A robust multiple heartbeats classification with weight-based loss based on convolutional neural network and...

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine
• Cardiology

Background:

• Electrocardiogram (ECG) analysis is crucial for diagnosing cardiac arrhythmias but is labor-intensive and prone to subjective errors.
• Deep learning shows promise for ECG analysis, yet struggles with imbalanced datasets and computationally intensive preprocessing.
• A need exists for efficient, lightweight algorithms for real-time ECG analysis on portable devices.

Purpose of the Study:

• To develop a robust and efficient deep learning method for classifying heartbeats suitable for wearable ECG monitors.
• To create an automated system that minimizes reliance on manual feature extraction and noise reduction.

Main Methods:

• A novel, lightweight deep learning architecture combining Convolutional Neural Network (CNN) and Bidirectional Long Short-Term Memory (Bi-LSTM) was proposed.
• A weight-based loss function was implemented to mitigate classification bias from imbalanced ECG datasets.
• The k-fold cross-validation method was employed to enhance model reliability and avoid validation set division bias.

Main Results:

• The algorithm achieved high performance on the MIT-BIH Arrhythmia Database.
• Key metrics include 99.33% accuracy, 93.67% sensitivity, 99.18% specificity, 89.85% positive prediction, and 91.65% F1-score.
• The model processed raw ECG signals without extensive preprocessing, demonstrating efficiency.

Conclusions:

• The developed CNN-BiLSTM model offers an efficient and accurate solution for automated heartbeats classification.
• The weight-based loss function effectively addresses class imbalance in ECG data.
• This lightweight approach is suitable for deployment on portable ECG sensors, advancing remote cardiac monitoring.