Qualitative and quantitative detection and identification of two benzodiazepines based on SERS and convolutional neural network technology. Issue 24 (25th November 2022)

• Record Type: Journal Article
• Authors: Sha, Xuanyu; Fang, Guoqiang; Cao, Guangxu; Li, Shuzhi; Hasi, Wuliji; Han, Siqingaowa
• Is Part Of: Analyst. Volume 147:Issue 24(2022)
• Page Start: 5785
• Page End: 5795
• Publication Date: 2022-11-25
• Subjects: Chemistry, Analytic -- Periodicals; 543
• DOI: 10.1039/d2an01277d
• Languages: English
• ISSNs: 0003-2654

Abstract

Abstract : CNN-SERS is used for the qualitative and quantitative analysis of benzodiazepines midazolam and diazepam. Deep learning can make full use of the information in the SERS spectrum and has advantages over traditional analysis strategies. Abstract : Drug abuse is a global social issue of concern. As the drug market expands, there is an urgent need for technological methods to rapidly detect drug abuse to meet the needs of different situations. Here, we present a strategy for the rapid identification of benzodiazepines (midazolam and diazepam) using surface-enhanced Raman scattering (SERS) combined with neural networks (CNN). The method uses a self-assembled silver nanoparticle paper-based SERS substrate for detection. Then, a SERS spectrum intelligent recognition model based on deep learning technology was constructed to realize the rapid and sensitive distinction between the two drugs. In this work, a total of 560 SERS spectra were collected, and the qualitative and quantitative identification of the two drugs in water and a beverage (Sprite) was realized by a trained convolutional neural network (CNN). The predicted concentrations for each scenario could reach 0.1–50 ppm (midazolam in water), 0.5–50 ppm (midazolam in water and diazepam in Sprite), and 5–150 ppm (diazepam in Sprite), with a strong coefficient of determination ( R 2 ) larger than 0.9662. The advantage of this method is that the neural network can extract data features from the entire SERS spectrum, which makes up for information loss when manually identifying the spectrum and selecting a limited number of characteristic peaks. This work clearly clarifies that the combination of SERS and deep learning technology has become an inevitable development trend, and also demonstrates the great potential of this strategy in the practical application of SERS.