Pith. sign in

REVIEW 4 major objections 5 minor 38 references

Trustworthy Medical Segmentation: Uncertainty-Aware U-Net Evaluation Under Clinical Image Degradation

T0 review · 4 major / 5 minor · reviewed 2026-08-01 · deepseek-v4-flash

Pith's one-line read A model's uncertainty signal can flag when brain-tumor segmentation fails under image degradation, achieving an AUROC of 0.843 under severe Gaussian noise.

desk verdict Solid, honest benchmark undermined by overbroad safety claim and a few un-reconciled numbers. read the letter →

arxiv 2607.22727 v1 pith:BXU5ZZHC submitted 2026-07-22 cs.CV cs.LG

classification cs.CVcs.LG
keywords uncertaintyquantificationMonteCarlodropoutmedicalimagesegmentationbraintumorMRIclinicaldegradationfailuredetectionselectivepredictionU-Net
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

This paper argues that uncertainty from Monte Carlo dropout can act as a practical safety layer in physician-in-the-loop radiology workflows, catching segmentation failures that would otherwise go unnoticed. Using a synthetic multimodal brain-tumor MRI cohort and eight clinically motivated corruptions at five severities, the authors show that slice-level predictive entropy rises with degradation and tracks segmentation error, detecting failed slices under severe Gaussian noise with AUROC 0.843. On clean data, Attention U-Net attains whole-tumor Dice 0.990; under the same severe noise its deterministic output collapses to 0.089, yet the stochastic predictive mean recovers to 0.911. The paper also shows selective prediction can route a quarter of slices to expert review while raising retained-case Dice to 0.994. The broader claim is that making uncertainty visible and acting on it is a viable route to trustworthy automated segmentation.

What carries the argument

Monte Carlo dropout: the segmentation model is run T=30 times at inference with dropout enabled; the softmax outputs are averaged, and per-voxel predictive entropy is computed, then averaged over a slice to produce a scalar uncertainty score. This score serves as the failure detector, driving both failure-detection AUROC and selective-prediction coverage by routing high-entropy slices to human review.

What would settle it

Take a real clinical brain-tumor MRI dataset with expert ground-truth segmentations, run the trained Attention U-Net with MC-dropout inference, and measure slice-level mean predictive entropy under naturally occurring or synthetically added Gaussian noise; if the failure-detection AUROC (Dice<0.65) drops well below 0.843 or the Pearson r with error approaches zero, the transfer claim fails.

Watch

Extended reading notes

Core claim

The central discovery is a measurable relationship between model uncertainty and segmentation failure under clinical-style degradation. Using predictive entropy averaged over slices, the authors report that uncertainty separates failed from successful segmentations with AUROC 0.843 under severity-3 Gaussian noise, and correlates with segmentation error at Pearson r=0.53. This relationship breaks for spatially localized artifacts—occlusion yields AUROC 0.293—suggesting uncertainty signals are strongest for global intensity degradations. The paper further finds that MC-dropout model averaging itself improves robustness (Dice 0.911 vs 0.486 deterministic), and that selective prediction converts

Load-bearing premise

The synthetic brain-tumor MRI cohort and its corruption functions reproduce real clinical imaging faithfully enough that the measured uncertainty–error relationship carries over to actual patient scans.

