DE-PNN: Differential Evolution-Based Feature Optimization with Probabilistic Neural Network for Imbalanced Arrhythmia...

Area of Science:

• Cardiology
• Biomedical Engineering
• Machine Learning

Background:

• Accurate electrocardiogram (ECG) heartbeat classification is crucial for diagnosing cardiac arrhythmias.
• Existing methods often struggle with imbalanced datasets and detecting minority heartbeat classes.
• The need for computationally efficient and highly accurate classification algorithms persists.

Purpose of the Study:

• To develop and validate a novel ECG heartbeat classification method.
• To enhance the detection of both normal and various arrhythmic heartbeats.
• To address the challenge of imbalanced datasets in arrhythmia classification.

Main Methods:

• Utilized direct signal amplitude points from ECG holter devices, bypassing traditional feature extraction.
• Employed differential evolution (DE) for evolutionary feature optimization and reduction.
• Implemented a probabilistic neural network (PNN) for heartbeat classification.
• Incorporated the Matthews correlation coefficient (MCC) as a fitness function to prioritize minority classes.

Main Results:

• Achieved a significant 85.77% reduction in features, optimizing from 253 to 36 direct amplitude features.
• Demonstrated high classification performance with overall 99.33% accuracy.
• Obtained a 94.56% F1 score, 93.84% sensitivity, and 99.21% specificity across eight heartbeat classes.
• Successfully classified normal, left bundle branch block, right bundle branch block, premature ventricular contraction, atrial premature, ventricular escape, ventricular flutter, and paced beats.

Conclusions:

• The proposed DE-PNN scheme offers a robust and efficient solution for ECG heartbeat classification.
• The method effectively handles imbalanced datasets and improves the detection of minority arrhythmia classes.
• This approach provides a promising tool for automated cardiac arrhythmia diagnosis.