Skip to main page content
U.S. flag

An official website of the United States government

Dot gov

The .gov means it’s official.
Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

Https

The site is secure.
The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
. 2025 Aug 28:23969873251362712.
doi: 10.1177/23969873251362712. Online ahead of print.

Automated DWI-FLAIR mismatch assessment in stroke using DWI only

Joseph Benzakoun  1   2   3 , Lauranne Scheldeman  3   4 , Anke Wouters  3   4 , Bastian Cheng  5 , Martin Ebinger  6 , Matthias Endres  7 , Jochen B Fiebach  8 , Jens Fiehler  9 , Ivana Galinovic  8 , Keith W Muir  10 , Norbert Nighoghossian  11   12 , Salvador Pedraza  13 , Josep Puig  13 , Claus Z Simonsen  14 , Vincent Thijs  15 , Götz Thomalla  5 , Emilien Micard  16 , Bailiang Chen  16 , Bertrand Lapergue  17 , Grégoire Boulouis  18   19 , Alice Le Berre  1   2 , Jean-Claude Baron  20   21 , Guillaume Turc  20   21 , Wagih Ben Hassen  1   2 , Olivier Naggara  1   2 , Catherine Oppenheim  1   2 , Robin Lemmens  3   4
Affiliations

Automated DWI-FLAIR mismatch assessment in stroke using DWI only

Joseph Benzakoun et al. Eur Stroke J. .

Abstract

Introduction: In Acute Ischemic Stroke (AIS), mismatch between Diffusion-Weighted Imaging (DWI) and Fluid-Attenuated Inversion-Recovery (FLAIR) helps identify patients who can benefit from thrombolysis when stroke onset time is unknown (15% of AIS). However, visual assessment has suboptimal observer agreement. Our study aims to develop and validate a Deep-Learning model for predicting DWI-FLAIR mismatch using solely DWI data.

Patients and methods: This retrospective study included AIS patients from ETIS registry (derivation cohort, 2018-2024) and WAKE-UP trial (validation cohort, 2012-2017). DWI-FLAIR mismatch was rated visually. We trained a model to predict manually-labeled FLAIR visible areas (FVA) matching the DWI lesion on baseline and early follow-up MRIs, using only DWI as input. FVA-index was defined as the volume of predicted regions. Area under the ROC curve (AUC) and optimal FVA-index cutoff to predict DWI-FLAIR mismatch in the derivation cohort were computed. Validation was performed using baseline MRIs of the validation cohort.

Results: The derivation cohort included 3605 MRIs in 2922 patients and the validation cohort 844 MRIs in 844 patients. FVA-index demonstrated strong predictive value for DWI-FLAIR mismatch in baseline MRIs from the derivation (n = 2453, AUC = 0.85, 95%CI: 0.84-0.87) and validation cohort (n = 844, AUC = 0.86, 95%CI: 0.84-0.89). With an optimal FVA-index cutoff at 0.5, we obtained a kappa of 0.54 (95%CI: 0.48-0.59), 70% sensitivity (378/537, 95%CI: 66-74%) and 88% specificity (269/307, 95%CI: 83-91%) in the validation cohort.

Discussion and conclusion: The model accurately predicts DWI-FLAIR mismatch in AIS patients with unknown stroke onset. It could aid readers when visual rating is challenging, or FLAIR unavailable.

Keywords: Ischemic stroke; artificial intelligence; decision support techniques; diffusion magnetic resonance imaging; magnetic resonance imaging.

PubMed Disclaimer

Conflict of interest statement

J.B.F reports consulting and advisory board fees from AbbVie, AC Immune, Alzheon, Artemida, BioClinica/Clario, Biogen, Bristol Myers Squibb, Brainomix, Cerevast, C2N Diagnostics, Daiichi-Sankyo, EISAI, Eli Lilly, F. Hoffmann-LaRoche AG, GlaxoSmithKline, Guerbet, Ionis Pharmaceuticals, IQVIA, Janssen, Julius Clinical, jung diagnostics, Lantheus Medical Imaging, Merck, Novo Nordisk, Octapharma AG, Premier Research, ProPharma Group, Prothena Biosciences, Regeneron Pharmaceuticals, Roche, Syneos, Tau Rx, Vertex Pharmaceuticals, and Worldwide Clinical Trials outside the submitted work.

G.Th reports personal consulting fees from Acandis, personal consulting fees from Astra Zeneca, personal consulting fees from Bayer, personal consulting fees from Boehringer Ingelheim, personal consulting fees from Stryker, personal payment or honoraria for lectures, presentations, speakers bureaus, manuscript writing or educational events from: Acandis, Alexion, Amarin, Bayer, Boehringer Ingelheim, BristolMyersSquibb/Pfizer, Daiichi Sankyo, Stryker, outside the submitted work.

C.Z.S is supported by a research grant from Health Research Foundation of Central Denmark Region. Also reports speaker fee from Pfizer.

V.T. reports consulting fees from Bayer, Boehringer Ingelheim, Medtronic.

