Clinically-Informed Preprocessing Improves Stroke Segmentation in Low-Resource Settings

Caitlin M. Neher; Hitender Oswal; Juampablo E. Heras Rivera; Mehmet Kurt; Tianyi Ren; William Henry; Yutong Pan

REVIEW 4 major objections 4 minor 29 references

Clinically-Informed Preprocessing Improves Stroke Segmentation in Low-Resource Settings

T0 review · 4 major / 4 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read This paper reports that a clinically informed preprocessing pipeline, including vessel segmentations extracted from CTA, improved CT-based ischemic stroke lesion segmentation by 38% in Dice over baseline preprocessing, with a further 21% ga

desk verdict The paper has a plausible, useful idea and a suggestive but under-reported result; the 38% figure should not be taken at face value without seeing variance and fold details. read the letter →

arxiv 2508.16004 v1 pith:NVKDG3NO submitted 2025-08-21 eess.IV cs.CV

Juampablo E. Heras Rivera , Hitender Oswal , Tianyi Ren , Yutong Pan , William Henry , Caitlin M. Neher , Mehmet Kurt This is my paper · ORCID

classification eess.IVcs.CV

keywords ischemicstrokesegmentationCTimagingDWInnU-Netclinically-informedpreprocessingCTAvessellow-resourcesettings

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

Stroke is usually diagnosed with DWI, but DWI is often unavailable in low-resource settings; hospitals there rely on CT. The paper tries to show that CT-alone stroke segmentation can get much closer to DWI quality if the preprocessing is chosen the way a radiologist reads the scan, rather than using generic intensity normalization. In 10-fold cross-validation, an nnU-Net (a self-configuring deep segmentation model) trained with this clinically-informed pipeline improved Dice by 38% over the same model with baseline preprocessing. Adding vessel segmentations extracted from CTA improved the best model by another 21% over 5 folds. If true, low-resource centers could use arrival CT—with or without CTA—to produce follow-up lesion maps that are currently only practical with MRI.

What carries the argument

The key mechanism is the clinically-informed preprocessing pipeline: a sequence of input transformations applied to admission CT and CTA before they enter the nnU-Net, chosen from clinical reading practice rather than generic intensity normalization. The pipeline carries the argument: the experimental claim is that with this preprocessing, the same nnU-Net jumps 38% in Dice over baseline preprocessing, and extracting vessel segmentations from CTA maps contributes another 21%.

What would settle it

Run the same 10-fold comparison but change only the ground truth: replace follow-up DWI annotations with same-day DWI or expert CT lesion contours. If the 38% improvement from clinically-informed preprocessing shrinks or disappears, the reported gain is specific to predicting delayed lesion evolution rather than to preprocessing generalizing.

Watch

Extended reading notes

Core claim

The paper's central claim is that clinically-informed preprocessing is largely responsible for the accuracy of CT-based ischemic stroke lesion segmentation, not the model architecture alone. In 10-fold cross-validation, replacing baseline preprocessing with the proposed pipeline improved Dice by 38%; adding vessel segmentations extracted from CTA improved the best model by a further 21% over 5 folds. The training target is the lesion volume annotated on follow-up DWI taken 2–9 days after the admission CT, so the model is learning to anticipate delayed infarct burden from early CT.

Load-bearing premise

The load-bearing premise is that lesions annotated on follow-up DWI taken 2–9 days after admission are a valid ground truth for what a model should predict from the admission CT; if the delayed DWI labels are noisy or the admission CT carries too little lesion signal, the reported Dice gains rest on a target that does not reflect the acute scan.

Editorial extensions

If this is right

CT-only stroke segmentation can approach DWI-level lesion maps if preprocessing is treated as part of the model design.
Routine CTA acquisitions can be leveraged by extracting vessel segmentations as inputs, yielding an additional 21% Dice improvement.
The 38% gain over baseline preprocessing suggests architecture choice is not the only driver; input representation matters as much.
Models trained on admission CT to predict follow-up DWI lesions could support early estimates of final infarct size in hospitals without MRI.

