The Role of MER Processing Pipelines for STN Functional Identification During DBS Surgery: A Feature-Based Machine...

Area of Science:

• Neurosurgery
• Biomedical Engineering
• Computational Neuroscience

Background:

• Microelectrode recording (MER) is crucial for validating subthalamic nucleus (STN) targeting in deep brain stimulation (DBS) for Parkinson's Disease (PD).
• Machine learning (ML) models are increasingly used to enhance STN localization from MER data.
• The influence of preprocessing steps on ML classifier performance for MER data remains underexplored.

Purpose of the Study:

• To systematically evaluate the impact of various preprocessing pipelines on the accuracy of ML-based STN localization using MER data.
• To identify optimal artifact removal, outlier handling, and feature normalization strategies for MER data preprocessing.
• To assess the interpretability of ML models using SHAP analysis to understand feature importance.

Main Methods:

• Investigated 24 distinct preprocessing pipelines combining artifact removal (COV, BCK), outlier handling (ORH), and feature normalization techniques.
• Evaluated pipeline performance using three ML models (Random Forest, Support Vector Machine, Logistic Regression).
• Utilized SHAP (SHapley Additive exPlanations) analysis to determine feature importance across different preprocessing configurations.

Main Results:

• Artifact rejection (COV, BCK) and optimized outlier management (ORH) consistently improved classification performance.
• Hemisphere-specific feature normalization prior to classification led to performance degradation across all evaluated metrics.
• The Random Forest model, applied to data preprocessed with COV artifact rejection and ORH outlier management, achieved the highest accuracy (0.945).

Conclusions:

• Effective artifact rejection and outlier treatment are critical for accurate MER-based STN identification in DBS surgery.
• Preliminary feature normalization strategies can potentially impair ML model performance in this context.
• SHAP-based interpretability provides valuable insights for refining ML pipelines and developing robust protocols for MER-guided DBS targeting.