Deep Learning‐Powered Nanoplasmonic Biosensing Approach Enables Ultrasensitive Extracellular Vesicles Profiling for Cancer Screening

Nanoplasmonic metasurface technology, known for its high sensitivity, has garnered significant attention in the field of cancer detection. However, its potential is currently hindered by the inefficient data processing and analysis of conventional biosensing approaches. Herein, a biosensing strategy based on the Kolmogorov-Arnold network (KAN)-enabled metasurface chip (metaEVchip) for ultrasensitive small extracellular vesicles (sEV) analysis in serum is proposed. By analyzing full-spectrum data from 600 pancreatic ductal adenocarcinoma (PDAC) patients and 1200 controls via KAN-powered deep learning nanoplasmonic biosensing, the strategy achieves an exceptional area under the curve (AUC) of 0.99 in an external validation set, outperforming traditional methods. Further exploration of this enhanced performance reveals KAN's mechanism for the simultaneous capture of multi-dimensional spectral features, an advantage that enables efficient data processing and accuracy. This advancement significantly expands the applicability of nanoplasmonic metasurfaces in biosensing and establishes a new paradigm for cancer screening and improved clinical management of multiple malignancies.