Automated diagnosis of Coronary Artery Disease using pattern recognition approach

Area of Science:

• Cardiology
• Biomedical Engineering
• Signal Processing

Background:

• Coronary Artery Disease (CAD) is a leading cause of Cardiovascular Disease (CVD), often diagnosed using Electrocardiogram (ECG).
• Manual ECG interpretation is time-consuming and complex due to nonlinearities.
• Automated systems are crucial for efficient and accurate diagnosis.

Purpose of the Study:

• To develop and validate an automated system for discriminating between Coronary Artery Disease (CAD) and Normal Sinus Rhythm (NSR) heartbeats.
• To utilize Higher-Order Statistics (HOS) cumulants for feature extraction from ECG signals.
• To employ machine learning classifiers for robust CAD diagnosis.

Main Methods:

• Extracted Higher-Order Statistics (HOS) cumulants features from ECG signals.
• Reduced feature dimensionality using Principal Components Analysis (PCA), selecting statistically significant Principal Components (PCs) (p-value < 0.05).
• Classified CAD and NSR using Random Forest (RAF) and Rotation Forest (ROF) ensemble classifiers.

Main Results:

• The proposed system successfully discriminated between CAD and NSR heartbeats.
• The combination of HOS cumulants, PCA, and ensemble classifiers demonstrated robust diagnostic capability.
• Identified medically significant features for improved CAD screening.

Conclusions:

• The automated system using HOS cumulants and ensemble classifiers offers a robust method for CAD diagnosis.
• This approach can aid in screening CAD risk factors and support telemonitoring applications.
• The system provides a noninvasive and efficient tool for cardiovascular disease management.