Grounding Radiology Reports in 3D Chest CT: Results of the ReXGrounding Challenge
Segmenting each reported finding in a radiology scan would allow findings to be measured and tracked over time and let AI systems generate reports that are verifiable, but so far AI models have performed poorly at this task. To advance this task, we organized the ReXGrounding Challenge at MICCAI 2026, the first challenge dedicated to it. Given a non-contrast chest CT scan and the set of findings from its report, participants trained models to produce the corresponding set of 3D segmentations. The challenge used 2,992 training and 200 validation scans from the ReXGroundingCT dataset, and evaluated methods on a held-out test set of 300 scans with 582 findings, exhaustively annotated by board-certified radiologists. In total, 48 teams from 15 countries made 760 submissions, and the best method reached a Dice of 0.375 and a hit rate of 0.77, matching agreement between two independent radiologist groups (Dice 0.32, hit rate 0.76, on a 50-scan subset). On the 100 scans of the pre-challenge benchmark, the winner reached a Dice of 0.390, against 0.285 for VoxTell, the best model before the challenge, which five of the six best teams fine-tuned. Non-focal findings remained far harder to segment than focal ones, instance F1 stayed below inter-reader agreement (0.34 vs. 0.47), and models sometimes segmented the right abnormality in the wrong location. Top methods thus match radiologist agreement in Dice, but non-focal findings, findings with multiple instances, and adherence to the anatomical location described in the text remain open problems.
Authors
- Pranav Rajpurkar (ORCID: https://orcid.org/0000-0002-8030-3727)
- Maximilian Rokuss (ORCID: https://orcid.org/0009-0004-4560-0760)
- Fatima Jabali
- Anna Maria Hunkemöller (ORCID: https://orcid.org/0000-0002-5021-5697)
- Brecht Van Berkel (ORCID: https://orcid.org/0000-0002-3782-6223)
- Eric Bijnens
- Klaus Hermann Maier-Hein (ORCID: https://orcid.org/0000-0002-6626-2463)
- Reem Alwasiah (ORCID: https://orcid.org/0000-0002-3149-6673)
- Asma Khalid Alamoudi (ORCID: https://orcid.org/0000-0001-8101-8218)
- Xi Ouyang (ORCID: https://orcid.org/0000-0002-7841-5359)
- Hassan Alomaish (ORCID: https://orcid.org/0009-0006-9915-0966)
- Sezgin Er (ORCID: https://orcid.org/0000-0001-7266-9844)
- Adam Perer (ORCID: https://orcid.org/0000-0002-8369-3847)
- Ibrahim Alturki
- Kevin B. W. Groot Lipman (ORCID: https://orcid.org/0000-0003-3651-2529)
- Ayed Alaklubi
- İbrahim Ethem Hamamcı (ORCID: https://orcid.org/0000-0003-2932-3105)
- Katelyn Morrison (ORCID: https://orcid.org/0000-0002-2644-4422)
- Rolf Symons (ORCID: https://orcid.org/0000-0001-7889-6773)
- Kwang-Hyun Uhm (ORCID: https://orcid.org/0000-0002-0819-2575)
- Maarten Steyvers
- Junha Im
- Yannick Kirchhoff (ORCID: https://orcid.org/0000-0001-8124-8435)
- Tim Veenboer
- Mohammed Baharoon (ORCID: https://orcid.org/0009-0000-1467-1364)
- 晓丹 叶
- Weiqiu Jin
- Youngjun Jun (ORCID: https://orcid.org/0009-0001-8926-2327)
- Benjamin Hamm (ORCID: https://orcid.org/0009-0003-4818-8700)
- Moritz Langenberg
- Ali Alburkani
- Bryan Eelen
- Lucas Bijnens (ORCID: https://orcid.org/0009-0000-5765-1742)
- Michael Moritz (ORCID: https://orcid.org/0009-0001-6940-0518)
- Inhwa Son
- Luyang Luo
- Amirah Fahad Alshammeri
- Cees Snoek
- Shahad Alhomaiani
- Talal Yahyan
- Mashael Alshehri
- Dijia Wu
- Sung Eun Kim
- Saud Alhussain
- Tomas Gilvydis
- Maha Buraykan Alshammari
- Peter Youssef
- Mingyang Yu
- Ali Piracha
- Mahmoud Hussain Alabbad
- Jiayu Wang
- Daesung Kim
- Hannah Hancock
- Monerah Almohideb
- Zirong Chen
- Afnan Aljumayan
- Rachel van 't Hull
- Sangwon Lee
- Xiaoman Zhang
- Mijin Yun
- Rayan Alsager
- Mona A. Alhammad
Publication Details
- Journal
- arXiv (Cornell University)
- Published
- 2026-11-18
- DOI
- https://doi.org/10.48550/arxiv.2611.23862
- Primary Topic
- Medical Image Segmentation Techniques
- Type
- preprint