Related Experiment Video
Updated: Sep 5, 2025
08:51
Statistical Modelling of Cortical Connectivity Using Non-invasive Electroencephalograms
Published on: November 1, 2019
5.7K
Global ECG Classification by Self-Operational Neural Networks With Feature Injection
IEEE Transactions on Bio-Medical Engineering
|July 5, 2022
Summary
This study introduces a novel 1D Self-ONN for accurate global Electrocardiogram (ECG) classification, achieving state-of-the-art arrhythmia detection without patient-specific data. The compact model excels in identifying normal and abnormal heart rhythms with high precision and recall.
Area of Science:
- Biomedical Engineering
- Artificial Intelligence in Medicine
- Cardiology
Background:
- Automated Electrocardiogram (ECG) classification for arrhythmia detection is crucial, especially with the rise of wearable sensors.
- Existing deep learning models show a performance gap in global (inter-patient) ECG classification compared to patient-specific approaches.
- Accurate, automated arrhythmia detection is highly desirable for widespread clinical and personal health monitoring.
Purpose of the Study:
- To propose a novel, compact 1D Self-Organizing Neural Network (Self-ONN) for inter-patient ECG classification.
- To enhance classification accuracy by integrating morphological and timing information from heart cycles.
- To demonstrate the efficacy of the proposed model in achieving high-performance global ECG analysis.
Main Methods:
- Utilized compact 1D Self-ONN layers to automatically learn morphological representations from ECG waveforms around R peaks.
- Incorporated temporal features derived from RR intervals for enhanced timing characterization.
- Employed a feature injection strategy combining morphological and temporal data for arrhythmia classification.
Main Results:
- Achieved record-breaking classification performance on the MIT-BIH arrhythmia benchmark database.
- Reported 99.21% precision, 99.10% recall, and 99.15% F1-score for normal segments.
- Obtained high scores for supra-ventricular ectopic beats (SVEBs) and ventricular-ectopic beats (VEBs), demonstrating robust detection capabilities.
Conclusions:
- The proposed compact 1D Self-ONN with feature injection surpasses state-of-the-art deep models in global ECG classification.
- The method achieves high performance with minimal computational complexity, making it suitable for resource-constrained environments.
- This study confirms that effective global ECG classification is achievable without relying on patient-specific data, offering a promising approach for arrhythmia detection.