Editorial extensions

If this is right

  • If the uncertainty–error relationship holds, a clinical workflow can set a threshold on slice-level entropy and automatically send the most uncertain predictions to radiologists, keeping automated Dice on retained cases near 0.99.
  • Uncertainty-guided failure detection is artifact-specific: it works well for global intensity changes (AUROC 0.843 for Gaussian noise, 0.992 for contrast shift) but poorly for localized occlusion (0.293), so deployment must calibrate per degradation type.
  • MC-dropout averaging offers a free robustness boost under severe noise, implying that uncertainty quantification is not just a diagnostic tool but also improves the segmentation output itself.
  • Attention U-Net's head-to-head advantage on 7 of 8 corruptions means attention gates modestly improve robustness, but the collapse under noise and contrast shift shows attention is not a substitute for explicit uncertainty safeguards.
  • The benchmark's deterministic pipeline allows direct reproduction, making it possible for other groups to test the same uncertainty protocol on their own segmentation models and data.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If the synthetic-cohort relationship transfers, the same corruption-uncertainty evaluation could be applied to other 2D medical segmentation tasks—lung nodule, cardiac MRI, etc.—since the eight corruption types are imaging-level rather than tumor-specific, and the uncertainty scoring is task-agnostic.
  • The poor failure detection on occlusion hints that dropout-based entropy captures global confidence but not localized anomaly; a testable extension is to combine entropy with feature-space out-of-distribution scores, which may catch spatially isolated artifacts.
  • The reported r=0.53 means roughly 28% of the variance in segmentation error is predictable from uncertainty alone; in practice a site could calibrate its own entropy threshold using a small set of reviewed scans, an adoption path the paper does not spell out.
  • Because the absolute Dice values come from a synthetic cohort, the safety-layer claims need external validation on real patient MRI before the uncertainty signal can be trusted clinically; the paper itself defers this to future work.
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.

Referee Report

4 major / 5 minor

Summary. The paper presents a reproducibility-oriented benchmark for uncertainty-aware brain tumor segmentation under simulated clinical image degradation. The authors generate a synthetic 60-volume multimodal MRI cohort with BraTS-style labels, train U-Net and Attention U-Net baselines, add MC-dropout inference, and evaluate segmentation accuracy, calibration, failure detection, and selective prediction under eight corruption types at five severity levels. On clean data, Attention U-Net reaches a whole-tumor Dice of 0.990. Under severity-3 Gaussian noise, the paper reports that slice-level predictive entropy tracks segmentation error (Pearson r = 0.53) and detects failed slices with AUROC 0.843. The authors conclude that uncertainty-aware inference can act as a practical safety layer in physician-in-the-loop radiology workflows, and they release code and trained models.

Significance. If the central uncertainty–error relationship transfers to real clinical scans, the framework would be a useful addition to robustness evaluation for medical segmentation: it combines an ImageNet-C-style corruption taxonomy with uncertainty calibration and selective prediction metrics, and the reproducible pipeline is a concrete strength. The paper's main empirical finding, however, is demonstrated only on a synthetic phantom cohort and is strongly corruption-type dependent: several degradation types show negative or near-zero uncertainty–error correlations, and the abstract's broad safety-layer claim goes beyond the reported evidence. The released code and deterministic evaluation protocol are valuable assets, but the clinical transferability of the headline AUROC remains untested.

