Objective: This study aims to evaluate a novel, machine learning approach to volume of activation (VTA) prediction for real-time use in the clinical setting and to compare its accuracy to contemporary methods.

Background: Deep Brain Stimulation (DBS) is used to treat a variety of neurological and movement disorders. DBS programming aims to stimulate therapeutic and spare side-effect tissue, however programmers cannot directly determine which areas are being stimulated. Display of VTA is one solution to aid in visualization of stimulation area.

Full VTA models are computationally expensive and not apt for use in clinical systems. A variety of compressed models, such as the Blum two-variable lookup table [1], and Howell driving-force predictor [2,3] were developed to support rapid VTA display, and prior commercial software supported rapid VTA for select leads and settings. Concurrently, directional DBS leads with multiple independent current control (MICC) and greater stimulation parameter flexibility have been developed – demanding new compressed models which account for additional complexity but remain appropriate for use in the clinic.

Method: Our group devised, implemented, and evaluated a novel method using machine learning to select relevant features of stimulation(E.g activating function) to create an approximation algorithm trained(1200 configurations) on full models using the MRG model and McIntyre & Butson methods [4].  Both linear and a variety of directional leads were used to train and evaluate the method. The method was compared to contemporary approximation methods, using full biophysical models (NEURON) as the ground truth.

Results: The proposed method was both very sensitive (>95%)[figure 1] and specific (>99%)[figure 2] across tested amplitudes (2mA – 9mA) and faithfully matched the shape of neuronal modeling VTAs (volume error < 4% at all amplitudes).

Conclusion: The novel method was found to be sufficiently accurate and portable for use in computationally limited systems – such as those used in the clinic.

References: 1.Blum, DA. (SfN, 2014). Novel Activation Volume Methodology for Deep Brain Stimulation. 2.Howell et al 2019 – A Driving-Force Predictor for Estimating Pathway Activation in Patient-Specific Models of DBS 3.Noecker etal 2020 – StimVision v2; Examples and Applications in STN DBS for PD 4.Butson CR, McIntyre CC. Current steering to control the volume of tissue activated during deep brain stimulation. Brain Stimul. 2008;1(1):7-15.

« Back to MDS Virtual Congress 2021

MDS Abstracts - https://www.mdsabstracts.org/abstract/a-machine-learning-approach-to-vta-prediction-in-dbs/