A novel surface-enhanced raman scattering strategy integrating spectral and mapping data for stable and rapid...

Surface-enhanced Raman scattering (SERS) holds great potential for aflatoxin B1 (AFB1) monitoring, but its reliability is hampered by inconsistent hotspot uniformity and signal fluctuations from single-spot measurements. To overcome these limitations, this study proposed a strategy combining substrates with high hotspot uniformity and multi-dimensional data fusion to enhance detection stability. Three-dimensional (3D) gold nanotrees were fabricated via electrodeposition, which served as SERS substrates for mapping signal generation. Quantitative analysis incorporated both one-dimensional (1D) spectral data and two-dimensional (2D) SERS mapping, with normalization applied to eliminate dimensional discrepancies. The fused data was modeled through variable combination population analysis-iteratively retaining informative variables-partial least squares (VCPA-IRIV-PLS) with exploration in exponentially decreasing function and binary matrix sampling. The supervised learning VCPA-IRIV-PLS algorithm compresses the variable space holistically while preserving key information through local iterations, enabling stable and accurate prediction. The results showed a prediction correlation coefficient of 0.9641. Compared to non-fusion strategies, the fusion strategy improved the relative prediction deviation by 14.5 % and reduced the relative standard deviation of recovery by 6.28 %. The optimized model achieved rapid prediction within 30-50 ms, with a 10 pg mL^-1 detection limit and a 98.70-104.10 % recovery for AFB1 in wheat. This integrated mapping-spectra SERS approach shows potential for stable and rapid AFB1 quantification in complex food matrices.