The human brain is a wrinkled, interconnected mass of tissue weighing slightly less than three pounds. It it is one of the least understood portions of the human body. But that is changing. ASU computer scientists are helping medical researchers develop methods to see and study the brain better than ever before.

Encased within the skull’s protective shell, the brain and the secrets of its workings have eluded scientists for centuries. As recent as 100 years ago, one of the few ways to study the brain was to expose it. Medical researchers cut through skin and bone, peeled back the covering and began to poke, prod, and observe. They examined cadavers, cut out pieces of monkey brains, and applied electric shock to see the effects of brain damage and stimulation.

Today, brain researchers also study living human beings. But in the present era of medical imaging technology, new understanding increasingly comes from the electronic slice of a scanner in addition to the surgeon’s scalpel.

Medical imaging technology provides new research challenges for a team of Arizona State University computer scientists and engineers. Using computer science, graphics, and new mathematical methods, members of ASU’s Computer Aided Geometric Design Group are solving unique research problems and enhancing the utility of brain imaging techniques.

Thomas Foley with graduate students Yu-Kuang Chang, Ping Ning, and Lang-Sheng Yun

Sharing their technical expertise with physicians at the Good Samaritan Regional Medical Center in Phoenix, and with brain researchers at the University of Arizona, the ASU scientists are helping medical researchers to see, model, and study the brain better than ever before.

Positron emission tomography (PET) is the primary focus of the CAGD group’s research. PET is one of the medical imaging world’s most innovative tools. Most imaging techniques show what the body looks like on the inside. PET depicts body function.

Eric Reiman and the detector ring of Good Samaritan’s PET scanner

When aimed at the brain, a PET scanner can show nerve cells processing the image of a bright red car or sending directions for the hand to pick up a coffee cup. PET can trace thought patterns or follow the formation of sentences in the mind; and PET can highlight the happiness, rage, ecstasy, or doubt of the psyche. Doctors also can use PET images to pinpoint the abnormal cells of a tumor or illustrate the uncontrolled firing of nerve cells in an epileptic seizure.

PET was first introduced to the medical community about 15 years ago. The technique holds immense promise to provide insights into the nature of human brain function and disease. Researchers hope to use PET to characterize uniquely human features, such as language and emotion, and to investigate psychiatric disorders such as depression, anxiety, and schizophrenia. They also hope to improve the diagnosis and treatment of such conditions as learning disorders, cancer, epilepsy, and Alzheimer’s disease.

Initial studies have been promising. But according to Dr. Eric Reiman, director of the PET Center at Good Samaritan, a number of challenges must be addressed before PET’s full potential is reached. And that is why he has hooked up with the CAGD research team.

Completing The Picture
One of the problems with images provided by techniques like PET is that they are, in a sense, incomplete. A scanner usually takes a series of images from the top of the head to the base of the skull. When viewed on a computer screen, they appear as flat sections resembling slices of an apple. Each slice represents one plane of the brain at a certain moment.

But looked at individually, none of the slices offers any hint of what comes in front of or behind it in space, nor before or after it in time. To piece together a concept of the whole brain from such snapshots, doctors must view consecutive images side-by-side and estimate the relationships among them. The exercise is, in the words of one scientist, mostly guesswork.

The puzzle is much easier to assemble with the help of mathematics and computers. Powerful algorithms, or sets of instructions, allow the computer to stitch together a set of stacked slices, and transform them into a three-dimensional representation of the brain.

“We can shade (the brain) as if you were shining a spotlight on it. You get an idea of what the brain would look like if you took it out, dried it off, and illuminated it straight on,” says ASU’s Alyn Rockwood, an associate professor of computer science and engineering who specializes in computer graphics.

Computers also increase the precision of comparison, one of science’s most essential tools.

“If we can’t compare data across subjects and across labs, we can’t progress,” Reiman says.

