One-Step Urinary EV Capture-to-SERS on a Temperature-Responsive AuEIH Substrate with Transformer-Based Urologic Cancer Classification

Urinary extracellular vesicles (EVs) are promising biomarkers for noninvasive diagnosis of urologic cancers, yet current workflows often require labor-intensive EV preisolation and multistep assays that limit clinical translation. Here we develop AuEIH, a temperature-responsive EV-imprinted hydrogel integrated with a monolayer AuNP array, enabling one-step urinary EV capture and in situ SERS profiling on the same substrate. At 25 °C, the boronic-acid-functionalized imprinted hydrogel selectively captures EVs from urine. Raising the temperature to 37 °C triggers hydrogel contraction, decreases AuNP interparticle gaps, and generates abundant plasmonic hot spots, thereby switching the substrate to a detection state for enhanced SERS acquisition at a physiological temperature. We evaluated AuEIH using 56 clinical urine samples (20 healthy volunteers; 12 bladder cancer, 12 prostate cancer, and 12 renal cancer). The platform achieved 100% accuracy in distinguishing cancer patients from healthy volunteers in this cohort. To further enable robust multiclass tumor typing from label-free spectra, we implemented a CNN-embedded Transformer model, which yielded accuracies of 98% (healthy), 98% (bladder), 94% (prostate), and 94% (renal). This capture-to-detection integrated AuEIH platform, coupled with attention-based spectral learning, provides a practical route toward a high-accuracy, noninvasive urologic cancer diagnosis from urinary EVs.