This article is reproduced with the permission of New Scientist for exclusive use by Nova users.

The mind readers
21 September 2002
From New Scientist Print Edition.
Laura Spinney

Enlarge A portrait of the brain

Scientists have identified spots in the brain that "light up" when people contemplate murder, recognise their best friend, feign illness or feel the urge to skip. It's a formula that has crept into the vernacular ever since functional magnetic resonance imaging, or fMRI, revolutionised the field of non-invasive brain scanning in the early 1990s. The amazing images pinpoint our most intimate thoughts and emotions, reveal the basis of brain diseases and personality, and the neural pathways that allow us to move, see, hear and learn. Or do they?

According to Nikos Logothetis of the Max Planck Institute for Biological Cybernetics in Tübingen, brain researchers were not sure exactly why particular areas "lit up" during fMRI. He warns that they have vastly overestimated the reliability of the technique, and produced oversimplified interpretations of the images. The last decade of fMRI work has ushered in a "new phrenology", he says, more insidious than the 19th century variety because of the false respectability it gains from being computerised.

His comments have not gone down well. Although he's not alone in his criticisms, he won't win many friends by suggesting that with a scanner or three in every hospital, almost any psychological researcher can test a pet theory without worrying about the strengths or weaknesses of the technology. But it's worth remembering that what these stunning brain images actually show is variations in blood and oxygen supply in different parts of the brain, not neural activity. The whole fMRI edifice rests on the assumption - albeit a reasonable one - that these things are closely related, so the greater the blood flow and oxygen consumption, the harder a region is working.

Logothetis is busy testing that assumption. He produced the first direct evidence supporting it just last year. But discrepancies have come to light, too, and it's these that are now dividing researchers.

He has found that the fMRI signal is a less reliable representation of neuron activity than was assumed, and by no means tells the whole story. For every area that glows, there might be a whole network of vital regions working away in the background that don't show up because the technique isn't sensitive enough. And the signal can be affected by the position of blood vessels as well as neural activity.

Some have dismissed Logothetis's findings, saying they detract nothing from the studies that have been carried out using fMRI. But others say his results mean many such studies are at best misleading, at worst useless.

In pointing out the technique's deficiencies, however, Logothetis has begun to find ways to get around them. He believes we are on the threshold of a second generation of fMRI imaging, one that will leave today's "phrenologists" standing. At the moment, the most widely used scanners lump together signals from several hundred thousand neurons- the number represented by a single "voxel", a three-dimensional pixel, of an image. In the new era, he foresees researchers mixing fMRI with measurements from tiny electrodes placed inside single cells and a whole range of other high-resolution techniques to zoom in on individual neurons, synapses, even ultimately molecules of neurotransmitters, while simultaneously scanning the whole brain.

fMRI relies on the fact that molecules of oxygenated haemoglobin in the blood behave differently in a strong magnetic field compared with haemoglobin molecules that have given up their oxygen - they have a different "magnetic resonance". An MRI scanner can be tuned to pick up these different signals in the brain of a patient lying in the machine and superimpose them on a three-dimensional anatomical image (see "A portrait of the brain"). What is actually measured is a complex combination of blood flow, blood volume and oxygen consumption. But what everyone in the field would love to know is how this "blood-oxygen-level-dependent" (BOLD) signal relates to neural activity.

Logothetis and his colleagues realised that the only way to be sure was to listen in to individual neurons and measure their activity while carrying out a scan. It might sound simple, until you think about the problems of measuring the activity of a single cell in a powerful, alternating magnetic field. Even the tiniest bit of metal, such as an electrode, induces voltages of up to 30 volts that swamp the puny microvolt-sized "spikes" produced by the cells. The team's unique achievement was to devise a way to detect and compensate for this interference while scanning a monkey's brain.

Although the interference is huge and destructive, it is a predictable and repeatable pattern of voltages, so it can be subtracted from the signal. The researchers use sensors to measure interfering voltages as they arise during recording, then invert these and add them back into the mix. What's left is the part of the signal they're interested in. Software accounts for only 20 per cent of the job, says Logothetis - the rest is skill.

