Pith. sign in

REVIEW 5 major objections 5 minor 59 references

Generalizable automated ischaemic stroke lesion segmentation with vision transformers

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

Pith's one-line read A SWIN-UNETR trained on 3,563 curated DWI scans achieves state-of-the-art stroke lesion segmentation, and adding lesion-free controls cuts false positives more than tenfold.

desk verdict A valuable dataset and evaluation framework with a plausible internal comparison, but the headline SOTA Dice is optimistically biased by epoch selection on the validation set and the absence of external benchmarks. read the letter →

arxiv 2502.06939 v1 pith:SQV44HHM submitted 2025-02-10 eess.IV cs.CVcs.LG

classification eess.IVcs.CVcs.LG
keywords ischemicstrokediffusion-weightedimaginglesionsegmentationvisiontransformerSWIN-UNETRfalsepositivecontrolmedicalimageevaluationframework
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 claims that a vision-transformer-based segmentation model, trained on 3,563 expert-curated diffusion-weighted MRI images of acute ischaemic stroke, outperforms U-Net baselines and reaches state-of-the-art out-of-sample accuracy, with a mean Dice of 0.8952 and a Hausdorff distance of 2.6498 voxels. It further claims that adding 6,900 lesion-free control images to training, combined with a Thresholded Average loss that penalises false positives, cuts average false-positive voxels per control image from 126.80 to 3.31 without sacrificing segmentation fidelity. The paper also introduces a four-part evaluation framework that checks performance across brain anatomy, lesion morphology, image noise, and false-positive rates, and reports that the transformer models show no significant anatomical inequity, unlike the U-Net. If these claims hold, automated DWI lesion mapping could be reliable enough for clinical and research use on heterogeneous, real-world scanning data.

What carries the argument

The central machinery is the SWIN-UNETR architecture—a U-Net with Swin-transformer stages that capture long-range spatial dependencies—which the paper identifies as the main source of the accuracy gain over the U-Net baseline. Around it sit a data augmentation suite covering intensity shifts, geometric transforms, and DWI-typical noise; a fold-balancing scheme that splits the dataset by lesion phenotype and volume; a combined Dice-plus-Focal training loss; and, for the control-trained variant, a Thresholded Average loss that penalises any voxel with probability above 0.5 in lesion-free images. The evaluation framework uses voxel-wise general linear models, a low-dimensional UMAP embedding of lesion morphology, parametric noise corruption, and false-positive counting on held-out controls.

What would settle it

Manually and densely segment a random sample of roughly 200 of the 3,563 lesions, then compare SWIN-UNETR predictions against those dense manual labels instead of the curated semi-automated labels; if the mean Dice drops below 0.85 or voxel-level errors align with the original label errors, the central claim of state-of-the-art fidelity is not supported.

Watch

Extended reading notes

Core claim

The paper reports that a SWIN-UNETR vision transformer, trained with data augmentation and a statistically balanced five-fold split on 3,563 curated DWI lesion images, achieves state-of-the-art out-of-sample segmentation with a mean Dice of 0.8952 and a Hausdorff distance of 2.6498 voxels, surpassing the U-Net baseline at a Dice of 0.8408. It further reports that training a second SWIN-UNETR with 6,900 DWI-negative control images and a Thresholded Average loss reduces average false-positive voxels per control from 126.80 to 3.31, makes the model significantly more resilient to Rician and bias-field noise, and does not degrade standard segmentation metrics. The accompanying evaluation framework, using voxel-wise statistical mapping, low-dimensional morphological embedding, parametric noise corruption, and control-image false-positive counting, shows that the transformer models are equitable across brain anatomy while the U-Net performs significantly better within middle-cerebral-artery territory.

Load-bearing premise

The results depend on the semi-automated ground-truth labels being accurate enough for training and evaluation; if the machine-generated labels that experts curated carry residual errors that correlate with the models' errors, the reported Dice, equity, and false-positive numbers would all be inflated.