major comments (4)
  1. [§VI.D, Table 5 vs Tables 3–4] The headline comparison for MC-dropout stability is confounded by different evaluation subsets. The text states that under severity-3 Gaussian noise the stochastic predictive mean attains Dice 0.911 'compared with 0.486 for deterministic inference,' but the 0.486 comes from the full 400-slice test set (Tables 3–4) while Table 5 is computed on the 120 tumor-bearing slices used for uncertainty analysis (§V-B). The Dice gain from MC averaging may be partly or largely a slice-selection effect. The same issue likely explains the contrast-shift Dice discrepancy: 0.241 in Table 3 versus 0.205 in Table 5. Please report all metrics on a common subset or clearly label the subset for each table and avoid cross-table comparisons without stating the subset difference.
  2. [§VI.D, Table 5, Fig. 6] The uncertainty–error relationship is not robust across the corruption taxonomy. For motion blur, defocus, low resolution, brightness shift, and JPEG compression the Pearson r is negative or near zero (−0.03 to −0.29), and occlusion has a failure-detection AUROC of 0.293, below chance. Only Gaussian noise and contrast shift show the positive relationship highlighted in the abstract and conclusion. The paper acknowledges this in one sentence, but the abstract and §VII still conclude that uncertainty-aware inference is a 'practical safety layer' for clinical workflows. This generalization is not supported by the paper's own data unless the authors restrict the claim to global intensity degradations or provide evidence that the other corruption types are clinically irrelevant in the target workflow.
  3. [§IV-A, §VII-D] The clinical transferability of the central claim is untested. The uncertainty–error correlation and AUROC values are measured on a purpose-built synthetic phantom cohort, and the paper concedes that the simulator 'cannot capture the full variability of real anatomy, pathology, scanner hardware, and protocol drift.' Validation on real BraTS or other clinical data is deferred to future work. Since the abstract's safety-layer claim concerns real clinical radiology workflows, the current evidence is at most suggestive. Please either soften the abstract and conclusion to a benchmark-level claim, or add a small real-data validation (even on a public dataset) to test whether the uncertainty–error relationship survives domain shift.
  4. [§VI.F, Fig. 8, Table 6] The selective prediction analysis is performed only on clean data, where the model is already highly accurate (Dice 0.990, 0/120 failures). The reported improvement from 0.990 to 0.994 at 75% coverage has little practical meaning, and it does not demonstrate that selective prediction rescues the corrupted-input cases where the safety layer is allegedly needed. Please report selective prediction coverage–accuracy curves under severity-3 (or higher) corruptions, especially Gaussian noise and contrast shift, where failures actually occur, and state the number of failed slices in each condition. This is necessary to support the workflow-level claim in §VII-A.
minor comments (5)
  1. [§III-C] The severity calibration is described only as 'calibrated on held-out data.' Please specify the held-out set, the calibration criterion, and the resulting parameter values for each corruption severity so the benchmark is reproducible.
  2. [§V-B, Tables 3–5] The paper should explicitly state in each table caption whether the metrics are computed on the full 400-slice test set or the 120 tumor-bearing slice subset. Currently the reader must infer this from the protocol section.
  3. [§VI.D, Fig. 6] For low-failure conditions such as motion blur or defocus, failure AUROC values may be unstable. Report the number of failed slices per condition and consider confidence intervals on the AUROC estimates.
  4. [§I, Contribution 6; §VIII] The phrase 'paper assembly' as a pipeline contribution is unusual and not a scientific contribution. Please remove or clearly describe it as a LaTeX-generation artifact.
  5. [Abstract] The abstract states that under severe Gaussian noise Attention U-Net's performance 'falls to 0.089.' This is severity-5 on the full test set; please add the severity label to avoid confusion with the severity-3 results that immediately follow.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the uncertainty–error relationship is measured on held-out test slices, not constructed from fitted inputs or self-citations.

full rationale

The paper is an empirical evaluation, not a derivation. The central result—that slice-level mean predictive entropy detects failed slices under severity-3 Gaussian noise with AUROC 0.843 and Pearson r = 0.53—is computed on held-out test slices using a fixed failure threshold (whole-tumor Dice < 0.65) and a fixed uncertainty score (slice-level mean predictive entropy, Eq. 5). Nothing in the procedure fits the uncertainty predictor to the failure labels: the corruption severity scales are calibrated on held-out data for corruption magnitude, not for uncertainty or error, and the Dice threshold is chosen independently of entropy. The MC-dropout stochastic passes are used both to form a predictive mean and to estimate entropy, but this shared source is not a definitional reduction because the relationship between entropy and Dice is measured, not imposed. The paper contains no self-citations in the reference list, and no uniqueness theorem or prior-work ansatz is invoked to force the conclusion. Section VII-D's concession that the synthetic phantom 'cannot capture the full variability of real anatomy, pathology, scanner hardware, and protocol drift' is an external-validity limitation, not circularity: it weakens clinical transferability but does not make the measured AUROC equal to its inputs. The negative or near-zero uncertainty–error correlations under motion, defocus, low resolution, and occlusion (Table 5) also show that the reported Gaussian/contrast successes are not artifacts of the evaluation metric by construction. Hence no circular step can be exhibited with the required specificity.

