Characterization of clot composition in acute cerebral infarct using machine learning techniques.

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Cardiovascular Medicine

Background:

• Clot characteristics in cerebral artery occlusion can indicate stroke etiology.
• Identifying the cause of stroke is crucial for guiding treatment and prevention.
• Endovascular treatment is a key intervention for acute ischemic stroke.

Purpose of the Study:

• To develop and validate a machine learning (ML) system for automated clot analysis.
• To predict atrial fibrillation (AF) as the cause of ischemic stroke using ML.
• To compare the accuracy of ML clot analysis with expert visual inspection.

Main Methods:

• Utilized gradient echo (GRE) images from patients with middle cerebral artery occlusion.
• Developed an ML algorithm to predict AF based on clot signal profiles.
• Compared ML accuracy to neuroimaging specialists' visual assessment of blooming artifact.
• Analyzed endovascular procedures and outcomes in patients with and without AF.

Main Results:

• The ML system achieved >75.4% accuracy in predicting AF via cross-validation.
• External validation showed an area under the curve >0.87 for AF probability.
• ML analysis took approximately 3 minutes per patient.
• Visual inspection had lower accuracy (sensitivity 0.79, specificity 0.63) for AF detection.

Conclusions:

• Machine learning-based rapid clot analysis is feasible for identifying AF in stroke patients.
• The ML system demonstrates high accuracy in predicting AF, outperforming visual inspection.
• Automated clot analysis can inform the selection of endovascular treatment strategies.