Immunochromatographic SERS sensor-deep learning combined strategy for intelligent diagnosis of chronic kidney disease

A combined "immunochromatographic surface-enhanced Raman scattering (SERS) sensor-deep learning" detection strategy is proposed. Specifically, we first prepared an immunochromatographic SERS sensor with excellent surface enhancement capability for the specific recognition of neutrophil gelatinase-associated lipocalin (NGAL) molecules in serum, significantly enhancing the detection sensitivity of the target molecule. Subsequently, we integrated deep learning algorithms to perform preprocessing, feature extraction, and pattern recognition analysis on complex SERS spectral signals, overcoming the limitations of traditional algorithms in handling high-dimensional nonlinear data. This approach enables rapid, sensitive, and highly specific detection of NGAL. The platform achieves a detection limit as low as 0.1059 ng/mL with a broad linear detection range. For clinical validation, serum samples from 14 volunteers were analyzed. Six deep learning models were employed to classify the acquired clinical spectral data, with the Residual Network (ResNet) model achieving a classification accuracy of 98.81%, a loss value of only 0.0877, and an area under the receiver operating characteristic curve (AUC) of 99.97%. In addition, in the analysis of new samples, the classification and prediction of data from patients and normal individuals were successfully achieved. The proposed "immunoassay surface-enhanced Raman scattering sensor - deep learning" combined detection strategy has demonstrated extraordinary potential in the rapid, sensitive and specific detection of NGAL, providing crucial technical support for the intelligent discrimination of chronic kidney disease.