pith. sign in

arxiv: 2606.24883 · v1 · pith:YMW2MDHJnew · submitted 2026-06-23 · 💻 cs.CV

BenchX: Benchmarking AI Models for Cancer Detection and Localization with Demographic and Protocol Biases

Qi Chen , Wenxuan Li , Pedro R. A. S. Bassi , Xinze Zhou , Jakob Wasserthal , Ibrahim Ethem Hamamci , Sezgin Er , Ashwin Kumar
show 9 more authors
Yiwen Ye Yuhan Wang Yuyin Zhou Akshay S. Chaudhari Curtis Langlotz Kang Wang Yang Yang Alan L. Yuille Zongwei Zhou
This is my paper

Pith reviewed 2026-06-26 00:06 UTC · model grok-4.3

classification 💻 cs.CV
keywords benchmarkcancer detectionCT scansdemographic biasmedical imagingtumor detectionsubgroup evaluationAI robustness
0
0 comments X

The pith

AI models for cancer detection in CT scans show large performance drops on underrepresented patient subgroups like young female African Americans.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper introduces a benchmark of 85,355 CT scans to test 12 tumor-detection AI models across tumor size, location, patient demographics, and imaging protocols. Large language models extract subgroup labels from clinical reports to enable this scalable analysis. Results indicate that models tuned for average accuracy perform poorly on rare groups where data collection is impractical. This setup reveals the limits of average-case optimization for real-world medical imaging use.

Core claim

The paper claims that current state-of-the-art AI models for tumor detection and localization, when optimized for average accuracy, exhibit poor performance in rare or underrepresented subgroups such as young, female African Americans, as measured across a large benchmark of CT scans with LLM-derived demographic and protocol labels.

What carries the argument

The BenchX benchmark dataset and evaluation protocol that uses LLM-extracted labels to measure model performance across demographic subgroups, tumor characteristics, and imaging protocols.

If this is right

  • AI models for medical imaging must be evaluated at the subgroup level rather than by average accuracy alone.
  • Collecting enough labeled data for every rare subgroup is often not feasible, so other methods are needed to improve consistency.
  • Reliable clinical use of these models requires testing across variations in patient age, sex, race, and scan protocols.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same LLM-labeling approach could be applied to other imaging modalities or diseases to check for similar subgroup gaps.
  • Targeted data synthesis or model adaptation techniques might reduce the observed drops without needing massive new annotations.
  • Hospitals could use subgroup monitoring to decide when to override or retrain an AI model for specific patient populations.

Load-bearing premise

The large language model extraction of demographic and protocol subgroup labels from clinical reports is accurate and unbiased enough to support the reported performance differences.

What would settle it

Manual re-annotation of a random sample of the clinical reports showing that the LLM labels contain systematic errors large enough to erase or reverse the reported accuracy gaps between subgroups.

Figures

Figures reproduced from arXiv: 2606.24883 by Akshay S. Chaudhari, Alan L. Yuille, Ashwin Kumar, Curtis Langlotz, Ibrahim Ethem Hamamci, Jakob Wasserthal, Kang Wang, Pedro R. A. S. Bassi, Qi Chen, Sezgin Er, Wenxuan Li, Xinze Zhou, Yang Yang, Yiwen Ye, Yuhan Wang, Yuyin Zhou, Zongwei Zhou.

