Early diagnosis of transient ischemic attack facilitated by SERS-based artificial intelligence sensors

Transient ischemic attack (TIA) serves as a critical early warning sign for ischemic stroke. Its timely identification holds significant clinical value in reducing recurrence risk and improving patient prognosis. However, existing detection methods exhibit limitations in sensitivity and specificity. This study developed a surface-enhanced Raman scattering (SERS) detection platform based on a lotus leaf (LL)/copper (Cu)/silicon dioxide (SiO₂)/silver (Ag) multilayer film. By coupling Interleukin-6 (IL-6) coating antibodies (Ab₁), it achieved highly specific detection of inflammatory markers in serum. We collected serum SERS spectra from TIA patients and healthy controls (HC), then performed feature extraction and classification analysis using six intelligent algorithm models: Light Gradient Boosting Machine (LightGBM), Artificial Neural Network (ANN), One-Dimensional Convolutional Neural Network (1D-CNN), Bidirectional Long Short-Term Memory (BiLSTM), Transformer and AlexNet with four convolutional layers (AlexNet4). The results showed that the sensor was able to accurately distinguish between TIA and healthy samples, with the LightGBM model achieving an accuracy of 99.64 % and the other models exceeding 90.78 %. Predictions on independent unlabeled sample showed high consistency with clinical diagnose, validating the sensor's robustness and generalization capability. This study demonstrates the feasibility of combining SERS with deep learning for TIA identification and highlights its potential applicability in future studies on early stroke diagnosis.