Ischemic Stroke Lesion Segmentation Challenge 2024

Accurate segmentation of ischemic lesions in stroke is essential both during acute stages to guide treatment decisions (e.g., determine a patient's eligibility for thrombectomy treatment), and during sub-acute and chronic stages for evaluating disease outcomes, clinical follow-up, and for defining optimal therapeutic and rehabilitation strategies to maximize critical windows for recovery. The Ischemic Stroke Lesion Segmentation (ISLES) Challenge (https://www.isles-challenge.org/), an initiative aimed at advancing stroke image analysis, involves neurointervention, radiology, and computer science professionals and researchers from multiple leading institutions in the field. The ISLES challenge has been hosted at five MICCAI conferences (2015, 2016, 2017, 2018, 2022). In its inaugural edition, ISLES'15 [1], participants were tasked with segmenting sub-acute ischemic stroke lesions from post-interventional MRI and acute perfusion lesions from pre-interventional MRI. Subsequent editions, ISLES'16 and ISLES'17 [2], focused on stroke outcome prediction, requiring the segmentation of follow-up stroke lesions from acute multimodal MR imaging and the estimation of patient outcome disability scores. In ISLES'18 [3], acute stroke segmentation was approached indirectly and in a cross-modality fashion, with teams predicting the core tissue delineated in concomitant MRI from acute perfusion CT series. The most recent challenge, ISLES'22, addressed acute, sub-acute, and chronic ischemic stroke segmentation in over 2000 MRI scans [4,5]. Over the years, ISLES events have garnered significant attention from the research community. There were 120 database downloads until the ISLES'15 challenge day with 14 participating teams, and the number of participating teams was roughly duplicated in the ISLES'18 edition. The datasets released in ISLES'22 have been downloaded over 2000 times each, serving as reference datasets in prominent clinical and image analysis research [6,7,8]. The ISLES challenge has played a pivotal role in stroke image analysis for over eight years, contributing to the development of open stroke imaging datasets and benchmarking state-of-the-art image processing algorithms. Based on the experience gained from these previous editions, ISLES'24 seeks to benchmark final infarct segmentation algorithms. Unlike ISLES'16-17 editions, which share common points with this year's one, ISLES'24 makes use of standard-of-care acute stroke CT imaging (including non-contrast CT, perfusion CT, and CT angiography) and sub-acute stroke MRI (follow-up DWI with delineated infarct labels), coupled with clinical and demographic tabular data, highlighting its clinical relevance. Furthermore, we will release four times more imaging data through this challenge than ISLES'17. ISLES'24 aims to identify prominent final infarct segmentation algorithms from pre-interventional data, thus providing outputs that could, from a clinical standpoint, help optimize reperfusion treatment decision-making. Overall, the diversity of ISLES'24 imaging and clinical data and the clinical relevance of the task will provide participants with a unique challenge.

References

1. Maier, Oskar, et al. "ISLES 2015-A public evaluation benchmark for ischemic stroke lesion segmentation from multispectral MRI." Medical image analysis 35 (2017): 250-269.
2. Winzeck, Stefan, et al. "ISLES 2016 and 2017-benchmarking ischemic stroke lesion outcome prediction based on multispectral MRI." Frontiers in neurology 9 (2018): 679.
3. Hakim, Arsany, et al. "Predicting infarct core from computed tomography perfusion in acute ischemia with machine learning: Lessons from the ISLES challenge." Stroke 52.7 (2021): 2328-2337.
4. Hernandez Petzsche, Moritz R., et al. "ISLES 2022: A multi-center magnetic resonance imaging stroke lesion segmentation dataset." Scientific data 9.1 (2022): 762.
5. Liew, Sook-Lei, et al. "A large, curated, open-source stroke neuroimaging dataset to improve lesion segmentation algorithms." Scientific data 9.1 (2022): 320.
6. Iglesias, Juan E., et al. "SynthSR: A public AI tool to turn heterogeneous clinical brain scans into high-resolution T1-weighted images for 3D morphometry." Science advances 9.5 (2023): eadd3607.
7. Liew, Sook-Lei, et al. "Association of Brain Age, Lesion Volume, and Functional Outcome in Patients With Stroke." Neurology 100.20 (2023): e2103-e2113.
8. Ashtari, Pooya, et al. "Factorizer: A scalable interpretable approach to context modeling for medical image segmentation." Medical image analysis 84 (2023): 102706.