Artefact Detection in Impedance Pneumography Signals: A Machine Learning Approach

Area of Science:

• Biomedical Engineering
• Respiratory Medicine
• Signal Processing

Background:

• Ambulatory impedance pneumography is crucial for monitoring respiratory diseases.
• Noisy signals in ambulatory recordings can significantly impair data-driven decision support tools.
• Accurate identification and removal of artefacts are essential for reliable respiratory monitoring.

Purpose of the Study:

• To evaluate the effectiveness of machine learning algorithms in distinguishing clean from noisy bio-impedance signals.
• To compare the performance of a heuristic algorithm, a Support Vector Machine (SVM), and a Convolutional Neural Network (CNN) for artefact detection.
• To assess the added value of data-driven approaches in improving the quality of respiratory impedance data.

Main Methods:

• A dataset of 47 chronic obstructive pulmonary disease patients undergoing inspiratory threshold loading was used.
• Respiration was recorded simultaneously using a bio-impedance device and a spirometer (gold standard).
• Signals were annotated for artefacts by four experts based on the reference spirometer data. Three methods were compared: heuristic algorithm, SVM, and CNN.

Main Results:

• Both machine learning approaches (SVM and CNN) demonstrated significantly higher accuracy (87.77% ± 2.64% and 87.20% ± 2.78%, respectively) compared to the heuristic method (84.69% ± 2.32%).
• No significant performance difference was found between the SVM and CNN models.
• The area under the curve (AUC) for the feature-based (SVM) and neural network (CNN) models were 92.77% ± 2.95% and 92.51% ± 1.74%, respectively.

Conclusions:

• Data-driven approaches, specifically machine learning algorithms like SVM and CNN, are highly beneficial for artefact detection in respiratory thoracic bio-impedance signals.
• These advanced methods offer superior performance over traditional heuristic techniques for ensuring signal quality in ambulatory monitoring.
• The findings support the integration of machine learning for robust artefact removal, enhancing the reliability of respiratory monitoring systems.