Assumptions & free parameters 4 free parameters · 3 assumptions · 0 invented entities

The paper is an empirical benchmark; it invents no new physical entities. It does rely on a self-built simulator (not externally validated), calibrated corruption severities, and several hand-chosen thresholds. Those are the free parameters and domain assumptions that carry the results.

free parameters (4)
  • Corruption severity calibration = Severity 1-5 ranges per corruption (e.g., Gaussian sigma 0.15-1.40, gamma 0.8-0.22)
    Section III-C: severity scales are 'calibrated on held-out data so that severity 3 represents moderate clinical degradation.' The chosen ranges affect every severity-conditional metric.
  • Failure threshold for 'failed' slices = Whole-tumor Dice < 0.65
    Section VI-D: this threshold defines the failure label used for AUROC calculation; changing it would change AUROC values.
  • MC dropout hyperparameters = p=0.1, T=30, dropout in deep blocks only
    Sections III-B, V-A: chosen to balance quality and cost; all uncertainty estimates depend on these choices.
  • Selective prediction operating point = 75% coverage
    Section VI-F and Table 6: the operating point is selected from the coverage-Dice curve post hoc to demonstrate a Dice improvement from 0.990 to 0.994.
assumptions (3)
  • domain assumption The synthetic phantom simulator generates anatomically and pathologically realistic BraTS-protocol MRI volumes with correct ground-truth labels.
    Section IV-A: the entire evaluation and label ground truth depend on this simulator; it is purpose-built by the authors and not validated against real expert-annotated scans.
  • domain assumption MC dropout approximates the Bayesian predictive distribution well enough for uncertainty estimation.
    Section III-B: follows Gal & Ghahramani [7]; this is a known approximation, and the paper itself notes deep ensembles are stronger.
  • domain assumption The eight corruption functions quantitatively simulate clinical MRI degradation.
    Section III-C: motivated by 'clinical shadowing' but no quantitative validation against real corrupted clinical images.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Trustworthy Medical Segmentation: Uncertainty-Aware U-Net Evaluation Under Clinical Image Degradation." pith.science (2026). https://pith.science/paper/BXU5ZZHC

@misc{pith2026260722727,
  author       = {Pith},
  title        = {Pith review of: Trustworthy Medical Segmentation: Uncertainty-Aware U-Net Evaluation Under Clinical Image Degradation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/BXU5ZZHC}},
  note         = {Machine review of arXiv:2607.22727}
}
read the original abstract

Medical image segmentation models often report high benchmark accuracy under ideal imaging conditions, yet their failures under clinical degradation can be quiet: sensor noise, patient motion, low- resolution acquisition, and contrast variability may all alter model behavior without producing an obvious warning. We present a reproducible framework for evaluating uncertainty-aware segmentation under con- trolled clinical degradation. Our experiments use a synthetic multimodal brain tumor MRI cohort generated with a biophysical phantom simulator that follows the BraTS protocol. We train U-Net and Attention U-Net baselines for multi-class tumor sub-region segmentation and augment both models with Monte Carlo dropout to estimate per-voxel uncertainty. Across eight clinically motivated corruption types at five severity levels, we measure segmentation accuracy, calibration, failure detection, and selective prediction coverage. On clean data, Attention U-Net achieves a whole-tumor Dice of 0.990; under severe Gaussian noise, its performance falls to 0.089. Predictive uncertainty rises with degradation and tracks segmentation error (Pearson r = 0.53 under severity-3 Gaussian noise), allowing us to flag failures with an AUROC of 0.843. These results argue for uncertainty-aware inference as a practical safety layer in physician-in-the-loop radiology workflows. We release the code, trained models, and evaluation protocol to support direct reproduction.

Figures

Figures reproduced from arXiv: 2607.22727 by the authors.

