LETTER Communicated by Tzyy Ping Jung Filtering Out Deep Brain Stimulation Artifacts Using a Nonlinear Oscillatory Model Tatyana I. Aksenova tatyana.aksyonova@ujf-grenoble.fr Unit 318, INSERM, 38043 Grenoble, Cedex 09, France, and Institute of Applied System Analysis, Ukrainian Academy of Sciences, Kiev 03056, Ukraine Dimitri V. Nowicki nowicki@mail.ru Unit 318, INSERM, 38043 Grenoble, Cedex 09, France, and Institute of Mathematical Machines and Systems, 03187 Kiev Ukraine Alim-Louis Benabid AlimLouis@aol.com Unit 318, INSERM, 38043 Grenoble, Cedex 09, France This letter is devoted to the suppression of spurious signals (artifacts) in records of neural activity during deep brain stimulation. An approach based on nonlinear adaptive model with self-oscillations is proposed. We developed an algorithm of adaptive filtering based on this approach. The proposed algorithm was tested using recordings collected from patients during the stimulation. This was then compared to existing methods and showed the best performance. 1 Introduction This letter presents an approach for filtering signals of neuronal activity during deep brain stimulation (DBS) using nonlinear oscillatory models. High-frequency (100–300 Hz) DBS is a surgical procedure for treating a variety of disabling neurological symptoms, in particular, those due to Parkinson’s disease. In spite of its clinical efficiency over 20 years, the mechanism of action of DBS is still a matter of debate (Benabid et al., 2005; McIntyre, Grill, Sherman, & Thakor, 2004). Understanding how DBS at high frequency works is of paramount im- portance, as this will provide an understanding of the circuitry of basal ganglia. The major difficulty in studying the mechanism of action of DBS is that the appropriate signal of neuronal activity during the stimulation, the extracellular microelectrode recording of action potentials (spikes), cannot be analyzed directly due to stimulation artifacts present in the records (see Figures 1a and 2a). The artifacts are induced by the periodically repeated Neural Computation 21, 2648–2666 (2009) C  2009 Massachusetts Institute of Technology