Exploratory study of extracellular matrix biomarkers for non-invasive liver fibrosis staging: A machine learning approach with XGBoost and explainable AI.

BACKGROUND

Novel circulating markers for the non-invasive staging of chronic liver disease (CLD) are in high demand. Although underutilized, extracellular matrix (ECM) components offer significant diagnostic potential. This study evaluates ECM-related markers in hepatitis C virus (HCV)-positive patients across varying fibrosis stages.

METHODS

Sixty-eight patients with mild-to-moderate fibrosis (F1-F2), sixty-six with advanced fibrosis (F3-F4), and thirty healthy donors were recruited. Inclusion criteria were detectable HCV-RNA and no other liver diseases or co-infections. Levels of ECM markers-hyaluronic acid (HA), laminin (LN), collagen-III N-peptide (PIIIP N-P), collagen-IV (C-IV)-along with cholylglycine (CG) and Golgi protein-73 (GP73), were measured in serum using the MAGLUMI 800 CLIA platform.

RESULTS

Levels of LN, HA, C-IV, PIIIP N-P (p < 0.001), and GP73 (p < 0.01) increased from controls to F1-F2 and F3-F4. CG levels were higher in pathological subjects compared to controls (p < 0.001), but no significant differences emerged between fibrosis stages. These trends persisted after adjusting for age and sex. A multivariate ordinal regression identified LN, PIIIP N-P, and C-IV as promising markers, with an accuracy of 0.77. An XGBoost model improved accuracy to 0.87 and enhanced other metrics. SHAP analysis confirmed these variables as key contributors to the model's predictions.

CONCLUSION

This study underscores the potential of ECM biomarkers, particularly LN, PIIIP N-P, and C-IV, in non-invasively staging CLD. Furthermore, our preliminary data suggest that a machine learning approach, combined with explainable AI, could further enhance diagnostic accuracy, potentially reducing the need for invasive biopsies.