An accurate, low-cost BCI (brain-to-computer interface) can help realize the science-fiction ideal in which there’s no need to speak, gesture, or type into a keyboard to communicate with machinery: You just think—and the machine responds. BCI technology is not just the domain of sci-fi junkies: An obvious use for BCI control is in medical therapeutic equipment for paralyzed patients or for research into brain conditions, such as Parkinson’s disease or epilepsy. Other possible applications include game-control interfaces and military equipment. For example, the Defense Advanced Research Projects Agency’s bionic-arm project to improve the state of the art for prosthetics partially funded research into BCIs at the University of Utah. Plus, the growing use by the military of remotely piloted aircraft highlights the potential the military sees for BCI.
The brain is a 3-lb bag of fluids and neurons that communicate by firing off tiny electrical pulses. There are several ways to track these electrical signals: One requires going beneath the skull and implanting electrodes onto or into the brain itself. This approach is risky business and so far has found use only for therapeutic purposes—for example, for the study and treatment of epilepsy. Another method, ECOG (electrocorticography), dates back to the 1950s. It places electrodes directly on the exposed surface of the brain but still beneath the skull to record electrical activity from the cerebral cortex.
More recent work by researchers at the University of Utah uses silicon electrodes the size of baby aspirin that float above the brain but still under the skull. In microECOG, the device comprises an array of electrodes rather than just one (Figure 1 and Reference 1). The researchers placed arrays of tiny electrodes between the skull and the brain. They found that these electrodes can accurately detect the brain’s signals that control arm movements. Surgeons placed two kinds of microECOGs on the brains of severely epileptic patients. Parts of the patients’ skulls had been temporarily removed for placement of the larger ECOG electrodes, which locate and treat the brain area responsible for epileptic seizures. These larger, metallic, button-like electrodes are numbered in the figure. Figure 1 (left) also shows two micro-ECOG arrays, each with 16 microelectrodes that connect to microwires that pass through the orange and green tubes. Photo-editing software outlined the electrodes in the figure. Figure 1 (right) shows one microECOG array with 32 microelectrodes that connects with microwires entering through a clear tube at the bottom of the figure. The green wires connect to the large, conventional ECOG electrodes.
ECOG and microECOG are intermediate steps between electrodes that penetrate the brain and EEG (electroencephalography), which places electrodes outside the skull on the scalp. Compared with the risky, surgical nature of ECOG, EEG is a relatively simple procedure that relies on electrodes anchored through an adhesive to the scalp. However, to reach electrodes on the scalp, the brain’s electrical signals must travel through the skull. Bone conductivity is low, and signals attenuate rapidly. By the time the brain’s signals make it through the surrounding membrane, skull, skin, and hair, these already-faint signals are vanishingly small. EEGs for medical purposes use electrodes that require a conductive jelly that can be messy to apply and remove. These medical-grade EEG-sensor systems can cost tens of thousands of dollars, keeping research into BCIs within the realm of academia and medical research (Figure 2).
However, the lucrative gaming market, in which thought control of games is a novel gimmick, and military applications are driving the interest in BCI devices, which are starting to appear at prices far below the tens of thousands of dollars you can expect to pay for medical-research-quality EEG. Recently, products such as Emotiv’s Epoc and NeuroSky’s Mindset have become available for approximately $150 to $300.
How likely is it that EEG-based headsets can contribute to robust BCI-hardware approaches? Following a BCI workshop early this year at the Massachusetts Institute of Technology, Rod Furlan, Singularity University founder, summarized his thoughts on invasive versus noninvasive BCIs (Reference 2). “As noninvasive interfaces are generally limited to reading brain states, it is unlikely they will be able to evolve into robust input and output solutions,” he said. “Consensus among the experts in the room was that EEG is probably a dead end because, while it provides great temporal resolution, its maximum achievable spatial resolution will probably fall short of the requirements of future applications.”
