An All‐in‐One Nanohole Array for Size‐Exclusive Trapping and High‐Throughput Digital Counting of Single Extracellular Vesicles for Non‐Invasive Cancer Screening

The analysis of single small extracellular vesicles (sEVs) could distinguish the heterogeneity of sEVs thus better extract tumor-related signatures. Current protocols for the analysis of single sEV rely mainly on the advanced techniques and require lengthy isolation procedures, limiting applications in clinical diagnosis. Herein, we developed a one-step procedure for rapid isolation of single sEVs from urine, along with an analytical pipeline for the diagnosis of early bladder cancer (BCa). Single sEVs are isolated by an EV-imprinted gold nanohole (EI-AuNH) array that selectively traps individual sEVs and spatially enhances their Raman spectra. After the invalid spectral data from incomplete or absent sEVs was eliminated using Smart-Filter, a convolutional neural network model identifies the origin of the spectra and generates a digital count matrix for each patient. By integrating the digital count data of both tumor-associated and normal sEVs, our model achieves an accuracy of 97.37% in early diagnosis of BCa. Feature extraction using explainable AI identified nine BCa-related signatures, with noticeable reduction on cholesterol and lipids in BCa-associated sEVs. These signatures could further distinguish BCa from other cancers. Overall, the present non-invasive and highly accurate diagnosis platform may revolutionize clinical disease diagnostics through simplified single sEV isolation and advanced modeling.