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

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 .

Authors

Institutions

Publication Details

Journal
Nature Communications
Published
2025-08-09
DOI
https://doi.org/10.1038/s41467-025-62373-x
Citations
5
Primary Topic
Medical Image Segmentation Techniques
Type
article
Field-Weighted Citation Impact
6.38

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
article

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

Anne Mickan, Arsany Hakim, Bjoern Menze, Chun‐Jung Juan et al.
5 citations
Nature Communications
Medical Image Segmentation Techniques
6.38
article

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

Anne Mickan, Arsany Hakim, Bjoern Menze, Chun‐Jung Juan, Moona Mazher, Andriy Myronenko, Rachel Sparks, Jan S. Kirschke, Domènec Puig, Shengbo Gao, Arne Brys, Pooya Ashtari, Liam Chalcroft, Suprosanna Shit, Md Mahfuzur Rahman Siddiquee, Johannes Kaesmacher, Uta Hanning, Jiayu Huo, Christian Heitkamp, Yi‐Jui Liu, Florian Kofler, Chiho Yoon, Arnau Oliver, J. Moreau, James Meakin, Jeroen Bertels, David Robben, Sook‐Lei Liew, Ivan Ezhov, Benedikt Wiestler, Moritz Roman Hernández Petzsche, A. Hutton, Sabine Van Huffel, Tao Hu, Maged Goubran, Gabriel Broocks, Waldo Valenzuela, Roland Wiest, Albert Clèrigues, Mauricio Reyes, Lyndon Boone, Xavier Lladó, Vincenzo Anania, Richard Zubal, Carole Frindel, Kongming Liang, Ezequiel de la Rosa, Nima Hatami, Sébastien Ourselin, Abdul Qayyum, Diana M. Sima, Susanne Wegener, Ioannis Pappas, Chulhong Kim, Michael Müller, Shao-Chieh Lin, Ziji Zhang, Evelien Heylen, Hyunsu Jeong
article en
5 citations

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 .

Nature CommunicationsVol. 16(1)
Université Claude Bernard Lyon 1 (FR), University of Southern California (US), University of Bern (CH), Pohang University of Science and Technology (KR), Centre National de la Recherche Scientifique (FR), Sunnybrook Health Science Centre (CA), Beijing University of Posts and Telecommunications (CN), Radboud University Nijmegen (NL), Inserm (FR), Universität Hamburg (DE), Wellcome Centre for Human Neuroimaging (GB), King's College London (GB), China Medical University (TW), University of Toronto (CA), University of Zurich (CH), Universitat de Girona (ES), University Hospital of Bern (CH), TUM Klinikum (DE), Radboud University Medical Center (NL), Helmholtz Zentrum München (DE), Centre Hospitalier Universitaire de Tours (FR), Quantitative BioSciences (US), University Hospital of Zurich (CH), Clinical Investigation Center Plurithematic Tours (FR), University Medical Center Hamburg-Eppendorf (DE), China Medical University Hospital (TW), Nvidia (United States) (US), Centre de Recherche en Acquisition et Traitement de l'Image pour la Santé (FR), Centre d'Investigation Clinique - Innovation Technologique (FR), Icometrix (Belgium) (BE), Sunnybrook Research Institute, University College London (GB), Imperial College London (GB), Institut National des Sciences Appliquées de Lyon (FR), Feng Chia University (TW), Arizona State University (US), Universitat Rovira i Virgili (ES), Technical University of Munich (DE), KU Leuven (BE)
National Science Foundation, National Research Foundation, Schweizerischer Nationalfonds zur Förderung der Wissenschaftlichen Forschung, Pohang University of Science and Technology, Ministry of Trade, Industry and Energy, National Research Foundation of Korea, KU Leuven, Schweizerische Herzstiftung, Vlaamse regering, Helmut Horten Stiftung, Ministry of Science and ICT, South Korea, National Institutes of Health, Korea Evaluation Institute of Industrial Technology, Institute for Information and Communications Technology Promotion, National Institute of Neurological Disorders and Stroke
Industry, innovation and infrastructure
Openalex Percentile: Top 3%
Medical Image Segmentation Techniques
6.38
DOIOpen Access PDFOpenAlex
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.