Abstract: Surface-enhanced Raman spectroscopy(SERS)enables highly sensitive and specific label-free detection by capturing the vibrational signatures of molecular chemical bonds. It offers significant advantages and broad application potential in the analysis of biological samples. However, its clinical translation remains limited due to challenges such as spectral complexity, variability, limited reproducibility, and overlapping characteristic peaks. Machine learning(ML), which allows computational models to learn patterns from data and make informed predictions or decisions, has shown great promise in deciphering complex SERS spectra and advancing their biomedical applications. This review highlights recent advances in the integration of ML with SERS, detailing how ML algorithms enhanced analytical performance. It also outlines common data preprocessing techniques for SERS, describes core ML workflows, and examines their roles in classification, quantitative analysis, and disease diagnosis. Furthermore, the review explores emerging applications such as molecular structure prediction, Raman spectral database development, and discrimination between DNA and RNA. Finally, the current challenges and future directions of ML-assisted SERS are discussed, with an emphasis on improving robustness, interpretability, and clinical applicability.

Key words: SERS, machine learning, biomedicine, feature extraction, disease diagnosis