Reading between the lines

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

A natural next test is to apply the same clinically-informed preprocessing to other segmentation backbones; if the gain is backbone-independent, the effect likely lives in the input representation.
Because the target is delayed DWI, the model may be learning both lesion detection and infarct growth; separating the two would require same-day DWI labels, which this study does not report.
The CTA vessel-segmentation gain hints that vascular anatomy is a strong spatial prior for downstream infarction; injecting vessel maps as an attention or auxiliary channel could further improve accuracy.
Standardizing this preprocessing could let low-resource radiology departments obtain DWI-like lesion burden estimates from CT without changing hardware or adding MRI.

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

I’ve only got the abstract in front of me, so this is a preliminary read. The core idea is appealing: use CT on admission to predict the DWI lesion seen days later, and shape the preprocessing with clinical knowledge rather than just throwing the raw scans at nnU-Net. That’s a genuinely practical question for low-resource settings, and the reported 38% Dice gain over baseline is the kind of result that would matter if it holds. Credit where it’s due: the task is well-motivated, the baseline is a strong and sensible one (nnU-Net), and the cross-validation scheme at least acknowledges generalizability.

The soft spots are real, though. The headline number is an aggregate over 10 folds with no standard deviation, no per-fold breakdown, and no paired statistical test. In nnU-Net segmentation, fold-to-fold variance is often large; a 38% mean gain could be driven by a few easy folds. The additional 21% is from 5 folds, not the same 10, so it’s not directly comparable to the baseline number. The abstract doesn’t say whether the same hyperparameters, training epochs, or model selection were used for the pipelines, nor whether the preprocessing steps were frozen before looking at validation performance—so some selection on the validation data can’t be ruled out. The CTA vessel segmentations also raise a leakage question: if those vessels are derived from the same scans that inform the outcome, part of the gain could be an artifact. And the ground truth itself, DWI annotations taken 2–9 days after the CT, is likely noisy; the abstract doesn’t say whether it was independently verified. None of this proves the claim is wrong, but it means the current evidence is not yet convincing.

If the full paper addresses these points—per-fold results, significance tests, preprocessing choices made a priori, and annotator agreement—this could be a solid contribution. The stress-test note’s concerns are fair and should be taken as requests for details, not as a dismissal.

My recommendation: let it go to peer review. The question is important enough, and the initial result is promising enough, that referees should see the full methodology. I wouldn't cite the 38% figure in my own work until I’ve seen the variance and the leakage checks, but I’d happily read a revised version.

Referee Report

4 major / 4 minor

Summary. The paper proposes a set of clinically motivated preprocessing steps for CT-based ischemic stroke lesion segmentation, aiming to improve performance in low-resource settings where MRI/DWI is unavailable. The authors train nnU-Net models on admission CT to predict follow-up DWI-derived lesion volumes and report a 38% mean Dice improvement over 10 folds relative to baseline preprocessing, and a further 21% improvement over 5 folds when CTA vessel segmentations are added. The evaluation is performed on held-out folds, which is appropriate for avoiding circularity in the core comparison.

Significance. If the reported gains hold under rigorous statistical evaluation, the work addresses a practically important problem: automated ischemic stroke lesion segmentation from CT alone, with preprocessing informed by clinical knowledge rather than brute-force augmentation. The use of follow-up DWI as ground truth is clinically meaningful, and the low-resource framing is timely. The main contributions—delineating a preprocessing pipeline and demonstrating its benefit—are potentially useful, but the evidence presented in the abstract is not yet sufficient to establish the magnitude or reliability of the improvement. I credit the authors for evaluating on held-out folds, but the absence of variance estimates, paired comparisons, and explicit method details leaves the headline claim under-supported.

major comments (4)

[Abstract] The headline '38% improvement in Dice score over 10 folds' is reported without absolute Dice values, per-fold results, error bars, or any statistical test. In nnU-Net-based segmentation, fold-to-fold variance is often substantial; a mean improvement of this size could be driven by one or two favorable folds, especially if the fold splits differ between baseline and proposed pipelines. The authors must report the per-fold Dice distributions for both methods and a paired test (e.g., Wilcoxon signed-rank) across the same folds.
[Abstract] The two reported gains are not directly comparable: the 38% improvement is over 10 folds, while the additional 21% improvement is over 5 folds. It is unclear whether the 5-fold subset is the same patient cohort as the 10-fold set, whether the baseline for the second comparison is the same, and whether the improvement is measured on the same test folds. The authors should report both gains on the same set of held-out folds/patients, or explicitly justify why the different fold counts do not affect comparability.
[Abstract / Methods (not provided in the accessible text)] The abstract does not specify whether the same hyperparameters, training epochs, model selection criterion, and data splits were used for baseline and proposed pipelines. More importantly, if the 'clinically motivated preprocessing' includes masks or vessel segmentations derived from the same CTA scans that define the outcome, part of the gain could reflect label leakage. The paper must describe the preprocessing steps in detail, state the provenance of all derived masks, and clarify whether any annotation-derived information is used only at training time or also at inference.
[Abstract] The ground truth is described as 'lesion volumes annotated from DWI taken 2-9 days later.' No information is given about the number of annotators, inter-observer variability, or independent verification of the DWI annotations. If the annotations are noisy, a preprocessing method that implicitly regularizes toward the annotator's style could inflate Dice scores without improving true clinical accuracy. Reporting annotation reliability would strengthen the validity of the ground truth and the interpretation of the reported gains.

