AI-Assisted SERS Decoding of Dual Enzymatic Activity for Intraoperative Chemical Mapping of Epilepsy

Abstract

Precision resection remains the definitive treatment for drug-resistant epilepsy, yet intraoperative demarcation of epileptic foci is hindered by the lack of rapid molecular imaging. Herein, we report a rapid, artificial intelligence (AI)-accelerated surface-enhanced Raman scattering (SERS) platform for the simultaneous mapping of caspase-3 and matrix metalloproteinase-2 (MMP-2) activities at epileptic foci. To overcome the kinetic mismatch that traditionally stalls dual-enzyme tracking, we employed an AI-driven workflow to identify an MMP-2 peptide substrate whose catalytic kinetics are synchronized with those of its caspase-3 counterpart, compressing measurement times from hours to minutes. A bespoke dual-branch cross-attention Kolmogorov–Arnold Network (DBC-KAN) further accelerated the diagnostic speed by capturing subtle early-stage spectral evolution patterns through learnable edge activations, achieving 98.21% predictive accuracy within 4 min of sample loading, a >7-fold improvement relative to commercial activity assays (30–60 min). Notably, these substrates were integrated into a ratiometric SERS array featuring biological-silent-region internal standards, enhancing quantitative robustness while eliminating spectral cross-talk. Leveraging this platform, we established a logic-gated diagnostic framework (caspase-3/MMP-2 ratio) that successfully discriminated epileptic foci from healthy and inflamed regions in transgenic murine models and resected human cortical tissues. This AI-driven SERS strategy delivers ultrafast intraoperative molecular profiling, providing a high-fidelity roadmap for precision neurosurgery.