CS-based multi-task learning network for arrhythmia reconstruction and classification using ECG signals.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence

Background:

• Deep learning excels at electrocardiograph (ECG) arrhythmia classification but struggles with large datasets in real-time systems.
• Limited bandwidth and real-time processing hinder the application of current deep learning methods in long-term ECG monitoring.

Purpose of the Study:

• To develop a novel deep learning approach for efficient ECG arrhythmia classification and reconstruction.
• To address the challenges of limited bandwidth and real-time processing in ECG monitoring systems.

Main Methods:

• Proposed a multi-task network, CSML-Net, combining compressed sensing (CS) and convolutional neural networks (CNNs).
• ECG signals are compressed using a learned measurement matrix and then simultaneously recovered and classified via shared layers and dual task branches.
• Incorporated a multi-scale feature module to enhance model performance.

Main Results:

• CSML-Net demonstrated superior reconstruction quality and classification performance on the MIT-BIH arrhythmia dataset compared to existing methods.
• The model achieved effective ECG arrhythmia reconstruction and classification within the compressed domain.

Conclusions:

• The proposed CSML-Net offers a promising solution for ECG arrhythmia reconstruction and classification in resource-constrained environments.
• This approach enhances the applicability of deep learning in real-time and limited-bandwidth ECG monitoring systems.