Stroke Risk Prediction With Hybrid Deep Transfer Learning Framework.

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Cardiovascular Disease Research

Background:

• Stroke is a major global cause of death and disability with limited treatment options.
• Current deep learning models for stroke risk prediction require large, labeled datasets, which are challenging to obtain due to privacy concerns and data fragmentation across institutions.
• Health data often suffers from extreme class imbalance between positive and negative instances, further complicating model development.

Purpose of the Study:

• To develop a novel Hybrid Deep Transfer Learning-based Stroke Risk Prediction (HDTL-SRP) framework.
• To address the challenges of limited data availability, data privacy, and class imbalance in stroke risk prediction.
• To improve the accuracy and applicability of AI-driven stroke risk prediction models.

Main Methods:

• Proposed a Hybrid Deep Transfer Learning-based Stroke Risk Prediction (HDTL-SRP) scheme.
• Integrated knowledge from multiple correlated data sources, including external stroke datasets and chronic disease data (hypertension, diabetes).
• Utilized transfer learning to overcome limitations of small and imbalanced datasets.

Main Results:

• The HDTL-SRP framework demonstrated superior performance compared to existing state-of-the-art stroke risk prediction models.
• The model was validated through extensive testing in both synthetic and real-world scenarios.
• Achieved significant improvements in stroke risk prediction accuracy despite data limitations.

Conclusions:

• The proposed HDTL-SRP scheme offers an effective solution for stroke risk prediction in data-scarce and privacy-sensitive healthcare environments.
• The framework shows potential for real-world deployment across multiple hospitals, enhanced by advanced communication infrastructures like 5G/B5G.
• This approach advances the application of artificial intelligence in managing cardiovascular disease risk.