Explainable deep learning combined with SERS for simultaneous detection of nicotine and etomidate isomers in...

The incorporation of illegal drugs into e-cigarettes poses significant threats to both human health and societal safety. Therefore, the development of accurate and rapid detection methods for illicit drug additives in e-cigarettes is of significant importance. This study integrates deep learning with surface-enhanced Raman spectroscopy (SERS) to develop a highly sensitive detection method for nicotine and four etomidate isomers. First, Raman and SERS detection were performed on five drug standards. Furthermore, density functional theory calculations were employed to analyze the characteristic peak vibration modes of the various substances. Additionally, a liquid-liquid extraction pretreatment method for commercially available e-cigarettes was developed, facilitating the simultaneous detection of nicotine and etomidate isomers. Trace detection was successfully achieved for all five substances. Subsequently, an interpretable deep network model was constructed for the classification and identification of these substances, attaining an accuracy of 97.8%. To further enhance the model's transparency and interpretability, gradient-weighted class activation mapping (Grad-CAM) was utilized to reveal the critical SERS peaks necessary for molecular discrimination. This research provides robust technical support for the rapid on-site screening of illicit drugs in e-cigarettes.