Who Gets Missed in the Tail? Thresholded Subgroup Underdiagnosis in Long-Tailed Chest X-ray Classification

Dang P. M. Cao; Ha-Hieu Pham; Hai-Dang Nguyen; Huy-Hieu Pham; Min Xu; Thanh-Huy Nguyen; Trung-Nghia Le; Ulas Bagci

REVIEW 3 major objections 4 minor 14 references

Who Gets Missed in the Tail? Thresholded Subgroup Underdiagnosis in Long-Tailed Chest X-ray Classification

T0 review · 3 major / 4 minor · reviewed 2026-07-11 · grok-4.5

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read Rare chest X-ray findings can rank well yet still miss subgroup patients once scores become yes/no decisions; the threshold, not just the model, decides who is left out.

desk verdict Clean audit paper: rare CXR fairness is a finding imes subgroup imes threshold problem, and the M5–M6 isolation is the real contribution—not a new loss. read the letter →

arxiv 2607.07717 v1 pith:HQ7K25NS submitted 2026-07-04 cs.LG cs.CV

Ha-Hieu Pham , Hai-Dang Nguyen , Dang P. M. Cao , Thanh-Huy Nguyen , Min Xu , Trung-Nghia Le , Ulas Bagci , Huy-Hieu Pham This is my paper · ORCID

classification cs.LGcs.CV

keywords fairnesslong-tailedlearningmulti-labelclassificationchestX-raysubgroupbiasunderdiagnosisthresholdedFNRgrouprobustness

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

Chest X-ray models are often judged by ranking scores, but a real triage system only acts when a score crosses a threshold. This paper shows that rare findings can still leave true-positive patients below that line, and that the misses concentrate in particular disease–subgroup cells rather than following label frequency alone. Using a diagnostic ladder on VinDr-CXR and MIMIC-CXR/CXR-LT, the authors separate class-level long-tail losses, subgroup-aware weighting, group-robust training, and the final threshold rule. On VinDr, group-tail weighting plus a tail-aware operating point cuts tail false-negative rate from 0.665 to 0.269 and sharply lowers worst-group rates for sex and age while macro-mAP rises slightly; the same score-to-threshold step also lowers worst-group rates across sex, age, race, and insurance on MIMIC, though residual miss rates stay high. The claim is deliberately narrow: rare-label fairness in this setting is jointly determined by finding, subgroup, and operating threshold, so fairness reports must include support, subgroup-conditioned misses, workload trade-offs, and the policy that turns scores into action.

What carries the argument

A diagnostic ladder of model probes (BCE, asymmetric loss, class-balanced weighting, GroupDRO, group-tail weighting, then tail-aware thresholds on the same scores) that isolates class imbalance, subgroup support, group robustness, and the score-to-decision step, with the audit unit defined as a positive rare-label case within a subgroup after a class-specific threshold.

What would settle it

On the same VinDr and MIMIC test splits, re-run the M5–M6 comparison with alternative clinically plausible threshold policies (for example equalized opportunity per subgroup, fixed alert-rate budgets, or higher precision floors) and check whether the large reductions in tail and worst-group false-negative rates disappear or reverse while ranking metrics stay fixed.

Watch

Extended reading notes

Core claim

Rare-label fairness in long-tailed multi-label chest X-ray classification depends jointly on the finding, the subgroup, and the operating threshold, not on label frequency or ranking metrics alone. Class-level long-tail methods and aggregate group robustness do not fully determine who remains below threshold; changing the tail threshold on an unchanged score model materially changes subgroup false-negative rates.

Load-bearing premise

The paper treats its chosen validation F1 thresholds and the specific tail-aware rule (maximize worst-group recall under a 0.05 precision floor) as representative enough operating policies that the observed score-to-threshold gaps generalize beyond these choices and the very small VinDr tail cells.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

The useful takeaway is narrow and real: after you turn long-tailed CXR scores into decisions, who gets missed is not fixed by macro-mAP or by aggregate group robustness. The paper’s best move is the diagnostic ladder that holds scores fixed between M5 and M6, so the FNR drop is pure threshold policy. That isolation is cleaner than most fairness-in-imaging papers manage.