Editorial extensions

If this is right

  • DWI lesion segmentation at clinical scale is achievable: both transformer models clear the 0.85 Dice threshold that the literature associates with expert-level performance.
  • Reporting only mean Dice is insufficient: the new anatomical, morphological, noise, and false-positive calibrations expose differences among models that average metrics hide.
  • Adding DWI-negative controls with a false-positive-penalising loss is a cheap, replicable way to cut artefact-driven over-segmentation by more than an order of magnitude.
  • Balanced five-fold cross-validation on a large, multi-site routine dataset provides more dependable out-of-sample performance estimates than small held-out test sets.

Reading between the lines

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

  • If the semi-automated labels contain residual error that correlates with model error, the paper's absolute Dice values may overstate true accuracy, while its between-model comparisons are likely to remain informative.
  • Because the U-Net baseline was trained without the augmentation and balancing scheme, part of the observed gap may reflect training recipe rather than the Transformer architecture itself; a U-Net trained with the same recipe would isolate the architectural contribution.
  • The open-source release of code and weights invites independent replication on other DWI datasets, including non-UK populations where scanner and protocol differences are even larger than those captured here.
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

5 major / 5 minor

Summary. The manuscript presents a pipeline for automated ischaemic stroke lesion segmentation from diffusion-weighted imaging, built around the SWIN-UNETR vision transformer and trained on a large (n=3563) curated set of lesion images plus 6900 controls. The authors compare SWIN-UNETR variants against a U-Net baseline using 5-fold cross-validation, reporting an out-of-sample Dice of 0.8952 and HD of 2.6498 voxels for SWIN-UNETR, and propose a multi-axis evaluation framework covering anatomical equity, morphological variation, false-positive control, and noise robustness. They also introduce a Thresholded Average loss for training with negative images and provide open-source code and model weights.

Significance. If the reported results are valid, the paper would make a useful contribution: it introduces a very large multi-site DWI lesion dataset, an open-source training pipeline, a practical control-training loss, and an evaluation framework that goes beyond mean Dice to examine equity and robustness. These are strengths that the field would benefit from. However, the central claim of state-of-the-art performance is not supported by comparison with published methods or public benchmarks, and the reported absolute Dice figures are vulnerable to a model-selection bias in the cross-validation protocol. The paper's significance therefore depends on the outcome of the methodological fixes requested below.

