ISLES 2015 - A public evaluation benchmark for ischemic stroke lesion segmentation from multispectral MRI

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neurology

Background:

• Ischemic stroke diagnosis and research heavily depend on non-invasive magnetic resonance imaging (MRI).
• Current stroke lesion segmentation algorithms lack comparability due to diverse datasets and evaluation methods.
• The Ischemic Stroke Lesion Segmentation (ISLES) challenge aimed to standardize evaluation.

Purpose of the Study:

• To establish a common evaluation framework for stroke lesion segmentation algorithms.
• To present results from the Sub-Acute Stroke Lesion Segmentation (SISS) and Stroke Perfusion Estimation (SPES) sub-challenges.
• To critically evaluate the state-of-the-art in stroke lesion segmentation and identify future research directions.

Main Methods:

• Organized the ISLES challenge in conjunction with MICCAI 2015.
• Developed a common evaluation framework and utilized publicly available datasets.
• Collected and analyzed results from 16 participating research groups using various automatic segmentation algorithms.

Main Results:

• Segmentation of acute perfusion lesions (SPES) was found to be feasible.
• Algorithms for sub-acute lesion segmentation (SISS) demonstrated insufficient accuracy.
• No specific algorithmic characteristic consistently outperformed others; lesion appearance and evolution are critical factors.

Conclusions:

• Standardized evaluation frameworks are crucial for comparing stroke segmentation algorithms.
• Further research is needed to improve sub-acute stroke lesion segmentation accuracy.
• Understanding stroke lesion characteristics and evolution is key to advancing segmentation techniques.