Integrating Machine Learning and Physics-Based Flow Models for Population-Level Respiratory Disease Simulation

This study presents a novel framework for simulating respiratory disease transmission across diverse populations using advanced machine learning and fluid-based reduced-order modeling. Our approach leverages computational efficiency to integrate a wide range of facial shapes, mask sizes, and fluid dynamics, modeling the intricate interactions between facial movement, mask fit, and varying dynamic conditions. We represent the space between the face and mask as interconnected channels with porous boundaries and imposed compatibility conditions to accurately predict airflow leakage patterns. By incorporating facial deformations linked to specific phonemes, we analyze how different speech scenarios affect mask efficacy, providing a nuanced understanding of how verbal communication impacts leakage. We will then compare these results with our breathing simulations inside a large cohort of subjects, quantifying the differential impact of these two distinct respiratory events on mask leakage patterns. Finally, we discuss how this methodology contributes to the identification of more effective mitigation strategies for respiratory disease transmission.