By reducing the uncertainty inherent in eyeballing relationships between PET images, and by providing better ways to evaluate those relationships, the CAGD team is helping brain researchers refine knowledge gained by other means, says Reiman, who also serves as an associate professor of psychiatry at the University of Arizona.

For example, Reiman notes that most current psychiatric treatments were discovered accidentally. That is, when used to treat disorders unrelated to mental state, such treatments were found to help psychiatric symptoms as well.

“If you find things by accident, they are more likely to have side effects and be less effective,” Reiman says.

Once found, the treatments were developed using animal models of human behavior. While such studies are useful, treatments based on such models are prone to side effects in humans. That’s because animal models are imperfect imitations of the human condition.

“It’s difficult to study ‘happiness’ in an animal,” Reiman says.

PET, on the other hand, provides a direct “window” on human brain function. It allows researchers to study how the brain malfunctions in people with psychiatric disorders. PET also allows them to investigate, in humans, the effectiveness of various treatments.

Such direct methods should lead to “more benign and effective ways to treat and prevent” psychological disorders, says Reiman. The plan is to make PET itself as effective as possible.

Adapting Techniques
The primary instruments in such a plan are computer analysis techniques originally developed for the automobile and aviation industries. Beginning in the 1960s, principles of geometry were adapted for computers and used to design shapes such as car hoods, airplane wings, and wheel hubs.

These techniques now are used in areas as diverse as geographic sensing, pollution monitoring, robotics, and animation. Brain image analysis and modeling is one of the discipline’s newest applications.

“I think one of the exciting things about (ASU’s brain-imaging project) is that you can solve real problems,” says Robert Barnhill, ASU’s vice president for research and strategic initiatives.

In 1974, Barnhill and Richard Riesenfeld, both then at the University of Utah, coined the term Computer Aided Geometric Design to describe the mathematical aspects of computer-aided design. Barnhill heads the CAGD group at ASU.

The group specializes in techniques that are part of a broader new discipline called scientific visualization. Researchers use computer graphics and mathematical methods as tools to transform raw scientific data (usually numbers) into something easier to understand, such as pictures or images.

Using visualization techniques, a scatter of individual temperature readings from across the United States becomes a color-coded weather map that makes regional relationships and trends clear at a glance. In the case of PET images, color can be used to represent levels of brain cell activity.

PET can show body function by measuring a variety of biochemical processes. Researchers at Good Samaritan’s PET Center usually measure one of two indicators of nerve cell activity: blood flow and energy use. Both increase in areas where the brain is at work. Biological molecules chosen to reflect those processes are “tagged” with radioactive elements and either injected into the bloodstream or inhaled by a patient.

For example, an unstable form of oxygen (¹⁵O) can be attached to water or carbon dioxide and “traced” as it is carried through the bloodstream. Or a form of the sugar called glucose—the brain’s primary energy source—might be joined to radioactive fluorine (¹⁸F) to help scientists monitor its use by brain cells.

The radioactive elements “decay” by emitting small particles called positrons (named for the positive charge they carry). Positrons usually travel a short distance before they collide with negatively-charged electrons roaming free in the body’s tissues.

These subatomic masses annihilate one another when they collide, converting matter into energy. They also emit particles of light called photons. The photons zip away from the site of annihilation and are detected by sensors on opposite sides of the cylindrical PET scanner.

Millions of these annihilation reactions are recorded by the computer, which then determines the density of the radioactive molecules by calculating the number of photon bursts in a given area. Densities are recorded as numbers at points on a two-dimensional grid that represents that particular plane of the brain.

Like regions of the color-coded weather map, different areas are tinted according to the range their readings fall into. The result is an iridescent slice of the brain with the blotchy appearance of funky-colored marble cake. Very active areas might be red, inactive areas might be blue, and areas of intermediate activity might appear yellow or green.

Within such a range of activity, certain “hot spots” might stand out. For example, if a person were looking at an object when the scan was taken, images near the base of the brain would glow red along a small wedge penetrating inward from the back of the head. This is where the nerves that process visual images are located.

