A dataset of Solicited Cough Sound for Tuberculosis Triage Testing

Area of Science:

• Medical research
• Artificial Intelligence
• Public Health

Background:

• Cough is a prevalent but often overlooked symptom of lung diseases.
• Tuberculosis (TB) remains a leading global infectious killer, with current screening methods lacking accuracy.
• Existing artificial intelligence (AI) models for respiratory conditions have limited application in TB diagnosis.

Purpose of the Study:

• To compile a comprehensive, multi-country database of cough sounds for TB research.
• To facilitate the development of accurate, point-of-care AI-driven triage tools for TB detection.
• To support innovative approaches for improving TB diagnosis and combating the pandemic.

Main Methods:

• Collected over 700,000 cough sounds from 2,143 individuals undergoing TB evaluation across multiple countries.
• Included detailed demographic, clinical, and microbiological diagnostic data alongside cough recordings.
• Established a large-scale dataset to train and validate AI models for cough sound analysis.

Main Results:

• A substantial, diverse dataset of cough sounds and associated patient data has been created.
• The database provides a foundation for developing AI models capable of analyzing cough characteristics for TB.
• This resource aims to significantly advance the field of AI-assisted respiratory diagnostics for TB.

Conclusions:

• The compiled cough sound database is a critical resource for advancing AI-based TB diagnostic tools.
• Development of accurate, point-of-care AI solutions can enhance global TB screening efforts.
• This initiative is vital for improving TB diagnosis and addressing the ongoing global pandemic.