ECG arrhythmia detection in an inter-patient setting using Fourier decomposition and machine learning

Area of Science:

• Cardiology
• Biomedical Engineering
• Artificial Intelligence

Background:

• Electrocardiogram (ECG) beat classification is crucial for monitoring cardiac abnormalities.
• Existing AI algorithms for arrhythmia detection face challenges with unseen data and superventricular ectopic beat (SVEB) detection accuracy.
• An inter-patient validation approach is necessary for robust AI model performance assessment.

Purpose of the Study:

• To develop an AI-based ECG beat classification methodology that addresses limitations in existing algorithms.
• To improve the accuracy of superventricular ectopic beat (SVEB) detection in arrhythmia monitoring.
• To validate the proposed AI model using an inter-patient paradigm for enhanced generalizability.

Main Methods:

• Utilized the Fourier Decomposition Method (FDM) for multi-scale analysis of ECG signals.
• Extracted time-domain and statistical features from narrow-band signal components derived via FDM.
• Employed Kruskal-Wallis test and Minimum Redundancy Maximum Relevance (mRMR) for feature selection and redundancy reduction.
• Applied a Support Vector Machine (SVM) classifier with a linear kernel for beat classification.

Main Results:

• The proposed methodology achieved an F1 score of 89.35% for SVEB detection, outperforming existing algorithms.
• An overall accuracy of 98.03% and a Mathew's Correlation Coefficient (MCC) of 91.84% were obtained on the MIT-BIH arrhythmia dataset.
• The inter-patient validation paradigm demonstrated the model's effectiveness on unseen data.

Conclusions:

• The developed AI approach effectively classifies ECG beats and detects arrhythmias, particularly SVEBs, with improved accuracy.
• The inter-patient validation strategy ensures a more reliable assessment of AI model performance in clinical settings.
• This research contributes a robust and accurate AI solution for cardiac abnormality monitoring using ECG data.