A Study on Dictionary Selection in Compressive Sensing for ECG Signals Compression and Classification.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Data Science

Background:

• Compressive Sensing (CS) offers efficient signal acquisition.
• Electrocardiographic (ECG) signal analysis requires accurate reconstruction for diagnosis.
• Existing CS methods for ECG may involve complex preprocessing or suboptimal reconstruction accuracy.

Purpose of the Study:

• To comparatively analyze projection matrices and dictionaries for CS of ECG signals.
• To propose and test application-specific dictionaries derived from cardiac patterns.
• To evaluate the trade-offs between preprocessing complexity and reconstruction accuracy in ECG CS.

Main Methods:

• Developed application-specific dictionaries based on R-wave patterns (centered/non-centered).
• Analyzed various projection matrices for ECG signal CS.
• Quantitatively and qualitatively assessed reconstructed ECG signals using distortion measures.
• Utilized k-nearest neighbors (KNN) and multilayer perceptron (MLP) neural networks for signal classification and evaluation.

Main Results:

• Application-specific dictionaries improved ECG signal reconstruction accuracy.
• The proposed CS method demonstrated superior performance compared to other compression techniques.
• KNN and MLP classifiers confirmed the high quality of reconstructed signals.

Conclusions:

• Optimized CS dictionaries and projection matrices enhance ECG signal reconstruction.
• The developed method offers a superior alternative for ECG compression and analysis.
• Reduced preprocessing complexity while maintaining high reconstruction fidelity is achievable.