Automated DWI-FLAIR mismatch assessment in stroke using DWI only

Joseph Benzakoun ^1,2,3, Lauranne Scheldeman ^3,4, Anke Wouters ^3,4

¹IMA-BRAIN, INSERM U1266, Institute of Psychiatry and Neuroscience of Paris (IPNP), Université Paris Cité, Paris, France.
²Neuroradiology Department, Hôpital Sainte Anne, GHU-Paris Psychiatrie et Neurosciences, Paris, France.
³Department of Neurosciences, Experimental Neurology, KU Leuven - University of Leuven, Leuven, Belgium.
⁴Department of Neurology, University Hospitals Leuven, Leuven, Belgium.
⁵Department of Neurology, University Medical Center Hamburg-Eppendorf, Hamburg, Germany.
⁶Department of Neurology, Medical Park Berlin Humboldtmühle, Berlin, Germany.
⁷Department of Neurology, Charité - Universitätsmedizin Berlin, Berlin, Germany.
⁸Center for Stroke Research Berlin, Charité - Universitätsmedizin Berlin, Berlin, Germany.
⁹Department of Diagnostic and Interventional Neuroradiology, University Medical Center Hamburg-Eppendorf, Hamburg, Germany.
¹⁰School of Cardiovascular & Metabolic Health, University of Glasgow, Glasgow, United Kingdom.
¹¹Hospices Civils de Lyon, Department of Neurology and Stroke Center, Lyon University, Lyon, France.
¹²Université Claude Bernard Lyon 1, Bron, France.
¹³Department of Radiology (CDI) and IDIBAPS, Hospital Clinic, Barcelona, Spain.
¹⁴Department of Neurology, Aarhus University Hospital, Aarhus, Denmark.
¹⁵Stroke Division, University of Melbourne, Florey Institute of Neuroscience and Mental Health, Melbourne, VIC, Australia.
¹⁶CIC, Innovation Technologique, Université de Lorraine, INSERM 1433, Nancy, France.
¹⁷Department of Neurology, Foch Hospital, Versailles Saint-Quentin-en-Yvelines University, Suresnes, France.
¹⁸Diagnostic and Interventional Neuroradiology Department, CHRU de Tours, Tours, Centre Val de Loire, France.
¹⁹CIC-IT 1415, INSERM 1253 iBrain, Tours, Centre Val de Loire, France.
²⁰Stroke Team, INSERM U1266, Institute of Psychiatry and Neuroscience of Paris (IPNP), Université Paris Cité, Paris, France.
²¹Neurology Department, Hôpital Sainte Anne, GHU-Paris Psychiatrie et Neurosciences, Paris, France.

⁰IMA-BRAIN, INSERM U1266, Institute of Psychiatry and Neuroscience of Paris (IPNP), Université Paris Cité, Paris, France.

European Stroke Journal +

|

August 28, 2025

View abstract on PubMed

Summary

A deep-learning model accurately predicts Diffusion-Weighted Imaging-Fluid-Attenuated Inversion-Recovery mismatch in Acute Ischemic Stroke patients using only DWI data. This tool aids in identifying candidates for thrombolysis when stroke onset is unknown.

Area Of Science

Radiology
Artificial Intelligence
Neurology

Background

Diffusion-Weighted Imaging (DWI) and Fluid-Attenuated Inversion-Recovery (FLAIR) mismatch is crucial for identifying Acute Ischemic Stroke (AIS) patients eligible for thrombolysis, especially when stroke onset is unknown.
Visual assessment of DWI-FLAIR mismatch has limitations due to suboptimal observer agreement, impacting treatment decisions for approximately 15% of AIS cases.

Purpose Of The Study

To develop and validate a Deep-Learning (DL) model capable of predicting DWI-FLAIR mismatch using only DWI data.
To improve the accuracy and consistency of DWI-FLAIR mismatch identification in AIS patients.

Main Methods

A retrospective study utilizing AIS patient data from the ETIS registry (derivation) and WAKE-UP trial (validation).
A DL model was trained to predict FLAIR Visible Areas (FVA) using only DWI input, defining an FVA-index.
Model performance was evaluated using Area Under the ROC Curve (AUC) and optimal FVA-index cutoff for predicting DWI-FLAIR mismatch.

Main Results

The DL model demonstrated strong predictive value in both derivation (AUC=0.85) and validation cohorts (AUC=0.86).
An optimal FVA-index cutoff of 0.5 achieved 70% sensitivity and 88% specificity in the validation cohort.
The model showed good agreement with visual ratings, indicated by a kappa of 0.54.

Conclusions

The developed DL model accurately predicts DWI-FLAIR mismatch in AIS patients with unknown stroke onset.
This AI tool can assist clinicians in challenging visual assessments or when FLAIR sequences are unavailable, potentially optimizing thrombolysis decisions.

Keywords:

Ischemic stroke artificial intelligence decision support techniques diffusion magnetic resonance imaging magnetic resonance imaging