Related Experiment Videos
Possibilities of using neural networks for ECG classification
G Bortolan1, C Brohet, S Fusaro
1LADSEB-CNR, Padova, Italy.
Journal of Electrocardiology
|January 1, 1996
Summary
This study validates neural network approaches for diagnosing conditions like ventricular hypertrophy and myocardial infarction using electrocardiography (ECG) data. The research compares two neural network architectures across two distinct ECG databases.
Area of Science:
- Cardiology
- Medical Informatics
- Artificial Intelligence
Background:
- Computerized electrocardiography (ECG) is crucial for diagnosing cardiac conditions.
- Neural networks offer potential for automated diagnostic classification in ECG analysis.
- Standardized validation on diverse datasets is essential for reliable AI in healthcare.
Purpose of the Study:
- To evaluate the efficacy of neural network approaches for diagnostic classification in computerized electrocardiography.
- To analyze the impact of normalization, pruning, and fuzzy preprocessing on network performance.
- To compare results from two distinct ECG databases (CORDA and ECG-UCL).
Main Methods:
- Utilized two independent ECG databases: CORDA and ECG-UCL.
- Focused on electrocardiographic signals with single diagnoses and no conduction abnormalities.
- Analyzed two neural network architectures, examining normalization, pruning, and radial basis function-based fuzzy preprocessing.
Main Results:
- Neural network characteristics were tested and validated for diagnostic classification.
- Seven diagnostic classes, including ventricular hypertrophy and myocardial infarction, were considered.
- Performance variations between the two databases were analyzed in detail.
Conclusions:
- Neural network approaches show promise for diagnostic classification in computerized electrocardiography.
- The choice of network architecture and preprocessing techniques influences diagnostic accuracy.
- Validation across multiple datasets is key to robust ECG diagnostic tools.