Machine Learning analysis of human pulmonary extracellular matrix architecture identifies disease-specific remodeling patterns

ABSTRACT

Background: Normal organ function is critically dependent on an intact three-dimensional architecture. Structural abnormalities induced by pathological situations instruct cells to behave abnormally and promoting disease progression oftentimes leading to organ failure. Current approaches do not allow for high-resolution (HR) threedimensional (3D) visualisation and analysis of human organ structure. Method(s) Here, we develop a method to perfuse human tissue segments to remove cells and study the 3D structural scaffold, which could be applied to any organ. Our approach enables HR-3D imaging of organ architecture, which we apply to study healthy and diseased human lung, specifically emphysema, usual interstitial pneumonia, pulmonary sarcoidosis, and COVID-19. Result(s) Our imaging reveals major structural abnormalities previously unseen by existing methodologies. Furthermore, we identify disease-specific patterns of structural remodelling using machine learning, including the altered spatial relationship between extracellular matrix (ECM) proteins collagen type IV, elastin and fibrillar collagen present across all diseases. Conclusion(s) Given the importance of organ structure on function, our approach opens the possibility to understand human physiology in a new way, which may assist in future disease diagnosis and explain the detrimental pulmonary effects of the diseases studied here.