An Interpretable SERS–AI Platform for Rapid and Quantitative Diagnosis of Polymicrobial UTIs: Powered by Positively Charged Plasmonic Nanoparticles and Attention‐Based Deep Learning

Abstract Polymicrobial urinary tract infections (UTIs) present diagnostic challenges due to overlapping symptoms and limitations of conventional methods. Although SERS and AI have shown potential for microbial diagnostics, existing approaches lack reproducibility, quantification capability, and interpretability—especially in complex clinical samples. Here, a label‐free, interpretable SERS‐AI platform for rapid identification and quantification of mixed urinary tract pathogens is proposed. A plasmonic substrate is engineered by combining Au@Ag core–shell nanoparticles with a positively charged bPEI surface, enabling electrostatic bacterial capture and stable SERS signal generation across diverse microbial mixtures. A convolutional neural network (CNN) enhanced with a convolutional block attention module (CBAM) to enable both accurate classification (95.8%, AUC = 0.9774) and reliable bacterial proportion prediction (R 2 = 0.9112), surpassing traditional models, is developed. Importantly, the attention mechanism offers mechanistic interpretability, highlighting biologically relevant spectral features related to nucleic acids, proteins, and virulence factors. Validation with clinical urine samples demonstrates strong predictive performance (accuracy = 86.9%, R 2 = 0.8626), supporting real‐world applicability. Overall, this work not only delivers a high‐throughput and explainable framework for polymicrobial diagnostics, but also contributes to the mechanistic understanding of Raman‐based microbial phenotyping, paving the way for clinical deployment and microbiome‐informed interventions.