Figure 1
Figure 1. (a) Volume-level tumor presence distribution and (b) slice-level pixel class distribution in the synthetic cohort. Background pixels dominate, creating severe class imbalance typical of medical segmenta￾tion tasks. VI Results A Clean Test Set Performance [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
Figure 2
Figure 2. Whole-tumor Dice coefficient under eight corruption types at severity level 3. Dashed lines show clean-data performance for each model [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Progressive performance degradation as corruption severity increases from 1 (mild) to 5 (severe). Solid lines: Attention U-Net; dashed lines: U-Net. failure: under severity-3 Gaussian noise, pre￾dictive entropy separates failed from successful segmentations with AUROC 0.843. On clean data, the model is uniformly confident and cor￾rect (minimum whole-tumor Dice 0.887, with 0 of 120 slices below the 0.65 failure thres… view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: (a) Radar chart comparing model robustness across the corruption spectrum at severity 3. (b) Heatmap of segmentation and calibration metrics (Dice, UCE, failure AUROC, uncertainty–error correlation) across corruption types for Attention U-Net at severity 3. 0.07 0.08 0…
Figure 5
Figure 5. Figure 5: Scatter plots of slice-level mean predictive entropy versus segmentation error for (a) U-Net (r = 0.88) and (b) Attention U-Net (r = 0.53) under Gaussian noise at severity 3. Each point represents one test slice; the red dashed line shows the linear regression fit. inf…
Figure 6
Figure 6. Figure 6: Failure detection ROC curves using slice-level mean predictive entropy as the failure predictor (Attention U-Net). All corruption conditions are evaluated at severity 3; conditions with no failed slices are omitted because an ROC cannot be defined for them. (a) T2-FLAI…
Figure 7
Figure 7. Figure 7: Monte Carlo dropout uncertainty visualization: (a) T2-FLAIR input slice with hyperintense tumor, (b) expert-convention ground truth annotation, (c) Attention U-Net prediction mask, (d) predictive uncertainty map (hot colors = high uncertainty). The map emphasizes tumor…
Figure 8
Figure 8. Figure 8: Selective prediction tradeoff curves for U-Net and Attention U-Net on clean test data. As the uncertainty threshold tightens, more predictions are routed to human review (decreasing coverage) but the Dice on retained predictions increases. The starred operating point (…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

38 extracted references · 2 linked inside Pith

  1. [1]

    Ronneberger, P

    O. Ronneberger, P. Fischer, and T. Brox. U-Net: Convolutional networks for biomedi- cal image segmentation. InProc. Med. Im- age Comput. Comput.-Assist. Intervent. (MIC- CAI), pages 234–241, 2015

  2. [2]

    Oktay, J

    O. Oktay, J. Schlemper, L. L. Folgoc, M. Lee, M. Heinrich, K. Misawa, K. Mori, S. McDon- agh, N. Y. Hammerla, B. Kainz, B. Glocker, and D. Rueckert. Attention U-Net: Learn- ing where to look for the pancreas, 2018. arXiv:1804.03999

  3. [3]

    Mehrtash, W

    A. Mehrtash, W. M. Wells, C. M. Tempany, P.Abolmaesumi, andT.Kapur. Confidencecal- ibration and predictive uncertainty estimation for deep medical image segmentation.IEEE Trans. Med. Imag., 39(12):3868–3878, Decem- ber 2020

  4. [4]

    Hendrycks and T

    D. Hendrycks and T. Dietterich. Benchmarking neural network robustness to common corrup- tions and perturbations. InProc. Int. Conf. Learn. Represent. (ICLR), 2019

  5. [5]

    Kamnitsas, C

    K. Kamnitsas, C. Ledig, V. F. J. Newcombe, J. P. Simpson, A. D. Kane, D. K. Menon, D. Rueckert, and B. Glocker. Efficient multi- scale 3D CNN with fully connected CRF for accurate brain lesion segmentation.Med. Im- age Anal., 36:61–78, 2017

  6. [6]

    F.-C.Ghesu, B.Georgescu, E.Gibson, S.Guen- del, M. K. Kalra, R. Singh, S. R. Digumarthy, S. Grbic, and D. Comaniciu. Quantifying and leveraging classification uncertainty for chest radiograph assessment. InProc. MICCAI, pages 676–684, 2019

  7. [7]

    Gal and Z

    Y. Gal and Z. Ghahramani. Dropout as a Bayesian approximation: Representing model uncertaintyindeeplearning. InProc. Int. Conf. Mach. Learn. (ICML), pages 1050–1059, 2016

  8. [8]

    W. Jin, M. Fatehi, K. Abhishek, M. Mallya, et al. Artificial intelligence in glioma imaging: Challengesandadvances.J. Neural Eng., 17(4), 2020

