ECG Classification Using Orthogonal Matching Pursuit and Machine Learning

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Signal Processing

Background:

• Electrocardiogram (ECG) is crucial for diagnosing heart disease.
• Existing ECG monitoring systems are rapidly increasing in number.
• Orthogonal Matching Pursuit (OMP) algorithms are often underestimated for ECG analysis.

Purpose of the Study:

• To investigate the impact of concise representation parameters on ECG classification performance.
• To develop and evaluate cardiovascular disease classification models using classical Machine Learning.
• To introduce novel methods for R-wave detection, QRS complex localization, and ECG signal aggregation.

Main Methods:

• Utilized the PTB-XL database for ECG signal analysis.
• Applied Orthogonal Matching Pursuit (OMP) for signal representation.
• Developed new algorithms for R-wave detection, QRS complex extraction, and signal resampling.
• Employed classical Machine Learning classifiers, particularly Decision Trees, for classification.

Main Results:

• Achieved a highest accuracy of 90.4% in classifying 2 cardiac disease classes.
• Demonstrated significant improvement in classification performance through optimized signal processing.
• Classification accuracy for 5 and 15 classes was lower (78% and 71%, respectively), indicating class complexity impact.

Conclusions:

• Optimized ECG signal processing, including R-wave detection and QRS complex extraction, enhances classification accuracy.
• Structured ECG signals using the proposed methods perform effectively with Decision Tree classifiers.
• The study highlights the potential of OMP algorithms and signal structuring for improved cardiovascular disease diagnosis.