The Ischemic Stroke Lesion Segmentation Challenge (ISLES)'24 Dataset: A Multimodal Stroke Imaging Dataset with...

Area of Science:

• Neurology
• Medical Imaging
• Artificial Intelligence

Background:

• Stroke is a significant global health issue, with outcomes improving due to advanced imaging and reperfusion therapies.
• Current imaging methods for estimating stroke lesion size have limitations, impacting treatment decisions.
• While AI shows promise in stroke lesion detection, clinical application needs large, well-annotated datasets, particularly those combining different imaging modalities.

Purpose of the Study:

• To address the limited availability of datasets pairing acute CT with follow-up MRI in stroke patients.
• To provide a publicly accessible dataset for analyzing infarct evolution.
• To support the development of AI models for postinterventional stroke care.

Main Methods:

• Compiled a dataset combining hyperacute CT scans (within 24 hours of onset) with acute postinterventional MRI scans (2-9 days post-reperfusion).
• Included follow-up clinical data for up to 3 months post-stroke.
• Ensured data compatibility for analyzing infarct progression and AI model training.

Main Results:

• A novel, publicly available dataset has been created.
• The dataset integrates multimodal imaging (CT and MRI) with clinical follow-up.
• Facilitates research into infarct evolution and AI development in stroke.

Conclusions:

• The new dataset bridges a critical gap in stroke research data.
• It enables detailed analysis of infarct changes over time.
• Supports advancements in AI-driven stroke management and patient outcomes.