Source-linked AI summary
ISLES 2022: A multi-center magnetic resonance imaging stroke lesion segmentation dataset
Moritz Roman Hernandez Petzsche, Ezequiel de la Rosa, Uta Hanning, Roland Wiest, Waldo Enrique Valenzuela Pinilla, Mauricio Reyes, Maria Ines Meyer, Sook-Lei Liew, Florian Kofler, Ivan Ezhov, David Robben, Alexander Hutton, Tassilo Friedrich, Teresa Zarth, Johannes Bürkle, The Anh Baran, Bjoern Menze, Gabriel Broocks, Lukas Meyer, Claus Zimmer, Tobias Boeckh-Behrens, Maria Berndt, Benno Ikenberg, Benedikt Wiestler, Jan S. Kirschke
TL;DR
Existing stroke-lesion segmentation benchmarks have been limited in scale and diversity, despite the clinical importance of MRI-based lesion delineation. ISLES 2022 introduces a multicenter, multi-vendor dataset of 400 expert-annotated cases to benchmark acute and subacute stroke segmentation, with 250 training and 150 test cases.
Problem
Prior ISLES benchmarks used smaller datasets, motivating broader evaluation of MRI stroke-lesion segmentation across varied infarct patterns and clinical stages.
Method
The paper constructs a multicenter, multi-vendor MRI dataset with expert-reviewed hybrid human-algorithm voxel-level lesion annotations and heterogeneous acquisition protocols.
Results
The dataset contains 400 cases, split into 250 training and 150 test cases, spanning acute to subacute lesions with varied size, quantity, location, and intervention status.
Takeaways & Limitations
ISLES 2022 provides the foundation for benchmarking segmentation robustness and generalization across centers, infarct patterns, and pre- versus post-intervention MRI.
Takeaways & Limitations
The dataset is based on retrospectively collected imaging data approved by the local ethics boards of participating centers.
Abstract
from arXiv · showhide
Magnetic resonance imaging (MRI) is a central modality for stroke imaging. It is used upon patient admission to make treatment decisions such as selecting patients for intravenous thrombolysis or endovascular therapy. MRI is later used in the duration of hospital stay to predict outcome by visualizing infarct core size and location. Furthermore, it may be used to characterize stroke etiology, e.g. differentiation between (cardio)-embolic and non-embolic stroke. Computer based automated medical image processing is increasingly finding its way into clinical routine. Previous iterations of the Ischemic Stroke Lesion Segmentation (ISLES) challenge have aided in the generation of identifying benchmark methods for acute and sub-acute ischemic stroke lesion segmentation. Here we introduce an expert-annotated, multicenter MRI dataset for segmentation of acute to subacute stroke lesions. This dataset comprises 400 multi-vendor MRI cases with high variability in stroke lesion size, quantity and location. It is split into a training dataset of n=250 and a test dataset of n=150. All training data will be made publicly available. The test dataset will be used for model validation only and will not be released to the public. This dataset serves as the foundation of the ISLES 2022 challenge with the goal of finding algorithmic methods to enable the development and benchmarking of robust and accurate segmentation algorithms for ischemic stroke.
1 Background & Summary
ISLES’22 builds on prior ISLES benchmarks to evaluate acute and sub-acute ischemic stroke lesion segmentation on MRI at greater scale. The challenge addresses clinically important embolic infarct patterns and a broad spectrum of lesion sizes and burdens.
- Clinical motivation: Stroke causes substantial global morbidity and mortality, with up to two thirds of survivors experiencing permanent disability.Endovascular reperfusion therapy and image-based CT and MRI guidance have improved outcomes for selected patients.
- Prior challenges: Earlier ISLES’15 and ISLES’18 challenges helped identify prominent methods and establish benchmark datasets for acute and sub-acute ischemic stroke segmentation.The first ISLES challenge in 2015 included 64 training and testing cases across two sub-challenges.
- ISLES’22 contribution: 400 cases form the basis of ISLES’22, which benchmarks acute and sub-acute ischemic stroke MRI segmentation through DWI infarct segmentation.The MICCAI 2022 edition targets both acute and sub-acute stroke.
- Clinical scope: ISLES’22 focuses clinically on acute embolic infarct patterns before intervention and typical post-interventional sub-acute infarct patterns.The challenge exposes participants to a wider ischemic stroke spectrum with variable lesion size and burden.
2 Methods
The ISLES 2022 methods used multicenter MRI data from adults with suspected or diagnosed stroke, emphasizing diverse lesion patterns and challenging cases. Images were standardized and annotated with a hybrid human-algorithm workflow, then divided into center- and stage-varied training and test sets.
- Cohort and imaging: Adults with suspected or diagnosed stroke underwent brain MRI including at least FLAIR and DWI with corresponding ADC maps.Acquisitions used multiple 3T Philips and Siemens MRI scanners.
- Cohort and imaging: The dataset deliberately included diverse vascular territories, many posterior-circulation ischemias, punctiform infarcts, and five suspected-stroke cases without ischemia.These patterns were selected because infratentorial and small multifocal lesions are difficult for human raters and algorithms to segment.
- Preprocessing and annotation: MRI data were anonymized and converted to NIfTI according to BIDS, with Center #1 DWI and ADC resliced to 2×2 mm2 axial isotropic voxels.Additional processing included FLAIR-to-DWI rigid registration and skull stripping using registered brain masks.
- Dataset split and evaluation: The dataset contained 250 training and 150 test subjects, with training data from centers #1 and #2 and test data distributed equally across three centers.The test set combined acute to early sub-acute post-treatment scans from centers #1 and #3 with hyper-acute to acute pre-treatment scans from center #2.
- Preprocessing and annotation: Voxel-level segmentation masks were produced through a hybrid human-machine algorithm with hierarchical manual checks and corrections.The workflow was applied to imaging data exported in NIfTI format, with masks also saved in NIfTI format.