NeuroPCF-ML: An Explainable Machine Learning Approach for Brain Tissue Characterization through Photonic Crystal Fiber Sensing
摘要
The early diagnosis of neurological disorders and the advancement of non-invasive clinical sensing both demand accurate brain tissue characterization, a task conventional optical techniques struggle to deliver due to limitations in sensitivity, stability, and interpretability. To address these gaps, we propose an approach that integrates PCF-based biosensing with machine learning, all within an explainable AI paradigm, to achieve highly accurate brain tissue analysis. The designed PCF incorporates a gold-coated plasmonic structure, numerically optimized using the Finite Element Method (FEM), to generate high-fidelity optical datasets. Six regression models including, Multiple Linear Regression (MLR), Decision Tree Regressor (DTR), Support Vector Regressor (SVR), Random Forest Regressor (RFR), Gradient Boosting Regressor (GBR), and XGBoost Regressor (XGBR) were evaluated to predict key optical parameters, including confinement loss (CLoss) and wavelength sensitivity (WS). Explainable AI methods, namely SHAP and LIME, were implemented to interpret global and local feature contributions, ensuring transparency and trustworthiness. Experimental outcomes demonstrate that the XGBR model achieved superior performance with