An effective brain stroke diagnosis strategy based on feature extraction and hybrid classifier

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neurology

Background:

• Stroke is a major cause of death and disability globally.
• Early detection of stroke is crucial but remains a clinical challenge.
• Current diagnostic methods require timely and accurate interpretation of medical images.

Purpose of the Study:

• To develop and validate an Effective Brain Stroke Diagnosis Strategy (EBDS) for accurate and interpretable stroke detection.
• To integrate a hybrid deep learning model for enhanced diagnostic performance.
• To improve clinical trust through explainability techniques.

Main Methods:

• A hybrid deep learning framework combining Vision Transformer (ViT) and VGG16 was developed.
• The model was trained and evaluated on a public Kaggle CT image dataset.
• Explainability techniques (Grad-CAM, LIME) were incorporated for decision transparency.

Main Results:

• The EBDS model achieved a test accuracy of 99.6%.
• High precision (1.00 normal, 0.98 stroke) and recall (0.99 normal, 1.00 stroke) were recorded.
• An overall F1-score of 0.99 demonstrated the model's robustness and reliability, outperforming state-of-the-art methods.

Conclusions:

• The EBDS framework offers a scalable, explainable, and clinically relevant solution for early brain stroke diagnosis.
• The model's high diagnostic performance and interpretability enhance its potential for real-time clinical application.
• This study addresses a critical gap in medical imaging diagnostics for stroke detection.