SRPNet: stroke risk prediction based on two-level feature selection and deep fusion network.

Area of Science:

• Neurology
• Artificial Intelligence
• Public Health

Background:

• Stroke is a leading cause of chronic non-communicable diseases (NCDs), characterized by high morbidity, disability, and mortality.
• Effective stroke prevention hinges on controlling risk factors, yet their identification and risk quantification remain challenging.

Purpose of the Study:

• To introduce a novel prediction model for stroke risk, named SRPNet (Stroke Risk Prediction Network).
• To address the challenges in screening stroke risk factors and quantifying patient risk levels.

Main Methods:

• Employed a two-level feature selection method to identify significant stroke risk factors and reduce redundant information.
• Developed SRPNet using a deep fusion network that integrates Transformer and fully connected neural network (FCN) architectures.

Main Results:

• Evaluated SRPNet using data from the China Stroke Data Center (CSDC) and affiliated hospital census data.
• Demonstrated that SRPNet effectively selects stroke-related features and outperforms benchmark methods in risk prediction accuracy.

Conclusions:

• SRPNet facilitates rapid identification of high-quality stroke risk factors.
• The model enhances the accuracy of stroke risk prediction, serving as a valuable tool for clinical diagnosis and management.