Asthma Detection Research Based on Voice Signal Processing and Machine Learning.

Area of Science:

• Biomedical Engineering
• Computational Linguistics
• Pulmonology

Background:

• Asthma diagnosis often relies on subjective assessments and invasive tests.
• Objective and non-invasive methods for early asthma detection are highly desirable.

Purpose of the Study:

• To develop and validate a machine learning-based approach for identifying asthma patients using voice signal analysis.
• To explore the efficacy of different voice features and machine learning models for asthma classification.

Main Methods:

• Collected voice signals from 50 asthma patients and 50 healthy controls.
• Performed multi-dimensional voice signal analysis using MATLAB, identifying significant differential phonetic features.
• Applied dimensionality reduction and utilized Support Vector Machine (SVM) and Random Forest (RF) models for classification.

Main Results:

• Identified over 400 voice feature indicators, with 20 showing significant differences (P < 0.01) between asthma patients and controls.
• Both SVM and RF models achieved 87% accuracy on the test set.
• SVM achieved an Area Under the Curve (AUC) of 0.95, and RF achieved 0.93, indicating strong classification performance.

Conclusions:

• Voice signal analysis combined with machine learning offers a promising non-invasive method for asthma detection.
• The SVM model demonstrated a potentially better balance between sensitivity and specificity.
• This approach provides a foundation for real-world application and optimization in early asthma detection.