Early cancer detection by serum biomolecular fingerprinting spectroscopy with machine learning

Label-free surface-enhanced Raman scattering (SERS) technique with ultra-sensitivity becomes more and more desirable in biomedical analysis, which is yet hindered by inefficient follow-up data analysis. Here we report an integrative method based on SERS and Artificial Intelligence for Cancer Screening (SERS-AICS) for liquid biopsy such as serum via silver nanowires, combining molecular vibrational signals processing with large-scale data mining algorithm. According to 382 healthy controls and 1582 patients from two independent cohorts, SERS-AICS not only distinguishes pan-cancer patients from health controls with 95.81% overall accuracy and 95.87% sensitivity at 95.40% specificity, but also screens out those samples at early cancer stage. The supereminent efficiency potentiates SERS-AICS a promising tool for detecting cancer with broader types at earlier stage, accompanying with the establishment of a data platform for further deep analysis.