Wide-range and multi-target trace-level quantification of renal biomarkers in serum via a chromatography-SERS-AI...

Trace blood analysis (TBA) holds irreplaceable value in chronic disease management, elderly care monitoring, infant disease screening, anemia emergency care, and home-based dynamic monitoring. However, existing TBA technologies typically suffer from narrow quantitative ranges, poor batch-to-batch consistency, and single-target detection. Although surface-enhanced Raman scattering (SERS) technology is suitable for trace detection, SERS-based TBA has consistently failed to cover the full clinical concentration range for kidney disease.

We present a label-free, wide-dynamic-range SERS platform for multi-target trace-level quantification of renal indicators in 20 μL serum samples. A chromatographic paper-SERS substrate integrating separation, in-situ enrichment, and SERS enhancement of the target analytes is developed to resolve matrix interference, competitive adsorption, and timeliness issues without complex pretreatment. A ResUNet-MT multi-task deep learning algorithm with three independent regressors is established to correct concentration-dependent quantitative biases. As a result, the platform achieves quantitative ranges of 0.016-1 mM for serum creatinine, 0.063-4 mM for serum uric acid, and 1.563-100 mM for blood urea nitrogen, covering clinical concentrations from healthy individuals to various kidney disease stages. Correlation coefficients are 0.96, 0.99, and 0.99, with average recoveries of 103.49%, 104.53%, and 101.37%, and average coefficients of variation of 4.41%, 2.06%, and 0.17%-all meeting clinical detection standards. The clinical translation potential of this platform has been validated using clinical serum samples from patients with different renal biomarker levels.

This work overcomes key limitations of existing TBA and SERS technologies, providing a foundation for the development of wide-range, multi-target, accurate, and cost-effective TBA technologies.