Electronic cigarettes have rapidly gained acceptance recently. Nicotine-containing electronic cigarette liquids (e-liquids) are prohibited in some countries, but are permitted and simply available online in others. A rapid detection method is therefore required for on-site inspection or screening of a large amount of samples. Our previous study demonstrated a surface-enhanced Raman scattering (SERS)-based approach to identify nicotine-containing e-liquids; without any pre-treatment, e-liquid can be directly tested on our solid-phase SERS substrates, made of silver nanoparticle arrays embedded in anodic aluminium oxide nanochannels (Ag/AAO). However, this approach required manual determination of spectral signatures and negative samples should be validated in the second round detection. Here, after examining 406 commercial e-liquids, we refined this approach by developing artificial intelligence (AI)-assisted spectrum interpretations. We also found that nicotine and benzoic acid can be simultaneously detected in our platform. This increased test sensitivity because benzoic acid is usually used in nicotine salts. Around 64% of nicotine-positive samples in this study showed both signatures. Using either cutoffs of nicotine and benzoic acid peak intensities or a machine learning model based on the CatBoost algorithm, over 90% of tested samples can be correctly discriminated with only one round of SERS measurement. False negative and false positive rates were 2.5-4.4% and 4.4-8.9%, respectively, depending on the interpretation method and thresholds applied. The new approach takes only 1 microliter of sample and can be performed in 1-2 min, suitable for on-site inspection with portable Raman detectors. It could also be a complementary platform to reduce samples that need to be analyzed in the central labs and has the potential to identify other prohibited additives.
Keywords: Benzoic acid; Electronic cigarette liquid; Machine learning; Nicotine; SERS.
Copyright © 2023 Elsevier B.V. All rights reserved.