Machine Learning-Enabled Surface-Enhanced Raman Scattering for Label-Free Tracking of Hepatic Fibrosis Progression

Abstract

Accurate monitoring of liver fibrosis (LF) is critical for the clinical management and prognostic evaluation of liver diseases. However, it remains challenging to develop a reliable and efficient method for tracking LF progression and identifying its etiological characteristics. Herein, we present an innovative strategy combining label-free surface-enhanced Raman scattering (SERS) technology and advanced machine learning (ML) for the etiological tracking and staging diagnosis of LF. Specifically, a highly sensitive, uniform Au cluster-structured nanoarray (AuCSNAs) substrate was fabricated via reactive ion etching (RIE) and gas–liquid interface self-assembly, obtaining high-quality serum SERS spectra from carbon tetrachloride (CCl4)-, dimethylnitrosamine (DMN)-, and thioacetamide (TAA)-induced LF mice at weeks 0, 3, and 6. Subsequently, the principal component analysis (PCA)-Meta-learning Random Forest-Lite (Meta-RF-Lite) model was constructed to extract, identify, and analyze these spectra. The PCA-Meta-RF-Lite model realized robust discrimination of SERS spectra from distinct LF etiologies and pathological stages, which can reach an accuracy of up to 98.3%, a sensitivity of up to 100.0%, a specificity of up to 98.7%, and an AUC of up to 0.989. Collectively, these results verify that the AuCSNAs/PCA-Meta-RF-Lite strategy enables reliable tracking and etiological differentiation of LF.