Validation of Parametric Models in Microelectrode Recordings Acquired from Patients with Parkinson’s Disease

• Published in: Applied Computer Sciences in Engineering pp. 323 - 334
• Main Authors: Roldan-Vasco, Sebastian, Restrepo-Agudelo, Sebastian, Lopez-Rios, Adriana Lucia, Hutchison, William D.
• Format: Book Chapter
• Language: English
• Published: Cham Springer International Publishing
• Series: Communications in Computer and Information Science
• Subjects: Deep brain stimulation (DBS); Microelectrode recordings (MER); Parametric study; Parkinson’s disease; Signal modeling; Subthalamic nucleus
• ISBN: 3030310183; 9783030310189
• ISSN: 1865-0929; 1865-0937
• DOI: 10.1007/978-3-030-31019-6_28

The Deep Brain Stimulation (DBS) has become one of the most common therapies applied to patients with Parkinson’s disease (PD). Microelectrode recordings (MER) are taken invasively to measure the electrical activity of neurons in the basal ganglia during DBS procedures. The modeling of the MER signals can help to characterize the electrical behavior of the different nuclei along the surgical trajectory. In this paper, we applied linear and nonlinear parametric structures to model MER signals in 19 patients with PD undergoing DBS in two targets of the basal ganglia. We compared the fits obtained by different orders of pure autoregressive (AR) and AR-moving average (ARMA) models, as well as three models with exogenous input – ARX, ARMAX and BoxJenkins. Furthermore, we evaluated the performance of a nonlinear ARX – NARX - model. All comparisons were made in both right and left hemispheres. We found that exogenous input reduces the fit of the model significantly. The best performance was achieved by the AR model, followed by ARMA. Although NARX had better behavior than ARX, ARMAX and BoxJenkins, it did not surpass the fit of the AR and ARMA. Furthermore, we could reduce the order of the AR reported previously by increasing the database considerably. These results are an approach to characterize the MER signals, and could serve as a feature for pattern recognition-based algorithms intended for intraoperative identification of basal ganglia.