Ischemic Stroke Lesion Segmentation Challenge 2022: Acute, sub-acute and chronic stroke infarct segmentation

Infarct segmentation in ischemic stroke is crucial at i) acute stages to guide treatment decision making (whether to
reperfuse or not, and type of treatment) and at ii) sub-acute and chronic stages to evaluate the patients' disease
outcome, for their clinical follow-up and to define optimal therapeutical and rehabilitation strategies to maximize
critical windows for recovery.

In ISLES'18 [1], our last organized event, acute stroke segmentation (i) was addressed in an indirect and crossmodality fashion, where teams predicted the MRI-delineated infarcts from a perfusion CT series. In addition, the segmentation of sub-acute ischemic stroke lesions (ii) was addressed in ISLES'15 (task 'SISS', [2]), where teams directly delineated infarcts in post-interventional MRI images. Both ISLES events received major attention from the research community: there were 120 database downloads until the ISLES15 challenge day with 14 participating teams, and the number of participating teams was roughly duplicated in our latest ISLES'18 edition. The ISLES'15 and ISLES'18 challenges played a crucial role in identifying prominent methods for acute and sub-acute ischemic stroke lesion segmentation. These datasets have since served as important benchmarks for the scientific community.

Based on the experience gained from these previous editions [3], ISLES'22 aims to benchmark acute, sub-acute and chronic ischemic stroke MRI segmentation using more than a thousand MRI scans. This MICCAI 2022 challenge edition is organized as a two-task event, namely: 1) DWI infarct segmentation in acute and sub-acute stroke and 2) Single channel T1-weighted lesion segmentation in acute, sub-acute and chronic stroke. ISLES'22 differs in several ways from the previous challenge editions in ischemic stroke by: 1) targeting the delineation of not only large infarct lesions, but also of multiple embolic and/or cortical infarcts (typically seen after mechanical recanalization), and 2) by evaluating both pre- and post- interventional MRI images and by including for the very first time chronic stroke images. In addition, the ISLES'22 challenge introduces a new challenge of segmenting T1-weighted MRIs across acute, sub-acute, and chronic stroke. Altogether, ISLES'22 provides a considerably larger dataset than before
(more than 20x scans than in the ISLES'15 edition).

From a clinical perspective, ISLES'22 focuses on the clinical growing interest of acute embolic infarct patterns, both pre-intervention (i.e. at a very early disease state) and post-intervention (typical, post-interventional sub-acute infarct patterns not restricted to a single vessel territory). This clinical problem also challenges the participants from a technical perspective: teams will deal with a wider ischemic stroke disease spectra, involving variable lesion size and burden, more complex infarct patterns and variable anatomically located lesions in data from multiple centers. In addition, for the first time, ISLES'22 will also examine lesions in chronic stroke, which are pathologically different from acute and sub-acute scans, have a different follow-up regimen than acute cases, and introduce additional complexity. The diversity of the ISLES'22 dataset will provide a unique challenge for participants.