Flexible Microfluidic Surface-Enhanced Raman Scattering Devices Integrated with Interpretable Machine Learning for Metabolic Fatigue Monitoring

Abstract The real-time, noninvasive assessment of metabolic fatigue is fundamentally constrained by the challenge of capturing multiple dynamic biomarkers with high temporal resolution. Herein, we develop a fully integrated wearable surface-enhanced Raman scattering device that synergizes chronometric microfluidics with interpretable machine learning to decode sweat biochemistry for a precise fatigue assessment. This platform integrates a microfluidic interface that employs capillary burst valves to sequentially direct and compartmentalize freshly secreted sweat into discrete compartments. This physical compartmentalization effectively eliminates sample contamination inherent to conventional sensors, enabling the high-fidelity, time-resolved quantification of lactate, glucose, urea, and pH in sweat samples. By coupling these multiplexed metabolic profiles with a machine learning framework, we achieve accurate classification of fatigue states, with SHapley Additive exPlanations (SHAP) analysis identifying the synergy of lactate accumulation and glucose depletion as the dominant predictive signature. This work establishes a robust, data-driven pathway for personalized metabolic fatigue management and precision sports medicine.