Decoding Spectral Perturbations from a SERS Artificial Nose in Serum for Early Lung Cancer Detection

Serum biomarkers for early cancer detection often suffer from limited sensitivity and specificity due to the biochemical complexity of blood. Here, we report a surface-enhanced Raman scattering (SERS) artificial nose that integrates a set of chemically distinct molecular receptors to generate multidimensional spectral responses to human serum. These receptors exhibit large Raman cross sections and well-defined vibrational signatures, enabling high signal-to-noise readouts and subtle yet reproducible spectral perturbations upon exposure to serum components. By applying machine-learning analysis to the resulting multivariate Raman patterns, we extract a diagnostic molecular fingerprint capable of distinguishing early-stage non-small-cell lung cancer (NSCLC) from controls. Using an optimized multireceptor array, the model achieves 100% sensitivity at 98% specificity, markedly outperforming conventional serological biomarkers and imaging-based screening approaches. This work establishes a chemically tunable SERS artificial nose as a powerful strategy for serum-based cancer detection and highlights the potential of multiplexed, receptor-driven sensing for disease-associated metabolic phenotyping.