Reiman says that such hot spots are fairly easy to pick out in brain areas as well developed as the visual cortex. But subtle differences—the ones that tend to be the most interesting—are more difficult to find.

Subtle Subtraction
To make the subtleties stand out, brain researchers employ a technique called subtraction, which produces a picture of the difference between two images. The images may compare a person’s brain activity under two different research conditions.

For example, one image might depict the brain activity of a person comfortably at rest. The second might depict the same area while the person is reacting in fear to a coiled snake. Or, the images might be used to follow the progress of a disease and the effectiveness of intervention—say, by comparing tumor cell activity before and after radiation treatment.

Because each image is built of density values representing points on a grid, each is a mathematical entity. For that reason, two superimposed images can be subtracted from one another, averaged, or manipulated in numerous other ways useful for comparison.

Subtraction will cancel out any brain activity that is similar between the pictures. Reiman says that differences that were imperceptible before will “pop right out at you.”

Subtraction is a relatively simple technique. And while it certainly is useful, it can only be performed if the images to be subtracted fit together correctly. If the head moves between scans, the images are not subtractable because their points appear in different spatial coordinates. In order to work, the images must first be aligned.

To do alignment, the CAGD group draws upon the expertise of Arunabha Sen, an ASU associate professor of computer science. Sen uses the power of a massively parallel computer (MASPAR) system. The MASPAR is a sophisticated system that joins more than 8,000 individual processors to carry out a multitude of similar tasks all at once.

Simultaneous problem-solving is important in operations like alignment. Every grid point of an “at rest” image must be turned in space to align with the corresponding coordinate of its “active” image. Once aligned, the values at each point can be subtracted.

On a regular computer, each point—and each mathematical operation—must be dealt with one after another in a linear fashion. The MASPAR can do the alignment and math on many points at once, making it thousands of times faster. Sen is working to increase the speed further so that researchers can have results within minutes rather than hours.

One problem with both subtracted and whole PET images is that they are blurry. “It’s difficult to know just by looking at a PET image (precisely) where you are in the brain,” Reiman says.

Combination Solutions
To make it easier to correlate levels of brain function with specific structures in the brain, PET images often are combined with magnetic resonance imaging (MRI) scans, which provide detailed pictures of brain anatomy. In principle, the combination is somewhat like laying one piece of paper on top of the other and lining up the edges; it marries objects that were originally in different spatial coordinates. The advantage for brain researchers is that they can see exactly where the PET activity is located.

“We make both (images) a little transparent—so you can see both structural and functional values,” says Yu-Kuang Chang, an ASU doctoral student who is helping combine the PET and MRI images. Chang is one of three doctoral students, including Lang-Sheng Yun and Qing He, working on the brain image analysis project.

Initial methods to combine structural and functional information were developed at other institutions. The CAGD group is helping to refine those techniques, and extending them into three dimensions.

In an application that combines visualization with modeling, MRI and PET images are stacked and mathematically stitched into two separate three-dimensional representations.

Usually, subtracted PET images are used, and the functional model appears as a volume suspended in space. That volume is then embedded into a model of the brain’s outer surface produced by combining MRI slices. In the resulting picture, the functional blob glows from the inside of a transparent brain exterior, somewhat like a suspension of luminescent gas inside a wrinkled neon tube. Researchers no longer have to imagine the position of functional activity within the brain as a whole.

Researchers use CAGD techniques to assemble a complete visualization of the brain and its activity. Brain activity can be more easily related to brain anatomy.

The three-dimensional model is produced using a “best fit” procedure developed by ASU computer scientist Gerald Farin. His algorithm tells the computer how to put the images together. It simply moves the objects in space until they match up.

“It’s similar to what you do when you fit on a pair of shoes,” Rockwood explains. His analogy equates brain structure with a foot and brain function with a shoe. “You move your foot around in the shoe to find the most comfortable position,” he says. “But you don’t actually change the shape of your foot.”

