Multiplexed nanophotonic biosensing and deep learning-driven protein quantification for traumatic brain injury diagnosis at the point of care.

Abstract

Traumatic brain injury (TBI) triage and monitoring demand rapid, sensitive, and deployable biomarker assays. We present an AI-integrated nanophotonic biosensor enabling ultrasensitive multiplexed quantification of S100B, GFAP, and UCH-L1 across diverse human biofluids. The sensing substrate consists of polystyrene single chains formed <i>via</i> liquid-confinement self-assembly and functionalized with antibodies. Selenium nanoparticles are pre-incubated with specimens to form SeNP-antigen complexes that co-localize on the chains, generating high-contrast elastic scattering. A standardized four-zone layout employs hydrophilic/hydrophobic patterning to passively isolate reaction domains without physical barriers. A deep learning pipeline facilitates device-agnostic quantification from images captured using various microscopy systems, including professional setups and smartphone-based adaptations. Validation with 195 clinical specimens from 75 individuals (TBI patients and controls) spanning serum, urine, saliva, and cerebrospinal fluid demonstrated a detection limit of 1 pg mL^-1 and strong agreement with ELISA (<i>R</i>² > 0.93) for all biomarkers, with consistent performance across imaging modalities. The workflow is completed in ∼30 min and requires only a smartphone in portable modes, demonstrating a proof-of-concept for point-of-care neurotrauma diagnostics and highlighting a potential pathway toward AI-assisted decentralized TBI management.