(12) Computationally Efficient Method to Predict Axonal Pathway Activation with Deep Brain Stimulation

Introduction: Deep brain stimulation (DBS) of the subthalamic nucleus (STN) is a well-established treatment for Parkinson’s disease motor symptoms. However, optimizing DBS settings to target specific motor signs, such as gait dysfunction, remains challenging. This study investigated methods to differentially activate pedunculopontine tegmental nucleus (PPTg) efferents and afferents that course through or around the STN, given their potential impact on balance and gait. Previous studies have shown the practical utility of using the so-called modified driving function (MDF) to predict pathway activation using a weighted second spatial difference of extracellular voltages. Here, we showed an additional modification to improve the fit between MDF and multi-compartment axonal activation thresholds, streamlining multi-objective optimization for targeting DBS more precisely in and around the STN.

Methods: Patient-specific finite element brain models and pathway activation models were developed from pre-operative high-field 7T MRI and post-operative CT scans across three participants with bilateral DBS implants. Three axonal pathways were modeled: PPTg efferent (1 µm diameter), PPTg afferent (2 µm), and internal capsule (IC) fibers (5.7 µm). The field-cable model thresholds for each pathway were calculated in the NEURON environment across four pulse widths (30, 60, 90, and 120 µsec) and monopolar settings on each DBS lead contact. We used a log-log transformation between NEURON thresholds and MDF thresholds, providing a more accurate linear fit compared to other nonlinear fitting methods. Additionally, we calculated a polynomial surface fit to identify the relationship between maximal MDF along an axon, NEURON threshold, and stimulation pulse width. We integrated this fit into a multi-objective optimization algorithm to identify electrode configurations and pulse widths that optimize activation of PPTg efferents, PPTg afferents, and IC.

Results: The results showed a consistent linear relationship between maximum MDF and activation thresholds for pathways with the same axon diameter and pulse width across all hemispheres. While slopes remained consistent across pulse widths, intercepts varied. The polynomial surface fit, with an R-squared value greater than 0.95, was integrated into the optimization algorithm, successfully identifying electrode configurations and pulse widths that maximize the disparity between activating PPTg efferents and afferents, while minimizing IC activation

Conclusion: This study presents a computationally efficient approach for optimizing STN-DBS settings on a patient-specific basis using a simple fitting adjustment to the relationship between MDF and NEURON-based activation thresholds and incorporating a polynomial surface fit across pulse widths. These findings enable more rapid personalization of DBS programming algorithms without extensive computational resources.