Figure 1
Figure 1. Figure 1: Existing tumor-detection models, including top leaderboard performers [4,7], show inconsistent performance across pa￾tient demographics (age, race, sex) and imaging protocols (contrast phases). This benchmark (BenchX) systematically mea￾sures and exposes these performance gaps. The main bottlenecks are the lack of training data and evaluation bench￾marks that explicitly account for de￾mographic and acquisi… view at source ↗
Figure 2
Figure 2. Figure 2: The BenchX Curation Pipeline. We unify 85,355 heterogeneous scans into a standardized benchmark using a scalable, semi-automated pipeline. Tumor Anno￾tation: A consistency-check loop where only discordant predictions between segmen￾tation and detection models trigger radiologist review, ensuring expert-level ground truth with minimal manual effort. LLM Metadata Extraction: An automated parser converts unst… view at source ↗
Figure 3
Figure 3. Figure 3: Summary of BenchX metadata. These metadata reveal the breadth and heterogeneity of BenchX, capturing extensive demographic diversity, protocol varia￾tion, and dataset-specific population biases. This comprehensive characterization un￾derscores the importance of subgroup-aware evaluation and motivates the need for robust, fair, and generalizable AI models in medical imaging. uate all submitted models. Its s… view at source ↗
Figure 4
Figure 4. Figure 4: Benchmark performance biases in BenchX. Performance varies substan￾tially across cohorts, revealing strong dataset and protocol shifts between institu￾tions. Accuracy is highest on E-Coast, PanTS, and N-Europe but drops on Merlin, N-California, and S-Europe. Arterial and non-contrast scans consistently reduce per￾formance relative to venous CT, and younger and minority populations show lower detection accu… view at source ↗
Figure 5
Figure 5. Figure 5: LLM Prompt for extracting structured attributes [PITH_FULL_IMAGE:figures/full_fig_p027_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Example CT slices across ct phase, sex and race subgroups in the [PITH_FULL_IMAGE:figures/full_fig_p028_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Example CT slices across age subgroups in the E-Coast dataset. [PITH_FULL_IMAGE:figures/full_fig_p029_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Example CT slices across ct phase and race subgroups in the Merlin [PITH_FULL_IMAGE:figures/full_fig_p030_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Example CT slices across age and sex subgroups in the Merlin [PITH_FULL_IMAGE:figures/full_fig_p031_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Example CT slices across ct phase subgroups in the PanTS dataset. [PITH_FULL_IMAGE:figures/full_fig_p032_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Example CT slices across age and sex subgroups in the PanTS [PITH_FULL_IMAGE:figures/full_fig_p033_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Example CT slices across ct phase and race subgroups in the N [PITH_FULL_IMAGE:figures/full_fig_p034_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Example CT slices across age and sex subgroups in the N-California [PITH_FULL_IMAGE:figures/full_fig_p035_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Example CT slices across ct phase and race subgroups in the S. Eu [PITH_FULL_IMAGE:figures/full_fig_p036_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Example CT slices across age and sex subgroups in the S. Europe [PITH_FULL_IMAGE:figures/full_fig_p037_15.png] view at source ↗
Figure 16
Figure 16. Figure 16: Example CT slices across age and sex subgroups in the N [PITH_FULL_IMAGE:figures/full_fig_p038_16.png] view at source ↗
Figure 17
Figure 17. Figure 17: Example CT slices across contrast-phase subgroups in the N-Europe [PITH_FULL_IMAGE:figures/full_fig_p039_17.png] view at source ↗
read the original abstract

Artificial intelligence (AI) has achieved remarkable success in medical imaging, but it is widely recognized that these models often perform inconsistently across real-world clinical settings. Such inconsistencies occur when patient demographics and imaging protocols vary, for example, in detecting small tumors, analyzing scans from different contrast phases, or evaluating patients of different ages or sexes. To quantify these inconsistencies, we develop a large-scale, open benchmark of 85,355 CT scans that systematically evaluates 12 tumor-detection AI models across tumor size, location, patient subgroup, and imaging protocol. We leverage large language models (LLMs) to extract and organize subgroup information from clinical data, which makes the analysis both scalable and reproducible. Our benchmark reveals that current state-of-the-art AI models, optimized for average accuracy, perform poorly in rare or underrepresented subgroups, such as young, female African Americans. However, collecting sufficient annotated data for these rare cases is often impractical. The benchmark provides a foundation for building more reliable and robust AI models for tumor detection and highlighting the need for rigorous, subgroup-level evaluation in medical imaging and computer vision. Datasets, code

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 1 minor

Summary. The paper introduces BenchX, a benchmark of 85,355 CT scans that evaluates 12 tumor-detection AI models across tumor size, location, patient demographic subgroups, and imaging protocols. Subgroup labels are obtained by applying LLMs to clinical reports; the central empirical finding is that average-optimized SOTA models exhibit poor performance on rare or underrepresented subgroups such as young female African Americans, motivating the need for subgroup-level evaluation in medical imaging.

