An Automatic DWI/FLAIR Mismatch Assessment of Stroke Patients

Area of Science:

• Medical Imaging
• Neurology
• Artificial Intelligence in Medicine

Background:

• Diffusion-weighted imaging (DWI) and FLAIR MRI are crucial for ischemic stroke assessment.
• DWI/FLAIR mismatch aids in determining eligibility for recombinant tissue-type plasminogen activator (r-tPA) therapy.
• Current mismatch assessment faces challenges with binary classification and inter-observer subjectivity.

Purpose of the Study:

• To develop an automated method for segmenting parenchymal hyperintensities on FLAIR images.
• To enable automatic and continuous DWI/FLAIR mismatch assessment.
• To reduce subjectivity and variability in clinical stroke assessment.

Main Methods:

• A simple, automatic image segmentation algorithm for FLAIR hyperintensities.
• Quantitative comparison of automated segmentation with neuro-radiologist assessments using DICE scores.
• Evaluation of method robustness to hyper-parameter variations and correlation with clinical data.

Main Results:

• Automated segmentation achieved comparable inter-rater agreement (DICE 0.820) to neuro-radiologists (DICE 0.856).
• The method demonstrated robustness to hyper-parameter choices, indicating good generalizability.
• Continuous DWI/FLAIR mismatch assessment correlated with clinical evaluations of wake-up stroke patients.

Conclusions:

• The proposed automated segmentation method effectively quantifies parenchymal hyperintensity in ischemic stroke.
• This technique has the potential to decrease inter-observer variability in clinical DWI/FLAIR mismatch assessments.
• It offers a continuous, objective alternative to the current binary assessment, aiding r-tPA treatment decisions.