major comments (5)
  1. [Section 2.3.3 and 2.4.1] The reported 5-fold Dice/HD values (Section 3.1) are not unbiased out-of-sample estimates. Section 2.3.3 states that training was stopped when Dice and HD had not improved together for 150 epochs and that the best model was selected by best Dice between the last epoch of joint improvement and 150 further epochs; Section 2.4.1 then claims 'no hyperparameter tuning or model selection was performed during the cross-validation process.' This is contradictory. Early stopping with epoch selection is a form of model selection, and because the reported metrics are computed on the same validation folds used for that selection, the reported Dice is a maximum over a window of candidate epochs, biased upward. Please re-evaluate with a separate held-out test set, or use nested cross-validation where the inner loop selects the epoch, and report the variance of the selection.
  2. [Section 3.1 and Discussion] The state-of-the-art claim is not supported by the evidence presented. The only comparison is against an internal U-Net without augmentation; no comparison is made with published DWI lesion segmentation methods or with public benchmarks such as ISLES. The reference to the 0.85 Dice threshold (Discussion, citing Liew et al., 2022) is not a direct method comparison. Please add results on a public benchmark (e.g., ISLES 2015/2022) or a table comparing with recent published Dice/HD on comparable data, and temper the 'state-of-the-art' wording until such evidence is provided.
  3. [Section 2.1.5] The reference standard is machine-generated and only manually edited. Section 2.1.5 describes an iterative process in which initial labels came from the zeta anomaly method, then a residual U-Net, then UNETR, with manual selection and modification. The Limitations state that dense manual segmentation is infeasible at this scale and that there is no mechanism to quantify the advantage of dense manual labels. This is a load-bearing issue because if residual errors in these iterative labels correlate with the errors of the trained models, the reported Dice, HD, and false-positive counts are inflated. Please provide an independent assessment on a random subset of, say, 50-100 images, with expert manual segmentations, and report inter-rater agreement and the agreement of the final model with that independent reference.
  4. [Section 3.1] Standard performance metrics are reported as fold means without any confidence intervals or variance. For example, the difference between SWIN-UNETR+Ctr (Dice 0.8915, HD 2.6636) and SWIN-UNETR (Dice 0.8952, HD 2.6498) may not be meaningful with only five folds, yet the Discussion treats these as distinguishing the two models. Please report per-fold results, standard deviations, and bootstrapped confidence intervals for all primary metrics, and avoid over-interpreting small differences without statistical support.
  5. [Section 2.4.1 and Sections 2.2.2-2.3.3] The claim that 'no hyperparameter tuning or model selection was performed' is at odds with the methods. Section 2.2.2 states that augmentation probabilities were 'chosen empirically between 5, 10 and 20%'; Section 2.2.3 sets the focal loss focusing parameter to 2; Section 2.2.4 assigns a weight to the Thresholded Average loss 'in response to training performance'; and Section 2.3.3 fixes the early-stopping patience at 150 epochs based on prior tests. These are hyperparameter choices that affect all reported results. Please disclose all such choices and treat them as part of model selection when discussing generalizability, or remove the claim in Section 2.4.1.
minor comments (5)
  1. [Section 2.2.5] The phrase 'Hausdorff Distance in the 95% confidence interval' is ambiguous; standard practice is the 95th percentile of surface-to-surface distances (HD95). Please define the metric explicitly.
  2. [Section 2.2.4] The Thresholded Average loss is described in prose but no equation is given. Please provide the mathematical definition, including how the 0.5 threshold is applied and how the loss is averaged over the control set.
  3. [Section 2.3.2] There is a typo: 'Kruskall-Wallis' should be 'Kruskal-Wallis'. Also, the description of the balancing criterion (highest p-value and lowest variability in average volume and standard deviation) would benefit from an explicit equation or pseudocode.
  4. [Sections 2.4.4 and 3.4] The bootstrap procedure used for the paired t-tests in Table 1 and Figure 6 is not fully specified. Please state how the 100-image subsamples are drawn, whether the resampling is paired across models, and how the t-statistic and p-value are computed from the bootstrap distribution.
  5. [Conclusion] The claim of 'the largest DWI dataset of delineated acute ischaemic stroke lesions known to us' should be placed in context of public resources such as ATLAS or ISLES. If the claim is intended as a strong comparative statement, please provide a citation or explicit comparison of sizes.

Circularity Check

1 steps flagged · score 6.0 of 10

Early-stopping epoch selection on the validation set makes the reported 'out-of-sample' Dice a selected maximum, so the headline SOTA claim is partly fitted to its own evaluation data.

  1. fitted input called prediction [Section 2.3.3 (Early stopping); Section 2.4.1 (5-fold cross-validation); Section 3.1 (Standard performance evaluation)]
    "We selected the best model with the best Dice score between the last epoch in which both Dice and HD improved and 150 further epochs. ... This strategy avoids the selection biases of a held-out testing set without sacrificing learning potential, as no hyperparameter tuning or model selection was performed during the cross-validation process. ... by combining the performance of the five folds, we obtain an out-of-sample performance for the whole 3563 training dataset."