Significance. If the extracted labels prove reliable, the work supplies a large-scale, open benchmark that directly quantifies demographic and protocol biases in tumor-detection models, a practically important gap in the field. The scale (85k scans) and the attempt to make subgroup analysis scalable via LLMs are genuine strengths that could support follow-on research on robust model development.

major comments (1)
  1. [Methods (LLM label extraction)] Methods section on LLM-based label extraction: no human validation, inter-annotator agreement, or error analysis is reported for the demographic and protocol labels extracted from the 85,355 clinical reports. Because the headline performance gaps (e.g., on young female African-American patients) rest entirely on the correctness of these per-scan labels, the absence of any accuracy quantification on a held-out sample makes the reported disparities impossible to interpret; systematic mislabeling correlated with image features or demographics would artifactually create or inflate the observed differences.
minor comments (1)
  1. [Abstract] Abstract ends with the fragment 'Datasets, code' without supplying repository URLs or DOIs.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for highlighting the critical need for validation of the LLM-extracted labels. This is a substantive methodological point that directly affects interpretability of the subgroup results. We address it below and will incorporate the requested analysis in revision.

read point-by-point responses

  1. Referee: Methods section on LLM-based label extraction: no human validation, inter-annotator agreement, or error analysis is reported for the demographic and protocol labels extracted from the 85,355 clinical reports. Because the headline performance gaps (e.g., on young female African-American patients) rest entirely on the correctness of these per-scan labels, the absence of any accuracy quantification on a held-out sample makes the reported disparities impossible to interpret; systematic mislabeling correlated with image features or demographics would artifactually create or inflate the observed differences.

    Authors: We agree that the absence of validation makes the reported disparities difficult to interpret with full confidence and that systematic label errors could artifactually influence results. In the revised manuscript we will add a dedicated validation subsection: a random sample of 500 reports will be independently annotated by two human experts (blinded to model outputs and to each other) for all demographic and protocol fields. We will report per-field accuracy, precision, recall, F1, and inter-annotator agreement (Cohen’s kappa). We will also conduct an error analysis stratified by demographic group and protocol to test for systematic mislabeling. If validation accuracy falls below a pre-specified threshold we will discuss the implications for the main findings and, where feasible, provide sensitivity analyses. This addition directly addresses the concern while preserving the scalability argument for LLM extraction. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical benchmark with no derivations or self-referential steps

full rationale

The paper is a large-scale empirical benchmark evaluating 12 existing tumor-detection models on 85k CT scans, using LLM extraction for subgroup labels. No equations, fitted parameters, or derivations are present that could reduce to inputs by construction. Claims rest on external models, data, and the (unvalidated) LLM labeling step, but this is a data-quality assumption rather than circularity per the enumerated patterns. No self-citation load-bearing or ansatz smuggling occurs in the provided text.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The benchmark relies on standard assumptions about CT image quality and tumor annotation validity; no new free parameters or invented entities are introduced. The LLM labeling step is treated as a domain tool rather than a fitted component.

axioms (1)
  • domain assumption LLM extraction of demographic and protocol labels from clinical text is accurate enough for subgroup analysis
    Invoked to scale the benchmark; no validation numbers given in abstract.

pith-pipeline@v0.9.1-grok · 5802 in / 1114 out tokens · 17424 ms · 2026-06-26T00:06:54.130397+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

49 extracted references · 12 canonical work pages · 2 internal anchors

  1. [1]

    zenodo (2024) 8

    Alves, N., Schuurmans, M., Rutkowski, D., et al.: The panorama study protocol: Pancreatic cancer diagnosis-radiologists meet ai. zenodo (2024) 8

    2024

  2. [2]

    The Lancet Oncology27(1), 116–124 (2026) 4

    Alves, N., Schuurmans, M., Rutkowski, D., Saha, A., Vendittelli, P., Obuchowski, N., Liedenbaum, M.H., Haldorsen, I.S., Molven, A., Yakar, D., et al.: Artificial intelligence and radiologists in pancreatic cancer detection using standard of care ct scans (panorama): an international, paired, non-inferiority, confirmatory, obser- vational study. The Lancet...

