Microelectrode Signal Dynamics as Biomarkers of Subthalamic Nucleus Entry on Deep Brain Stimulation: A Nonlinear Feature Approach

TL;DR

This study introduces a nonlinear and entropy-based feature framework for accurately classifying microelectrode signals to improve intraoperative localization of the subthalamic nucleus during deep brain stimulation in Parkinson's disease, demonstrating high accuracy and robustness.

Contribution

It presents a novel quantitative approach using nonlinear dynamics and entropy metrics for real-time neural signal classification during DBS surgery, outperforming traditional methods.

Findings

Entropy and nonlinear features provided the best classification performance.

The Extra Trees classifier achieved a cross-validated F1-score of 0.902.

Robust generalization was confirmed with a hold-out test set (F1=0.922).

Abstract

Accurate intraoperative localization of the subthalamic nucleus (STN) is essential for the efficacy of Deep Brain Stimulation (DBS) in patients with Parkinson's disease. While microelectrode recordings (MERs) provide rich electrophysiological information during DBS electrode implantation, current localization practices often rely on subjective interpretation of signal features. In this study, we propose a quantitative framework that leverages nonlinear dynamics and entropy-based metrics to classify neural activity recorded inside versus outside the STN. MER data from three patients were preprocessed using a robust artifact correction pipeline, segmented, and labelled based on surgical annotations. A comprehensive set of recurrence quantification analysis, nonlinear, and entropy features were extracted from each segment. Multiple supervised classifiers were trained on every combination of feature domains using stratified 10-fold cross-validation, followed by statistical comparison using paired Wilcoxon signed-rank tests with Holm-Bonferroni correction. The combination of entropy and nonlinear features yielded the highest discriminative power, and the Extra Trees classifier emerged as the best model with a cross-validated F1-score of 0.902+/-0.027 and ROC AUC of 0.887+/-0.055. Final evaluation on a 20% hold-out test set confirmed robust generalization (F1= 0.922, ROC AUC = 0.941). These results highlight the potential of nonlinear and entropy signal descriptors in supporting real-time, data-driven decision-making during DBS surgeries