    The early-stopping rule selects, per fold, the epoch whose validation-fold Dice is highest within a 150-epoch window. The same validation-fold predictions are then pooled across folds and reported as 'out-of-sample' Dice (0.8952 for SWIN-UNETR). Consequently the reported figure is the maximum of a noisy validation curve, not an unbiased estimate for an independent test set; the selection procedure has already fitted the epoch to the very labels used to compute the metric. The paper's statement that 'no ... model selection was performed' is contradicted by its own early-stopping rule. The absolute SOTA claim and the 0.85-threshold argument in the Discussion therefore rest on a selected maximum rather than on a prediction from an independent sample.

full rationale

The central performance claim is not a fitted equation, and most of the paper's evaluation framework is self-contained: the three models are trained and compared on the same curated labels, the control false-positive evaluation uses held-out folds even for the control-trained model, and the proposed equity/noise analyses are independent methodological additions. However, one load-bearing 'prediction' does reduce to model selection on the same data: the reported out-of-sample Dice is obtained after choosing the best epoch by maximizing Dice on the validation fold, and that same validation-fold Dice is then averaged and presented as out-of-sample performance. This makes the headline SOTA number and the 0.85-threshold conclusion optimistically biased by construction. The label-curation chain (Section 2.1.5) is a real validity limitation, but because manual curation and expert review intervene between the machine-generated segmentations and the final labels, it is not a by-construction circularity. Self-citations such as Mah et al. 2014, Xu et al. 2018, and Bonkhoff et al. 2021 supply data sources and phenotype definitions rather than an unverified uniqueness or equivalence argument, so they do not independently raise the circularity score. Overall, the paper has substantial independent content, but the primary absolute performance estimate is partly fitted to its own evaluation set, warranting a score of 6.

Assumptions & free parameters 6 free parameters · 6 assumptions · 0 invented entities

The central claims rest mainly on assumptions about label quality, control-image negativity, registration fidelity, and morphological archetypes. Hyperparameters such as augmentation probabilities, early stopping patience, and the control-loss weight are chosen empirically, and several numeric values are not reported. No new physical or ontological entities are postulated. The heaviest unquantified burden is that all performance and equity analyses are measured against semi-automated labels whose independent validity is not established.

free parameters (6)
  • Augmentation probabilities = 5%, 10%, or 20% depending on augmentation type
    Chosen empirically in Section 2.2.2; not optimized systematically and vary by augmentation category.
  • Focal loss focusing parameter (gamma) = 2
    Fixed to the standard Lin et al. value; affects loss weighting for hard and easy voxels.
  • Thresholded Average loss weight = Not reported numerically
    Section 2.2.4 states the weight was assigned in response to training performance; the value is not disclosed, so the final loss is not exactly reproducible from the paper.
  • Early stopping patience = 150 epochs
    Empirically determined in Section 2.3.3; affects which checkpoint is selected.
  • Data balancing search size and selection criterion = 50,000 permutations; highest Kruskal-Wallis p-value and lowest volume variability
    Section 2.3.2: split selection is a fitted choice that could influence cross-validation estimates.
  • Noise calibration maxima = Not specified numerically
    Section 2.4.5: parameters for bias field, Gibbs, and Rician noise were empirically selected at the limit of interpretability; used to define robustness curves.
assumptions (6)
  • domain assumption Manually edited machine-generated segmentations approximate expert ground truth sufficiently for training and evaluation.
    Section 2.1.5 and Limitations: dense manual labels are infeasible at this scale, and the authors state the advantage of dense manual labels cannot be quantified.
  • domain assumption DWI-negative classifications from radiological reports and manual inspection are correct, with no covert ischaemic lesions.
    Section 2.1.1: controls are curated from reports; false negatives would contaminate the false-positive evaluation.
  • domain assumption SPM12 nonlinear registration to MNI correctly maps lesioned brains, and reslicing to 2 mm preserves lesion boundaries for voxel-wise Dice.
    Section 2.1.4 and Limitations: registration is assumed robust; on lesioned brains this is a known vulnerability.
  • domain assumption Bonkhoff phenotype masks thresholded at 10% are valid morphological archetypes for fold balancing.
    Section 2.3.2: membership is defined by maximum Dice with thresholded archetype masks.
  • domain assumption Mass-univariate FWE inference is sensitive enough that null results imply equity.
    Section 3.2: the paper relies on the well-established high sensitivity of mass-univariate inference to argue that SWIN-UNETR models show no anatomical inequity.
  • domain assumption Noise types and intensity ranges used in calibration represent clinical DWI corruption.
    Section 2.4.5: parameters are empirically selected, not grounded in measured clinical noise distributions.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Generalizable automated ischaemic stroke lesion segmentation with vision transformers." pith.science (2026). https://pith.science/paper/SQV44HHM