    2026

  3. [3]

    Nature communications13(1), 1–13 (2022) 3, 4

    Antonelli, M., Reinke, A., Bakas, S., Farahani, K., Kopp-Schneider, A., Landman, B.A., Litjens, G., Menze, B., Ronneberger, O., Summers, R.M., et al.: The medical segmentation decathlon. Nature communications13(1), 1–13 (2022) 3, 4

    2022

  4. [4]

    arXiv preprint arXiv:2106.05735 (2021) 2, 4

    Antonelli, M., Reinke, A., Bakas, S., Farahani, K., Landman, B.A., Litjens, G., Menze, B., Ronneberger, O., Summers, R.M., van Ginneken, B., et al.: The medical segmentation decathlon. arXiv preprint arXiv:2106.05735 (2021) 2, 4

    work page arXiv 2021

  5. [5]

    Qwen Technical Report

    Bai, J., Bai, S., Chu, Y., Cui, Z., Dang, K., Deng, X., Fan, Y., Ge, W., Han, Y., Huang, F., et al.: Qwen technical report. arXiv preprint arXiv:2309.16609 (2023) 2

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  6. [6]

    In: International Conference on Medical Image Computing and Computer-Assisted Intervention

    Bassi, P.R., Li, W., Chen, J., Zhu, Z., Lin, T., Decherchi, S., Cavalli, A., Wang, K., Yang, Y., Yuille, A.L., Zhou, Z.: Learning segmentation from radiology reports. In: International Conference on Medical Image Computing and Computer-Assisted Intervention. pp. 305–315. Springer (2025),https://github.com/MrGiovanni/R- Super5

    2025

  7. [7]

    Bassi, P.R., Li, W., Tang, Y., Isensee, F., Wang, Z., Chen, J., Chou, Y.C., Kirch- hoff, Y., Rokuss, M., Huang, Z., Ye, J., He, J., Wald, T., Ulrich, C., Baumgartner, M., Roy, S., Maier-Hein, K.H., Jaeger, P., Ye, Y., Xie, Y., Zhang, J., Chen, Z., Xia, Y., Xing, Z., Zhu, L., Sadegheih, Y., Bozorgpour, A., Kumari, P., Azad, R., Merhof, D., Shi, P., Ma, T.,...

    2024

  8. [8]

    arXiv preprint arXiv:2510.14803 (2025),https://github

    Bassi, P.R., Zhou, X., Li, W., Płotka, S., Chen, J., Chen, Q., Zhu, Z., Prządo, J., Hamamci, I.E., Er, S., Chen, X., Yavuz, M.C., Chou, Y.C., Lin, T., Wang, K., Tang, Y., Cwikla, J.B., Decherchi, S., Cavalli, A., Yang, Y., Yuille, A.L., Zhou, Z.: Scaling artificial intelligence for multi-tumor early detection with more re- ports, fewer masks. arXiv prepri...

    work page arXiv 2025

  9. [9]

    arXiv preprint arXiv:1901.04056 (2019) 4

    Bilic, P., Christ, P.F., Vorontsov, E., Chlebus, G., Chen, H., Dou, Q., Fu, C.W., Han, X., Heng, P.A., Hesser, J., et al.: The liver tumor segmentation benchmark (lits). arXiv preprint arXiv:1901.04056 (2019) 4

    work page arXiv 1901

  10. [10]