What is new is the audit unit—rare class imes subgroup imes thresholded decision—plus the controlled comparison on VinDr and MIMIC/CXR-LT. Group-tail weighting plus tail-aware thresholds cut VinDr tail FNR 0.665 o0.269 and worst-group sex/age FNR sharply, with paired bootstrap support; MIMIC shows the same direction across sex, age, race, and insurance, with high residual miss rates left on the table. They also show GroupDRO alone does not clear rare subgroup misses, and they report missed positives per 100 and alert burden so the workload trade-off is visible. Citations are appropriate; the math is ordinary empirical comparison, not circular.

Soft spots are real but the paper mostly owns them. VinDr tail cells are tiny (3–19 positives; many subgroup cells ≤5), so the headline magnitudes rest on low-resolution estimates and a deliberately lenient τ=0.05 worst-group-recall rule. MIMIC reductions are smaller and residual FNR stays high, which actually supports their measurement claim more than a “solved” claim. Single backbone, incomplete VinDr metadata, no released code. The stress-test concern about overstated isolation is fair as a caution on magnitude, not a refutation of the joint claim—the paper already frames itself as an audit tool, not a deployable fix.

This is for people who audit clinical multi-label models and care about operating points. Worth a serious referee. I would engage, cite the audit framing, and push for code plus multi-architecture checks.

Referee Report

3 major / 4 minor

Summary. The paper studies pre-deployment fairness in long-tailed multi-label CXR classification as an audit of who is missed after scores become decisions. Using a diagnostic ladder (BCE, ASL, class-balanced ASL, GroupDRO, group-tail ASL, and the same scores with tail-aware thresholds) on VinDr-CXR and MIMIC-CXR/CXR-LT, it separates class-level long-tail losses, subgroup-aware weighting, group robustness, and threshold selection. The central claim is that rare-label fairness depends jointly on finding, subgroup, and operating threshold, not ranking metrics or label frequency alone. On VinDr, group-tail weighting plus tail-aware thresholding reduces tail FNR from 0.665 to 0.269 and worst-group FNRs for sex and age substantially, with modest macro-mAP gain; on MIMIC the same score-to-threshold step lowers tail and worst-group FNR across sex, age, race, and insurance, though residual FNRs remain high. Paired bootstrap contrasts and GroupDRO references support that aggregate group robustness alone does not remove rare subgroup misses after thresholding.

Significance. If the result holds, the paper makes a useful methodological contribution to medical imaging fairness: it reframes underdiagnosis as a thresholded class-subgroup cell rather than a pure ranking or class-frequency problem, and shows that the score-to-action step can dominate observed tail harm. Strengths include the clean M5–M6 isolation (identical scores, ECE, and Brier; only the threshold rule changes), dual-cohort evaluation with richer MIMIC axes, paired bootstrap contrasts, explicit positive-support reporting, and workload/alert-burden trade-offs. The narrow audit framing and limitations discussion are appropriately cautious. The work is more measurement than remedy, but that is a legitimate and needed contribution for CXR triage fairness reporting.

major comments (3)

