Research on a rapid and accurate diagnosis platform for liver fibrosis based on machine learning-assisted SERS technology

Rapid and accurate diagnosis of liver fibrosis (LF) is essential for timely clinical intervention, yet traditional diagnostic methods are frequently constrained by low sensitivity and limited accuracy, representing a major obstacle to early liver disease screening. In this work, we present a label-free, non-invasive detection strategy that integrates surface-enhanced Raman spectroscopy (SERS) technology with an optimized machine learning (ML), principal component analysis (PCA)-K-means-+. In detail, Au colony-like nanoarray substrate (AuCLNAs) was fabricated as a SERS active platform to acquire high-quality SERS spectra of serum from LF mice. The PCA-K-means-+ algorithm was then employed to construct and train a classification model. The results demonstrate that the substrate exhibits excellent uniformity, stability, and SERS enhancement. The PCA-K-means-+ model effectively extracted key spectral features, achieving robust classification with an accuracy of 99.0%, a sensitivity of 98.8%, a specificity of 100%, and an area under the curve (AUC) of 0.992. These findings highlight the significant potential of this SERS technology combined with the PCA-K-means-+ model in identifying subtle spectral variations at different stages, offering a promising and reliable tool for the precise clinical diagnosis of LF.