Audio Cough Analysis by Parametric Modelling of Weighted Spectrograms to Interpret the Output of Convolutional Neural...

Area of Science:

• Medical Informatics
• Artificial Intelligence in Medicine
• Respiratory Medicine

Background:

• Persistent cough is a common symptom across various respiratory diseases.
• Analyzing cough patterns can provide valuable diagnostic information.
• Current methods for cough analysis may lack detailed interpretability.

Purpose of the Study:

• To investigate the feasibility of using eXplainable Artificial Intelligence (XAI) for analyzing cough patterns.
• To determine if XAI can differentiate between different respiratory disease groups based on cough characteristics.
• To explore the insights gained from XAI interpretability in respiratory health.

Main Methods:

• Collected 24-hour smartphone audio recordings from 20 adult patients with persistent cough.
• Applied frequency domain transformation to generate 1-second spectrograms.
• Utilized Convolutional Neural Networks (CNNs) for cough event detection and occlusion maps for feature highlighting.

Main Results:

• Quantitative analysis of spectrogram regions identified significant differences between patient groups.
• XAI highlighted distinctions in cough patterns between Chronic Obstructive Pulmonary Disease (COPD) patients and those with other respiratory pathologies.
• Interpretability analysis provided insights into cough-related variations across chronic and non-chronic respiratory conditions.

Conclusions:

• XAI methodologies are feasible for analyzing cough patterns in respiratory diseases.
• Interpretability of neural networks offers valuable insights into cough sound distinctions.
• This approach shows potential for differentiating between COPD and other respiratory conditions using cough analysis.