The detection of sweat metabolites is crucial for health monitoring, disease screening, and personalized medicine. Traditional methods encounter challenges like low metabolite concentrations, complex biological matrices, and difficulty in achieving multitarget simultaneous detection, limiting sensitivity, stability, and multiplexing capabilities. This study introduces an innovative, label-free surface-enhanced Raman spectroscopy (SERS) method integrated with machine learning (ML) algorithms, using a portable Raman spectrometer. For the first time, this method enables simultaneous quantitative detection of glucose, uric acid (UA), and lactate in real sweat, as well as classification of physiological states. Nanostructure-enhanced amplification boosts SERS sensitivity and accuracy, mitigating interference from complex biological matrices. Quantitative analysis and physiological state classification were performed using seven models, including K-Nearest Neighbors (KNN), Support Vector Machine (SVM), Convolutional Neural Networks (CNN), Deep Neural Networks (DNN), and other ML models. The KNN model achieved the best performance in metabolite detection, while the SVM model achieved 94.7% accuracy and a 94.5% F1 score in state classification. By integrating advanced ML techniques, this study significantly improves sensitivity, accuracy, and reliability in multitarget metabolite detection and physiological state classification, overcoming the limitations of traditional methods. This approach provides valuable data for health assessments, disease screening, exercise optimization, and personalized health management, advancing biosensing technologies for clinical and personalized medicine.
Zhu et al. (2025) studied this question.