Intelligent Stroke Disease Prediction Model Using Deep Learning Approaches.

Area of Science:

• Neurology
• Artificial Intelligence
• Biomedical Engineering

Background:

• Stroke is a leading cause of death and disability worldwide.
• Early detection of stroke warning signs is crucial for timely intervention and reducing severity.
• Traditional methods struggle with complex physiological data and sample imbalance.

Purpose of the Study:

• To develop an accurate and robust stroke prediction model using deep neural networks.
• To address the challenge of imbalanced datasets in stroke prediction.
• To leverage physiological characteristics for predicting stroke risk.

Main Methods:

• Utilized Wasserstein generative adversarial networks with gradient penalty (WGAN-GP) for high-fidelity data augmentation.
• Designed a deep regression network to model nonlinear relationships between physiological parameters and stroke risk.
• Compared the proposed model against traditional machine learning algorithms (Decision Tree, Random Forest, SVM, ANN).

Main Results:

• The proposed deep learning model achieved optimal performance based on the F-measure index.
• WGAN-GP effectively generated high-fidelity synthetic data to overcome sample imbalance.
• Ablation experiments confirmed the robustness and effectiveness of the developed stroke prediction model.

Conclusions:

• Deep neural networks, particularly the proposed WGAN-GP and regression network, offer a powerful approach for stroke risk prediction.
• Data augmentation techniques are vital for improving the performance of models trained on imbalanced datasets.
• The developed model shows significant potential for clinical application in early stroke detection and prevention.