Cardiac arrhythmia beat classification using DOST and PSO tuned SVM.

Area of Science:

• Biomedical Engineering
• Cardiology
• Signal Processing

Background:

• Cardiovascular diseases (CVDs) are a leading cause of death, necessitating accurate and efficient diagnostic tools.
• Electrocardiogram (ECG) signal analysis is crucial for diagnosing CVDs, but manual interpretation of long-term recordings is time-consuming and challenging.
• Existing signal processing techniques for ECG analysis face limitations due to the non-stationary nature of these signals.

Purpose of the Study:

• To develop an automated diagnostic solution for cardiac arrhythmia detection.
• To improve the classification accuracy rate of ECG signal analysis.
• To address the limitations of current methods in handling non-stationary ECG data.

Main Methods:

• A four-stage methodology involving filtering, R-peak detection, feature extraction, and classification.
• Wavelet-based filtering and the Pan-Tompkins algorithm for R-peak detection.
• Discrete Orthogonal Stockwell Transform (DOST) for time-frequency feature extraction, combined with Principal Component Analysis (PCA) and dynamic features, classified using Particle Swarm Optimization (PSO)-tuned Support Vector Machines (SVM).

Main Results:

• The proposed method achieved a 99.18% accuracy for 16 classes in a category-based assessment and 89.10% accuracy for 5 classes in a patient-based assessment on the MIT-BIH arrhythmia database.
• These results demonstrate improved performance compared to state-of-the-art diagnostic methods.
• The methodology was validated on the benchmark MIT-BIH arrhythmia database.

Conclusions:

• The novel feature representation and PSO-optimized SVM classifier significantly enhance classification accuracy for cardiac arrhythmias.
• The developed system offers a promising automated computer-aided diagnosis (CAD) solution for cardiac arrhythmia beats.
• The approach effectively handles the non-stationary characteristics of ECG signals.