DeepISLES: a clinically validated ischemic stroke segmentation model from the ISLES'22 challenge

E Ezequiel de la Rosa M Mauricio Reyes S Sook-Lei Liew A Alexandre Hutton R Roland Wiest J Johannes Kaesmacher U Uta Hanning A Arsany Hakim R Richard Zubal W Waldo Valenzuela D David Robben D Diana M. Sima V Vincenzo Anania A Arne Brys J James A. Meakin A Anne Mickan G Gabriel Broocks C Christian Heitkamp S Shengbo Gao K Kongming Liang Z Ziji Zhang M Md Mahfuzur Rahman Siddiquee A Andriy Myronenko P Pooya Ashtari S Sabine Van Huffel H Hyunsu Jeong C Chiho Yoon C Chulhong Kim J Jiayu Huo S Sebastien Ourselin R Rachel Sparks A Albert Clèrigues A Arnau Oliver X Xavier Lladó L Liam Chalcroft I Ioannis Pappas J Jeroen Bertels E Ewout Heylen J Juliette Moreau N Nima Hatami C Carole Frindel A Abdul Qayyum M Moona Mazher D Domenec Puig S Shao-Chieh Lin C Chun-Jung Juan T Tianxi Hu L Lyndon Boone M Maged Goubran Y Yi-Jui Liu S Susanne Wegener F Florian Kofler I Ivan Ezhov S Suprosanna Shit M Moritz R. Hernandez Petzsche M Michael Müller (Julius-Maximilians-Universität Würzburg, Institute of Inorganic Chemistry, Institute for Sustainable Chemistry & Catalysis with Boron (ICB), Am Hubland, 97074 Würzburg, Germany) B Bjoern Menze J Jan S. Kirschke B Benedikt Wiestler

Abstract

Abstract Diffusion-weighted MRI is critical for diagnosing and managing ischemic stroke, but variability in images and disease presentation limits the generalizability of AI algorithms. We present DeepISLES, a robust ensemble algorithm developed from top submissions to the 2022 Ischemic Stroke Lesion Segmentation challenge we organized. By combining the strengths of best-performing methods from leading research groups, DeepISLES achieves superior accuracy in detecting and segmenting ischemic lesions, generalizing well across diverse axes. Validation on a large external dataset (N = 1685) confirms its robustness, outperforming previous state-of-the-art models by 7.4% in Dice score and 12.6% in F1 score. It also excels at extracting clinical biomarkers and correlates strongly with clinical stroke scores, closely matching expert performance. Neuroradiologists prefer DeepISLES’ segmentations over manual annotations in a Turing-like test. Our work demonstrates DeepISLES’ clinical relevance and highlights the value of biomedical challenges in developing real-world, generalizable AI tools. DeepISLES is freely available at https://github.com/ezequieldlrosa/DeepIsles.

Article Details

Volume / Issue Vol. 16, Issue 1
Published August 09, 2025
ISSN 2041-1723
Publisher Nature Portfolio

Journal Info

Nature Communications

Nature Portfolio

ISSN: 2041-1723 Open Access Life Sciences

Authors (59)

E

Ezequiel de la Rosa

M

Mauricio Reyes

S

Sook-Lei Liew

A

Alexandre Hutton

R

Roland Wiest

J

Johannes Kaesmacher

U

Uta Hanning

A

Arsany Hakim

R

Richard Zubal

W

Waldo Valenzuela

D

David Robben

D

Diana M. Sima

V

Vincenzo Anania

A

Arne Brys

J

James A. Meakin

A

Anne Mickan

G

Gabriel Broocks

C

Christian Heitkamp

S

Shengbo Gao

K

Kongming Liang

Z

Ziji Zhang

M

Md Mahfuzur Rahman Siddiquee

A

Andriy Myronenko

P

Pooya Ashtari

S

Sabine Van Huffel

H

Hyunsu Jeong

C

Chiho Yoon

C

Chulhong Kim

J

Jiayu Huo

S

Sebastien Ourselin

R

Rachel Sparks

A

Albert Clèrigues

A

Arnau Oliver

X

Xavier Lladó

L

Liam Chalcroft

I

Ioannis Pappas

J

Jeroen Bertels

E

Ewout Heylen

J

Juliette Moreau

N

Nima Hatami

C

Carole Frindel

A

Abdul Qayyum

M

Moona Mazher

D

Domenec Puig

S

Shao-Chieh Lin

C

Chun-Jung Juan

T

Tianxi Hu

L

Lyndon Boone

M

Maged Goubran

Y

Yi-Jui Liu

S

Susanne Wegener

F

Florian Kofler

I

Ivan Ezhov

S

Suprosanna Shit

M

Moritz R. Hernandez Petzsche

M

Michael Müller

Julius-Maximilians-Universität Würzburg, Institute of Inorganic Chemistry, Institute for Sustainable Chemistry & Catalysis with Boron (ICB), Am Hubland, 97074 Würzburg, Germany

B

Bjoern Menze

J

Jan S. Kirschke

B

Benedikt Wiestler