Predicting Pulmonary Function From the Analysis of Voice: A Machine Learning Approach

Area of Science:

• * Biomedical Engineering
• * Artificial Intelligence in Medicine
• * Respiratory Medicine

Background:

• * Current asthma self-monitoring tools like spirometers require specialized equipment and have low patient adherence.
• * Voice recordings present a non-invasive, accessible alternative for assessing lung function.
• * This study explores the potential of machine learning (ML) to analyze voice data for asthma management.

Purpose of the Study:

• * To develop and evaluate ML models for predicting lung function and its abnormality severity in asthma patients using voice recordings.
• * To compare the performance of different ML algorithms (Random Forest, Support Vector Machine, linear regression) for these prediction tasks.
• * To establish the feasibility of using voice analysis as a surrogate for traditional lung function tests.

Main Methods:

• * A dataset of 323 voice recordings was analyzed, with a mechanism to separate speech and breathing sounds.
• * Features from voice recordings were combined with biological factors for ML model development.
• * Three types of ML models were trained and validated: regression for lung function, multi-class classification for severity, and binary classification for abnormality detection.

Main Results:

• * The Random Forest (RF) regression model achieved the lowest root mean square error (10.86) for predicting lung function.
• * For multi-class classification of lung function severity, the Support Vector Machine (SVM) model yielded the highest accuracy (73.20%).
• * The RF model demonstrated superior performance in binary classification for detecting abnormal lung function, with an accuracy of 85%.

Conclusions:

• * ML models utilizing voice recordings can effectively predict lung function and its abnormality in asthma patients.
• * The developed voice-based prediction methods show superior performance compared to previously published approaches.
• * This technology holds significant potential for developing accessible telehealth solutions, including smartphone applications, to enhance asthma self-management and clinical decision-making.