    Research Square pp

    Blankemeier, L., Cohen, J.P., Kumar, A., Van Veen, D., Gardezi, S.J.S., Paschali, M., Chen, Z., Delbrouck, J.B., Reis, E., Truyts, C., et al.: Merlin: A vision language foundation model for 3d computed tomography. Research Square pp. rs–3 (2024) 3, 7

    2024

  11. [11]

    In: Conference on fairness, accountability and transparency

    Buolamwini, J., Gebru, T.: Gender shades: Intersectional accuracy disparities in commercial gender classification. In: Conference on fairness, accountability and transparency. pp. 77–91. PMLR (2018) 2 BenchX17

    2018

  12. [12]

    Nature medicine29(12), 3033–3043 (2023) 2, 8

    Cao,K.,Xia,Y.,Yao,J.,Han,X.,Lambert,L.,Zhang,T.,Tang,W.,Jin,G.,Jiang, H., Fang, X., et al.: Large-scale pancreatic cancer detection via non-contrast ct and deep learning. Nature medicine29(12), 3033–3043 (2023) 2, 8

    2023

  13. [13]

    MONAI: An open-source framework for deep learning in healthcare

    Cardoso, M.J., Li, W., Brown, R., Ma, N., Kerfoot, E., Wang, Y., Murrey, B., Myronenko, A., Zhao, C., Yang, D., et al.: Monai: An open-source framework for deep learning in healthcare. arXiv preprint arXiv:2211.02701 (2022) 19

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  14. [14]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision

    Chen, J., Xia, Y., Yao, J., Yan, K., Zhang, J., Lu, L., Wang, F., Zhou, B., Qiu, M., Yu, Q., et al.: Cancerunit: Towards a single unified model for effective detection, segmentation, and diagnosis of eight major cancers using a large collection of ct scans. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. pp. 21327–21338 (2023) 2

    2023

  15. [15]

    In: IEEE/CVF conference on computer vision and pattern recognition (CVPR)

    Chen, Q., Chen, X., Song, H., Xiong, Z., Yuille, A., Wei, C., Zhou, Z.: To- wards generalizable tumor synthesis. In: IEEE/CVF conference on computer vision and pattern recognition (CVPR). pp. 11147–11158 (2024),https://github.com/ MrGiovanni/DiffTumor14

    2024

  16. [16]

    In: Proceedings of the IEEE International Conference on Computer Vision (ICCV)

    Chen, Q., Zhou, X., Liu, C., Chen, H., Li, W., Jiang, Z., Huang, Z., Zhao, Y., Yu, D., He, J., Zheng, Y., Shao, L., Yuille, A., Zhou, Z.: Scaling tumor segmen- tation: Best lessons from real and synthetic data. In: Proceedings of the IEEE International Conference on Computer Vision (ICCV). pp. 24001–24013 (2025), https://github.com/BodyMaps/AbdomenAtlas2....

    2025

  17. [17]

    Cancer120, 1091–1096 (2014) 2

    Chen Jr, M.S., Lara, P.N., Dang, J.H., Paterniti, D.A., Kelly, K.: Twenty years post-nih revitalization act: enhancing minority participation in clinical trials (em- pact): laying the groundwork for improving minority clinical trial accrual: renewing the case for enhancing minority participation in cancer clinical trials. Cancer120, 1091–1096 (2014) 2

    2014

  18. [18]

    Radiology: Artificial Intelligence5(6), e230060 (2023) 3

    Glocker, B., Jones, C., Roschewitz, M., Winzeck, S.: Risk of bias in chest radio- graphy deep learning foundation models. Radiology: Artificial Intelligence5(6), e230060 (2023) 3

    2023

  19. [19]

    In: Proceedings of the Com- puter Vision and Pattern Recognition Conference

    Gutbrod, M., Rauber, D., Nunes, D.W., Palm, C.: Openmibood: Open medical imaging benchmarks for out-of-distribution detection. In: Proceedings of the Com- puter Vision and Pattern Recognition Conference. pp. 25874–25886 (2025) 4

