Developing a multi-feature fusion model for exacerbation classification in asthma and COPD

Area of Science:

• Respiratory Medicine
• Biomedical Engineering
• Data Science

Background:

• Exacerbations significantly impact chronic respiratory diseases like COPD and asthma.
• Vocal changes during respiratory distress suggest speech analysis as a monitoring tool.

Purpose of the Study:

• To develop a remote, speech-based method for automatic detection of exacerbations in COPD and asthma patients.

Main Methods:

• Utilized optimized Mel-Frequency Cepstral Coefficients (MFCC) and multi-domain acoustic features for speech analysis.
• Employed SHapley Additive exPlanations (SHAP) for model transparency and identification of key speech biomarkers.

Main Results:

• Achieved high classification performance with 0.892 accuracy and 0.955 AUC on the TACTICAS dataset.
• MFCC-related and energy features were identified as key contributors to exacerbation detection via SHAP analysis.

Conclusions:

• Speech-based monitoring shows significant potential for real-world respiratory condition management.
• The proposed method surpasses existing approaches, offering efficient exacerbation diagnosis and monitoring.