Interpretable machine learning models for COPD ease of breathing estimation | JoVE Visualize

Area of Science:

Biomedical Engineering
Digital Health
Respiratory Medicine

Background:

Chronic obstructive pulmonary disease (COPD) is a major global health concern, significantly impacting patient quality of life and increasing mortality rates.
Remote monitoring technologies offer potential for improving COPD management by reducing exacerbations and enhancing patient well-being, though further research is needed.
Current remote monitoring methods for COPD often lack non-intrusive, accurate, and interpretable ways to assess breathing difficulties.

Purpose of the Study:

To introduce a novel, non-intrusive remote monitoring method for estimating breathing ease changes in COPD patients using wearable sensor data.
To evaluate the effectiveness of different physiological signals, feature engineering techniques, and classification models for predicting breathing changes.
To enhance clinical adoption by incorporating interpretable machine learning techniques for signal analysis.

Main Methods:

Collected physiological data from COPD patients using wearables, including respiratory airflow, acceleration, audio, and bio-impedance signals.
Developed and compared various signal processing and machine learning models, analyzing the impact of signal selection and feature engineering.
Implemented novel saliency map methods and imputation techniques for time-series data to improve model interpretability and faithfulness.

Main Results:

Acceleration and audio signals were identified as the most effective for monitoring breathing ease in COPD patients.
The optimal classification model achieved a high F1-score of 0.83, demonstrating strong predictive performance.
Novel interpretability methods were developed, providing insights into signal importance and improving the faithfulness of explanations.

Conclusions:

The developed wearable-based system offers a promising non-intrusive method for remote monitoring of breathing ease in COPD patients.
The integration of interpretable AI enhances the potential for clinical translation and trust in the monitoring system.
This approach can contribute to improved quality of life and reduced exacerbations for individuals living with COPD through effective remote management.