Detecting paroxysmal coughing from pertussis cases using voice recognition technology.

Area of Science:

• Medical technology
• Computational biology
• Pediatrics

Background:

• Pertussis (whooping cough) is a highly contagious respiratory illness requiring prompt identification for outbreak control.
• Classic pertussis symptoms include coughing fits followed by gasps and a "whooping" sound, but many clinicians lack experience with these presentations.
• Difficulty in identifying early pertussis cases can hinder effective outbreak management.

Purpose of the Study:

• To develop and evaluate voice-recognition software for distinguishing pertussis coughs from other respiratory sounds in children.
• To create a tool that assists clinicians in the early and accurate diagnosis of pertussis.

Main Methods:

• Collected and manually categorized audio recordings of pertussis, croup, and other coughs from children.
• Extracted Mel-frequency cepstral coefficients (MFCCs) to create 39-element feature vectors for each cough.
• Employed machine learning algorithms including Neural Networks, K-Nearest Neighbor (KNN), and Random Forest (RF) for classification.

Main Results:

• The study analyzed 16 non-pertussis and 31 pertussis cough recordings.
• Voice-recognition models achieved over 90% accuracy in classifying pertussis coughs.
• Error rates varied among algorithms, with the Neural Network showing 7% Type I and 8% Type II errors.

Conclusions:

• A robust cough classifier can be developed using voice-recognition technology.
• This technology shows promise in assisting clinicians and the public to identify pertussis in children.
• Automated cough analysis offers a potential solution for improving pertussis diagnosis and control.