Conceptual and Technical Approaches to Human Neural Ensemble Recordings

Abstract

The ability to perform either multineuron or local field/EEG recordings from the nervous system is a critical requirement to develop a new generation of neuroprosthetics that can sense the brain’s intent for action (Nicolelis 2001, 2003). This form of sensing neuroprosthesis builds upon the concept of current neuroprosthetic devices, which are primarily for macrostimulation of neural elements, such as deep brain stimulation (DBS); (Abosch, Hutchison et al. 2002; Rodriguez-Oroz, Obeso et al. 2005). A key aspect of this evolving technology is the translation of preclinical multineuron recording and analysis technology into the clinical arena (Donoghue 2002; Carmena, Lebedev et al. 2003; Mussa-Ivaldi, Miller et al. 2003). This translation requires the use of medical-grade components at all levels of electrodes, connections, and electronics, and the stabilization of technology and software for the long process of Food and Drug Administration (FDA) approval.Human sensing neuroprosthetic devices currently depend upon control signals from residual nerve or muscle activity to restore motor functions lost due to disease or trauma. It has been proposed that these devices could be significantly improved by directly harnessing brain activity from central motor-related regions to drive artificial actuators (Chapin 2000; Nicolelis 2001; Caves, Shane et al. 2002; Nicolelis, Chapin et al. 2002; Chapin and Chapin 2004). Recently, laboratory studies involving nonhuman primates have made considerable advances toward the development of such devices. For example, neuronal ensemble recordings from motor areas of cerebral cortex in nonhuman primates have been demonstrated to accurately predict three-dimensional arm movements (Chapin, Moxon et al. 1999; Wessberg, Stambaugh et al. 2000; Taylor, Tillery et al. 2002; Carmena, Lebedev et al. 2003; Nicolelis, Dimitrov et al. 2003) and to successfully control a robotic arm neuroprosthetic device. Despite these interesting advances, primate studies have yet to address the fundamental question as to whether current brain–machine interface (BMI) technology and approaches may be successfully applied to human patients, in particular, those who are naive regarding the eventual tasks (Wolpaw, Birbaumer et al. 2002; Patil, Carmena et al. 2004; Gage, Ludwig et al. 2005). Nonhuman primate BMI studies suggest that multineuronal recordings are critical for neuroprosthetic applications, and may require a minimum of 50–100 recorded neurons to drive a real-time neuroprosthesis (Nicolelis 2001, 2003; Sanchez, Carmena et al. 2004). In addition to cortical motor regions, subcortical regions, such as the motor thalamus and subthalamic nucleus, are also involved in motor planning and execution, and could serve as alternative multineuron recording sites (Lenz, Kwan et al. 1990; Cheruel, Dormont et al. 1996; Abosch, Hutchison et al. 2002; Guillery, Sherman et al. 2002; MacMillan, Dostrovsky et al. 2004; Patil, Carmena et al. 2004).Devices utilizing control signals from the nervous system have also been developed recently to enhance functional independence, using external reflections of brain events rather than direct neuronal recordings, such as electroencephalogram (EEG), direct cortical surface recordings (ECoG), or evoked potentials (Kubler, Kotchoubey et al. 1999; Donchin, Spencer et al. 2000; Pfurtscheller, Guger et al. 2000; Birch, Bozorgzadeh et al. 2002; Wolpaw, Birbaumer et al. 2002; Scherberger, Jarvis et al. 2005). These external signals suffer considerable information loss, because the control signal is derived from thousands or millions of neurons averaged across time and space. For example, scalp EEG signals can enable the control of approximately 6–7 characters per minute on an optimized keyboard, for a short period, but this is very limited for most purposes (Wolpaw, Birbaumer et al. 2002). Although a large variety of devices and approaches to neuroprosthetics are currently available, there is not at present a r