minor comments (4)

[Abstract] Please report absolute Dice values alongside relative improvements. A 38% improvement from Dice 0.10 to 0.14 has very different clinical significance than from 0.50 to 0.69.
[Abstract] The phrase 'clinically informed preprocessing' is vague. State briefly in the abstract what the preprocessing steps are (e.g., skull stripping, intensity standardization, vessel segmentation), so readers can assess whether the gains are plausibly mechanism-based.
[Abstract] The sentence 'we further improve our best model by 21% over 5 folds' should clarify whether this is a relative or absolute Dice increase, and whether the 5 folds are a subset of the 10 folds.
[Abstract] Consider reporting the Dice score on the baseline preprocessing in addition to the relative improvement; this would make the contribution easier to benchmark against existing literature.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity identified from the provided text; the claim is an empirical benchmark comparison against a baseline on held-out folds.

full rationale

The provided manuscript text contains only the abstract. The central claim is that a clinically motivated preprocessing pipeline improves Dice score for ischemic stroke lesion segmentation over a nnU-Net baseline (38% over 10 folds, +21% over 5 folds). This is an empirical performance claim evaluated on held-out folds, not a derivation that reduces to its inputs. No equation, fitted parameter, or self-citation is available in the text to exhibit a specific circular reduction. The preprocessing is described as 'clinically motivated,' which, if taken at face value, means it was not selected on the validation folds, though the abstract alone does not provide the details needed to confirm that. Without access to the methods, there is no basis for asserting that the reported gains are forced by construction. Concerns about differing fold counts, lack of paired statistics, and possible label leakage are correctness risks, not circularity. Per the hard rules, speculation about intent or hidden selection does not constitute circularity. Therefore score 0.

Assumptions & free parameters 2 free parameters · 3 assumptions · 0 invented entities

Only the abstract is available. No new physical or conceptual entities are introduced. The central claim depends on the clinical validity of CT-to-DWI mapping and on preprocessing not being fitted to the test set.

free parameters (2)

preprocessing hyperparameters = not reported
The abstract does not specify the preprocessing steps or their parameters; they are likely tuned, which could inflate the reported improvement.
nnU-Net training hyperparameters = not reported
The model configuration and training details are not provided; these affect the baseline and improved performance.

assumptions (3)

domain assumption DWI-annotated lesion volumes at 2-9 days are a valid ground truth for final infarct
The paper uses follow-up DWI to define the outcome, assuming it represents the true lesion.
domain assumption Admission CT images contain signal predictive of follow-up DWI lesions
The entire method relies on CT-to-DWI prediction being learnable.
domain assumption The reported Dice improvements are computed with fair cross-validation and no data leakage
Abstract reports 10-fold and 5-fold results but gives no details on splits or leakage prevention.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Clinically-Informed Preprocessing Improves Stroke Segmentation in Low-Resource Settings." pith.science (2026). https://pith.science/paper/NVKDG3NO