[Sec. 2 Diagnostic ladder; Tables 2, 4; Fig. 2] Sec. 2 (Diagnostic ladder) and Tables 2/4: the load-bearing M5–M6 FNR reductions rest on a deliberately lenient tail-aware rule (maximize worst-group recall subject to precision floor τ=0.05). With VinDr tail support of only 3–19 positives per class and many subgroup-tail cells ≤5 (Table 1; Limitations), both F1 baselines and the worst-group objective are low-resolution. The paper already shows τ=0.10 still helps but with different alert rates (Sec. 3, Fig. 2); a fuller sensitivity analysis over τ and non-worst-group objectives (e.g., equalized opportunity or fixed-FPR) is needed to show that the isolation of a general score-to-action fairness failure is not an artifact of this particular sparse-cell objective.
[Sec. 3.1–3.2; Limitations; Table 1] Sec. 3.1–3.2 and Limitations: VinDr supplies the large absolute FNR drops that make the joint finding–subgroup–threshold claim vivid, while MIMIC shows only smaller absolute reductions with high residual FNR (0.866→0.741). Incomplete VinDr metadata (informative sex on 49% of test images; PatientAge missing in 69%) further weakens worst-group estimates. The manuscript should either strengthen VinDr evidence (more seeds already help; disease-level support checks are good) or rebalance the claim so that MIMIC’s directional but modest effect, not VinDr point estimates, carries the primary quantitative weight.
[Sec. 2; Tables 2–3] Sec. 2 and Tables 2–3: GroupDRO is correctly labeled a reference rather than a ladder rung, but the claim that “aggregate group robustness alone does not remove rare subgroup misses” would be stronger if GroupDRO groups were defined at the same class–subgroup granularity used for M5/M6 audit cells, or if this mismatch were quantified. As written, it is unclear whether the residual misses reflect a true distinction between group robustness and thresholded rare-positive recall, or a coarser group definition.

minor comments (4)

[Fig. 1–2; Sec. 3.1] Fig. 1 and Fig. 2 are informative but dense; adding explicit n+ per tail class (or a small support table) next to the disease-level discussion in Sec. 3.1 would help readers judge resolution without hunting through the text.
[Table 1; Sec. 2 Data] Clarify the exact training-frequency cutoff used to define “tail” classes within each cohort (Table 1 states counts but not the rule).
[Sec. 2; Limitations] A short note on why ConvNeXt-Tiny alone is sufficient for the audit claim (vs. at least one alternative backbone) would address architecture-generality concerns already flagged in Limitations.
[Abstract; Introduction] Minor prose issues: occasional missing spaces after periods in the abstract/intro (“thresholdselection”, “Existinglong-tailmethods”) and inconsistent hyphenation of “worst-group” / “worst group”.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: empirical diagnostic ladder with controlled M5–M6 score reuse; FNR is measured on held-out positives, not forced by definition or self-citation.

full rationale

The paper is a deployment-oriented empirical audit, not a first-principles derivation. Its central claim—that rare-label fairness in CXR depends jointly on finding, subgroup, and operating threshold rather than ranking metrics alone—is supported by held-out test comparisons across a diagnostic ladder (M1–M6) on VinDr-CXR and MIMIC-CXR/CXR-LT. M6 reuses M5 scores and changes only the tail-class threshold objective (worst-group recall subject to precision floor τ=0.05), so the M5–M6 FNR gap isolates the score-to-decision step by experimental design; the FNR itself is still computed from true positives that fall below the chosen threshold on the test split, not redefined as the threshold rule. Tail classes are defined from training frequency, thresholds are tuned on validation, and metrics (macro-mAP, tail FNR, worst-group FNR, missed positives per 100) are reported with bootstrap CIs on the test set. Citations are to standard external methods (ASL, effective-number reweighting, GroupDRO, equal opportunity) and public datasets; none is a load-bearing uniqueness theorem or ansatz from the present authors that forces the result. Sparse VinDr cells and the policy choice of τ affect statistical reliability and generalizability, but those are correctness/robustness concerns, not circular reductions of claim to input. No equation or definition makes the headline FNR reductions true by construction.