Tan Le, co-founder and president of EEG-headset maker Emotiv, explains the human brain, the limitations of EEG, and Emotiv’s approach (Reference 3). “Our brain is made up of billions of neurons, around 170,000 km of combined axon length,” she says. “When these neurons interact, the chemical reaction emits an electrical impulse, which can be measured. The majority of our functional brain is distributed over the outer surface layer of the brain. To increase the area that’s available for mental capacity, the brain surface is highly folded. This folding presents a significant challenge for interpreting surface electrical impulses because everyone’s cortex is folded differently. Even though a signal comes from the same functional part of the brain, by the time the structure has been folded, its physical location is very different between individuals, even identical twins.
“[Emotiv created] an algorithm that ‘unfolds’ the cortex [to] map the signal closer to its source and make it able to work across a mass population. EEG measurements typically involve a hair net with an array of sensors. The Emotiv headset is a 14-channel, high-fidelity EEG-acquisition system and requires no scalp prep [and] no conductive gel. It only takes a few minutes to put on and for the signals to settle. It’s wireless and costs only a few hundred dollars.”
Pull it out of the box, connect it to your PC, place it on your head, spend a few moments on the canned exercises that let the headset algorithms learn your brain-wave pattern, and you can begin manipulating virtual images on your PC with your brain (Figure 3). Pretty neat, huh?
Hacker Cody Brocious thought so, too. New to the world of BCI, Brocious was impressed by the simplicity of the Emotiv device, and he wanted to delve deeper. He asked for donations within the hacker community to buy one and quickly raised the money. He discovered the key to the encrypted data coming over the USB connection and built a decryption routine. So far, his library of code hacks to the device just pulls raw data from the unit; there’s no ability to filter the signals or tell which sensor corresponds to each data stream. Brocious created Cody’s Emokit project, an open-source library for reading data directly from the headset, and posted about his project on the Emotiv user forum, which the company runs (Reference 4).
Emotiv officials didn’t like the fact that Brocious had cracked the encryption and posted his library. They claimed that doing so could force the company out of business (Reference 5). Emotiv sells a $700 developer’s version of the headset that allows access to the data, but it is not an open environment; the company controls access. Apparently, Emotiv is working to close the encryption hole and thus end Cody’s project.
Alternatives to EEG exist for measuring small signals on the surface of the head. One such technology, EOG (electrooculography), employs eye polarization. The back of the eye is more negative than the front of the eye because of the large populations of neurons on the retina. As the eye moves, the electric field surrounding the eye also moves. Electrodes on the left and right side of the face and above and below the eyes can measure these fields. An electrode behind the ear serves as a reference voltage.
The inputs to the EOG are the bipotential signals measured at the electrodes; these signals are smaller than those of background environmental noise, such as RF-communication signals or 60-Hz ac-mains noise. Fortunately, each electrode picks up the same noise, making the interference the common signal. Amplifying the difference between the electrodes and rejecting everything common to them yield the EOG’s signal. The Waterloo Labs engineers used an Analog Devices 8221 instrumentation amplifier because of its low noise and high common-mode rejection.
Whenever metal, such as that composing an electrode, touches an electrolytic solution, such as human skin, a potential difference—the half-cell potential— results. The 8221 amplifies the half-cell potential along with the EOG signal. If the amplifier’s gain is too high, the half-cell potential swamps the EOG’s signal. The half-cell potential cannot overpower a gain of 10, which is enough to distinguish it from the noise. The circuit next uses a highpass RC filter set to 0.1 Hz to reject the half-cell voltage. Without the half-cell constant, the circuit can further amplify the signal, carefully adding no noise back in after all that work rejecting it. It uses two low-noise, high-offset amplifiers to keep the signal clean. The final stage uses a lowpass filter set to 50 Hz to remove any high-frequency noise, including 60-Hz ac-mains voltage, and then an ADC.
As for electrical isolation, the EOG receives its power from a 9V battery, so it has no dangerous voltages. However, the video-game controller, the TV display, and the RIO all use 120V wall power, so if there’s a short in the transformer, the EOG user will get a 120V jolt of electricity across the face. To avoid this painful scenario, the circuit uses an isolated AD7401 ADC that magnetically couples the signal across a dielectric gap so that there is no electrical connection between the EOG and the RIO. The AD7401 can withstand 120V wall power for an indefinite amount of time and more than 3000V for as long as a minute.