A Deep Learning Regression Model for Photonic Crystal Fiber Sensor With XAI Feature Selection and Analysis.

Area of Science:

• Photonics and Optical Sensing
• Biomedical Engineering
• Artificial Intelligence in Material Science

Background:

• Photonic Crystal Fibers (PCFs) offer unique light-guiding properties for sensing applications.
• Asymmetric Twin Elliptical Core PCF (TEC-PCF) designs enable multi-parameter sensing, including biological analytes.
• Accurate modeling of TEC-PCF optical design parameters is crucial for optimizing biosensor performance.

Purpose of the Study:

• To develop a Deep Learning Multi-output regression model for predicting TEC-PCF sensing performances.
• To investigate the relationships between optical design parameters and biosensing capabilities of TEC-PCF for blood glucose detection.
• To enhance model efficiency and interpretability using Explainable AI (XAI).

Main Methods:

• Utilized a Deep Learning Multi-output regression model trained on Finite Element Method (FEM) data.
• Employed Gretel.ai for synthetic data augmentation to improve training efficiency.
• Applied Explainable AI (XAI) with the Shapley Additive Explanations (SHAP) framework for feature selection and interpretability.
• Evaluated sensor performance metrics including effective index difference, transmission spectrum, coupling length, and sensitivity.

Main Results:

• The proposed model accurately predicts super modes for TEC-PCF biosensors across a specified wavelength range.
• The model demonstrates significantly faster prediction times compared to traditional numerical simulations (FEM).
• XAI facilitated feature selection, leading to an optimal model with reduced computational demands and enhanced interpretability.

Conclusions:

• A computationally efficient and interpretable deep learning model is presented for TEC-PCF biosensor design and analysis.
• The model accurately predicts sensing performance for blood glucose detection, considering hemoglobin influence.
• This approach accelerates the development and optimization of advanced photonic biosensors.