Deep Learning-Assisted Cactus-Inspired Osmosis-Enrichment Patch for Biosafety-Isolative Wearable Sweat Metabolism Assessment

Sweat, which contains a rich array of biomarkers, serves as a vital biological fluid for non-invasive biosensing. Wearable sweat sensors have garnered significant interest owing to their portability and capacity for continuous monitoring. However, there are safety concerns regarding the direct contact of sweat sensors with the skin during the detection process. The chemical substances in the sensor patches may cause contamination of the epidermis when in contact with the skin, leading to skin allergic reactions. Sample collection and biosafety isolation are critical issues in wearable sweat detection. To address this, we develop a cactus-inspired biomimetic Janus membrane capable of unidirectionally transporting and concentrating sweat toward a designated detection zone. Through unidirectional transport from the hydrophobic layer to the hydrophilic layer of the Janus membrane, sweat droplets are enriched at the designated detection point of the conical hydrophilic pattern via Laplace pressure. The bionic osmosis-enrichment sensing patch effectively inhibits direct contact between indicators and skin, eliminating potential epidermal contamination. This achieved the effect of in situ perspiration collection under the premise of biosafety isolation. To rapidly and accurately analyze sweat biomarkers, we employ a deep learning (DL)-assisted fluorescence sensor for efficient and precise detection of biomarker concentrations. A dataset of 4500 fluorescence images are constructed and used to evaluate two DL and seven machine learning (ML) algorithms. The convolutional neural network (CNN) model could easily and accurately classify and quantitatively analyze the total concentration of the amino acid mixture, Ca2+ and Cl−, with 100% classification accuracy. The consistency between the detection results of actual sweat by the DL-assisted fluorescence method and fluorescence spectroscopy was 91.4–96.0%. This approach demonstrates high reliability in sweat collection and analysis, offering a practical tool for clinical health monitoring, early disease intervention, and diagnosis.