Cardiovascular disease detection from cardiac arrhythmia ECG signals using artificial intelligence models with...

Area of Science:

• Cardiology and Biomedical Engineering
• Machine Learning for Healthcare
• Signal Processing for Medical Diagnostics

Background:

• Cardiovascular diseases (CVDs) are linked to irregular cardiac electrical activity detectable via ECG.
• Automated ECG analysis is crucial for timely arrhythmia detection, including Ventricular Tachycardia (VT), Premature Ventricular Contraction (PVC), and ST Change (ST).
• Current methods necessitate improved accuracy and efficiency in identifying these critical cardiac conditions.

Purpose of the Study:

• To enhance the detection accuracy of specific CVDs (VT, PVC, ST) using ECG data.
• To evaluate the effectiveness of various dimensionality reduction techniques combined with advanced feature selection and hyperparameter optimization algorithms.
• To identify the optimal combination of methods for superior classification performance in cardiac arrhythmia detection.

Main Methods:

• Employed dimensionality reduction techniques: Local Linear Embedding (LLE), Diffusion Maps (DM), and Laplacian Eigen (LE).
• Utilized Cuckoo Search (CS) and Harmonic Search Optimization (HSO) for feature selection on reduced ECG data.
• Classified cardiac conditions using seven classifiers (GMM, EM, NLR, LR, BDLC, Detrended FA, Firefly) and optimized hyperparameters with Adam and Grid Search Optimization (GSO).

Main Results:

• Feature selection significantly improved accuracy: HSO yielded 75.39% average accuracy, outperforming CS (64.36%) and no selection (55.65%).
• Hyperparameter tuning further boosted performance: Adam optimization with HSO feature selection achieved an average accuracy of 93.77%.
• The LLE dimensionality reduction with HSO feature selection and GMM classifier using Adam tuning reached a peak accuracy of 98.92% for ST vs. NSR detection.

Conclusions:

• The integration of advanced dimensionality reduction, heuristic feature selection (HSO), and hyperparameter optimization (Adam) substantially enhances ECG-based CVD detection.
• The GMM classifier with Adam tuning, combined with LLE and HSO, demonstrates superior performance for specific arrhythmia classification tasks.
• This methodology offers a promising avenue for developing more accurate and reliable automated systems for cardiovascular disease diagnosis.