Ischemic Stroke Lesion Segmentation Challenge (ISLES) 2026

Accurate infarct segmentation in brain stroke is one of the critical requirements across the disease trajectory, from acute-stage treatment guidance (e.g., reperfusion eligibility) to sub-acute and chronic-stage evaluation of patient outcome, clinical follow-up, and optimized therapeutic/rehabilitation strategies. The Ischemic Stroke Segmentation Challenge (ISLES) has served as a foundational benchmark for the scientific community since 2015, systematically addressing infarct segmentation across diverse modalities (CT/MR), disease stages (hyper-acute to chronic), and multicenter cohorts. The established open ISLES datasets have attracted significant community interest, evidenced by over 6,000 downloads of our ISLES’22 [1] and ATLAS [2] datasets, respectively. While initial challenges focused on technical segmentation tasks (e.g., ISLES’2015 [3]), recent editions have increasingly emphasized clinical relevance and real-world transferability. For instance, ISLES’2022 algorithms were subsequently stress-tested on real-world cohorts and implemented into a standalone software for clinical utility [4]. Furthermore, our ISLES’24 challenge setting envisioned personalized patient phenotyping through detailed multimodal datasets, enabling algorithms to capture a comprehensive patient picture for final infarct prediction [5,6].

The ISLES'26 edition specifically aims to establish a robust benchmark for generalized stroke infarct segmentation on T1-weighted MRI scans, covering acute, sub-acute, and chronic disease stages. Participants will develop methodologies on the largest-ever annotated stroke infarct dataset, comprising approximately 2,000 scans sourced from over 60 international centers. This initiative directly addresses generalization gaps identified in ISLES’22-ATLAS (a prior attempt to this task we organized), by providing a 40% increase in data diversity and a greater than 2-fold expansion in training data. Furthermore, the expanded dataset also includes crucial clinical variables, enabling the organizing team to conduct post-MICCAI evaluations of algorithmic clinical utility through analyses of patient outcome measures.

The complexity of ISLES’26 requires algorithms to robustly manage the full clinical spectrum, including diverse lesion sizes, complex anatomical patterns, and inherent heterogeneity from multi-center, multi-scanner, un-harmonized MR scans. Our ultimate goal is to facilitate the direct transfer of successful algorithms into standalone software for practical clinical utility. For MICCAI’26, ISLES would be happy to collaborate with the SWITCH+ workshop, TopBrain2, and TopAneu to organize a joint, comprehensive full-day satellite event focusing on the advances and persistent challenges in neurovascular image analysis.