Assumptions & free parameters 4 free parameters · 5 assumptions · 2 invented entities

The central audit claim rests on standard multi-label training assumptions, public cohort label/subgroup definitions, and a small set of hand-chosen operating-policy parameters (especially the precision floor τ). No new physical entities are postulated; the invented constructs are measurement units and a ladder of probes. Free parameters are policy knobs, not fitted constants that secretly define the reported FNR.

free parameters (4)

precision floor τ for tail-aware thresholds = 0.05 (default)
Hand-chosen operating constraint (default τ=0.05, ablated at 0.03/0.10/0.20) that defines the M6 threshold objective; the headline FNR reductions depend on this policy choice.
F1-optimal per-class validation thresholds (M1–M5)
Thresholds selected on validation to maximize F1; they define the baseline operating points against which tail-aware thresholds are compared.
clipped class-subgroup weight multiplier in M5 (GT-ASL)
Training-count-based reweighting with clipping; the exact clip and functional form are design choices that affect M5 scores before thresholding.
tail-class definition (training-frequency cutoff within each cohort) = 5/15 VinDr; 13/40 MIMIC
Which labels count as 'tail' is defined from training-set frequency per cohort (5 of 15 on VinDr; 13 of 40 on MIMIC); all tail FNR aggregates depend on this partition.

assumptions (5)

domain assumption Public CXR label spaces and DICOM/EHR subgroup metadata are sufficiently accurate for within-cohort fairness auditing of false negatives.
Sec. 2 Cohort-specific observability and Limitations: race/insurance unavailable on VinDr; EHR race/insurance treated as imperfect proxies on MIMIC.
domain assumption A single ConvNeXt-Tiny backbone with dataset-specific linear head is adequate to probe loss and threshold effects without architecture confounds.
Sec. 2 Diagnostic ladder; Limitations explicitly note single-backbone scope.
domain assumption False-negative rate after class-specific thresholding is the primary safety-relevant underdiagnosis metric for triage/screening.
Introduction and Sec. 4 Clinical relevance: FN removes true positives from downstream attention; missed positives per 100 used as interpretable unit.
standard math Bootstrap resampling of test rows yields interpretable uncertainty for sparse tail cells.
Sec. 2 Endpoints: 200 independent and 1000 paired resamples; authors still flag low resolution for n+=3–19.
ad hoc to paper M5 and M6 share identical ranked scores so any FNR difference is threshold-mediated.
Sec. 2: M6 reuses M5 checkpoint and changes only tail threshold objective; used to isolate score-to-action step.

invented entities (2)

thresholded class-subgroup underdiagnosis (audit unit)
purpose: Defines the fairness object as a rare-label positive within a subgroup after a class-specific threshold is applied.
Introduced in Contributions and Sec. 1 as the unit that ranking metrics alone miss; measurement construct rather than a physical entity.
diagnostic ladder (M1–M6 probes)
purpose: Separates class-level losses, subgroup weighting, GroupDRO, and threshold selection while holding architecture fixed.
Sec. 2; experimental design device, not an external phenomenon with independent evidence.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Who Gets Missed in the Tail? Thresholded Subgroup Underdiagnosis in Long-Tailed Chest X-ray Classification." pith.science (2026). https://pith.science/paper/HQ7K25NS

@misc{pith2026260707717,
  author       = {Pith},
  title        = {Pith review of: Who Gets Missed in the Tail? Thresholded Subgroup Underdiagnosis in Long-Tailed Chest X-ray Classification},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/HQ7K25NS}},
  note         = {Machine review of arXiv:2607.07717}
}

read the original abstract

In chest X-ray (CXR) classification, acceptable ranking performance can still leave rare-positive patients below threshold, especially within subgroups. We study this pre-deployment fairness problem as an audit question: after a long-tailed multi-label CXR model is converted from scores into decisions, who is missed? Across VinDr-CXR and MIMIC-CXR/CXR-LT, we use a diagnostic ladder to separate class-level long-tail losses, subgroup-aware weighting, group robustness, and threshold selection. On VinDr-CXR, group-tail weighting followed by tail-aware thresholding reduces tail FNR from 0.665 to 0.269, sex worst-group FNR from 0.705 to 0.157, and age worst-group FNR from 0.822 to 0.133, while macro-mAP increases from 0.611 to 0.635. On MIMIC-CXR/CXR-LT, the same score-to-threshold comparison reduces tail FNR from 0.866 to 0.741 and lowers worst-group FNR across sex, age, race, and insurance; residual missed-positive rates nevertheless remain high. Paired bootstrap contrasts on VinDr support the thresholded FNR reductions, and GroupDRO reference runs indicate that aggregate group robustness alone does not remove rare subgroup misses in this setting. The study supports a narrow audit claim: rare-label fairness in CXR depends jointly on the finding, subgroup, and operating threshold, not on label frequency or ranking metrics alone.

