Deep Learning Algorithm Classifies Heartbeat Events Based on Electrocardiogram Signals.

Area of Science:

• Cardiology
• Biomedical Engineering
• Artificial Intelligence

Background:

• Cardiovascular diseases (CVDs) are a leading global health threat, necessitating advanced diagnostic tools.
• Electrocardiogram (ECG) is a critical non-invasive method for CVD screening and diagnosis.
• Existing methods for ECG analysis face challenges in accuracy and efficiency.

Purpose of the Study:

• To develop and evaluate a novel deep learning algorithm for heartbeat event classification using ECG signals.
• To compare the proposed deep learning approach with an evolutionary neural system method.
• To assess the model's performance on single- and multiple-lead ECG datasets.

Main Methods:

• A deep learning model combining Convolutional Neural Network (CNN) with Bidirectional Long Short-Term Memory (BiLSTM) was proposed.
• The CNN-BiLSTM model (Method II) was compared against an evolutionary neural system approach (Method I).
• Experiments were conducted using single-lead (Database I) and 12-lead (Database II) ECG data from multiple challenge datasets.

Main Results:

• The CNN-BiLSTM model (Method II) achieved significantly faster training times (1 hour) compared to Method I (28.3 hours).
• Method II demonstrated high accuracy, reaching 80%, 82.6%, and 85% on the China Physiological Signal Challenge 2018, PhysioNet Challenge 2017, and MIT-BIH Arrhythmia datasets, respectively.
• While Method I showed slightly better performance, Method II offered a more practical and efficient solution for heartbeat classification.

Conclusions:

• The novel CNN-BiLSTM deep learning approach provides an efficient and accurate method for heartbeat event classification from ECG signals.
• This approach holds significant potential for improving cardiovascular disease screening and diagnosis.
• The reduced training time makes the CNN-BiLSTM model a viable and scalable solution for clinical applications.