@misc{pith2026250206939,
  author       = {Pith},
  title        = {Pith review of: Generalizable automated ischaemic stroke lesion segmentation with vision transformers},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/SQV44HHM}},
  note         = {Machine review of arXiv:2502.06939}
}
read the original abstract

Ischaemic stroke, a leading cause of death and disability, critically relies on neuroimaging for characterising the anatomical pattern of injury. Diffusion-weighted imaging (DWI) provides the highest expressivity in ischemic stroke but poses substantial challenges for automated lesion segmentation: susceptibility artefacts, morphological heterogeneity, age-related comorbidities, time-dependent signal dynamics, instrumental variability, and limited labelled data. Current U-Net-based models therefore underperform, a problem accentuated by inadequate evaluation metrics that focus on mean performance, neglecting anatomical, subpopulation, and acquisition-dependent variability. Here, we present a high-performance DWI lesion segmentation tool addressing these challenges through optimized vision transformer-based architectures, integration of 3563 annotated lesions from multi-site data, and algorithmic enhancements, achieving state-of-the-art results. We further propose a novel evaluative framework assessing model fidelity, equity (across demographics and lesion subtypes), anatomical precision, and robustness to instrumental variability, promoting clinical and research utility. This work advances stroke imaging by reconciling model expressivity with domain-specific challenges and redefining performance benchmarks to prioritize equity and generalizability, critical for personalized medicine and mechanistic research.

Figures

Figures reproduced from arXiv: 2502.06939 by the authors.

Figure 2
Figure 2. Lesions overlap of the ground truths and SWIN-UNETR+Ctr predictions. [PITH_FULL_IMAGE:figures/full_fig_p022_2.png] view at source ↗
Figure 5
Figure 5. UMAP embeddings of lesion predictions for different models, evaluated with [PITH_FULL_IMAGE:figures/full_fig_p025_5.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

59 extracted references · 51 canonical work pages

  1. [1]

    Feigin, V. L. et al. The global burden of neurological disorders: translating evidence into policy. Lancet Neurol. 19, 255–265 (2020)

  2. [2]

    Foroushani, H. M. et al. The Stroke Neuro-Imaging Phenotype Repository: An Open Data Science Platform for Stroke Research. Front. Neuroinformatics 15, (2021)

  3. [3]

    Liew, S.-L. et al. A large, open source dataset of stroke anatomical brain images and manual lesion segmentations. Sci. Data 5, 180011 (2018)

  4. [4]

    Román, L. S. et al. Imaging features and safety and efficacy of endovascular stroke treatment: a meta-analysis of individual patient-level data. Lancet Neurol. 17, 895–904 (2018)

  5. [5]

    Pinto, A. et al. Stroke Lesion Outcome Prediction Based on MRI Imaging Combined With Clinical Information. Front. Neurol. 9, (2018)

  6. [6]

    Roberts, T. P. L. & Rowley, H. A. Diffusion weighted magnetic resonance imaging in stroke. Eur. J. Radiol. 45, 185–194 (2003)

  7. [7]

    Artifacts in Diffusion MRI

    Pierpaoli, C. Artifacts in Diffusion MRI. in Diffusion MRI: Theory, Methods, and Applications (ed. Jones, P., Derek K.) 0 (Oxford University Press, 2010). doi:10.1093/med/9780195369779.003.0018

  8. [8]

    Bonkhoff, A. K. et al. Reclassifying stroke lesion anatomy. Cortex 145, 1–12 (2021)

