Response-Function Driven and Consistency-Gated High-Reliability Urine-Based Cancer Screening via Surface-Enhanced Raman Scattering (SERS) Biosensing

Abstract Surface-enhanced Raman scattering (SERS) is attractive for trace-level analysis in complex systems owing to its nondestructive readout and molecular fingerprint information. However, direct and reliable SERS analysis remains challenging because signals from trace analytes are readily obscured by matrix interference and competitive adsorption, complicating analysis in complex biofluids. Here, we introduce a SERS sensing approach that amplifies spectral differences through a novel response function, which enhances inter-sample separability and facilitates subsequent classification. As a practical demonstration, we apply this framework to urine-based cancer detection using three sensing molecules functionalized on silver nanorods (AgNRs). The sensing molecules report urine components through spectral variations of their own functional groups, while controlled experiments with representative urinary-related molecules revealed receptor-dependent response patterns across the three sensing interfaces, indicating partially shared but non-identical molecular-response information. Response functions derived from these responses are used to establish separate CNN-MLP models to generate receptor-specific cancer/healthy predictions, which are combined through a three-principal-substrate consistency gating strategy. In a dataset of 226 urine samples, the method achieved 92.0% screening coverage with 100% accuracy within the covered set. This work provides a practical response-function-based SERS framework for extracting molecular-response information from complex samples and integrating multi-receptor responses for reliable classification.