Comparison of JADE and canonical correlation analysis for ECG de-noising

Area of Science:

• Biomedical Engineering
• Signal Processing
• Medical Informatics

Background:

• Electrocardiogram (ECG) signals are often corrupted by noise, necessitating effective de-noising techniques.
• Blind source separation (BSS) methods offer potential for separating desired ECG signals from interference.
• Evaluating different BSS algorithms is crucial for optimizing ECG de-noising performance.

Purpose of the Study:

• To compare the performance of two BSS methods: Joint Approximate Diagonalization of Eigenmatrices (JADE) and Canonical Correlation Analysis (CCA).
• To assess their efficacy in de-noising long-term ECG signals, specifically targeting power line interference and unstructured noise.
• To evaluate the computational efficiency of each method within a decision tree-based framework.

Main Methods:

• Implementation of JADE based on fourth-order cross-cumulant tensor estimation and diagonalization.
• Implementation of CCA based on correlation matrix estimation between multidimensional variables.
• Integration of both BSS algorithms with a decision tree for ECG de-noising and evaluation on a database of 382 long-term ECG signals.

Main Results:

• CCA completely failed to remove 50 Hz power line interference from ECG signals.
• CCA demonstrated effectiveness in estimating and removing unstructured noise from ECG recordings.
• JADE exhibited higher computational complexity compared to CCA, resulting in slower component estimation.

Conclusions:

• CCA is more suitable for identifying and removing unstructured noise in ECG signals, whereas JADE may be more robust for structured interference like power line noise.
• The choice between JADE and CCA depends on the specific noise characteristics and computational resource availability.
• Further research could explore hybrid approaches to leverage the strengths of both JADE and CCA for comprehensive ECG de-noising.