#brain-machine interfaces

Most frightening and/or depressing story: We are most likely on a path to the AMOC tipping point. I distinguished between the tipping point, which is when collapse becomes inevitable, and the actual collapse itself. These are separated in time, which means the tipping point may only be called in retrospect when it is too late to prevent the collapse. This why being “on the path to the tipping…

HEY SO after a totally planned and foreseen "summer hiatus," I am back to existing in the blog-world. I can hear your cries of resounding joy from here.

More reasons to get a good night's sleep after your Late Beginner Adult Ballet 1 class—it's not just a rest for your muscles, it's a chance for your brain to keep practicing.

Slow-wave sleep (SWS; the last stage of the sleep cycle before REM) is already known to be important for the consolidation of memories, including motor memories. One mechanism for this might be enhanced activity of motor control areas in the brain during SWS, which has been shown to correlate with better learning of motor skills. A new Nature Neuroscience paper suggests that this is not just a diffuse increase in motor cortex activity, but offline reactivation of the specific activity patterns involved in performing a newly learned task. The study is also notable for being one of the first (at least, the first I've seen...) to use a brain-machine interface (BMI) as a means to a scientific end other than just the development of the devices themselves.

A BMI is any of a wide variety of devices that can be stuck in the brain to read out neural activity and use it to control things like computer cursors or artificial limbs. In this study, an array of tiny electrodes recorded spikes—little electrical pulses that are the basic units of neural computation—produced by a small population of neurons in the rat primary motor cortex (M1). A computer then changed the position of a small feeding tube based on the activity of a preselected set of neurons. With some practice, the rats learned to increase the firing rate of the task-related (TR) neurons, moving the tube into the correct position and triggering the delivery of a small reward.

(I'll just pause and let you absorb this, in case you're new to the idea of BMIs. This is a study involving rats controlling machines with their brains. And that's not even the main finding. The future is now, you guys.)

There's a huge benefit to using a BMI to study motor learning—not only can you measure the activity at a very fine spatial scale, you can also define a task with respect to a particular, known, set of neurons. That makes it much easier to determine whether subsequent changes in neural activity are reflecting a general boost to the motor area, or are specific to the processing of the newly learned task (i.e., the TR neurons).

The interesting findings in this study deal with TR neuron activity during a period of SWS following the learning session, as compared to their SWS activity before. What matters is not so much the number of times neurons fire but when they fire, with respect to the local field potential (LFP). The LFP is a measure of the average electrical activity in the neuron's cellular neighborhood, which tends to oscillate in fairly regular cycles. (Probably important to note that it's still a bit of a mystery exactly what neural activity is reflected in the LFP—it's likely some aggregate including incoming and outgoing signals as well as local processing.) Phase locking analysis determines whether an individual neuron's spikes occur at the same phase (e.g., near the peak or the trough) of the LFP waves. You might think of the LFP as a kind of neural background music—neurons who're clapping with the beat are phase-locked. Neurons who are just randomly active (even if they have a high average firing rate), aren't.

Phase-locking is a common metric in memory studies. It's been shown that phase-locking of individual neurons increases during the learning portion of various different memory tasks, and that performance on these tasks correlates with the amount of phase-locking. So the alignment of individual neurons' spikes with the LFP might be one mechanism for encoding new memories—one hypothesis is that certain phases of the LFP facilitate the formation of new neural connections (synapses); or, the rhythmicity of the LFP might act like a time-keeper that helps coordinate the widespread brain activity necessary to retrieve a memory. Again, bit of a mystery, but the data are consistent: phase-locking is a key mechanism in memory formation.

But back to the rats: after they learned the task, TR neurons exhibited increased phase-locking during SWS, as compared to before the training. Phase-locking of task-unrelated (TU) neurons didn't change, nor did that of the TR neurons in rats who failed to actually learn the task. That's the first indication that task-related activity was being reactivated during sleep.

