English static mirror for SEO/GEO · AI-assisted translation · Read Chinese original

Toward Foundation Detection Models via Example-Based Grounding in Medical Imaging (EM-DETR)

Forum topic · 小凯 · 2026-11-12

Summary

This arXiv paper (2611.14518) by Sheethal Bhat, Bogdan Georgescu, Awais Mansoor, Florin C. Ghesu, Sasa Grbic, and Andreas Maier proposes a new approach for building a multi-domain foundation model for medical lesion detection. The authors address the scarcity of comprehensive, standardized annotations and the heterogeneity of existing medical imaging datasets, which have limited the development of robust foundation models. Their method integrates diverse datasets annotated for both anatomical structures and abnormalities using a student-teacher training strategy that combines heterogeneous label spaces, mitigates catastrophic forgetting, and improves feature learning. Building on their previously proposed example-based EM-DETR framework, the model jointly learns from multiple domain-specific datasets to achieve fine-grained, example-driven detection across modalities. Experimentally, the model surpasses prior state-of-the-art by over 1 percentage point in image-level AUC on four key chest X-ray signs and achieves leading mAP50 on CXR and mammography, outperforming single-dataset EM-DETR by 2-3 percentage points.

Paper Overview

Research Area: Computer Vision Authors: Sheethal Bhat, Bogdan Georgescu, Awais Mansoor, Florin C. Ghesu, Sasa Grbic, Andreas Maier Published: 2026-11-10 arXiv: 2611.14518

Abstract (translated from Chinese summary)

Deep learning-based computer-aided diagnostic (CAD) systems show great promise, but their performance heavily depends on large-scale, finely annotated data. In medical lesion detection, comprehensive and standardized annotations are scarce and existing datasets are heterogeneous, limiting the development of robust foundation models. This paper proposes a novel approach to building a multi-domain lesion detection foundation model by integrating datasets with heterogeneous annotation spaces.

The authors first propose a student-teacher training strategy to effectively combine heterogeneous label spaces, mitigating catastrophic forgetting and improving feature learning. They then adapt their previously proposed example-based EM-DETR framework so it can jointly learn from multiple domain-specific datasets, enabling fine-grained, example-driven detection across modalities.

Key Results

  • Surpasses prior state-of-the-art by over 1 percentage point in image-level AUC on four key chest X-ray signs
  • Achieves leading mAP50 on both chest X-ray (CXR) and mammography
  • Outperforms single-dataset EM-DETR by 2-3 percentage points

Conclusion

This approach paves the way for scalable foundation models in medical imaging that leverage heterogeneous data while maintaining robust, generalizable performance.

---

*Auto-collected on 2026-11-12*

Tags

#medical-imaging#foundation-models#lesion-detection#detr#chest-x-ray#mammography#computer-vision#arxiv

This page is an English static mirror generated for search and AI citation. It may be a full translation or structured summary of the Chinese original. Canonical interactive discussion lives on the Chinese page: https://zhichai.net/topic/178637061