Liver Viscosity Dominates Liver Stiffness in MASLD: A Multiparametric Ultrasound Study With IPTW Adjustment and Explainable Machine Learning

Liver stiffness measured by shear wave elastography (SWE) is widely used as a noninvasive surrogate for fibrosis in metabolic dysfunction-associated steatotic liver disease (MASLD). However, stiffness may be influenced by multiple biomechanical factors beyond fibrosis, and the relative contributions of inflammation-related viscosity, steatosis, and metabolic parameters remain unclear.

In this prospective study, 573 adults with MASLD underwent multiparametric ultrasound, including SWE, viscosity plane-wave ultrasound (ViPLUS), and attenuation (ATT). Participants were stratified into lower-body mass index (BMI) (BMI < 24 kg/m²) and higher-BMI (BMI ≥ 24 kg/m²) groups. Covariate balance was achieved using inverse probability of treatment weighting (IPTW). Multivariable linear regression, subgroup analyses, and interaction testing were performed to identify independent determinants of liver stiffness. Machine learning models (extreme gradient boosting [XGBoost]) with Shapley Additive Explanations (SHAP) were used for exploratory feature attribution to identify key determinants of liver stiffness rather than to develop a standalone clinical diagnostic or prediction tool.

After IPTW adjustment, prespecified clinical covariates were well balanced between BMI groups. Hepatic viscosity (ViPLUS) was the strongest independent determinant of liver stiffness (β = 5.80, 95% confidence interval: 5.33-6.27; p < 0.001), corresponding to an approximately 0.58 kPa increase in SWE per 0.1 Pa·s increase in ViPLUS. The association remained dominant in both BMI groups and was greater in the higher-BMI group (approximately 0.63 vs. 0.51 kPa per 0.1 Pa·s). The contribution of viscosity was consistently dominant across lower-BMI and higher-BMI subgroups and appeared stronger in the higher-BMI group. SHAP-based feature attribution identified ViPLUS as the most influential model-derived feature associated with SWE, consistent with the regression findings. Nonlinear associations were observed for aspartate aminotransferase/alanine aminotransferase ratio and high-density lipoprotein cholesterol, while triglycerides and glucose showed no significant nonlinear relationships. Notably, a subset of lower-BMI individuals exhibited high-risk hepatic phenotypes despite lower BMI.

Hepatic viscosity is the dominant determinant of liver stiffness in MASLD, with even a 0.1 Pa·s increase in ViPLUS corresponding to an approximately 0.58 kPa increase in SWE. These findings suggest that liver stiffness reflects viscosity-related viscoelastic alterations rather than fibrosis alone. Multiparametric ultrasound, particularly the integration of viscosity and stiffness, may provide incremental value for identifying high-risk individuals, including those with lower BMI.