Block-based neural networks for personalized ECG signal classification.

Area of Science:

• Artificial Intelligence
• Biomedical Engineering
• Signal Processing

Background:

• Electrocardiogram (ECG) signal analysis is crucial for diagnosing cardiac conditions.
• Personalized and adaptive classification methods are needed to account for individual variations and time-varying ECG characteristics.
• Existing ECG classification methods may lack robustness in dynamic operating environments.

Purpose of the Study:

• To introduce evolvable block-based neural networks (BbNNs) for personalized ECG heartbeat pattern classification.
• To develop an adaptive evolutionary algorithm for optimizing BbNN structure and weights.
• To enhance the accuracy and robustness of ECG arrhythmia detection.

Main Methods:

• Utilized a 2-D array of modular component neural networks (NNs) with flexible structures (BbNNs).
• Implemented adaptive evolutionary algorithms (EAs) combining local gradient-based search and evolutionary operators with adaptive rate updates.
• Employed Hermite transform coefficients and R-peak intervals as input features for the BbNN.
• Configured BbNNs using signal flow between blocks and reconfigurable hardware (FPGAs).

Main Results:

• Achieved high average detection accuracies: 98.1% for ventricular ectopic beats and 96.6% for supraventricular ectopic beats.
• Demonstrated superior performance compared to previously reported ECG classification results using the MIT-BIH arrhythmia database.
• Validated the effectiveness of the adaptive operator rate update scheme for improved fitness.

Conclusions:

• Optimized BbNNs provide personalized heartbeat classifiers that adapt to individual differences and time-varying ECG signals.
• The proposed EA-optimized BbNNs represent a significant advancement in ECG arrhythmia monitoring.
• The methodology offers a robust and accurate solution for real-time heartbeat classification.