A machine learning-driven SERS platform for precise detection and analysis of vascular calcification

Abstract

Vascular calcification (VC) significantly increases the incidence and mortality rates of cardiovascular diseases, severely threatening public health as a global issue. Currently, there are no effective methods to prevent and treat vascular calcification. This study proposes a machine learning-assisted surface-enhanced Raman scattering (SERS) technique for label-free, highly sensitive analysis of VC rat serum. We prepared gold nanobipyramid (GNBP) substrates using seed-mediated and liquid–liquid interface self-assembly methods and measured the SERS spectra of the serum. The collected spectral data were processed using a Principal Component Analysis (PCA)-Linear Discriminant Analysis (LDA) model to achieve effective sample differentiation. In this analysis model, GNBP substrates enabled rapid, sensitive, and label-free serum spectral detection, achieving classification accuracy, sensitivity, and specificity of 96.0%, and an AUC value of 0.98, significantly outperforming currently used machine learning methods. By analyzing the PCA loading plots, key spectral features that distinguished VC were successfully captured. This study demonstrates that combining SERS technology with machine learning provides a new method and foundation for real-time diagnosis and identification of VC, showcasing the significant advantages of GNBP substrates in high-sensitivity and high-specificity detection, potentially improving the early diagnosis and treatment of VC significantly.