Gastric Cancer Diagnosis by Deep Learning Based on Variation Trend of Exosomal SERS Signals

Gastric cancer with high incidence requires effective diagnostic methods. Existing machine learning diagnosis methods based on plasma Surface-Enhanced Raman Scattering (SERS) intensity are usually affected by factors such as Raman measurement instruments and reagents, which result in poor accuracy and adaptability. Given the great potential of exosome-based liquid biopsies for gastric cancer diagnosis, herein, we proposed a diagnosis approach by deep learning based on variation trend of exosomal SERS signals that can identify underlying interpretable factors in complex spectra. Then our approach was compared with existing common machine learning diagnosis methods. The results demonstrated that our approach can improve the performance of most machine learning models. Additionally, it can achieve the optimal Area Under Curve (AUC) (97.29% ± 0.97%) in deep learning networks. Overall, deep learning techniques combined with the variation trend of exosomal SERS signals can provide insights for gastric cancer diagnosis.