needed., Included all necessary figures and removed *unused* files in the zip archive., Removed special formatting, visual annotations, and highlights used during rebuttal., All special characters in the paper and .bib file use LaTeX commands (e.g., \'e for é)., No separate supplementary PDF uploads., Acknowledgements, references, and appendix must start after the main content.

Submission Number: 274