A Comprehensive Approach for Classification of the Cough Type

Area of Science:

• Biomedical Engineering
• Respiratory Medicine
• Acoustic Signal Processing

Background:

• Sputum detection in the lungs is crucial for diagnosing conditions like lung infections, pneumonia, and cancer.
• Traditional methods for classifying cough type (dry/wet) rely on subjective and often inaccurate clinical physical examinations.

Purpose of the Study:

• To develop an objective, acoustic-feature-based approach for classifying cough sounds.
• To improve the accuracy and reliability of dry versus wet cough classification.

Main Methods:

• Collected a dataset of 5971 cough sounds (5242 dry, 729 wet) from 131 subjects using smartphones.
• Utilized a novel multi-layer labeling platform for cough sound annotation.
• Achieved high inter-rater agreement scores (0.81 for the first layer).

Main Results:

• Developed a cough classification system based on acoustic features.
• Achieved high classification performance with 88% sensitivity and 86% specificity for distinguishing between wet and dry coughs.
• The achieved performance is reported as the highest in existing literature.

Conclusions:

• The proposed objective approach using smartphone-recorded cough acoustics offers a significant improvement over traditional subjective methods.
• This technology has the potential to enhance early and accurate detection of lung diseases.