Enhancing multiparametric optical biosensing with AI: a Python-based FDTD/ML framework for spectral feature extraction and optimization

Optical sensor research faces significant limitations due to reliance on single-parameter measurements and scarce experimental data. This study introduces an advanced physics-based framework employing rigorous finite-difference time-domain (FDTD) simulations to analyze electromagnetic wave propagation in optical grating structures. A comprehensive dataset of 5000 simulations with 11 parameters is systematically generated, closely mimicking experimental conditions and addressing critical biosensor data gaps. Moving beyond traditional peak-wavelength analysis, multiple spectral features including FWHM, peak reflectance, and integrated spectral area are extracted using Python-based post-processing algorithms. Advanced data visualization reveals non-trivial sensitivity patterns, particularly highlighting enhanced performance for 100-nm analyte layers at n = 2.500. A machine learning (ML) approach, utilizing a multi-layer perceptron (MLP), establishes a new measurement paradigm, achieving exceptional prediction accuracy (R²=0.9992 for wavelength, 0.9546 for FWHM). This demonstrates that multi-parametric analysis significantly outperforms conventional methods. The methodology is extendable to diverse optical sensor architectures through feature engineering, and the publicly available datasets provide a foundation for future computational photonics and intelligent sensor design. This work innovatively integrates physics-based simulations, data generation/processing, advanced visualization, and ML, enabled by Python’s computational power and physical insights. In contrast to conventional AI applications in photonics that typically rely on experimental data optimization or limited numerical simulations, this study establishes a novel paradigm by integrating large-scale FDTD-generated datasets with machine learning, enabling comprehensive multi-parameter spectral analysis previously unattainable with traditional methods.