Transformer-driven Ag@5/Ag-Cu@20/PSS SERS Biosensor for Early Diagnosis of Hepatobiliary Diseases

Accurate and sensitive detection of bilirubin, cholic acid, and Alanine aminotransferase (ALT) is essential for the early detection of hepatobiliary diseases in healthcare and the medical field. Integrating surface-enhanced Raman scattering (SERS) with deep learning algorithm has shown great application prospect in the early identification of markers of hepatobiliary diseases. Here, an regular Ag decorated Cu-Ag nanoisland-liked SERS biosensor (Ag@5/Ag-Cu@20/PSS) was constructed by using pattered sappire substrate as a template. An additional 5 mins of Ag nanofilm significantly improved the SERS signal intensity and homogeneity, and the sensitivity was enhanced by two orders of magnitude up to 10^-10 M for 4-aminothiophenol (4-ATP). Importantly, deep learning algorithms included temporal convolutional networks (TCN), convolutional neural networks (CNN), recurrent neural network (RNN) and Transformer were used to deeply mine SERS spectral data of bilirubin, cholic acid and ALT, extract effective features, and establish an accurate classification and prediction model. Results indicating that the developed SERS biosensor combined with Transformer model demonstrated excellent accuracy (ACC) of 0.9816±0.0090 in qualitatively and quantitatively analyzing of bilirubin, cholic acid, and ALT. The successful application of the SERS biosensor will help improve the accuracy of early identification of bilirubin, cholic acid, and ALT disease markers and promote the early intervention and treatment of hepatobiliary diseases.