Dynamic Cardiac Event Detection from Single-Arm Wearable ECG via a Contrastive Multitask Framework.

Area of Science:

• Biomedical Engineering
• Cardiology
• Artificial Intelligence in Medicine

Background:

• Wearable upper-limb electrocardiogram (ECG) devices offer a practical method for dynamic cardiac monitoring.
• Single-channel Arm-ECG presents challenges due to higher noise and distinct morphology compared to standard 12-lead ECG.
• Reliable arrhythmia detection from Arm-ECG is crucial for effective dynamic cardiac assessment.

Purpose of the Study:

• To evaluate the capability of single-channel Arm-ECG to reliably detect arrhythmias.
• To develop and validate a deep learning framework for analyzing Arm-ECG signals.
• To assess the generalizability of the proposed model across different datasets.

Main Methods:

• Proposed CLMF-Net, a contrastive multitask framework with multiscale convolutional layers for temporal-morphological pattern capture.
• Incorporated a fine-grained reconstruction branch to preserve subtle clinical features (e.g., P waves, ST segments).
• Employed a contrastive module for robustness against signal quality variations and trained end-to-end with a unified loss function.

Main Results:

• Feasibility of dynamic Arm-ECG acquisition confirmed in 132 elderly participants during six-minute walk tests.
• CLMF-Net achieved 95.19% accuracy on the Arm-ECG dataset for arrhythmia identification.
• Demonstrated generalizability with 95.55% and 84.23% accuracy on Chapman and CPSC2018 datasets, respectively.

Conclusions:

• CLMF-Net effectively detects arrhythmias from single-channel Arm-ECG, despite noise and morphological challenges.
• Learned representations may not always align with clinically emphasized intervals, requiring caution in clinical translation.
• Single-arm ECG monitoring shows clinical utility for real-world cardiac health assessment and translational potential.