Show all 59 references
  1. [9]

    Ryu, W.-S. et al. Deep Learning-Based Automatic Classification of Ischemic Stroke Subtype Using Diffusion-Weighted Images. J. Stroke 26, 300–311 (2024)

  2. [10]

    Sinha, S., Bhatt, M., Anand, A., Areeckal, A. S. & A. EnigmaNet: A Novel Attention-Ⅴ Enhanced Segmentation Framework for Ischemic Stroke Lesion Detection in Brain MRI. IEEE Access 12, 91480–91498 (2024)

  3. [11]

    Koska, I. O. et al. Voxel level dense prediction of acute stroke territory in DWI using deep learning segmentation models and image enhancement strategies. Jpn. J. Radiol. 42, 962–972 (2024)

  4. [12]

    Werdiger, F. et al. Clinical performance review for 3-D Deep Learning segmentation of stroke infarct from diffusion-weighted images. Neuroimage Rep. 4, 100196 (2024)

  5. [13]

    de la Rosa, E. et al. A Robust Ensemble Algorithm for Ischemic Stroke Lesion Segmentation: Generalizability and Clinical Utility Beyond the ISLES Challenge. Preprint at https://doi.org/10.48550/arXiv.2403.19425 (2024)

  6. [14]

    & R H, H

    Shedbalkar, J., K, P. & R H, H. UNet and Transformer-Based Model for Multi-Modality Brain Tumor Segmentation. Int. J. Electron. Commun. Eng. 10, 22–35 (2023)

  7. [15]

    & Viriri, S

    Magadza, T. & Viriri, S. Efficient nnU-Net for Brain Tumor Segmentation. IEEE Access 11, 126386–126397 (2023)

  8. [16]

    Sasank, V. V. S. & Venkateswarlu, S. An automatic tumour growth prediction based segmentation using full resolution convolutional network for brain tumour. Biomed. Signal Process. Control 71, 103090 (2022)

  9. [17]

    K., Mohinta, S., Gray, R., Hyare, H

    Ruffle, J. K., Mohinta, S., Gray, R., Hyare, H. & Nachev, P. Brain tumour segmentation with incomplete imaging data. Brain Commun. 5, fcad118 (2023)

  10. [18]

    Punn, N. S. & Agarwal, S. Modality specific U-Net variants for biomedical image segmentation: a survey. Artif. Intell. Rev. 55, 5845–5889 (2022)

  11. [19]

    Siddique, N., Paheding, S., Elkin, C. P. & Devabhaktuni, V. U-Net and Its Variants for Medical Image Segmentation: A Review of Theory and Applications. IEEE Access 9, 82031–82057 (2021)

  12. [20]

    Maier-Hein, L. et al. Metrics reloaded: recommendations for image analysis validation. Nat. Methods 21, 195–212 (2024)

  13. [21]

    & Nachev, P

    Xu, T., Rolf Jäger, H., Husain, M., Rees, G. & Nachev, P. High-dimensional therapeutic inference in the focally damaged human brain. Brain 141, 48–54 (2018)

  14. [22]

    & Nachev, P

    Mah, Y.-H., Husain, M., Rees, G. & Nachev, P. Human brain lesion-deficit inference remapped. Brain 137, 2522–2531 (2014)

  15. [23]

    Littlejohns, T. J. et al. The UK Biobank imaging enhancement of 100,000 participants: rationale, data collection, management and future directions. Nat. Commun. 11, 2624 (2020)

  16. [24]

    Ruffle, J. K. et al. Computational limits to the legibility of the imaged human brain. NeuroImage 291, 120600 (2024)

  17. [25]

    Alfaro-Almagro, F. et al. Image processing and Quality Control for the first 10,000 brain imaging datasets from UK Biobank. NeuroImage 166, 400–424 (2018)

  18. [26]

    & Ashburner, J

    Brudfors, M., Balbastre, Y., Nachev, P. & Ashburner, J. MRI Super-Resolution Using Multi-channel Total Variation. in Medical Image Understanding and Analysis (eds. Nixon, M., Mahmoodi, S. & Zwiggelaar, R.) 217–228 (Springer International Publishing, Cham, 2018). doi:10.1007/97...

