Automated recognition of cardiac arrhythmias using sparse decomposition over composite dictionary.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Cardiology

Background:

• Cardiovascular diseases (CVDs) are a leading global cause of death.
• Electrocardiogram (ECG) signal processing is crucial for diagnosing CVDs, but challenges remain due to signal non-stationarity.
• Classical methods for ECG analysis have limitations in achieving high diagnostic accuracy.

Purpose of the Study:

• To develop an efficient ECG signal representation technique for improved arrhythmia detection.
• To enhance the accuracy of cardiovascular disease diagnosis through advanced signal processing.

Main Methods:

• A novel sparse decomposition technique using a composite dictionary (CD) of analytical functions (Stockwell, sine, cosine) is proposed.
• ECG signals are decomposed into stationary and non-stationary components, with five features extracted per component.
• Multi-class least-square twin support vector machines (LST-SVM) classify heartbeats, optimized by the artificial bee colony (ABC) technique.

Main Results:

• The proposed method achieved high accuracy: 99.21% in the category scheme and 90.08% in the personalized scheme.
• Sensitivity, positive predictivity, and F-score also demonstrated superior performance in both schemes.

Conclusions:

• The developed methodology offers a highly accurate approach for monitoring cardiac arrhythmias.
• This technique can be integrated into computerized decision support systems for early CVD detection and treatment.