@misc{pith2026250816004,
  author       = {Pith},
  title        = {Pith review of: Clinically-Informed Preprocessing Improves Stroke Segmentation in Low-Resource Settings},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/NVKDG3NO}},
  note         = {Machine review of arXiv:2508.16004}
}

read the original abstract

Stroke is among the top three causes of death worldwide, and accurate identification of ischemic stroke lesion boundaries from imaging is critical for diagnosis and treatment. The main imaging modalities used include magnetic resonance imaging (MRI), particularly diffusion weighted imaging (DWI), and computed tomography (CT)-based techniques such as non-contrast CT (NCCT), contrast-enhanced CT angiography (CTA), and CT perfusion (CTP). DWI is the gold standard for the identification of lesions but has limited applicability in low-resource settings due to prohibitive costs. CT-based imaging is currently the most practical imaging method in low-resource settings due to low costs and simplified logistics, but lacks the high specificity of MRI-based methods in monitoring ischemic insults. Supervised deep learning methods are the leading solution for automated ischemic stroke lesion segmentation and provide an opportunity to improve diagnostic quality in low-resource settings by incorporating insights from DWI when segmenting from CT. Here, we develop a series of models which use CT images taken upon arrival as inputs to predict follow-up lesion volumes annotated from DWI taken 2-9 days later. Furthermore, we implement clinically motivated preprocessing steps and show that the proposed pipeline results in a 38% improvement in Dice score over 10 folds compared to a nnU-Net model trained with the baseline preprocessing. Finally, we demonstrate that through additional preprocessing of CTA maps to extract vessel segmentations, we further improve our best model by 21% over 5 folds.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

29 extracted references · 24 canonical work pages

[1]

Stroke 54(2), 558--566 (2023)

Alzahrani, A., Zhang, X., Albukhari, A., Wardlaw, J.M., Mair, G.: Assessing brain tissue viability on nonenhanced computed tomography after ischemic stroke. Stroke 54(2), 558--566 (2023)

work page 2023
[2]

Stroke 37(5), 1334--1339 (2006)

Bandera, E., Botteri, M., Minelli, C., et al.: Cerebral blood flow threshold of ischemic penumbra and infarct core in acute ischemic stroke. Stroke 37(5), 1334--1339 (2006). doi:10.1161/01.STR.0000217418.29609.22

work page arXiv 2006
[3]

Cochrane Database Syst Rev 4, CD007424 (2009)

Brazzelli, M., Sandercock, P., Chappell, F., Celani, M., Righetti, E., Arestis, N., et al.: Magnetic resonance imaging versus computed tomography for detection of acute vascular lesions in patients presenting with stroke symptoms. Cochrane Database Syst Rev 4, CD007424 (2009)

work page 2009
[4]

Lancet 369, 293--298 (2007)

Chalela, J., Kidwell, C., Nentwich, L., Luby, M., Butman, J., Demchuk, A., et al.: Magnetic resonance imaging and computed tomography in emergency assessment of patients with suspected acute stroke: a prospective comparison. Lancet 369, 293--298 (2007)

work page 2007
[5]

Neurology 97(20\_Supplement\_2), S42--S51 (2021)

Czap, A.L., Sheth, S.A.: Overview of imaging modalities in stroke. Neurology 97(20\_Supplement\_2), S42--S51 (2021)

work page 2021
[6]

Stroke 33, 2206--2210 (2002)

Fiebach, J., Schellinger, P., Jansen, O., Meyer, M., Wilde, P., Bender, J., et al.: CT and diffusion-weighted MR imaging in randomized order: diffusion-weighted imaging results in higher accuracy and lower interrater variability in the diagnosis of hyperacute ischemic stroke. Stroke 33, 2206--2210 (2002)

work page 2002
[7]

Radiology 210, 155--162 (1999)

Gonz \'a lez, R., Schaefer, P., Buonanno, F., Schwamm, L., Budzik, R., Rordorf, G., et al.: Diffusion-weighted MR imaging: diagnostic accuracy in patients imaged within 6 hours of stroke symptom onset. Radiology 210, 155--162 (1999)

work page 1999
[8]

Hasford, F., Mumuni, A.N., Trauernicht, C., et al.: A review of MRI studies in A frica with special focus on quantitative MRI : Historical development, current status and the role of medical physicists. Phys. Med. 103, 46--58 (2022)

work page 2022

Show all 29 references

[9]

In: International MICCAI Brainlesion Workshop

Hatamizadeh, A., Nath, V., Tang, Y., Yang, D., Roth, H.R., Xu, D.: Swin UNETR : Swin transformers for semantic segmentation of brain tumors in MRI images. In: International MICCAI Brainlesion Workshop. pp. 272--284. Springer (2021)

2021
[10]

arXiv preprint arXiv:2411.17617 (2024)

