A Non-Invasive Drug Use Screening Using Machine Learning and Spectral Trace Elements

Abstract

Diffuse reflectance spectroscopy exploits multiple scattering and absorption of light in superficial tissue to capture characteristic spectral signatures. When combined with machine learning, this approach enables accurate, real-time, non-invasive detection. Conventional drug screening methods, based on biochemical assays of urine, blood, or hair are limited by their invasiveness, long turnaround time, high cost, and poor portability that hinder their use in rapid, large-scale applications. To address these limitations, we propose a fast, efficient, and high-accuracy non-invasive drug screening framework that integrates diffuse reflectance spectroscopy with sex-stratified machine learning models. A skin spectral trace element database was established using data collected from over two thousand individuals. Two sequential classification pipelines were developed: support vector machines for male subjects and an Extreme Gradient Boosting for female subjects. In a case study focused on heroin detection, both models achieved high classification accuracy and demonstrated strong performance in terms of the area under the receiver operating curve. This approach provides a practical solution for preliminary drug screening and demonstrates the promise of combining optical spectroscopy with artificial intelligence for real-world drug surveillance.