On-site quantitative detection of fentanyl in heroin by machine learning- enabled SERS on superabsorbing metasurfaces

Abstract The global surge in opioid misuse, particularly fentanyl, presents a formidable public health challenge, highlighted by increasing drug-related mortalities. Our study introduces a novel approach for on-site quantitative detection of fentanyl in heroin, employing machine learning-enabled surface-enhanced Raman spectroscopy (SERS) on superabsorbing metasurfaces. The metasurface enables superior light absorption (> 90%) across a broad wavelength range (580–1100 nm). This architecture facilitates significant electromagnetic field enhancement, over 2.19 × 107, ensuring high sensitivity, uniformity, and reproducibility. Our method precisely captured SERS signals across a detection range of 1 to 100 µg/mL in fentanyl solutions, fentanyl-heroin mixtures, and fentanyl-spiked saliva, demonstrating its versatility and practical utility. Incorporation of partial least squares regression into our analysis achieved over 93% accuracy in concentration predictions, eliminating the need for pre-data processing or specialized personnel. This marks a key advancement in rapid, accurate fentanyl detection, aiding the fight against the opioid crisis and improving public health safety.