Until now, other researchers have had to compare fMRI data with electrical measurements taken from single neurons in the same animal at different times. Two years ago, however, a group led by fMRI pioneer Seiji Ogawa of Bell Laboratories in Murray Hill, New Jersey, succeeded in doing the two simultaneously in rats. But Logothetis's experiments were the first to manage it in primates, allowing him to begin to link decades of work on individual monkey neurons with the explosion in human fMRI studies of the past 10 years.

At first glance, their experiments appeared to legitimise the human studies. But the devil lay in the detail. For a start, the position of blood vessels had a noticeable effect on the signal. You might have two equally active brain sites, says Logothetis, but one lights up more than the other because it happens to be closer to a draining vein.

The BOLD signal also seems to be more closely related to what information feeds into a particular brain region, and the processing that goes on within that region, rather than its output or firing pattern. Regions can still be actively processing information even if the neurons there aren't madly firing- they could be receiving and weighing up both stimulatory and inhibitory inputs from elsewhere in the brain. Because it's more representative of the earlier input than the later output, the BOLD result is just part of the story behind the "function" being measured.

But perhaps the most crucial shortcoming of BOLD, he says, lies in the fact that it is an indirect measure of neuronal activity and therefore "noisy". Researchers were already aware that fMRI didn't reveal everything that was going on, but Logothetis showed just how limited it is. The BOLD signal was an order of magnitude more noisy than the electrode signal. In other words, the fMRI scans showed only a tiny fraction of the activation revealed by electrodes inserted into individual neurons. So the absence of a signal, he concludes, cannot safely be interpreted as an absence of neural activity.

The implication of all this, says Logothetis, is that just because an area of a patient's brain "lights up" during an fMRI scan doesn't mean it's the seat of the behaviour being studied. Rather, a large network of brain regions largely invisible to fMRI may be responsible. He believes that many researchers have fallen into the trap of making oversimplified interpretations of their data. "What you see with MRI is truly the very tip of the iceberg," he says. He has had letters from people who have reconsidered their data in the light of his findings. But others refuse to recant. "They will continue saying, 'I've discovered the soul area, the love area, the hate area, the blonde hair area.' There is absolutely nothing you can do about that."

According to Karl Friston of the Functional Imaging Laboratory at University College London, most questions that interest neuroscientists today don't call for fMRI and single-neuron recording to be performed simultaneously. He says it was well known that BOLD had shortcomings, but with careful experimental design it was sufficient for most neuroscientists. Now Logothetis has worked out more detail of what the signal really is, the simultaneous technique will have "little impact on the use or interpretation of brain-mapping experiments".

Others argue that Logothetis's work is of limited use because his monkeys were anaesthetised. Last November, a group led by neurophysiologist Guy Orban of the Catholic University of Leuven, Belgium, in collaboration with Massachusetts General Hospital in Boston, published the first ever report of an fMRI study in monkeys that were awake. Monkey fMRI is still in its infancy, and all the work to date has been done in anaesthetised animals because of the problems caused by even the tiniest movement of an animal in the scanner. Orban argues that comparing brain activation patterns in drugged monkeys and awake humans is like comparing apples and pears.

But Logothetis believes that combining scanning and single-neuron recording, where the animal has to be anaesthetised, will tell us things we could previously never know about the brain. As well as working out just what the BOLD signal is, he hopes that these techniques can be combined more routinely in experiments to get a much richer picture of how the brain works, from the cellular and even molecular level to the big picture of which networks are used to do what.

And once you have used animal models to get the detail, you can then use BOLD by itself to find the signature of the same activity in people. This might be the perfect way to bridge the huge gap between human studies and decades of animal studies. It's early days, but with a brand new centre dedicated to brain imaging due to open in Tübingen in the next few years, equipped with one of the most powerful fMRI machines in the world, some of the most elusive puzzles in neuroscience may become solvable.

Logothetis has already begun working on one application: brain plasticity. In a study now under way in collaboration with Bruce Rosen and his colleagues at Massachusetts General Hospital, the Tübingen group is using both fMRI and single-neuron recording - though not yet simultaneously - to look at how the visual cortex of macaques reorganises itself in response to retinal damage.

The part of the visual cortex fed by the damaged retinal area initially falls silent, but the group may have confirmed a suspicion that this "blind spot" steadily shrinks. Over several months, neurons on its periphery start to respond to visual stimuli just outside the destroyed part of the retina. Without the global picture from fMRI and the detail from single cells, that theory would be difficult to prove.

