A Hybrid Deep Learning Approach for ECG-Based Arrhythmia Classification.

Area of Science:

• Cardiology
• Biomedical Engineering
• Artificial Intelligence

Background:

• Irregular heart rhythms (arrhythmias) pose life-threatening risks, necessitating accurate detection.
• Electrocardiogram (ECG) data is crucial for diagnosis but complex for manual analysis.
• Automated systems are critical for efficient analysis of vast ECG datasets.

Purpose of the Study:

• To develop an automated system for detecting and classifying cardiac arrhythmias.
• To enhance the efficiency and accuracy of ECG data analysis.
• To reduce the need for extensive manual intervention by medical professionals.

Main Methods:

• A hybrid deep learning approach combining 2D Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) was developed.
• 1D ECG signals were converted into 2D Scalogram images for noise filtering and feature extraction.
• The proposed 2D-CNN-LSTM model was trained and evaluated using the MIT-BIH arrhythmia database.

Main Results:

• The 2D-CNN-LSTM model achieved high accuracy rates: ≈98.7% for Cardiac Arrhythmias (ARR), 99% for Congestive Heart Failure (CHF), and 99% for Normal Sinus Rhythm (NSR).
• The model demonstrated an average sensitivity of 98.33% and specificity of 98.35% across all three arrhythmia types.
• Results indicate superior performance compared to existing techniques for arrhythmia classification.

Conclusions:

• A robust deep learning approach for arrhythmia classification using ECG 2D Scalogram images and a CNN-LSTM model has been established.
• The proposed method offers significant improvements over current techniques, potentially reducing physician workload.
• Future research includes applying the method to live ECG signals and exploring bidirectional LSTM (Bi-LSTM).