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

Accurate detection of diaphragmatic electromyography (EMGdi) activity is critical for optimizing respiratory support in COPD patients undergoing nocturnal non-invasive ventilation (NIV). Traditional signal processing approaches have demonstrated limitations in handling low signal-to-noise ratio (SNR) conditions, often leading to misclassification of respiratory events. In this work, we propose a deep neural network-based method for automatic detection of EMGdi onset and offset. Specifically, a convolutional neural network (CNN) is trained on simulated EMGdi signals generated with a variable envelope model that incorporates realistic cycle variability and additional low- and high-frequency noise. A fuzzy label function is derived from these synthetic signals and used as the training target. The CNN-based detector is then applied to simulated data and annotated data from COPD patients undergoing nocturnal NIV, and its performance was compared against a conventional detector that incorporates a correction based on the ratio between the RMS values of the inspiratory and expiratory phases. Results indicate that the CNN-based approach achieves lower detection errors and reduced variability, supporting its potential for improving patient-ventilator synchrony.Clinical Relevance- Automated EMGdi detection enhances patient-ventilator synchrony assessment, enabling more personalized respiratory therapies.