Functional localization and visualization of the subthalamic nucleus from microelectrode recordings acquired during DBS surgery with unsupervised machine learning.

Overview

Microelectrode recordings are a useful adjunctive method for subthalamic nucleus localization during deep brain stimulation surgery for Parkinson's disease. Attempts to quantitate and standardize this process, using single computational measures of neural activity, have been limited by variability in patient neurophysiology and recording conditions. Investigators have suggested that a multi-feature approach may be necessary for automated approaches to perform within acceptable clinical standards. We present a novel data visualization algorithm and several unique features that address these shortcomings. The algorithm extracts multiple computational features from the microelectrode neurophysiology and integrates them with tools from unsupervised machine learning. The resulting colour-coded map of neural activity reveals activity transitions that correspond to the anatomic boundaries of subcortical structures. Using these maps, a non-neurophysiologist is able to achieve sensitivities of 90% and 95% for STN entry and exit, respectively, to within 0.5 mm accuracy of the current gold standard. The accuracy of this technique is attributed to the multi-feature approach. This activity map can simplify and standardize the process of localizing the subthalamic nucleus (STN) for neurostimulation. Because this method does not require a stationary electrode for careful recording of unit activity for spike sorting, the length of the operation may be shortened.