Avision of the future

The next phase of the study will be to compare the monkey findings with fMRI scans of patients with permanent retinal damage caused by a condition called ischaemic optic neuropathy. This is relatively rare, but the researchers' findings could help them work out ways to improve the brain's recovery after other kinds of damage, such as strokes, trauma and degenerative conditions like Parkinson's. Plasticity is also the key to learning, so maybe we could find out why some things stick in our memory and others don't, and why our ability to learn declines as we age.

The team plans to combine other techniques to look at fluctuations in neurotransmitters as reorganisation proceeds. In a year or so, Logothetis predicts they will be generating "undreamed of" data, revealing the whole chemical and molecular underpinning of the recovery process, which could lead to new treatments to speed it up.

Another phenomenon the lab has its sights on is "binocular rivalry". When each of your eyes is simultaneously shown a different image, you perceive one at a time, oscillating between them every few seconds, rather than mixing the two together. This phenomenon continues to fascinate researchers 160 years after it was first described, because of the bearing it has on how the brain generates conscious awareness.

Single-cell techniques have been used on anaesthetised cats and monkeys, but that's only half the equation, because there's no way to find out what the animal is experiencing. Logothetis's group is among the very few to have carried out similar experiments in alert monkeys, which can be trained to indicate which picture they can see, and when it changes. Currently, one of the big puzzles is the discrepancy between human fMRI results - which seem to suggest that control lies with the first stages of visual processing -and single-cell studies which don't.

"It's an example of a broader class of questions we're beginning to realise is fundamentally difficult," says David Heeger, co-leader of a vision and imaging group at Stanford University. Current scanning methods simply don't reveal the extent to which, almost from the moment the stimuli are presented, both the earliest areas of visual processing and a huge network of higher-level cognitive regions are activated.

The brain uses parallel and distributed information-processing to build consciousness, just the sort of spreading networks that BOLD is not good at revealing. One of the problems researchers are grappling with is how this widespread activity is integrated to produce a coherent perceptual state, and one suggestion is that the integration occurs through the timing. In other words, by synchronising their firing, widely distributed neurons create a single working population whose activity contributes to the build-up of perceptual awareness.

In future, researchers could use Logothetis's technique to test this "temporal binding" theory, by showing what is changing in the brain, where and when, as the brain switches between the two images during binocular rivalry. Hence, it might show whether the early visual areas are driving the illusion, or merely obeying orders issued by other brain areas.

The Holy Grail of a real-time, non-invasive, high-resolution brain imaging technique carried out on conscious subjects may still be out of reach, and may always remain so. But the picture is getting less murky as researchers start to read the signals that we never become aware of. Sooner or later, Logothetis and his colleagues are going to know far more about our own minds than we do.

Laura Spinney is a freelance science writer

From issue 2361 of New Scientist magazine, 21 September 2002, page 38

A portrait of the brain Functional magnetic resonance imaging, or fMRI, uses radio waves and a magnetic field tens of thousands of times stronger than the Earth's field to pick out variations in blood and oxygen supply to the different tissues of the brain. In this strong magnetic field, the nuclei of all the hydrogen atoms in the tissues line up. The scanner hits them with a precisely timed pulse of radio waves, which knocks them out of alignment. As they return to line, the hydrogen atoms emit a radio signal which the MRI machine detects. It uses a secondary magnetic field to locate the signal in space, to map its position. And tissue differences are visible because the size of the overall signal varies according to how many hydrogen atoms are in each region- in body tissues, this generally reflects the amount of water in the tissue. That yields information about the brain's structure. To study its function, scientists take advantage of the fact that oxygenated and deoxygenated haemoglobin molecules behave slightly differently in a magnetic field. It is therefore possible to measure transient, local differences in the oxygen content of the blood. What is actually measured is a combination of blood flow, blood volume and oxygen use: the blood-oxygen-level-dependent (BOLD) signal. These signals are mapped onto the detailed anatomical picture using colour to represent the strength of the BOLD signal. A series of these "functional" images is taken that compare two conditions: an experimental condition and a control. For example, this might be the difference between the subject looking at a face and looking at a blank screen. The only problem left is knowing exactly how BOLD relates to brain cell activity.

For the latest from New Scientiist visit www.newscientist.com