Brain-Controlled Robots on the March to Reality.

T. Lucas,R. Ellenbogen

DOI: https://doi.org/10.1227/01.NEU.0000308736.88678.C4

IF: 5.315

2005-05-01

Neurosurgery

Abstract:VOLUME 56 NUMBER 5 MAY 2005 be possible in the setting of an existing neurological deficit. Several laboratories have made recent advances toward clearing the hurdle of neuronal signal processing. Nicolelis and colleagues recently demonstrated that accurate predictions of intended hand position in space could be obtained in real time using a multiple linear regression model of spike trains (J Cognitive Neuroscience 16(6):1022-1035, 2004). These predictions could be used as command signals for a robotic arm using only 1 sec of neuronal data. Such an algorithm could serve as a starting point for implantable devices. Schwartz and colleagues have recently shown that trained monkeys can make virtual arm movements in a 3D virtual environment using control algorithms that can track changes in neuronal response tuning (Science 296:1829-1832, 2002). This finding suggests that BMI algorithms may adapt to changes in neuronal tuning, a complication which has plagued several previous models. In addition to advances in neuronal signal processing, significant strides have been made toward selecting appropriate cortical targets for electrode arrays. Prior cortical targets have ranged widely, from primary motor (MI) and dorsal premotor cortex (PMd), to parietal cortex. Indeed, sing brain signals to control machines is a recurring theme in science fiction. Recent advances in the field of neuroprosthetics and brain-machine interface devices (BMI) have made this science fiction theme more tangible. The potential clinical applications of BMI are numerous, and include its use in stroke and spinal cord injury patients. The basic concept underlying BMI is that neuronal populations generate signals that govern coordinated movements and ultimately, define motor behavior. In principle, it should be possible to record this activity and translate it into command sequences for robotic arms or other mechanized devices. Moreover, cortical neurons have been shown to modulate their activity before movements are initiated, thus giving scientists a potential source for signals to drive neuroprosthetic devices in the absence of overt movements (as in the case of the paralyzed patient). Recent advances in signal processing and neuroprosthetics have brought the possibility of using these devices in the clinical setting one step closer. Several hurdles must be overcome, however, before BMI is applied in clinical neurosurgery. First, reliable, long-term use of BMI requires the selection of an appropriate cortical target, followed by implantation of microelectrode arrays that must record high fidelity singleand multiunit activity indefinitely. Second, this activity must be translated using appropriate signal processing algorithms to generate command signals for a prosthetic device. This translation must encode all of the degrees of freedom built into the robotic arm, as well as joint position, trajectory, movement velocity, force and displacement. Finally, the above requirements must Science Times Editorial