Viscoelastic Mechanical Signature-Based Machine Learning for Assessment of Liver Fibrosis and Its Regression

Abstract

Viscoelastic mechanical characteristics of the liver play a crucial role in fibrosis and cancer progression. Despite the evolution of liver fibrosis, as a pivotal step toward cirrhosis and hepatoma, originates from molecular dysfunctions and further extends to cellular and tissue levels, current clinical assessment is still largely based on tissue level stiffness. This study introduces a novel multiscale viscoelastic signature-based machine learning model for liver fibrosis evaluation. Utilizing a hybrid hierarchical theory-microrheology approach, we unveil a universal two-stage power-law rheology capturing dynamic mechanical variations in diverse liver conditions. Our analysis of multiscale viscoelastic disparities through a self-similar hierarchical theoretical framework enhances our understanding of fibrosis evolution. Distinct mechanical signatures observed across liver states provide valuable insights for assessing fibrotic individuals and treatment responses at different spatial scales. Furthermore, we propose a series of threshold values for each marker in the diagnosis of liver fibrosis. Notably, based on these new viscoelastic signatures, we eventually propose a Light Gradient Boosting Machine (LightGBM) diagnostic model that outperforms conventional stiffness-based classification, offering superior diagnostic precision for fibrotic treatment. This research contributes to the growing knowledge of viscoelastic characteristics in soft tissues and holds promise for innovative diagnostic strategies in various diseases and cancers.