C.G. declares, independent of the presented study, grants from Deutsche Forschungsgemeinschaft (DFG), Deutsches Zentrum f. Luft- und Raumfahrt (DLR), Hertie Foundation, Wegener Foundation, Schilling Foundation, Werner Otto Foundation, Merz Pharmaceuticals, Allergan, European Union; CG declares consulting fees from AlphaSights Ltd. and Life Science Praxis S.L., honoraria (for lectures, presentations) from AstraZeneca GmbH, Elements Communications Ltd., Boehringer Ingelheim, Streamedup GmbH, Abbott Medical, Bayer AG; CG declares participation in the DSMB of RESSTORE1, work as an editor of INFO Neurologie & Psychiatrie, Therapie und Verlauf neurologischer Erkrankungen (Textbook), and membership of the presidium of the German Neurological Society (DGN).

K.W.M. reports consultancy fees from Boehringer Ingelheim, Abbvie, Lumosa, Hyperfine; Lecture Fees – Boehringer Ingelheim, Brainomix, IschemaView; Trial Support – Boehringer Ingelheim (drug supply for ATTEST-2).

M.E. reports grants from Bayer and fees paid to the Charité from Amgen, AstraZeneca, Bayer Healthcare, Boehringer Ingelheim, BMS, Daiichi Sankyo, Sanofi, Pfizer, all outside the submitted work.

R.L. reports compensation from iSchemaView for other services.

All other authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Figures

Graphical abstract
Graphical abstract
Flowchart depicting inclusion and exclusion criteria in derivation and validation cohorts for ischemic stroke study.
Figure 1.
Flowchart for derivation and validation cohorts. “Extended DWI lesions” corresponded to lesions larger than 2/3 of middle cerebral artery territory (according to WAKE-UP trial inclusion criteria); “Non assessable FLAIR” included low-quality FLAIR. In the derivation cohort, in contrast to the validation cohort, the availability of sequences was checked before inclusion and all subjects had an acute ischemic stroke.
Comparison between manual and automated brain tumor detection methods in MRI scans
Figure 2.
Illustration of study design.
Here’s the alt text description based on the image and context data: ROC curves for DWI-FLAIR mismatch prediction based on FVA-index, with optimal cutoffs determined by Youden Index in derivation and validation cohorts, showing sensitivity and specificity for on-site DWI-FLAIR mismatch assessment in WAKE-UP trial.
Figure 3.
ROC curves for the prediction of DWI-FLAIR mismatch based on FVA-index. The optimal FVA-index cutoff was chosen using Youden Index on the day-0 MRI cohort (green dot in derivation cohort, and cyan dot in validation cohort). The sensitivity and specificity using the “on-site DWI-FLAIR mismatch assessment” in the WAKE-UP trial is shown as a purple dot.
The image shows three cases of brain scans with DWI, FLAIR, and prediction results. In the first case, a patient with a right middle cerebral artery infarct showed no DWI-FLAIR mismatch, matching the ground truth. The second case involved a patient with a left middle cerebral artery infarct, and the FVA-model predicted no DWI-FLAIR mismatch, aligning with the ground truth. Lastly, a patient with bilateral middle cerebral artery infarcts had no DWI-FLAIR mismatch, as predicted by the FVA-model.
Figure 4.
Examples of heatmaps generated by the FVA-model. Three illustrative cases. Left column shows b = 1000 s/mm2 DWI. Middle column shows the FLAIR sequence. Right column shows DWI with a transparent overlay of the FVA-map. In (a), a 67-year old woman presented a right middle cerebral artery infarct. FVA- predicted “DWI-FLAIR mismatch” (FVA-index = 0), matching the ground truth. In (b), a 76-year old woman presented a left middle cerebral artery infarct. The FVA-model predicted “no DWI-FLAIR mismatch” (FVA-index = 250), matching the ground truth. In (c), a 70-year old man presented with bilateral middle cerebral artery infarcts. The FVA-model predicted “no DWI-FLAIR mismatch” (FVA-index = 33), matching the ground truth. Heatmap showed “no DWI-FLAIR mismatch” on the right infarct, and “DWI-FLAIR mismatch” on the left infarct, matching the visual analysis of FLAIR, as the right infarct was clearly visible on FLAIR (arrow, middle column), whereas the left one was not.
See this image and copyright information in PMC

References

    1. Pu L, Wang L, Zhang R, et al. Projected global trends in ischemic stroke incidence, deaths and disability-adjusted life years from 2020 to 2030. Stroke 2023; 54: 1330–1339. - PubMed
    1. Saver JL. Time is brain-quantified. Stroke 2006; 37: 263–266. - PubMed
    1. Hacke W, Kaste M, Bluhmki E, et al. Thrombolysis with alteplase 3 to 4.5 hours after acute ischemic stroke. N Engl J Med 2008; 359: 1317–1329. - PubMed
    1. Mackey J, Kleindorfer D, Sucharew H, et al. Population-based study of wake-up strokes. Neurol 2011; 76: 1662–1667. - PMC - PubMed
    1. Thomalla G, Cheng B, Ebinger M, et al. DWI-FLAIR mismatch for the identification of patients with acute ischaemic stroke within 4·5 h of symptom onset (PRE-FLAIR): a multicentre observational study. Lancet Neurol 2011; 10: 978–986. - PubMed

LinkOut - more resources