Plugging In: Brain-Machine Interfaces

'If man in space, in addition to flying his vehicle, must continuously be checking on things and making adjustments
merely to keep himself alive, he becomes a slave to the machine. The purpose of the Cyborg, as well as his own
homeostatic systems, is to provide an organizational system in which such robot-like problems are taken care
of automatically and unconsciously, leaving man free to explore, to create, to think, and to feel.'
-Cline and Clyne (1960), on the creation of the word "cyborg"

During the summer, the Neurokids (University of Michigan Neuroscience pre-docs) have a journal club where we all talk about a review article that is related to one of our respective lines of research (we also stuff our selves silly and drink a lot of beer, but thats neither here nor there). Tomorrow is the kickoff meeting, and I figured if I read it, might as well blog it! This is good for you, dear reader, as this is a really facinating paper on how machines are able to communicate with the brain, titled Brain-machine interfaces:computational demands and clinical needs meet basic neuroscience and was published in Trends in Neuroscience in 2003.

From Jordi LeForge to the Matrix, Inspector Gadget to Star Wars, popular culture is captivated by the possibility that machines may one day directly interact with the human brain. This is a natural line of thought, as the computer science has mirrored neuroscience, and the architecture of the computer is modeled after the brain. Beyond wishful thinking, there is also a real clinical need (paralysis, blindness, deafness) and practical advantage (fighter pilots and all kinds of scary military uses) to be able to communicate with a computer on an instantaneous, neural level. This is especially relevant to me, as a device such as this currently exists as the only real treatment for deafness--the cochlear implant. This is the first success story of the brain-machine interface, but many more are in development

The review focused on two main challenges in the field:
1. Establishing a "closed loop" between sensory input and motor output; and
2. Controlling neural plasticity to achieve the desired behavior of the brain-machine system

(More under the fold.....)

I encourage those interested to real the actual article, as I will be just highlighting.

The goal of any brain-machine interface (hereto called BMI) is to replace or enhance a neurobiological function in an artificial system; specifically, coaxing neurons to grow onto a semiconductor or perhaps implanting electrodes into the brain to induce direct electrical stimulation. Much of the driving force behind this research is to enable "locked-in" patients (disabled physically but otherwise aware and healthy) to interact with the environment more efficiently by "extracting" control signals from either surface EEG signals or implanted electrodes.

A study by Wolpaw et al. was successful in training both healthy and disabled human subjects to move a cursor on a screen in one or two dimensions, based on the amplitude of a component of the subjects' EEGs. The u-rhythm is an 8-12 Hz oscillation detected over the sensorimotor cortex which is present during both REAL and IMAGINED movement. (Figure below, A) Others have also investigated this method, although a limitation of EEG-based methods is the low information rate-- only 20 to 30 bits/minute. This limitation has lead the field to a more invasive approach based on extracellular recordings by microelectrodes.

Chapin et al. trained rats to get drops of water from a robotic arm by pressing a lever which controlled the rotation of the robotic arm. They recorded the activity of 21-46 neurons in the rats' motor cortex (M1 region) which was used as input into a computer program which controlled the robotic arm. Astonishingly, several of the rats learned how to manipulate the robotic arm without moving any part of their body! Anderson et al. demonstrated real-time control of a cursor on a computer display by monkeys (Figure above, B). They implanted electrodes in the posterior parietal cortex (believed to plan movement) and trained the monkeys to physically reach towards one of two targets on a screen. An algorithm recorded the monkey's brain activity during the process to associate what pattern of activity in the parietal cortex indicated a preference for which target. The algorithm began to accurately predict which target the monkey was "thinking about" reaching for so well, that within 50 trials, all the monkeys learned to modulate their brains to indicate the intended target, without moving!

Only one chronically implanted electrode system has ever been used in humans, the neurotrophic cone electrode. This electrode (Picture below) is a microscopic glass cone which contains a neurotrophic factor that attracts neurons to synapse with the recording electrode. Signal conditioning and telemeric electronics are implanted under the skin of the scalp, and the implanted transmitted (TX) sends signals to an external receiver (RX) which is connected to a computer. Four paralyzed patients were implanted with this apparatus, although even the most successful patient was only able to communicate at a rate of three letters per minute.

Mussa-Ivaldi et al. have investigated using the feedback from a BMI for inducing controlled physical changes in the brain at a synaptic level. They connected a mobile robot to a lamprey brainstem (Picture below). Signals from the optical sensors of the robot were encoded by the BMI into electrical stimulations, with the frequency of the stimulations depending on the light intensity.

These stimulations were delivered by electrodes to the right and left vestibular pathways in the lamprey brain. The stimuli are specifically delivered to the axons of the intermediate and posterior octavomotor nuclei (nOMI and nOMP), and recording electrodes record responses to the stimuli which are subsequently decoded by the BMI. During decoding, first the recording artifacts are removed, then the population spikes (bursts of activity) are detected and an average firing rate is computed. This firing rate is translated into a command to the corresponding wheel of the robot, which the angular velocity of the wheel set to be proportional to the average firing rate.

Furthermore,

Recent studies by Nudo and co-workers have provided preliminary evidence that the combination of behavioral training and electrical stimulation of areas surrounding a cerebrovascular accident can lead to a significant acceleration of functional recovery. If results such as these find further support, one could envisage a future scenario in which the closed-loop interaction between a patient's brain and an external device will be used to facilitate the reorganization of neural circuits that is necessary for reestablishing normal movement patterns.

This is a fascinating field which holds immense clinical value, and I look forward to this area of neuroscience really blossoming as the BMIs become more and more sophisticated.