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

Bueß L, Kurin R, Panambur AB, Arias-Vergara T, Maier A (2026)


Publication Type: Conference contribution

Publication year: 2026

Publisher: ML Research Press

Book Volume: 315

Pages Range: 2998-3017

Conference Proceedings Title: Proceedings of Machine Learning Research

Event location: Chientan, TWN

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 un-derreport 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. Our code is publicly available1 .

Authors with CRIS profile

How to cite

APA:

Bueß, L., Kurin, R., Panambur, A.B., Arias-Vergara, T., & Maier, A. (2026). ALO: Addressing Class Imbalance in Radiology Report Generation through Anatomy-Level Oversampling. In Yuankai Huo, Mingchen Gao, Chang-Fu Kuo, Yueming Jin, Ruining Deng (Eds.), Proceedings of Machine Learning Research (pp. 2998-3017). Chientan, TWN: ML Research Press.

MLA:

Bueß, Lukas, et al. "ALO: Addressing Class Imbalance in Radiology Report Generation through Anatomy-Level Oversampling." Proceedings of the 9th International Conference on Medical Imaging with Deep Learning, MIDL 2026, Chientan, TWN Ed. Yuankai Huo, Mingchen Gao, Chang-Fu Kuo, Yueming Jin, Ruining Deng, ML Research Press, 2026. 2998-3017.

BibTeX: Download