High-Risk Prediction of Cardiovascular Diseases via Attention-Based Deep Neural Networks

Area of Science:

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

Background:

• Cardiovascular disease (CVD) risk prediction is crucial due to rising sub-health populations.
• Current pathological methods for prognosis are often expensive and inaccurate.
• Machine learning models using Electronic Health Records (EHRs) show promise but face challenges in feature selection and data representation.

Purpose of the Study:

• To develop an automated, end-to-end model for accurate high-risk prediction of cardiovascular disease.
• To address the challenges of feature selection and data integration from longitudinal and heterogeneous EHRs.
• To improve patient risk stratification for cardiovascular diseases.

Main Methods:

• Proposed an end-to-end deep learning model named DeepRisk.
• Utilized an attention mechanism and deep neural networks for automatic feature learning.
• Integrated heterogeneous and time-ordered medical data from EHRs.

Main Results:

• DeepRisk demonstrated significant improvements in high-risk cardiovascular disease prediction accuracy.
• The model effectively learned high-quality features from EHR data.
• Experimental results on a real medical dataset outperformed state-of-the-art approaches.

Conclusions:

• DeepRisk offers a robust and accurate solution for high-risk cardiovascular disease prediction.
• The model's ability to automatically learn features and integrate complex EHR data is a key advancement.
• This approach holds significant potential for improving early detection and management of cardiovascular diseases.