    2025

  20. [20]

    In: Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision

    Hatamizadeh, A., Tang, Y., Nath, V., Yang, D., Myronenko, A., Landman, B., Roth, H.R., Xu, D.: Unetr: Transformers for 3d medical image segmentation. In: Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision. pp. 574–584 (2022) 12, 16, 26, 27, 28, 29, 30, 31

    2022

  21. [21]

    arXiv preprint arXiv:1904.00445 (2019) 4

    Heller, N., Sathianathen, N., Kalapara, A., Walczak, E., Moore, K., Kaluzniak, H., Rosenberg, J., Blake, P., Rengel, Z., Oestreich, M., et al.: The kits19 challenge data: 300 kidney tumor cases with clinical context, ct semantic segmentations, and surgical outcomes. arXiv preprint arXiv:1904.00445 (2019) 4

    work page arXiv 1904

  22. [22]

    Holdcroft, A.: Gender bias in research: how does it affect evidence based medicine? (2007) 2

    2007

  23. [23]

    Nature Medicine pp

    Hu, C., Xia, Y., Zheng, Z., Cao, M., Zheng, G., Chen, S., Sun, J., Chen, W., Zheng, Q., Pan, S., et al.: Ai-based large-scale screening of gastric cancer from noncontrast ct imaging. Nature Medicine pp. 1–9 (2025) 2

    2025

  24. [24]

    Advances in Neural Information Processing Systems37, 24496–24522 (2024) 2

    Hu, K., Li, J., Zhang, Y., Ye, X., Gao, X.: One-to-multiple: A progressive style transfer unsupervised domain-adaptive framework for kidney tumor segmentation. Advances in Neural Information Processing Systems37, 24496–24522 (2024) 2

    2024

  25. [25]

    arXiv preprint arXiv:2304.06716 (2023) 12, 17, 19, 26, 27, 28, 29, 30, 31 18 Q

    Huang, Z., Wang, H., Deng, Z., Ye, J., Su, Y., Sun, H., He, J., Gu, Y., Gu, L., Zhang, S., et al.: Stu-net: Scalable and transferable medical image segmen- tation models empowered by large-scale supervised pre-training. arXiv preprint arXiv:2304.06716 (2023) 12, 17, 19, 26, 27, 28, 29, 30, 31 18 Q. Chen et al

    work page arXiv 2023

  26. [26]

    Nature Methods18(2), 203–211 (2021) 12, 17, 19, 26, 27, 28, 29, 30, 31

    Isensee, F., Jaeger, P.F., Kohl, S.A., Petersen, J., Maier-Hein, K.H.: nnu-net: a self-configuring method for deep learning-based biomedical image segmentation. Nature Methods18(2), 203–211 (2021) 12, 17, 19, 26, 27, 28, 29, 30, 31

    2021

  27. [27]

    arXiv preprint arXiv:2404.09556 (2024) 12, 17, 19, 26, 27, 28, 29, 30, 31

    Isensee, F., Wald, T., Ulrich, C., Baumgartner, M., Roy, S., Maier-Hein, K., Jaeger, P.F.: nnu-net revisited: A call for rigorous validation in 3d medical image segmen- tation. arXiv preprint arXiv:2404.09556 (2024) 12, 17, 19, 26, 27, 28, 29, 30, 31

    work page arXiv 2024

  28. [28]

    In: 2022 44th Annual International Conference of the IEEE Engineering in Medicine & Biology Society (EMBC)

    Jain, R.K., Sato, T., Watasue, T., Nakagawa, T., Iwamoto, Y., Han, X., Lin, L., Hu, H., Ruan, X., Chen, Y.W.: Unsupervised domain adaptation using adversarial learning and maximum square loss for liver tumors detection in multi-phase ct images. In: 2022 44th Annual International Conference of the IEEE Engineering in Medicine & Biology Society (EMBC). pp. ...

