Controlling Your Mind with the Help of an fMRI


by Jim Schnabel

December 3, 2010

Much of what goes on in our brains—and much of what goes wrong there—lies frustratingly beyond our conscious control. We try to sleep, or to stop feeling anxious, or to stop craving food or drink or drugs, but our brains won’t let us. Medicines can help, but they are usually blunt weapons, with unwanted side effects. Hypnosis, biofeedback, and similar techniques are meant to open a back door to precise therapeutic mind-control, but they require a great deal of training, and even then don’t work well for everyone.

Could brain-imaging technology enable a better solution? A growing number of researchers think so, and have begun to use functional magnetic resonance imaging (fMRI) to provide near-immediate feedback to people so they can control specific brain activity. At the Society for Neuroscience’s annual meeting in San Diego in November, University of Pennsylvania researcher Anna Rose Childress reported a proof-of-principle experiment using an efficient new variant of the technique:  It allowed her subjects—including even crack cocaine addicts—to learn to control a cursor on a computer screen using only their thoughts, and with almost no training.

“It’s extremely rapid, extremely accurate, and basically can be done by almost anyone,” Childress said.

An fMRI device uses high-strength magnetic fields to detect the tiny surges of oxygenated blood that help fuel working neurons. It produces a 3-D map in fine-grained volume units known as voxels that roughly represents the change in neuronal activity during a given task.

In this case, Childress and her colleagues recruited fourteen volunteers, each of whom was placed inside an fMRI device and asked to think of hitting a tennis ball for thirty seconds, then of navigating a familiar space for thirty seconds.

During each of these cycles, explained Childress, “the computer compared all the voxels in one brain state to those in the other brain state and essentially came up with a classifier that distinguishes between the two.” (Childress and her colleagues have received grants from the Dana Foundation for earlier imaging work.)

The imagining-hitting-a-tennis-ball and imagining-navigating-a-familiar-space tasks are performed with crucial help from two distinct brain areas, the supplementary motor area and the parahippocampal place area, respectively. But the classifier used in this case didn’t merely compare fMRI activity in these two regions; instead, it compared the pattern of activity throughout the brain.

In the original fMRI-based feedback approach, developed over the past five years or so, the goal was to use mental imagination exercises to enhance or reduce activity in specific brain areas, or “regions of interest” (ROI). But brain activity is inherently noisy, and even when activity in an ROI was known to correlate strongly with the desired subjective outcome—such as reduced pain or cigarette cravings—it often turned out that people needed many training sessions before they could start controlling their symptoms.

In this case, using the new, “whole-brain classifier” approach, Childress and her team needed only a few one-minute cycles to train their subjects—and the computer—to work usefully together. With this training, the subjects could move a cursor to the upper half of the computer screen by thinking of the tennis activity, or to the lower half of the screen by thinking of the navigation activity. “Every two seconds as the magnet was pulsing on and off, this computer algorithm was deciding whether you are in the tennis thought state or in the navigating from room to room state, and was able to make that distinction with a high degree of accuracy,” said Childress.

Whatever works

An immediate application of this technique would be to enable people with paralytic “locked-in” syndrome, who cannot otherwise communicate, to communicate simple messages to loved ones and caregivers. But as Childress emphasized, this was a proof of principle experiment, a foundation for more ambitious therapeutic applications.

The focus of Childress’s research has been on finding ways to prevent cocaine addiction relapse, and although eleven of her subjects were ordinary healthy volunteers, three were habitual cocaine users. “It was really encouraging for us that our cocaine patients could do this too,” she said. In further work, she hopes to use the new classifier approach to help her patients to reach a brain state associated with reduced drug cravings.

Dave Scott, a researcher at Omneuron, Inc., a Silicon Valley company that is also working on fMRI-based feedback applications, notes that Omneuron’s subjects have been able to reduce pain and cigarette cravings with a variety of experimental approaches. Some mental tasks still seem amenable to an ROI-based approach, while other mental tasks are so complicated, and knowledge of their neural correlates so limited, that approaches based on regions of interest might not work well. “For example, imagine creating a behavioral task that asks a schizophrenia patient to attend to or try to ignore internal voices. What parts of the brain mediate hearing voices in these patients, and are these brain regions amenable to overt cognitive control, as real-time-fMRI feedback would demand? Even after a great deal of study we are still very far from having the answers,” says Scott. “But [whole-brain] pattern classification approaches like that employed by the Childress group eliminate the need for such a selection process.”