At the population level, correlations and interactions among the studied neurons were also different during the SWS periods after learning. First, there was an increase in spike-spike coherence between pairs of TR neurons, meaning that when one TR neuron fired, others tended to fire as well. They also used principle component analysis (PCA) to probe more complex interactions (“components”) within the population activity. PCA involves a lot of fancy math (read: REALLY FUN MATH but too tangential for this post), but as a quick example, imagine that neuron A is a TR neuron that fires a bunch of spikes whenever the animal performs the task. Imagine neuron A is linked to neuron B (as many neurons within a small cortical region might be) such that A can only fire more if B also calms down and fires less than its average baseline rate. PCA would find this interaction—the combined decrease in B's firing rate with the increase in A's—and other possibly more complicated “components,” and highlight them as important aspects of task-related activity at the population level. You might have already guessed, but these components, too, were both reactivated during SWS and phase-locked to the LFP, just like the individual TR neurons' spikes.

Those results are really cool—they show that in the period of SWS after rats learn a motor task, the increased activity in motor cortex is not simply a diffuse boost in the spiking rates throughout M1, but the specific rehearsal of (approximately) the actual population activity required to do the task.

But to show that these neural changes are actually the mechanism behind sleep-related performance improvements, you also have to show that performance actually improves after sleep, and that this improvement depends on the neural changes. The researchers find exactly that—they tested the animals in two sessions, before and after a period of sleep, and found that rats whose TR neurons became more phase-locked during SWS improved more between the end of session 1 and beginning of session 2. And, the more time a rat spent in SWS during his nap, the more he improved in session 2.

All together, this evidence suggests the reactivation of task-related neural activity during SWS as an important mechanism for sleep-dependent improvements in motor learning. It also helps define the roles of sleep and plasticity with respect to brain-machine interfaces, which can be used to replace lost sensation (for example, cochlear implants) or even (someday) amputated limbs.

by Marsha Lewis, Inside Science TV

Three years ago Igor Spetic was working at his manufacturing job when his life changed forever.

"When I looked down and saw red everywhere, I knew something had happened," said Spetic.

A huge metal hammer had crushed Igor's right hand.

"Thirty minutes later, I was in surgery, and two hours after that, I was without a hand," said Spetic.

Getting detailed neural activity data is pretty hard. Present methods generally involve putting many electrodes into the skull to measure LFPs. These electrodes degrade, etc. This paper develops a possible system for "chronic" recordings - a system that can be inserted once and used indefinitely, or at least for long periods.

A major hurdle in brain-machine interfaces (BMI) is the lack of an implantable neural interface system that remains viable for a lifetime. This paper explores the fundamental system design trade-offs and ultimate size, power, and bandwidth scaling limits of neural recording systems built from low-power CMOS circuitry coupled with ultrasonic power delivery and backscatter communication. In particular, we propose an ultra-miniature as well as extremely compliant system that enables massive scaling in the number of neural recordings from the brain while providing a path towards truly chronic BMI. These goals are achieved via two fundamental technology innovations: 1) thousands of 10 - 100 µm scale, free-floating, independent sensor nodes, or neural dust, that detect and report local extracellular electrophysiological data, and 2) a sub-cranial interrogator that establishes power and communication links with the neural dust.

Basically it works like this: On the outside of your head (back of your neck), you have a control unit. Inside of your head, under your dura mater, you have some "interrogators", that communicate with many "dust" motes in the cortex - small mostly-cubic integrated circuits with piezoelectric equipment etc., less than 100 µm to a side - through ultrasound, which also powers the motes (through the piezo).

by Charles Q. Choi

Researchers have developed a novel brain radio implant that can wirelessly broadcast signals from up to 100 brain cells for well over a year. They say the apparatus could one day help lead to improved robotic limbs and perhaps even remote control of devices.

A Brown University research team developed the rechargeable implant that can record neuron activity and wirelessly broadcast that data like a cell phone for the brain.

by Michael Keller

Another scientific development is bringing modern life closer to the future world envisioned in Star Trek.

Researchers have taken a significant step toward making an implantable machine that could restore and augment vision by bypassing the optic system and training another of the brain's sensory processing centers to "see."

A Duke University team has successfully fitted rats with an infrared detector whose electrodes were wired into the part of their brains that sense touch. When the neuroprosthetic device was powered up, the rats felt infrared light as if they were touching it.

The scientists who performed the study say the experiment represents the first time that a brain-machine interface has augmented an adult animal’s senses.