Assessment of Heart Rhythm and Arrhythmia Detection from Electromyographic Signals Using a Convolutional Neural...

Area of Science:

• Biomedical Engineering
• Signal Processing
• Wearable Technology

Background:

• Surface electromyography (EMG) systems are typically used for muscle activity monitoring.
• Assessing cardiac rhythm usually requires electrocardiography (ECG).
• There is a need for integrated systems for non-invasive functional status monitoring, including cardiac arrhythmias.

Purpose of the Study:

• To develop and validate a convolutional autoencoder algorithm for heart rhythm assessment using EMG signals.
• To explore the potential of the Myosuit EMG system for cardiac monitoring.
• To enable comprehensive functional status assessment and arrhythmia screening.

Main Methods:

• Developed a fully convolutional autoencoder trained on R-wave masks.
• Recorded synchronous EMG, ECG, and heart rate monitor data from 6 healthy males.
• Evaluated model performance using F-score in pooled, leave-one-subject-out, and individual subject scenarios during rest and exercise.

Main Results:

• The convolutional autoencoder successfully extracted rhythmograms from EMG signals.
• High classification performance was achieved, particularly in the leave-one-subject-out scenario, indicating good generalization.
• Feasibility of using Myosuit EMG electrodes for simultaneous muscle and heart rate monitoring was confirmed.

Conclusions:

• The developed algorithm can reliably extract rhythmograms from EMG signals for heart rate variability analysis and arrhythmia screening.
• The Myosuit EMG system, combined with the algorithm, offers a feasible approach for non-invasive cardiac monitoring.
• This paves the way for cost-effective wearable systems for real-time arrhythmia screening without a dedicated ECG channel.