Deep Neural Network-Based Detection of Diaphragmatic EMG Activity: From Simulation to Clinical Data.

Area of Science:

• Biomedical Engineering
• Respiratory Medicine
• Artificial Intelligence in Healthcare

Background:

• Accurate diaphragmatic electromyography (EMGdi) detection is crucial for optimizing respiratory support in COPD patients using nocturnal non-invasive ventilation (NIV).
• Conventional signal processing methods struggle with low signal-to-noise ratio (SNR) conditions, leading to respiratory event misclassification.
• Improving patient-ventilator synchrony is essential for effective NIV therapy.

Purpose of the Study:

• To develop and validate a deep neural network (DNN)-based method for automatic detection of EMGdi onset and offset.
• To address limitations of traditional methods in low SNR environments.
• To enhance patient-ventilator synchrony assessment for personalized respiratory therapies.

Main Methods:

• A convolutional neural network (CNN) was trained using simulated EMGdi signals with realistic variability and noise.
• A fuzzy label function derived from synthetic signals served as the training target.
• The CNN detector was evaluated on simulated and real-world patient data from COPD patients undergoing nocturnal NIV.

Main Results:

• The CNN-based detector demonstrated lower detection errors compared to a conventional method.
• The proposed approach exhibited reduced variability in EMGdi detection.
• Results indicate improved accuracy in identifying respiratory events.

Conclusions:

• The deep learning-based method shows significant potential for accurate EMGdi detection in challenging conditions.
• Enhanced detection accuracy can lead to improved patient-ventilator synchrony.
• This technology can facilitate more personalized and effective respiratory support for COPD patients.