Show all 38 references
  1. [9]

    J. Wong, A. Fong, N. McVicar, S. Smith, et al. Comparing deep learning-based auto- segmentation of organs at risk and clinical tar- get volumes to expert inter-observer variability in radiotherapy planning.Radiother. Oncol., 144:152–158, 2020

  2. [10]

    Y. Xie, B. Yang, Q. Guan, J. Zhang, Q. Wu, et al. Attention mechanisms in medical image segmentation: A survey, 2023

  3. [11]

    R. Gu, G. Wang, T. Song, R. Huang, M. Aert- sen, J. Deprest, S. Ourselin, T. Vercauteren, and S. Zhang. CA-Net: Comprehensive atten- tion convolutional neural networks for explain- able medical image segmentation.IEEE Trans. Med. Imag., 40(2):699–711, 2021. 12

  4. [12]

    B. H. Menze, A. Jakab, S. Bauer, J. Kalpathy- Cramer, K. Farahani, J. Kirby, et al. The multi- modal brain tumor image segmentation bench- mark (BRATS).IEEE Trans. Med. Imag., 34(10):1993–2024, October 2015

  5. [13]

    Bakas, H

    S. Bakas, H. Akbari, A. Sotiras, M. Bilello, M. Rozycki, J. S. Kirby, J. B. Freymann, K. Farahani, and C. Davatzikos. Advancing the cancer genome atlas glioma MRI collections with expert segmentation labels and radiomic features.Sci. Data, 4:170117, 2017

  6. [14]

    Bakas, M

    S. Bakas, M. Reyes, A. Jakab, S. Bauer, M. Rempfler, A. Crimi, et al. Identifying the best machine learning algorithms for brain tu- mor segmentation, progression assessment, and overall survival prediction in the BRATS chal- lenge, 2018. arXiv:1811.02629

  7. [15]

    Isensee, P

    F. Isensee, P. F. Jaeger, S. A. A. Kohl, J. Pe- tersen, and K. H. Maier-Hein. nnU-Net: A self-configuring method for deep learning-based biomedical image segmentation.Nat. Methods, 18(2):203–211, 2021

  8. [16]

    Myronenko

    A. Myronenko. 3D MRI brain tumor segmenta- tion using autoencoder regularization. InProc. Int. MICCAI Brainlesion Workshop (Brain- Les), pages 311–320, 2018

  9. [17]

    Kwon, J.-H

    Y. Kwon, J.-H. Won, B. J. Kim, and M. C. Paik. UncertaintyquantificationusingBayesian neuralnetworksinclassification: Applicationto biomedical image segmentation.Comput. Stat. Data Anal., 142:106816, 2020

  10. [18]

    Jungo and M

    A. Jungo and M. Reyes. Assessing reliability and challenges of uncertainty estimations for medical image segmentation. InProc. MICCAI, pages 48–56, 2019

  11. [19]

    Carannante, D

    G. Carannante, D. Dera, N. C. Bouaynaya, R. Ghulam Rasool, and H. M. Shafee. Trust- worthy medical segmentation with uncertainty estimation, 2021

  12. [20]

    Kendall and Y

    A. Kendall and Y. Gal. What uncertainties do we need in Bayesian deep learning for computer vision? InProc. Adv. Neural Inf. Process. Syst. (NeurIPS), pages 5574–5584, 2017

  13. [21]

    Mårtensson, D

    G. Mårtensson, D. Ferreira, T. Granberg, L. Cavallin, K. Oppedal, A. Padovani, I. Rek- torova, L. Bonanni, M. Pardini, M. G. Kram- berger, et al. The reliability of a deep learn- ing model in clinical out-of-distribution MRI data: A multicohort study.Med. Image Anal., 66:101714, 2020

  14. [22]

    Panfilov, A

    E. Panfilov, A. Tiulpin, S. Klein, M. T. Niem- inen, and S. Saarakkala. Improving robustness of medical image diagnosis with denoising con- volutional neural networks. InProc. MICCAI, pages 846–854, 2019

