Evaluation and classification of fentanyl-related compounds using EC-SERS and machine learning

Area of Science:

• Analytical Chemistry
• Forensic Science
• Spectroscopy

Background:

• Conventional methods like GC-MS and LC-MS for drug screening are costly and time-consuming.
• Raman spectroscopy offers a faster, more affordable alternative for analyzing chemical compounds.
• Electrochemical surface-enhanced Raman scattering (EC-SERS) significantly enhances signal detection for low-concentration analytes.

Purpose of the Study:

• To evaluate fentanyl-related compounds and other drugs of abuse using EC-SERS.
• To process EC-SERS spectral data with machine learning, specifically convolutional neural networks (CNNs).
• To assess the accuracy of CNN models in identifying drugs in both simple and complex mixtures.

Main Methods:

• Utilized EC-SERS for rapid spectral acquisition of drug samples.
• Developed and trained a CNN model using Keras and TensorFlow.
• Validated the CNN model with in-house binary mixtures and authentic forensic case samples.

Main Results:

• The CNN model achieved an overall accuracy of 98.4% with 10-fold cross-validation.
• Correct identification rates were 92% for in-house binary mixtures and 85% for authentic case samples.
• The model demonstrated superior discrimination of drugs in multicomponent mixtures compared to traditional methods.

Conclusions:

• Machine learning, particularly CNNs, significantly enhances the accuracy of EC-SERS for drug screening.
• This integrated approach offers a powerful tool for analyzing seized drug materials, including complex mixtures.
• The method provides a rapid, cost-effective, and accurate solution for forensic identification of controlled substances.