Figures

Figures reproduced from arXiv: 2607.07717 by the authors.

**Figure 1.** M5 to M6 threshold effect across cohorts. Grey markers show M5 with F1 thresholds; colored markers show the same score model after tail-aware thresholding. Numbers are absolute reductions in missed positives per 100 true positives. threshold objective, the M5–M6 reduction isolates the score-to-action step. From M1 to M6, VinDr tail positives missed per 100 fall from 66.5 to 26.9, worst-sexsubgroup misses from 70.5 … view at source ↗

**Figure 2.** Operating-point trade-off for the M5 score model in both cohorts. Moving from the F1 threshold to tail-aware thresholds lowers missed tail positives but increases predicted tail-positive labels per 100 studies; the choice of τ is therefore a workflow policy rather than a universal setting. points. This seed analysis is still VinDr-only and should not be read as crosscohort stability. 4 Interpretation The diagnostic… view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

14 extracted references · 1 canonical work pages

[1]

In: 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

Cui, Y., Jia, M., Lin, T.Y., Song, Y., Belongie, S.: Class-balanced loss based on effective number of samples. In: 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). pp. 9260–9269 (2019). https://doi.org/10.1109/CVPR.2019.00949

doi:10.1109/cvpr.2019.00949 2019
[2]

Dong, H., Lin, Y., Zhou, P., Zhao, F., Legasto, A.C., Cho, J., Kim, D., Kim, J.N., Kim, M., Kwak, S., et al.: Cxr-lt 2026 challenge: Multi-center long-tailed and zero shot chest x-ray classification (2026)

2026
[3]

JAMA Internal Medicine178(11), 1544–1547 (Nov 2018)

Gianfrancesco, M.A., Tamang, S., Yazdany, J., Schmajuk, G.: Po- tential biases in machine learning algorithms using electronic health record data. JAMA Internal Medicine178(11), 1544–1547 (Nov 2018). https://doi.org/10.1001/jamainternmed.2018.3763

doi:10.1001/jamainternmed.2018.3763 2018
[4]

In: Lee, D., Sugiyama, M., Luxburg, U., Guyon, I., Garnett, R

Hardt, M., Price, E., Srebro, N.: Equality of opportunity in supervised learning. In: Lee, D., Sugiyama, M., Luxburg, U., Guyon, I., Garnett, R. (eds.) Advances in Neural Information Processing Systems. vol. 29. Curran Associates, Inc. (2016)

2016
[5]

PhysioNet (Nov 2019)

Johnson, A., Lungren, M., Peng, Y., Lu, Z., Mark, R., Berkowitz, S., Horng, S.: MIMIC-CXR-JPG - chest radiographs with structured labels. PhysioNet (Nov 2019). https://doi.org/10.13026/8360-t248, version 2.0.0

doi:10.13026/8360-t248 2019
[6]

Scientific Data10(1), 1 (2023)

Johnson, A.E.W., Bulgarelli, L., Shen, L., Gayles, A., Shammout, A., Horng, S., Pollard, T.J., Hao, S., Moody, B., Gow, B., Lehman, L.w.H., Celi, L.A., Mark, R.G.: Mimic-iv, a freely accessible electronic health record dataset. Scientific Data10(1), 1 (2023). https://doi.org/10.1038/s41597-022-01899-x, https://doi.org/10.1038/s41597-022-01899-x

doi:10.1038/s41597-022-01899-x 2023
[7]

