Wearable Devices and Explainable Unsupervised Learning for COVID-19 Detection and Monitoring.

Area of Science:

• Artificial Intelligence in Healthcare
• Wearable Technology for Disease Monitoring
• Infectious Disease Surveillance

Background:

• Global healthcare systems face challenges with ongoing COVID-19 infections, even in vaccinated populations.
• Existing Artificial Intelligence (AI) solutions for disease monitoring often rely on supervised learning, facing issues with data annotation and reliability.
• Adolescents and young adults (AYA) remain a demographic susceptible to COVID-19 challenges.

Purpose of the Study:

• To propose and evaluate an innovative unsupervised AI framework for detecting and monitoring COVID-19 infections using smartwatch data.
• To provide cost-effective and accessible solutions for disease screening, diagnosis, and monitoring.
• To leverage interpretable clusters and language models for enhanced data pattern insights.

Main Methods:

• Utilized longitudinal smartwatch data (heart rate, heart rate variability, step count) from volunteers.
• Developed an unsupervised learning framework to identify normal and abnormal physiological measures.
• Employed Davinci GPT-3 language model for enhanced interpretation of data patterns and relationships.

Main Results:

• The unsupervised framework achieved a Silhouette score of 0.55, demonstrating effective clustering.
• Validation using supervised learning yielded high performance metrics: accuracy (0.884), precision (0.80), and recall (0.817).
• The study successfully identified potential inflammatory markers through unsupervised techniques.

Conclusions:

• Unsupervised learning techniques show significant potential for efficient and reliable COVID-19 detection and monitoring.
• AI and wearable devices offer scalable, low-cost solutions for health monitoring, particularly for inflammatory diseases.
• This approach opens new avenues for accessible and widely applicable health monitoring systems.