Deep learning optimized dual-analyte detection-based biosensor for monitoring pregnancy stage using a urine sample

This study presents a hybrid deep learning (DL) approach for designing and optimizing a photonic crystal fiber (PCF)-based surface plasmon resonance (SPR) multi-analyte biosensor. Using a simulation based on the finite element method (FEM), we generated a comprehensive data set that captures various sensor parameters and refractive index (RI) values. A hybrid recurrent neural network long-short-term memory (RNN-LSTM) model was developed to predict confinement loss (CL), which showed superior performance with an MSE of 0.0014, an MAE of 0.0188, and an R 2 of 0.9510 compared to other DL and machine learning (ML) models. The proposed model shows a maximum amplitude sensitivity (AS) of 3102.41 RIU −1 , a wavelength sensitivity (WS) of 10,000 nm/RIU, and a sensor resolution (SR) of 1 × 10 −5 . The effectiveness of the model was validated through extensive analysis, including ablation studies and SHAP-based explainability analysis. Our findings highlight the potential of DL to improve multi-analyte biosensor design and performance prediction.