  19. [27]

    & Abdallah, F

    El Jurdi, R., Petitjean, C., Honeine, P. & Abdallah, F. CoordConv-Unet: Investigating CoordConv for Organ Segmentation. IRBM 42, 415–423 (2021)

  20. [28]

    Cardoso, M. J. et al. MONAI: An open-source framework for deep learning in healthcare. Preprint at https://doi.org/10.48550/arXiv.2211.02701 (2022)

  21. [29]

    Kerfoot, E. et al. Left-Ventricle Quantification Using Residual U-Net. in Statistical Atlases and Computational Models of the Heart. Atrial Segmentation and LV Quantification Challenges (eds. Pop, M. et al.) 371–380 (Springer International Publishing, Cham, 2019). doi:10.1007/...

  22. [30]

    Hatamizadeh, A. et al. UNETR: Transformers for 3D Medical Image Segmentation. Preprint at https://doi.org/10.48550/arXiv.2103.10504 (2021)

  23. [31]

    Hatamizadeh, A. et al. Swin UNETR: Swin Transformers for Semantic Segmentation of Brain Tumors in MRI Images. Preprint at https://doi.org/10.48550/arXiv.2201.01266 (2022)

  24. [32]

    Medical image data augmentation: techniques, comparisons and interpretations

    Goceri, E. Medical image data augmentation: techniques, comparisons and interpretations. Artif. Intell. Rev. 56, 12561–12605 (2023)

  25. [33]

    Perrone, D. et al. The effect of Gibbs ringing artifacts on measures derived from diffusion MRI. NeuroImage 120, 441–455 (2015)

  26. [34]

    Wiest-Daesslé, N., Prima, S., Coupé, P., Morrissey, S. P. & Barillot, C. Rician Noise Removal by Non-Local Means Filtering for Low Signal-to-Noise Ratio MRI: Applications to DT-MRI. in Medical Image Computing and Computer-Assisted Intervention – MICCAI 2008 (eds. Metaxas, D., ...

  27. [35]

    Graves, M. J. & Mitchell, D. G. Body MRI artifacts in clinical practice: A physicist’s and radiologist’s perspective. J. Magn. Reson. Imaging 38, 269–287 (2013)

  28. [36]

    H., Cardoso, M

    Sudre, C. H., Cardoso, M. J. & Ourselin, S. Longitudinal segmentation of age-related white matter hyperintensities. Med. Image Anal. 38, 50–64 (2017)

  29. [37]

    & Ahmadi, S.-A

    Milletari, F., Navab, N. & Ahmadi, S.-A. V-Net: Fully Convolutional Neural Networks for Volumetric Medical Image Segmentation. in 2016 Fourth International Conference on 3D Vision (3DV) 565–571 (2016). doi:10.1109/3DV.2016.79

  30. [38]

    & Dollár, P

    Lin, T.-Y., Goyal, P., Girshick, R., He, K. & Dollár, P. Focal Loss for Dense Object Detection. Preprint at https://doi.org/10.48550/arXiv.1708.02002 (2018)

  31. [39]

    Set Theory

    Hausdorff, F. Set Theory. (Chelsea Publishing, New York, 1962)

  32. [40]

    Gardner, S. J. et al. Contouring variability of human- and deformable-generated contours in radiotherapy for prostate cancer. Phys. Med. Biol. 60, 4429 (2015)

  33. [41]

    Kruskal, W. H. & Wallis, W. A. Use of Ranks in One-Criterion Variance Analysis. J. Am. Stat. Assoc. 47, 583–621 (1952)

  34. [42]

