Ischemic stroke segmentation in multi-sequence MRI by symmetry determined superpixel based hierarchical clustering

Area of Science:

• Neuroimaging
• Medical Image Analysis
• Radiology

Background:

• Automated estimation of ischemic stroke evolution is crucial for effective treatment.
• Multi-sequence MRI provides rich data for characterizing brain tissue abnormalities.
• Brain hemisphere asymmetry is a key indicator, but requires precise symmetry axis estimation to avoid errors.

Purpose of the Study:

• To develop and evaluate an automated method for ischemic stroke lesion delineation using MRI.
• To improve the accuracy of lesion segmentation by leveraging inter-hemispheric asymmetry.
• To eliminate manual thresholding in superpixel clustering for stroke lesion segmentation.

Main Methods:

• The proposed method, symmetry determined superpixel based hierarchical clustering (SSHC), estimates lesions from inter-hemispheric asymmetry.
• Asymmetry data is used to determine thresholding parameters for hierarchical clustering of superpixels.
• A multi-sequence MRI pipeline combines estimations from individual sequences for robust analysis.

Main Results:

• SSHC was evaluated on the LLU and ISLES'15 datasets, demonstrating high performance.
• The method achieved a Dice similarity score of 0.704±0.27 and a recall of 0.85±0.01.
• SSHC outperformed the state-of-the-art by 6% in Dice score and 35% in recall.

Conclusions:

• SSHC offers a reliable and automated approach for detecting (sub-)acute adult ischemic stroke lesions.
• The method eliminates the need for manual threshold determination, enhancing efficiency and reproducibility.
• SSHC shows significant promise in revolutionizing stroke treatment through improved lesion detection.