Deep learning-assisted SERS platform for label-free detection of celecoxib in serum using ag@Pt@porous silicon Bragg mirror composite substrate

Celecoxib (CXB) is a non-steroidal anti-inflammatory drug used to prevent and treat arthritis. However, overdosing or improper use can cause adverse reactions. SERS is an important tool for therapeutic drug monitoring (TDM).The present study proposes an innovative detection scheme based on SERS deep learning, aiming to achieve qualitative identification and quantitative detection of CXB at the Raman spectrum level. In this paper, a composite SERS substrate consisting of silver (Ag)@ platinum (Pt)@ porous silicon Bragg mirror (PSB) was synthesized. Crystal violet (CV) was used as probe to characterize the substrate properties, and the substrate showed excellent SERS properties and long-term stability. The substrate significantly enhanced the SERS response capacity to CXB, showing excellent reproducibility and high sensitivity. The LOD and the recovery rate were both within the reasonable range when the substrate was used to detect CXB concentration in serum. Three deep learning models, GoogleNet, ResNet and VGG, were used to classify five kinds of blood drug concentrations. Among them, the GoogleNet model based on spectral dataset achieved the highest classification accuracy of 84.17 %. The deep learning-SERS integrated system supports blood drug concentration discrimination, helping doctors personalize treatment strategies for arthritis patients and improving clinical efficiency.