20 May 2008—A group of mechanical engineers at Caltech have come up with a way to guide miniature robots in the task of inserting and positioning electrode arrays in brain tissue. What they propose would be the first robotic approach to establishing an interface between computers and the brain by positioning electrodes in neural tissue. Researchers say that this could enhance the performance and longevity of emerging neural prosthetics, which allow paralyzed people to operate computers and robots with their minds.
In the brain-machine interface, the brain is undeniably the more chaotic and unpredictable of the pair. Electrical impulses flow between supple bodies in an aqueous pulp whose architecture is constantly changing. Neurons grow, shrink, and die, as do the synapses that link them. In the best marriage, a neural prosthetic would create a permanent one-to-one connection between electrode and neuron, ensuring that recordings of electrical spike activity—the language of information transfer in the brain—represents the chatter of a single cell per electrode. But this doesn’t always happen.
With today’s experimental neuroprostheses, ”whether the tip of the electrode ends up next to a neuron that’s helping you is still dependent on luck,” says Michael Wolf, an engineer at Caltech who, with his colleagues, will present his robotic approach to the procedure today at the IEEE International Conference on Robotics and Automation, in Pasadena, Calif.
As it is now, in single-cell recording, researchers determine where in the brain the electrodes should go by analyzing the signals they get as they bore deeper into the brain tissue. The technique is a way to figure out how many neurons you’re listening to at a given time. But since all neurons relay information in the same way, with an all-or-none electrical event called an action potential, it can be very difficult to tell what signals are coming from where.
”The uncertainty is the key issue,” says Wolf. The Caltech team has designed a system that would make the procedure more predictable by attaching a tiny MEMS-based motor to each electrode on a multichannel electrode array and using an algorithm to direct the electrodes to individual neurons. The MEMS part is still a work in progress, but the software algorithm has been worked out and tested in Caltech neuroscience labs.
Here’s how the algorithm makes a neural connection: As the electrodes are driven into the tissue, the software starts taking sample recordings to detect spikes of electrical activity at the electrode tip. When the software detects spikes, it moves forward in small increments and tracks how the signals change. After determining whether the signal has improved or gotten worse, it the algorithm moves the electrode to a new position and does more recording and comparing, driving the electrode in further if necessary until it finds the best signal. If the signal wanes, the algorithm will automatically adjust the electrode position to improve the signal.
For all this to work, the program must be able not only to correctly discriminate between spikes from different neurons in the same recording but also to retain this information and track a neuron’s spiking patterns over time. The neuron-tracking algorithm was inspired by software the U.S. military uses to track planes, and Wolf expects that his formulas may be useful to other applications in robot and computer vision.
Yu-Chong Tai, a MEMS researcher at Caltech, is designing the hardware that would move the electrodes on a scale of microns. Each electrode in an array would have to connect to a tiny motor on the surface of the brain that would control all its movements. And there’s quite a bit of work left to be done on that. ”The idea of actually putting this in the [human] brain is far off,” says Wolf.
Brain-machine-interface developers are making progress without the ability to keep hold of the signal from a single neuron. ”It’s not yet clear that holding the same neurons, the exact same neurons, is a requirement,” says Lee Hochberg, a neurologist at the Massachusetts General Hospital who works with brain-machine-interface developer Cyberkinetics Neurotechnology Systems, in Foxborough, Mass.
But Wolf argues that a robotic interface could increase the life span of neural prosthetics. No one has yet left an electrode array in a human brain long enough to see how long the connection survives, but there’s wide speculation that the signal will deteriorate as cells from the brain’s immune system gather around the electrode tip and form a signal-blocking scar. A system like the one Wolf proposes would be constantly capable of moving to find better electrical connections when old ones fail.
The algorithms have gotten a few test runs in other labs at Caltech. Grant Mulliken, a neuroscientist at the university who records signals in the posterior parietal cortex of rhesus macaques, has used the algorithm to control a machine, called a microdrive, which most electrophysiologists use to position their electrodes. He began by giving the algorithm control of half his six electrodes. Now, he says, he often uses Wolf’s program to guide all six electrodes before taking over and doing his own fine-tune adjustments.
”I’m very meticulous when doing my experiments. If something doesn’t benefit me, I immediately trash it,” he says. ”I wouldn’t use [the robotic system] if it didn’t work.”