Eigenvector methods for automated detection of electrocardiographic changes in partial epileptic patients

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine

Background:

• Electrocardiographic (ECG) signal analysis is crucial for diagnosing cardiac and neurological conditions.
• Automated diagnostic systems are increasingly employed for complex pattern recognition tasks in healthcare.
• Combining diverse features from raw data presents a significant challenge in pattern recognition applications.

Purpose of the Study:

• To develop and evaluate automated diagnostic systems for detecting electrocardiographic changes in partial epilepsy patients.
• To compare the performance of different machine learning classifiers when trained on diverse or composite features.
• To identify the most effective method for classifying normal and partial epilepsy ECG signals.

Main Methods:

• Utilized two types of ECG beats (normal and partial epilepsy) from the Physiobank database (180 records each).
• Employed feature extraction using eigenvector methods followed by classification.
• Tested and benchmarked Multilayer Perceptron Neural Network (MLPNN), Combined Neural Network (CNN), Mixture of Experts (ME), and Modified Mixture of Experts (MME) classifiers.

Main Results:

• The Modified Mixture of Experts (MME) trained on diverse features achieved the highest classification accuracy.
• The MME system demonstrated a total classification accuracy of 99.44%.
• Performance of MME surpassed other tested automated diagnostic systems.

Conclusions:

• The Modified Mixture of Experts (MME) model, trained on diverse features, is highly effective for automated epilepsy detection from ECGs.
• This approach offers a promising tool for improving the diagnosis of partial epilepsy.
• Automated systems integrating diverse feature analysis can significantly enhance diagnostic accuracy in clinical practice.