Proceedings of the National Academy of Sci- ences of the United States of America117(23), 12592–12594 (Jun 2020)

Larrazabal, A.J., Nieto, N., Peterson, V., Milone, D.H., Ferrante, E.: Gen- der imbalance in medical imaging datasets produces biased classifiers for computer-aided diagnosis. Proceedings of the National Academy of Sci- ences of the United States of America117(23), 12592–12594 (Jun 2020). https://doi.org/10.1073/pnas.1919012117

doi:10.1073/pnas.1919012117 2020
[8]

Medical Image Analysis106, 103739 (2025)

Lin, M., Holste, G., Wang, S., Zhou, Y., Wei, Y., Banerjee, I., Chen, P., Dai, T., Du, Y., Dvornek, N.C., Ge, Y., Guo, Z., Hanaoka, S., Kim, D., Messina, P., Lu, Y., Parra, D., Son, D., Álvaro Soto, Urooj, A., Vidal, R., Yamagishi, Y., Yan, P., Yang, Z., Zhang, R., Zhou, Y., Celi, L.A., Summers, R.M., Lu, Z., Chen, H., Flanders, A., Shih, G., Wang, Z., Pe...

doi:10.1016/j.media.2025.103739 2024

Show all 14 references

[9]

Scientific Data 9(1), 429 (2022)

Nguyen, H.Q., Lam, K., Le, L.T., Pham, H.H., Tran, D.Q., Nguyen, D.B., Le, D.D., Pham, C.M., Tong, H.T.T., Dinh, D.H., Do, C.D., Doan, L.T., et al.: VinDr-CXR: An open dataset of chest X-rays with radiologist’s annotations. Scientific Data 9(1), 429 (2022). https://doi.org/10....

2022 doi
[10]

In: 2021 IEEE/CVF International Conference on Computer Vision (ICCV)

Ridnik, T., Ben-Baruch, E., Zamir, N., Noy, A., Friedman, I., Protter, M., Zelnik- Manor, L.: Asymmetric loss for multi-label classification. In: 2021 IEEE/CVF International Conference on Computer Vision (ICCV). pp. 82–91 (2021). https://doi.org/10.1109/ICCV48922.2021.00015

2021 doi
[11]

arXiv preprint arXiv:1911.08731 (2019) 10 H.-H

Sagawa, S., Koh, P.W., Hashimoto, T.B., Liang, P.: Distributionally robust neu- ral networks for group shifts: On the importance of regularization for worst-case generalization. arXiv preprint arXiv:1911.08731 (2019) 10 H.-H. Pham et al

1911 arXiv
[12]

Nature Medicine27(12), 2176–2182 (2021)

Seyyed-Kalantari, L., Zhang, H., McDermott, M.B.A., Chen, I.Y., Ghassemi, M.: Underdiagnosisbiasofartificialintelligencealgorithmsappliedtochestradiographs in under-served patient populations. Nature Medicine27(12), 2176–2182 (2021). https://doi.org/10.1038/s41591-021-01595-0

2021 doi
[13]

Nature Medicine31(10), 3283–3289 (2025)

Sounderajah, V., Guni, A., Liu, X., Collins, G.S., Karthikesalingam, A., et al.: The stard-ai reporting guideline for diagnostic accuracy stud- ies using artificial intelligence. Nature Medicine31(10), 3283–3289 (2025). https://doi.org/10.1038/s41591-025-03953-8

2025 doi
[14]

In: The Thirty-eighth Annual Conference on Neural Information Processing Systems (2024)

Yang, Y., Gandhi, M., Wang, Y., Wu, Y., Yao, M.S., Callison-Burch, C., Gee, J., Yatskar, M.: A textbook remedy for domain shifts: Knowledge priors for medical image analysis. In: The Thirty-eighth Annual Conference on Neural Information Processing Systems (2024)

2024

Pith tools

Reviewed July 11, 2026 · model on record in the stance chip above.

