AI-Enhanced Nanophotonic Heterochain Sensor Enables Multiplexed Biomarker Detection across Serum, Urine, and Saliva for Stroke Differentiation

Background: Rapid and accurate identification of stroke subtype is critical for timely intervention, yet current diagnostic assays are limited by long turnaround times, dependency on centralized laboratories, and insufficient sensitivity in the ultra-early stage. Methods: We developed a nanophotonic heterochain biosensing platform integrated with deep learning−assisted image analysis for multiplex detection of S100B, GFAP, and UCH-L1 in serum, urine, and saliva. The system consists of antibody-functionalized PS@Se heterochain biosensor, portable smartphone-based imaging, and a cloud-enabled analysis pipeline. Analytical validation was performed using recombinant protein standards, followed by clinical evaluation in ischemic stroke, intracerebral hemorrhage, and healthy controls. Results: The biosensor achieved picogram-level sensitivity with strong linearity (overall R2 = 0.9583), high recovery (104.7%), and acceptable reproducibility (CV = 24.4%). Clinical validation demonstrated significant elevations of biomarkers in patient groups compared to controls, with distinct profiles between ischemic and hemorrhagic stroke. Integration of biomarker panels using logistic regression improved diagnostic performance over individual assays, enabling accurate discrimination of stroke subtypes within a clinically actionable window. Conclusion: This portable, AI-integrated biosensing system enables ultra-early, multiplexed, and quantitative detection of brain injury biomarkers, with potential to transform point-of-care stroke diagnostics in both clinical and field settings.