    2022

  29. [29]

    Nature Medicine30(4), 1166–1173 (2024) 3

    Ktena, I., Wiles, O., Albuquerque, I., Rebuffi, S.A., Tanno, R., Roy, A.G., Azizi, S., Belgrave, D., Kohli, P., Cemgil, T., et al.: Generative models improve fairness of medical classifiers under distribution shifts. Nature Medicine30(4), 1166–1173 (2024) 3

    2024

  30. [30]

    In: Conference on Neural Information Processing Systems (NeurIPS) Datasets and Benchmarks Track (2025),https: //github.com/MrGiovanni/PanTS5, 7

    Li, W., Zhou, X., Chen, Q., Lin, T., Bassi, P.R., Chen, X., Ye, C., Zhu, Z., Ding, K., Li, H., Wang, K., Yang, Y., Tang, Y., Xu, D., Yuille, A.L., Zhou, Z.: Pants: The pancreatic tumor segmentation dataset. In: Conference on Neural Information Processing Systems (NeurIPS) Datasets and Benchmarks Track (2025),https: //github.com/MrGiovanni/PanTS5, 7

    2025

  31. [31]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision

    Liu, J., Zhang, Y., Chen, J.N., Xiao, J., Lu, Y., Landman, B.A., Yuan, Y., Yuille, A., Tang, Y., Zhou, Z.: Clip-driven universal model for organ segmentation and tumor detection. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. pp. 21152–21164 (2023),https://github.com/ljwztc/CLIP- Driven-Universal-Model12, 26, 27, 28, 29, 30, 31

    2023

  32. [32]

    Medical image analysis98, 103295 (2024) 3, 12, 18, 26, 27, 28, 29, 30, 31

    Liu, T., Bai, Q., Torigian, D.A., Tong, Y., Udupa, J.K.: Vsmtrans: A hybrid paradigm integrating self-attention and convolution for 3d medical image segmen- tation. Medical image analysis98, 103295 (2024) 3, 12, 18, 26, 27, 28, 29, 30, 31

    2024

  33. [33]

    Nature communications 15(1), 7465 (2024) 3

    Lotter, W.: Acquisition parameters influence ai recognition of race in chest x-rays and mitigating these factors reduces underdiagnosis bias. Nature communications 15(1), 7465 (2024) 3

    2024

  34. [35]

    Ma, J., Zhang, Y., Gu, S., Ge, C., Wang, E., Zhou, Q., Huang, Z., Lyu, P., He, J., Wang, B.: Automatic organ and pan-cancer segmentation in abdomen ct: the flare 2023 challenge (2024),https://arxiv.org/abs/2408.125344

    work page arXiv 2023

  35. [36]

    Mastroianni, A.C., Faden, R., Federman, D., et al.: Women and health research: Ethical and legal issues of including women in clinical studies, vol. 1. National Academies Press (1994) 2

    1994

  36. [37]

    IEEE Transactions on pattern analysis and machine intelligence22(10), 1090–1104 (2000) 2

    Phillips, P.J., Moon, H., Rizvi, S.A., Rauss, P.J.: The feret evaluation method- ology for face-recognition algorithms. IEEE Transactions on pattern analysis and machine intelligence22(10), 1090–1104 (2000) 2

    2000

  37. [38]

    In: International Conference on Medical Image Computing and Computer-Assisted Intervention

    Ronneberger, O., Fischer, P., Brox, T.: U-net: Convolutional networks for biomedi- cal image segmentation. In: International Conference on Medical Image Computing and Computer-Assisted Intervention. pp. 234–241. Springer (2015) 12, 16, 19, 26, 27, 28, 29, 30, 31

    2015

  38. [39]

    In: International Conference on Medical Image Computing BenchX19 and Computer-Assisted Intervention

    Roy, S., Koehler, G., Ulrich, C., Baumgartner, M., Petersen, J., Isensee, F., Jaeger, P.F., Maier-Hein, K.H.: Mednext: transformer-driven scaling of convnets for medi- cal image segmentation. In: International Conference on Medical Image Computing BenchX19 and Computer-Assisted Intervention. pp. 405–415. Springer (2023) 11, 12, 17, 19, 26, 27, 28, 29, 30, 31

