A telepath in a wheelchair: details
"Open the skull, insert the electrodes"
What are the microchips implanted in the brain and the nanopowder that will replace them capable of
Roman Fishman, N+1
Recently, a group of neurophysiologists from the laboratory of Professor Miguel Nicolelis of Duke University School of Medicine demonstrated the control of a wheeled platform using signals received from electrodes implanted in the motor cortex of the monkey brain. Such works open up prospects for the creation of prostheses and wheelchairs for paralyzed patients who will be able to control their movement simply by "the power of thought". On the other hand, surgical implantation of electrodes into the brain remains a frightening and dangerous procedure that can limit the use of these technologies. We talked with one of the authors of the work, a researcher from Duke University Mikhail Lebedev, about how the experiments with monkeys took place and what the future holds for invasive methods of registering the activity of neurons.
"N+1": In recent years, your group has carried out absolutely amazing work using microelectrodes: expanding the visible spectrum of rats by connecting IR sensors; controlling a pair of robot arms at once due to signals received from the monkey's motor cortex; connecting the brains of three monkeys into a single control device... Against such a bright background, does the new work with "mental" control over the wheeled platform stand out?
M.L.: It stands out, and quite noticeably. The fact is that before, reading signals from the motor cortex was considered only from the point of view of controlling a limb – arm, leg... At the same time, the neural mechanisms associated with the movement of the entire body remain undiscovered even at a fundamental level. Therefore, until the very beginning of the experiments, it remained unclear whether, in principle, by reading the control signals of the arms and legs, we could interpret them from the point of view of the movement of the whole body. The monkey had no lever or other means of guiding the wheeled platform – it simply represented the corresponding movements of the arms and legs in space.
You can recall how the brain is usually drawn with a motor homunculus: neurons that control the movement of the arm correspond to one area, and legs or neck correspond to the other... Please note: in this diagram, there is nowhere a group of neurons that would correspond to the body as a whole, its movement in space. Since the brain regions representing the whole body have not yet been described, we tried to get the right signal from the activities of the neurons that control the limbs, and they stand out brightly on the same homunculus.
Moreover, we are currently preparing an article related to the continuation of this story. In the new version of the experiment, we combined it with the approach that we used earlier, when working with the "union" of two brains: one monkey drove in his cart, and the other sat in the corner, watching her. The movements of the cart were reflected in the brain of both the driver and the observer. There is no big news in this: it is known that the movements of creatures that a person observes are mirrored in the activity of his own neurons. However, we managed to combine the signals of both the "driver" and the "observer", and then transfer wheel control to the final summing signal.
«N+1»:This raises the question of how exactly the desired signal is allocated. Did you need a neural network to interpret the data on the excitation of hundreds of neurons?
M.L.: In this part we decided to do as simply as possible and used a relatively simple method of signal extraction, the Wiener filter. Simply put, our model received data on the frequency of neuronal discharges, multiplied the activity of each by a certain weighting factor, summed it up and gave the desired parameter at the output – in our case, this is either the forward-backward speed or the angular velocity corresponding to the rotation of the cart around the vertical axis. Such a filter, of course, needs pre-training to select the correct coefficients.
At the stage of learning the filter, the monkey did not control the cart, but simply rode it, the movements of the wheels were randomly set by the computer. With the help of our electrodes, we recorded the activity that occurred in the monkey's motor cortex in response to different types of movements. The Wiener filter evaluated this activity and, depending on its magnitude, assigned two coefficients in each case: the "contribution" of the neuron to the forward-backward movement and to the rotation around its axis. That is, if the neuron was actively excited during such a movement, it received a large coefficient, if not, then a small one.
We know from experience that this is not a very difficult task, and for training a filter with the amount of input data that we have – and this is about 300 neurons from two hemispheres – a session of a maximum of 10 minutes is enough. Then we can already transfer control to the monkey.
When trying to move in a certain direction, its neurons are discharged, this data is removed and fed into a trained filter, where it is multiplied by the necessary coefficients, summed up – and we get two high-speed components at the output. At the same time, as the monkey learns such control, its brain adapts plastically, giving out an increasingly clear signal, and the monkey gets to the reward faster.
«N+1»:Having received such a trained filter, can we apply it on another animal?
M.L.: No, it won't work. Implanting microelectrodes, of course, we get into the right area very accurately, but not with the accuracy of a neuron. And the localization of neurons in each animal is individual. Therefore, even the same electrodes in different monkeys can record different indicators, in fact, randomly.
It can be added that the animals we used in the experiments have been walking with "chronically" implanted electrodes for years. They have already participated in many experiments, mainly on the control of a robotic arm.
"N+1": Sounds pretty brutal. In general, against the background of the rapid development of non–invasive methods for obtaining data on brain activity – such as EEG or tomography - are microelectrodes losing relevance? After all, they require surgical implantation, opening of the skull, and so on...
M.L.: No way. The same EEG, in fact, as it was a hundred years ago, remains the same today – progress in this area consists mainly in improving algorithms and data interpretation tools. But in general, it still does not allow you to receive accurate, high-quality signals, and this is unlikely to change in the future. EEG data is always data about the synchronous firing of a sufficiently large number of neurons.
There is even a kind of paradox. After all, when neurons do something meaningful - for example, control the hand – they discharge in small groups. And synchronous activity manifests itself just at times when they are not engaged in any meaningful activity that we need for interpretation. The largest synchronous triggers are observed in sleep, and EEGs in sleep demonstrate the largest amplitude. Therefore, EEG systems work "in reverse": a strong signal means that the subject is at rest, a weak signal means that he is busy with something. At the same time, the details of this lesson completely slip away.
For these reasons, I think that EEG and methods close to it will develop more towards neurorehabilitation, helping patients after strokes and similar things. And for more subtle control, in any case, invasive methods will be needed. Of course, from the outside today they seem quite scary: you need to open the skull, insert electrodes... But I think these problems will be solved in the next 10-20 years.
"N+1": I recall the recent work of Australian scientists who proposed a microelectrode that penetrates the brain through the vessels and "opens" in the right place, fixing and removing the activity of neighboring neurons. Do you mean such solutions?
M.L.: This is certainly a very good idea. The main thing is that the recording quality that this approach gives is comparable to electrocorticography data, in which electrodes are simply placed on the surface of the brain, but are not inserted into the nervous tissue, so it is quite safe. Recording "through the vessels" gives the same resolution, but at the same time allows you to get into the deep structures of the brain, working in subcortical areas as well. On the other hand, of course, there are questions about how safe it is to walk with such electrodes in the vessels, whether they will provoke blood supply disorders and all that – it is still worth carefully studying.
We must not forget that nanotechnology is actively developing, which theoretically will make the electrodes very, very small. There is even a concept of using "nanopoils" – particles that will penetrate into the brain and read the activity of individual neurons, transmitting this data to a device located outside the skull. We have enough opportunities to minimize the danger of an invasive approach.
Portal "Eternal youth" http://vechnayamolodost.ru
22.03.2016