Cough Classification of Unknown Emerging Respiratory Disease with Federated Learning

Area of Science:

• Artificial Intelligence
• Machine Learning
• Epidemiology

Background:

• Pandemic scenarios require rapid diagnostic testing, but data privacy concerns hinder large-scale medical data collection.
• Federated Learning (FL) offers a solution by enabling distributed model training without direct data access, preserving user privacy.

Purpose of the Study:

• To propose and evaluate a Federated Learning-based approach for classifying cough sounds of an unknown emerging disease in a hypothetical pandemic scenario.
• To address the challenge of collecting representative medical data while maintaining data security and privacy.

Main Methods:

• Developed a federated cough classifier algorithm using FL to distinguish a novel disease's cough from existing ones.
• Trained the model on 2535 cough samples from nine COVID-19 patients, with edge devices communicating gradient information to a central server model 77 times.

Main Results:

• The federated cough classifier achieved a 45% Matthews correlation coefficient (MCC) for COVID-19 classification.
• An overall MCC classification performance of 69% was attained, demonstrating effective multiclass classification.

Conclusions:

• The proposed FL-based approach successfully learns to classify the cough of an unknown disease in a privacy-preserving manner.
• This method holds potential for rapid, secure diagnostic testing during future pandemic events.