Heras Rivera, J.E., Chopra, A.S., Ren, T., Oswal, H., Pan, Y., Sordo, Z., Walters, S., Henry, W., Mohammadi, H., Olson, R., et al.: An ensemble approach for brain tumor segmentation and synthesis. arXiv preprint arXiv:2411.17617 (2024)

2024 arXiv
[11]

Scientific data 9(1), 762 (2022)

Hernandez Petzsche, M.R., de la Rosa, E., Hanning, U., Wiest, R., Valenzuela, W., Reyes, M., Meyer, M., Liew, S.L., Kofler, F., Ezhov, I., et al.: ISLES 2022: A multi-center magnetic resonance imaging stroke lesion segmentation dataset. Scientific data 9(1), 762 (2022)

2022
[12]

NeuroImage 260, 119474 (Oct 2022)

Hoopes, A., Mora, J.S., Dalca, A.V., Fischl, B., Hoffmann, M.: Synthstrip: skull-stripping for any brain image. NeuroImage 260, 119474 (Oct 2022). doi:10.1016/j.neuroimage.2022.119474, http://dx.doi.org/10.1016/j.neuroimage.2022.119474

2022
[13]

Nature Methods 18(2), 203--211 (2021)

Isensee, F., Jaeger, P.F., Kohl, S.A.A., Petersen, J., Maier-Hein, K.H.: nnu-net: a self-configuring method for deep learning-based biomedical image segmentation. Nature Methods 18(2), 203--211 (2021). doi:10.1038/s41592-020-01008-z

2021 doi
[14]

In: Proc

Isensee, F., Wald, T., Ulrich, C., et al.: nnU-Net revisited: a call for rigorous validation in 3D medical image segmentation. In: Proc. MICCAI. pp. 488--498 (2024)

2024
[15]

a hnel, S., D \

Knauth, M., von Kummer, R., Jansen, O., H \"a hnel, S., D \"o rfler, A., Sartor, K.: Potential of CT angiography in acute ischemic stroke. American Journal of Neuroradiology 18(6), 1001--1010 (1997)

1997
[16]

Medicina 59(11), 1908 (2023)

Nukovic, J.J., Opancina, V., Ciceri, E., Muto, M., Zdravkovic, N., Altin, A., Altaysoy, P., Kastelic, R., Velazquez Mendivil, D.M., Nukovic, J.A., et al.: Neuroimaging modalities used for ischemic stroke diagnosis and monitoring. Medicina 59(11), 1908 (2023)

1908
[17]

Stroke 40(2), 469--475 (2009)

Olivot, J., Mlynash, M., Thijs, V.N., et al.: Optimal Tmax threshold for predicting penumbral tissue in acute stroke. Stroke 40(2), 469--475 (2009). doi:10.1161/STROKEAHA.108.526954

2009 doi
[18]

Stroke 30, 502--507 (1999)

Pantano, P., Caramia, F., Bozzao, L., Dieler, C., von Kummer, R.: Delayed increase in infarct volume after cerebral ischemia: correlations with thrombolytic treatment and clinical outcome. Stroke 30, 502--507 (1999)

1999
[19]

Radiology 262(2), 593--604 (2012)

Pulli, B., Schaefer, P.W., Hakimelahi, R., et al.: Acute ischemic stroke: infarct core estimation on CT angiography source images depends on CT angiography protocol. Radiology 262(2), 593--604 (2012)

2012
[20]

Ren, T., Heras Rivera , J.E., Oswal, H., Pan, Y., Henry, W., Walters, S., Kurt, M.: How we won the ISLES '24 challenge by preprocessing (2025), https://arxiv.org/abs/2505.18424

2025 arXiv
[21]

In: Medical Imaging with Deep Learning (2024)

Ren, T., Sharma, A., Heras Rivera, J.E., Rebala, L.H., Honey, E., Chopra, A., Kurt, M.: Re-diffinet: Modeling discrepancy in tumor segmentation using diffusion models. In: Medical Imaging with Deep Learning (2024)

2024
[22]

arXiv:1505.04597 (2015)

Ronneberger, O., Fischer, P., Brox, T.: U-Net : Convolutional networks for biomedical image segmentation. arXiv:1505.04597 (2015)

2015 arXiv
[23]

arXiv:2408.10966 (2024)

de la Rosa, E., Su, R., Reyes, M., et al.: ISLES '24: Improving final infarct prediction in ischemic stroke using multimodal imaging and clinical data. arXiv:2408.10966 (2024)

