Exploring Longitudinal Cough, Breath, and Voice Data for COVID-19 Progression Prediction via Sequential Deep...

Area of Science:

• Digital Health
• Biomedical Signal Processing
• Machine Learning in Healthcare

Background:

• Current COVID-19 audio analysis focuses on one-off detection, neglecting disease progression and recovery monitoring.
• Longitudinal tracking of COVID-19 progression via audio biomarkers can provide critical insights for timely treatment adjustments and healthcare resource management.

Purpose of the Study:

• To explore the potential of sequential deep learning techniques for predicting COVID-19 progression and recovery trends using longitudinal audio samples.
• To develop and validate a deep learning-enabled tool for continuous COVID-19 monitoring.

Main Methods:

• Crowdsourced respiratory audio data (cough, breathing, voice) from 212 individuals over 5-385 days were analyzed alongside COVID-19 test results.
• A deep learning model utilizing gated recurrent units (GRUs) was developed to track audio biomarker dynamics for disease progression prediction.
• The study involved COVID-19 detection and longitudinal disease progression prediction, evaluating performance using AUROC, sensitivity, specificity, and correlation coefficients.

Main Results:

• Longitudinal audio biomarker analysis demonstrated effectiveness in COVID-19 detection, achieving an AUROC of 0.79, sensitivity of 0.75, and specificity of 0.71.
• The predicted disease progression trajectory showed high consistency with longitudinal test results (correlation of 0.75 in the test cohort and 0.86 in a recovery subset).
• Findings indicate that monitoring COVID-19 evolution through longitudinal audio data is promising for tracking disease progression and recovery.

Conclusions:

• An audio-based system using deep learning effectively monitors COVID-19 progression and recovery trends, showing high consistency with test results.
• This technology holds potential for post-pandemic healthcare, aiding treatment guidance and resource allocation.
• The framework demonstrates the applicability of telemonitoring for respiratory diseases, offering a flexible, affordable, and timely solution.