    & Nachev, P

    Thiebaut de Schotten, M., Foulon, C. & Nachev, P. Brain disconnections link structural connectivity with function and behaviour. Nat. Commun. 11, 5094 (2020)

  35. [43]

    & Nachev, P

    Xu, T., Jha, A. & Nachev, P. The dimensionalities of lesion-deficit mapping. Neuropsychologia 115, 134–141 (2018)

  36. [44]

    Giles, D. et al. InterSynth: A Semi-Synthetic Framework for Benchmarking Prescriptive Inference from Observational Data. in Machine Learning for Multimodal Healthcare Data (eds. Maier, A. K., Schnabel, J. A., Tiwari, P. & Stegle, O.) 172–188 (Springer Nature Switzerland, Cham,...

  37. [45]

    & Melville, J

    McInnes, L., Healy, J. & Melville, J. UMAP: Uniform Manifold Approximation and Projection for Dimension Reduction. Preprint at https://doi.org/10.48550/arXiv.1802.03426 (2020)

  38. [46]

    Carruthers, R. et al. Representational ethical model calibration. Npj Digit. Med. 5, 1–9 (2022)

  39. [47]

    & Zadeh, S

    Abbasi, H., Orouskhani, M., Asgari, S. & Zadeh, S. S. Automatic brain ischemic stroke segmentation with deep learning: A review. Neurosci. Inform. 3, 100145 (2023)

  40. [48]

    Liew, S.-L. et al. A large, curated, open-source stroke neuroimaging dataset to improve lesion segmentation algorithms. Sci. Data 9, 320 (2022)

  41. [49]

    Friston, K. J. & Penny, W. Posterior probability maps and SPMs. NeuroImage 19, 1240– 1249 (2003)

  42. [50]

    Engleitner, H. et al. GeoSPM: Geostatistical parametric mapping for medicine. Patterns 3, (2022)

  43. [51]

    Muckley, M. J. et al. Training a neural network for Gibbs and noise removal in diffusion MRI. Magn. Reson. Med. 85, 413–428 (2021)

  44. [52]

    D., Maule, F

    Papinutto, N. D., Maule, F. & Jovicich, J. Reproducibility and biases in high field brain diffusion MRI: An evaluation of acquisition and analysis variables. Magn. Reson. Imaging 31, 827–839 (2013)

  45. [53]

    Wichmann, F. A. & Hill, N. J. The psychometric function: I. Fitting, sampling, and goodness of fit. Percept. Psychophys. 63, 1293–1313 (2001)

  46. [54]

    D., Martinetz, T

    Maier, O., Schröder, C., Forkert, N. D., Martinetz, T. & Handels, H. Classifiers for Ischemic Stroke Lesion Segmentation: A Comparison Study. PLOS ONE 10, e0145118 (2015)

  47. [55]

    Hernandez Petzsche, M. R. et al. ISLES 2022: A multi-center magnetic resonance imaging stroke lesion segmentation dataset. Sci. Data 9, 762 (2022)

  48. [56]

    R., Kennard, C

    Nachev, P., Coulthard, E., Jäger, H. R., Kennard, C. & Husain, M. Enantiomorphic normalization of focally lesioned brains. NeuroImage 39, 1215–1226 (2008)

  49. [57]

    & Friston, K

    Ashburner, J. & Friston, K. J. Nonlinear spatial normalization using basis functions. Hum. Brain Mapp. 7, 254–266 (1999)

  50. [58]

    Crinion, J. et al. Spatial normalization of lesioned brains: Performance evaluation and impact on fMRI analyses. NeuroImage 37, 866–875 (2007)

  51. [59]

    & Friston, K

    Ashburner, J. & Friston, K. J. Unified segmentation. NeuroImage 26, 839–851 (2005). Fig. 1: Dataset construction and curation process. This flowchart summarises the curation and label correction steps from the raw datasets to the final lesion and control sets. Lesion images fr...

Pith tools

Reviewed August 8, 2026 · model on record in the stance chip above.