New label-free serum exosomes detection method based on hierarchical SERS substrate for diagnosis of pancreatic cancer using AI

Early diagnosis significantly enhances the 5-year survival rate of pancreatic cancer (PaC) patients. Obtaining information on molecular phenotypic changes in exosomes provides prospects for early non-invasive diagnosis of PaC. Unfortunately, current detection modes are time-consuming and still not sensitive enough, so methods that can directly obtain exosome information in complex biological fluids are urgently needed. In this study, we developed a new method for early diagnosis of PaC by obtaining a spectral set of serum exosomes on a hierarchical surface-enhanced Raman scattering (SERS) substrate. Then these spectra were analyzed with artificial intelligence (AI). Specifically, we designed a micro-lens array/silver nanowires/silver nanoparticles hierarchical SERS substrate (MLA/AgNWs/AgNPs H-SERS substrate) that exhibited a minimum detection concentration of 10 −9 M and a minimum relative standard deviation of 7.68 %. The performance of the substrate increased the strength and stability of exosome biological information acquisition. Furthermore, through the spectral analysis of exosome from 149 serum samples using AI, we performed PaCs diagnosis with an area under the receiver operating curve (AUROC) of 0.96 and successfully classified 24 cases of early PaCs. Moreover, the maximum diagnostic positive rate of 161 cases of non-pancreatic cancer was 4.44 %, supporting the fact that the model was specific. This label-free Raman spectral analysis can potentially be extended to identify multiple cancers, offering a non-invasive diagnostic approach for clinic. • Created a highly sensitive hierarchical-SERS substrate to get a subtle difference signal. • A deep learning model capable of distinguishing subtle difference of SERS spectrum was built. • The diagnostic model can accurately identify pancreatic cancer patients and has tumor specificity. • Complex SERS spectra from serum exosomes can be directly analyzed.