Artificial intelligence assisted label-free surface-enhanced Raman scattering detection of early-stage cancer-derived exosomes based on g-C3N4/Ag hybrid substrate prepared by electro-synthesis

Early surveillance and differentiation of cancer possesses remarkable clinical merit, but no effective protocol exists to reliably recognize multiple categories of early-stage cancers. In the present research, we investigate the viability in label-free inspection of cancers by combining surface-enhanced Raman spectroscopy (SERS) with principal component analysis (PCA)-deep learning (DL) to analyze exosomes. Based on a simple two-step electrosynthetical strategy, g-C3N4/Ag 3 N 4 /Ag hybrid substrate with chemical enhancement ability and homogeneous electromagnetic hotspots was developed to directly collect the SERS spectra of various cancer cell-original exosomes (prostate, colon, and bladder) with high intensity and perfect resolution. Furthermore, a PCA-DL algorithm was constructed to delicately analyze the SERS spectra with similar characteristic peaks. Particularly, these cancer- derived exosomes were successfully identified with an accuracy of over 97.5 % and a specificity of over 97.6 % after capturing several key spectral features. In conclusion, the present work not only develops a promising SERS sensor for cancer surveillance, but also exploits a novel method to evaluate the spectral characteristics of exosomes.