[1] [1]

In: 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

Cui, Y., Jia, M., Lin, T.Y., Song, Y., Belongie, S.: Class-balanced loss based on effective number of samples. In: 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). pp. 9260–9269 (2019). https://doi.org/10.1109/CVPR.2019.00949

doi:10.1109/cvpr.2019.00949 2019

[2] [2]

Dong, H., Lin, Y., Zhou, P., Zhao, F., Legasto, A.C., Cho, J., Kim, D., Kim, J.N., Kim, M., Kwak, S., et al.: Cxr-lt 2026 challenge: Multi-center long-tailed and zero shot chest x-ray classification (2026)

2026

[3] [3]

JAMA Internal Medicine178(11), 1544–1547 (Nov 2018)

Gianfrancesco, M.A., Tamang, S., Yazdany, J., Schmajuk, G.: Po- tential biases in machine learning algorithms using electronic health record data. JAMA Internal Medicine178(11), 1544–1547 (Nov 2018). https://doi.org/10.1001/jamainternmed.2018.3763

doi:10.1001/jamainternmed.2018.3763 2018

[4] [4]

In: Lee, D., Sugiyama, M., Luxburg, U., Guyon, I., Garnett, R

Hardt, M., Price, E., Srebro, N.: Equality of opportunity in supervised learning. In: Lee, D., Sugiyama, M., Luxburg, U., Guyon, I., Garnett, R. (eds.) Advances in Neural Information Processing Systems. vol. 29. Curran Associates, Inc. (2016)

2016

[5] [5]

PhysioNet (Nov 2019)

Johnson, A., Lungren, M., Peng, Y., Lu, Z., Mark, R., Berkowitz, S., Horng, S.: MIMIC-CXR-JPG - chest radiographs with structured labels. PhysioNet (Nov 2019). https://doi.org/10.13026/8360-t248, version 2.0.0

doi:10.13026/8360-t248 2019

[6] [6]

Scientific Data10(1), 1 (2023)

Johnson, A.E.W., Bulgarelli, L., Shen, L., Gayles, A., Shammout, A., Horng, S., Pollard, T.J., Hao, S., Moody, B., Gow, B., Lehman, L.w.H., Celi, L.A., Mark, R.G.: Mimic-iv, a freely accessible electronic health record dataset. Scientific Data10(1), 1 (2023). https://doi.org/10.1038/s41597-022-01899-x, https://doi.org/10.1038/s41597-022-01899-x

doi:10.1038/s41597-022-01899-x 2023

[7] [7]

Proceedings of the National Academy of Sci- ences of the United States of America117(23), 12592–12594 (Jun 2020)

Larrazabal, A.J., Nieto, N., Peterson, V., Milone, D.H., Ferrante, E.: Gen- der imbalance in medical imaging datasets produces biased classifiers for computer-aided diagnosis. Proceedings of the National Academy of Sci- ences of the United States of America117(23), 12592–12594 (Jun 2020). https://doi.org/10.1073/pnas.1919012117

doi:10.1073/pnas.1919012117 2020

[8] [8]

Medical Image Analysis106, 103739 (2025)

Lin, M., Holste, G., Wang, S., Zhou, Y., Wei, Y., Banerjee, I., Chen, P., Dai, T., Du, Y., Dvornek, N.C., Ge, Y., Guo, Z., Hanaoka, S., Kim, D., Messina, P., Lu, Y., Parra, D., Son, D., Álvaro Soto, Urooj, A., Vidal, R., Yamagishi, Y., Yan, P., Yang, Z., Zhang, R., Zhou, Y., Celi, L.A., Summers, R.M., Lu, Z., Chen, H., Flanders, A., Shih, G., Wang, Z., Pe...

doi:10.1016/j.media.2025.103739 2024