2024 arXiv
[24]

Neurosurgical focus 36(1), E3 (2014)

Tong, E., Hou, Q., Fiebach, J.B., Wintermark, M.: The role of imaging in acute ischemic stroke. Neurosurgical focus 36(1), E3 (2014)

2014
[25]

Stroke 37(4), 979--985 (2006)

Wintermark, M., Flanders, A.E., Velthuis, B., et al.: Perfusion‐ CT assessment of infarct core and penumbra. Stroke 37(4), 979--985 (2006). doi:10.1161/01.STR.0000209238.61459.39

2006
[26]

https://www.who.int/data/gho/data/indicators/indicator-details/GHO/total-density-per-million-population-magnetic-resonance-imaging, accessed 2025-07-02

World Health Organization : Medical devices: magnetic resonance imaging units per million population, total density. https://www.who.int/data/gho/data/indicators/indicator-details/GHO/total-density-per-million-population-magnetic-resonance-imaging, accessed 2025-07-02

2025
[27]

Aging (Albany NY) 11(9), 2542--2544 (2019)

Yousufuddin, M., Young, N.: Aging and ischemic stroke. Aging (Albany NY) 11(9), 2542--2544 (2019). doi:10.18632/aging.101931

2019 doi
[28]

, " * write output.state after.block = add.period write

ENTRY address author booktitle chapter doi edition editor eid howpublished institution journal key month note number organization pages publisher school series title type url volume year label INTEGERS output.state before.all mid.sentence after.sentence after.block FUNCTION in...
[29]

write newline

" write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 gl...

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.

[1] [1]

Stroke 54(2), 558--566 (2023)

Alzahrani, A., Zhang, X., Albukhari, A., Wardlaw, J.M., Mair, G.: Assessing brain tissue viability on nonenhanced computed tomography after ischemic stroke. Stroke 54(2), 558--566 (2023)

work page 2023

[2] [2]

Stroke 37(5), 1334--1339 (2006)

Bandera, E., Botteri, M., Minelli, C., et al.: Cerebral blood flow threshold of ischemic penumbra and infarct core in acute ischemic stroke. Stroke 37(5), 1334--1339 (2006). doi:10.1161/01.STR.0000217418.29609.22

work page arXiv 2006

[3] [3]

Cochrane Database Syst Rev 4, CD007424 (2009)

Brazzelli, M., Sandercock, P., Chappell, F., Celani, M., Righetti, E., Arestis, N., et al.: Magnetic resonance imaging versus computed tomography for detection of acute vascular lesions in patients presenting with stroke symptoms. Cochrane Database Syst Rev 4, CD007424 (2009)

work page 2009

[4] [4]

Lancet 369, 293--298 (2007)

Chalela, J., Kidwell, C., Nentwich, L., Luby, M., Butman, J., Demchuk, A., et al.: Magnetic resonance imaging and computed tomography in emergency assessment of patients with suspected acute stroke: a prospective comparison. Lancet 369, 293--298 (2007)

work page 2007

[5] [5]

Neurology 97(20\_Supplement\_2), S42--S51 (2021)

Czap, A.L., Sheth, S.A.: Overview of imaging modalities in stroke. Neurology 97(20\_Supplement\_2), S42--S51 (2021)

work page 2021

[6] [6]

Stroke 33, 2206--2210 (2002)

Fiebach, J., Schellinger, P., Jansen, O., Meyer, M., Wilde, P., Bender, J., et al.: CT and diffusion-weighted MR imaging in randomized order: diffusion-weighted imaging results in higher accuracy and lower interrater variability in the diagnosis of hyperacute ischemic stroke. Stroke 33, 2206--2210 (2002)

work page 2002

[7] [7]

Radiology 210, 155--162 (1999)

Gonz \'a lez, R., Schaefer, P., Buonanno, F., Schwamm, L., Budzik, R., Rordorf, G., et al.: Diffusion-weighted MR imaging: diagnostic accuracy in patients imaged within 6 hours of stroke symptom onset. Radiology 210, 155--162 (1999)

work page 1999

[8] [8]

Hasford, F., Mumuni, A.N., Trauernicht, C., et al.: A review of MRI studies in A frica with special focus on quantitative MRI : Historical development, current status and the role of medical physicists. Phys. Med. 103, 46--58 (2022)

work page 2022