  15. [23]

    N. Porz, S. Bauer, A. Pica, P. Schucht, J. Beck, R. K. Verma, J. Slotboom, M. Reyes, and R. Wiest. Multi-modal glioblastoma segmen- tation: Man versus machine.PLoS ONE, 9(5):e96873, 2014

  16. [24]

    J. G. Pipe. Motion correction with PRO- PELLER MRI: Application to head motion and free-breathing cardiac imaging.Magn. Reson. Med., 42(5):963–969, 1999

  17. [25]

    Maier-Hein, A

    L. Maier-Hein, A. Reinke, P. Godau, et al. Metrics reloaded: Recommendations for image analysis validation.Nat. Methods, 21(2):195– 212, 2024

  18. [26]

    American College of Radi- ology, Reston, VA, USA, 2024

    American College of Radiology.ACR Manual on Contrast Media. American College of Radi- ology, Reston, VA, USA, 2024

  19. [27]

    MRI for cancer

    American Cancer Society. MRI for cancer. [Online]. Available:https://www.cancer.org, 2025

  20. [28]

    NBME–Score clinical patient notes

    Kaggle. NBME–Score clinical patient notes. [Online]. Available:https://www.kaggle.com, 2022

  21. [29]

    Tordjman, I

    M. Tordjman, I. Bolger, M. Yuce, F. Restrepo, Z. Liu, et al. Large language models in cancer imaging: Applications and future perspectives. J. Clin. Med., 14(10):3285, 2025

  22. [30]

    E. Pons, L. M. M. Braun, M. G. M. Hunink, and J. A. Kors. Natural language processing in radiology: A systematic review.Radiology, 279(2):329–343, 2016

  23. [31]

    N. Bakx, D. Rijkaart, M. van der Sangen, J. Theuws, et al. Clinical evaluation of a deep learning segmentation model including man- ual adjustments afterwards for locally advanced breast cancer.Tech. Innov. Patient Support Ra- diat. Oncol., 2023

  24. [32]

    S. Kohl, B. Romera-Paredes, C. Meyer, J. De Fauw, J. R. Ledsam, K. Maier-Hein, S. M. A. Eslami, D. Jimenez Rezende, and O. Ron- neberger. A probabilistic U-Net for segmenta- tion of ambiguous images. InProc. Adv. Neural 13 Inf. Process. Syst. (NeurIPS), pages 6965–6975, 2018

  25. [33]

    Geifman and R

    Y. Geifman and R. El-Yaniv. Selective classifi- cation for deep neural networks. InProc. Adv. Neural Inf. Process. Syst. (NeurIPS), pages 2150–2158, 2017

  26. [34]

    Hatamizadeh, V

    A. Hatamizadeh, V. Nath, Y. Tang, D. Yang, H. R. Roth, and D. Xu. Swin UNETR: Swin transformersforsemanticsegmentationof brain tumors in MRI images. InProc. Int. MICCAI Brainlesion Workshop (BrainLes), pages 272– 284, 2022

  27. [35]

    P. M. Bruntha, S. I. A. Pandian, K. M. Sagayam, S. Bandopadhyay, M. Pomplun, and H. Dang. Lung_PAYNet: A pyramidal atten- tion based deep learning network for lung nod- ule segmentation.Sci. Rep., 12:20330, 2022

  28. [36]

    H. Cui, C. Yuwen, L. Jiang, Y. Xia, and Y. Zhang. Multiscale attention guided U-Net architecture for cardiac segmentation in short- axis MRI images.Comput. Methods Programs Biomed., 207, 2021

  29. [37]

    S. M. McKinney, M. Sieniek, V. Godbole, et al. International evaluation of an AI system for breast cancer screening.Nature, 577(7788):89– 94, January 2020

  30. [38]

    Food and Drug Administration

    U.S. Food and Drug Administration. Ar- tificial intelligence-enabled medical devices. [Online]. Available:https://www.fda.gov/ medical-devices, 2025. 14

Pith tools

Reviewed August 1, 2026 · model on record in the stance chip above.