Statistical SERS spectroscopy based on concentration-encoded intensity distributions for deep-learning-assisted quantitation

Reliable quantitative SERS analysis is limited by strong intensity fluctuations arising from stochastic molecule-substrate-light interactions. Here, we reveal that SERS intensity distributions carry concentration-dependent statistical information, exhibiting a characteristic evolution from an asymmetric long-tailed profile at low concentration to a nearly symmetric Gaussian-like pattern at high concentration. This concentration-dependent evolution originates from stochastic molecular adsorption at plasmonic “hot spots” and can be reproduced by Poisson-statistics-based Monte Carlo simulations. On this basis, we propose a statistical SERS strategy for concentration-level quantitation, in which concentration-encoded intensity distribution patterns are identified using a ResNet4 deep learning model. Accurate concentration identification of R6G is achieved, and the generality of this method is further demonstrated by the analysis of methamphetamine in methanol, thiram in bean sprout extract, and rosiglitazone maleate in serum. These results transform intrinsic SERS intensity fluctuations from a limitation of conventional intensity-based quantitation into useful statistical fingerprints, which certainly deepens our understanding about the stochastic nature of SERS and provides a new route toward reliable SERS quantitation for chemical and biological analysis.