Machine learning assisted dual-modal SERS detection for circulating tumor cells

Detecting circulating tumor cells (CTCs) from blood has become a promising approach for cancer diagnosis. Surface-enhanced Raman Spectroscopy (SERS) has rapidly developed as a significant detection technology for CTCs, offering high sensitivity and selectivity. Encoded SERS bioprobes have gained attention due to their excellent specificity and ability to identify tumor cells using Raman signals. Machine learning has also made significant contributions to biomedical applications, especially in medical diagnosis. In this study, we developed a detection strategy combining encoded SERS bioprobes and machine learning models to identify CTCs. Dual-modal SERS bioprobes were designed and co-incubated with tumor cells by the "cocktail" method. An identification model for CTCs was constructed using principal component analysis (PCA) and the Random Forest classification algorithm. This innovative strategy endows SERS bioprobes with both effective magnetic separation and highly sensitive identification of CTCs, even at low concentrations of 2 cells/mL. It achieved a high detection rate of 98% for CTCs and effectively eliminated interference from peripheral WBCs. This simple and efficient strategy provides a new approach for CTCs detection and holds important significance for cancer diagnosis.