Cough sound analysis - a new tool for diagnosing pneumonia

Area of Science:

• Medical Technology
• Pediatrics
• Artificial Intelligence in Healthcare

Background:

• Childhood pneumonia causes over 1.8 million deaths annually, particularly in remote regions lacking diagnostic facilities.
• Accurate diagnosis of pediatric respiratory illnesses is challenging in resource-limited settings due to limited access to imaging and laboratory tests.
• A scarcity of trained healthcare workers further complicates timely and reliable diagnosis of pneumonia in children.

Purpose of the Study:

• To introduce a novel, non-contact technology for the automated analysis of cough and respiratory sounds.
• To develop and validate a diagnostic tool for childhood pneumonia using machine learning on cough sound features.
• To address the critical need for accessible diagnostic solutions for acute respiratory illnesses in underserved areas.

Main Methods:

• Collected cough sounds from 91 pediatric patients suspected of acute respiratory illness using non-contact microphones.
• Extracted mathematical features from the recorded cough sounds.
• Trained and validated a Logistic Regression classifier using clinical diagnoses as the gold standard.

Main Results:

• The automated cough sound analysis method demonstrated a sensitivity of 94% and a specificity of 75% in differentiating pneumonia from other respiratory diseases.
• The classifier achieved accurate separation of pneumonia based solely on cough sound parameters.
• The developed technology shows significant potential for improving diagnostic capabilities.

Conclusions:

• Automated analysis of cough sounds presents a pioneering, field-deployable solution for diagnosing childhood pneumonia.
• This non-contact method can overcome limitations in remote regions, revolutionizing pneumonia management.
• The technology offers a scalable and accessible approach to improve pediatric respiratory healthcare globally.