Development of AI based behavioral feature patterns on influencing asymptomatic cardiovascular disease attributes: a
V Sangeetha1, Syed Muzamil Basha1, Syed Thouheed Ahmed2
1School of Computer Science and Engineering, REVA University, Bengaluru, India.
Insights
This study introduces a novel transfer learning framework to improve cardiovascular disease (CVD) prediction by analyzing interdependencies between patient attributes. The new method enhances CVD risk characterization and provides reliable, explainable clinical decision support.
Area of Science:
- Medical Informatics
- Machine Learning in Healthcare
- Cardiovascular Research
Background:
- Cardiovascular diseases (CVD) pose a significant global health burden.
- Heterogeneous clinical data and feature variability hinder accurate CVD predictive modeling.
- Current methods often neglect interdependencies between symptomatic and asymptomatic patient attributes.
Purpose of the Study:
- To develop a transfer learning framework for harmonizing multi-source data in CVD prediction.
- To model interdependencies between clinical attributes for improved risk characterization.
- To enhance the interpretability and reliability of CVD predictive models.
Main Methods:
- Integration of multi-source datasets (MIMIC-III, FHS, CHD) into a unified latent feature space.
- Application of a transfer learning approach for cross-domain feature harmonization.
- Implementation of an interdependency-driven attribute modeling technique with confidence-calibrated influence scoring.
Main Results:
- Achieved a sensitivity of 92.37% and an F1-score of 0.8962, demonstrating improved predictive performance.
- Successfully captured behavioral interdependencies between clinical attributes.
- Enhanced model interpretability and prediction reliability.
Conclusions:
- Interdependency-aware feature mapping significantly improves cardiovascular disease risk characterization.
- The proposed framework offers a standardized approach for robust and explainable clinical decision support in cardiology.
- Highlights the potential of transfer learning in addressing data heterogeneity for complex diseases.
Abstract:
Cardiovascular diseases (CVD) remain a major global health challenge, characterized by heterogeneous clinical attributes and multi-source data variability that limit reliable predictive modeling. Existing approaches primarily focus on independent feature analysis, often overlooking the interdependency between symptomatic and asymptomatic attributes. In this study, a transfer learning-based framework is proposed for cross-domain feature harmonization and interdependency-driven attribute modeling. Multi-source datasets, including MIMIC-III, Framingham Heart Study (FHS), and Cleveland Heart Disease (CHD), are integrated to construct a unified latent feature space. The proposed method systematically captures behavioral interdependencies between clinical attributes and introduces a confidence-calibrated influence scoring mechanism to enhance interpretability and reliability of predictions. Experimental results demonstrate improved performance, achieving a sensitivity of 92.37% and an F1-score of 0.8962. The findings highlight the effectiveness of interdependency-aware feature mapping in improving CVD risk characterization and provide a standardized framework for robust and explainable clinical decision support.
Related Concept Videos
01:23Assessment of the Cardiovascular System I: Subjective Data
Initial Enquiry
Ask the patient about their primary concern and thoroughly explore all reported symptoms.
Medical History
Investigate past illnesses affecting the cardiovascular system, such as angina, anemia, rheumatic fever, congenital heart disease, stroke, thrombophlebitis, dysrhythmias, varicosities
Inquire about symptoms...
01:30Coronary Artery Disease I: Introduction
01:26Coronary Artery Disease IV: Preventive Measures