    2023

  39. [40]

    Journal of the Optical Society of America A4(3), 519–524 (1987) 2

    Sirovich, L., Kirby, M.: Low-dimensional procedure for the characterization of hu- man faces. Journal of the Optical Society of America A4(3), 519–524 (1987) 2

    1987

  40. [41]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Tang, Y., Yang, D., Li, W., Roth, H.R., Landman, B., Xu, D., Nath, V., Hatamizadeh, A.: Self-supervised pre-training of swin transformers for 3d medi- cal image analysis. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 20730–20740 (2022) 12, 16, 26, 27, 28, 29, 30, 31

    2022

  41. [42]

    arXiv preprint arXiv:2311.02189 (2023) 3

    Tian, Y., Shi, M., Luo, Y., Kouhana, A., Elze, T., Wang, M.: Fairseg: A large-scale medical image segmentation dataset for fairness learning using segment anything model with fair error-bound scaling. arXiv preprint arXiv:2311.02189 (2023) 3

    work page arXiv 2023

  42. [43]

    In: CVPR

    Turk, M.A., Pentland, A., et al.: Face recognition using eigenfaces. In: CVPR. vol. 91, pp. 586–591 (1991) 2

    1991

  43. [44]

    EBioMedicine102(2024) 3

    Wang, R., Kuo, P.C., Chen, L.C., Seastedt, K.P., Gichoya, J.W., Celi, L.A.: Drop the shortcuts: image augmentation improves fairness and decreases ai detection of race and other demographics from medical images. EBioMedicine102(2024) 3

    2024

  44. [45]

    arXiv preprint arXiv:2406.01264 (2024) 14

    Wu, L., Zhuang, J., Ni, X., Chen, H.: Freetumor: Advance tumor segmentation via large-scale tumor synthesis. arXiv preprint arXiv:2406.01264 (2024) 14

    work page arXiv 2024

  45. [46]

    medRxiv (2022) 2, 8

    Xia, Y., Yu, Q., Chu, L., Kawamoto, S., Park, S., Liu, F., Chen, J., Zhu, Z., Li, B., Zhou, Z., Yuille, A.L., Fishman, E.K., Hruban, R.H.: The felix project: Deep networks to detect pancreatic neoplasms. medRxiv (2022) 2, 8

    2022

  46. [47]

    Journal of medical imaging5(3), 036501 (2018) 3

    Yan, K., Wang, X., Lu, L., Summers, R.M.: Deeplesion: automated mining of large- scale lesion annotations and universal lesion detection with deep learning. Journal of medical imaging5(3), 036501 (2018) 3

    2018

  47. [48]

    Yang, Y., Zhang, H., Gichoya, J.W., Katabi, D., Ghassemi, M., et al.: The limits of fairmedicalimagingaiinreal-worldgeneralization.NatureMedicine30,2838–2848 (2024).https://doi.org/10.1038/s41591-024-03113-43

    work page doi:10.1038/s41591-024-03113-43 2024

  48. [49]

    In: International Conference on Medical Image Computing and Computer-Assisted Intervention

    Ye, Y., Xie, Y., Zhang, J., Chen, Z., Xia, Y.: Uniseg: A prompt-driven universal segmentation model as well as a strong representation learner. In: International Conference on Medical Image Computing and Computer-Assisted Intervention. pp. 508–518. Springer (2023) 19

    2023

  49. [50]

    small-model

    Yin, Y., Tang, Z., Huang, Z., Wang, M., Weng, H.: Source free domain adaptation for kidney and tumor image segmentation with wavelet style mining. Scientific Reports14(1), 24849 (2024) 2 20 Q. Chen et al. Appendix Appendix A Dataset Curation Details........................................ 1 A.1 Tumor Annotations Details ................................. 1...

    2024