Micropore array-based SERS sensor assisted by convolutional neural networks for subsurface biotoxic-free detection of interstitial fluid in bioassays.

Accurate and sensitive detection of small-molecule metabolites such as uric acid, glucose, and lactic acid is critical in biomedical diagnostics and clinical applications. Traditional detection methods often face limitations such as complex procedures, prolonged processing times, and insufficient sensitivity. To address these challenges, we developed a micropore array-based surface-enhanced Raman scattering (SERS) sensor, assisted by a convolutional neural network (CNN), for biotoxic-free detection of interstitial fluid (ISF) in bioassays. The sensor employs a 3D-printed design with optimized micropore configurations of extraction micropores and sensing spaces, which enhance liquid extraction rates while minimizing interference from SERS substrates. The integrated CNN efficiently processes Raman spectra, enabling accurate identification of individual and mixed components. The sensor demonstrated a detection limit of 10 M for methylene blue, with relative standard deviation (RSD) values of approximately 7 %, ensuring high sensitivity and stability. Calibration curves for uric acid, glucose, and lactic acid exhibited excellent linearity (R ≈ 0.99). For multi-component samples, the CNN-assisted sensor achieved a classification accuracy exceeding 99.38 %, effectively identifying and quantifying components in complex mixtures. Practical validation on pig skin demonstrated the sensor's capability for minimally invasive, in situ detection of ISF analytes. This study highlights the potential of the micropore array-based SERS sensor as a robust, biocompatible platform for real-time biomedical detection and multi-component analysis. Its innovative integration of advanced sensing technology with machine learning paves the way for future advancements in non-invasive diagnostics and precision medicine.