A Framework with Transformer-Based Model for Cerebrovascular Stenosis Detection in Magnetic Resonance Angiography

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neurology

Background:

• Accurate identification of cerebrovascular stenosis is crucial for stroke prevention and management.
• Automated stenosis detection in 3D Magnetic Resonance Angiography (MRA) is challenging due to complex anatomy and imaging variations.

Purpose of the Study:

• To develop an automated, robust transformer-based deep learning framework for detecting cerebrovascular stenosis in 3D MRA scans.
• To enhance clinical utility for early stroke detection and risk reduction.

Main Methods:

• A transformer-based deep learning framework was proposed for cerebrovascular stenosis detection.
• The framework automatically localizes vessel centerlines and classifies 3D MRA regions as normal or narrowed.
• The model was trained and validated on an expert-annotated dataset.

Main Results:

• The framework demonstrated strong and stable performance across five-fold cross-validation.
• Achieved high metrics, including 0.9339 accuracy, 0.7998 F1-score, 0.9488 AUC, and 0.8313 Precision-Recall AUC, even with imbalanced data.
• Indicated robust discrimination capability and effective detection.

Conclusions:

• The proposed framework is a dependable tool for automated cerebrovascular evaluation.
• Its superior performance suggests significant clinical utility for early stroke detection and risk reduction.