Shape-changing serves a different purpose.

“Brains come in different shapes and sizes,” Rockwood says. As a result, simply overlaying and subtracting images taken from different people will probably give false results. Such comparisons are more likely to show irrelevant differences among individuals than to reflect the type of variation that interests most researchers.

Brain Warping
“We want to find (functional) differences between groups of subjects—the differences between patients and normal volunteers, differences between males and females, and the differences associated with aging,” Reiman says.

To make that possible, ASU computer scientist Tom Foley has developed a “brain deformation algorithm” that he says puts different brains “on a level playing field.”

The algorithm starts with several predetermined points inside a person’s brain. Each point was chosen for its known position on a “standard brain” that researchers use as a reference. Foley’s algorithm compares the positions of these points in MRI scans to those of the standard brain and then changes the shape of the MRI image to match the standard.

Functional PET values usually are attached to the MRI images before the procedure. When the structure is deformed, the function just goes along for the ride.

The procedure, which Foley calls “warping,” puts people’s brains in the same spatial coordinates so they can be compared using statistics and other analytical techniques. Any differences that appear among standardized images are likely to reflect characteristics common to a group rather than individual variance in brain shape or size.

Reiman calls Foley’s brain warping algorithm the CAGD group’s “great contribution” to PET image analysis. Other brain deformation algorithms have been developed before, but most of them adjusted only for brain size. As a result, they were useful only for limited kinds of comparisons.

“Linear deformation algorithms,” as such methods are called, allow researchers to compare the subtracted images of different people. They can be used to help pinpoint areas of the brain involved in fairly universal functions such as language or vision.

But the methods cannot be used to compare complete PET images from different people. Such comparisons are necessary to see differences in brain function between old people and young, those with and without psychological disorders, and between men and women. Foley’s “non-linear deformation algorithms” allow such comparisons.

These non-linear warping methods are much more challenging and have the potential for much greater scientific payoff, according to Reiman. “They make something possible that is not possible right now,” he says. For example, Foley’s algorithm not only adjusts for the shape of the brain, it also preserves certain structural characteristics that sometimes get shifted or lost using other warping methods.

“It’s very important that (an entity) on the left side (of the brain) doesn’t cross over to the right side,” Foley says. Standardizations are based on sets of mathematical operations. Their reliability rests on how closely the math describes the geometry of the brain—and that is up to the person who develops the math. As Foley explains, a mathematical equation doesn’t know there is supposed to be a line separating the two halves of the brain.

Three-dimensional brain image analysis techniques are so new that Barnhill says he couldn’t convince Reiman to give a presentation on them at a conference on the applications of geometric design held late last year. Still, Reiman is an enthusiastic supporter of the ASU research group.

“The CAGD team probably is the best in the country for mathematically describing three-dimensional objects,” Reiman says. But both he and the ASU scientists agree that some of their techniques are preliminary and need more refinement.

“Part of our research is directed at developing good ways to test whether certain deformations are better than others,” Foley says. “If you don’t stop and test how (the method) deforms (the image), you don’t know whether it does a good job.”

Foley also explains that members of the CAGD group are not simply working with images; they are creating and manipulating “three-dimensional volumes.” Images are simply two-dimensional pictures on a screen. At most, they require only a computer graphics program to be put together.

“Graphics is Nintendo,” says Barnhill, who explains that image display is only part of the group’s work. “CAGD is much more complex than just drawing pictures. There’s some substantial mathematics involved.”

Pictures are merely the end result.

Imaging Emotion
Despite the scholarly caution, pictures are the product that physicians and psychiatrists need. Some images produced by methods the CAGD team is working to improve already are being used in Reiman’s own research.

In a recent study on emotion, Reiman and his colleagues used subtraction and a linear standardization method to compare the brain activity of college-age women in two different cases: when they watched a film that elicited strong emotion, and when they remembered an intense emotional experience.

Images representative of each state were prepared by comparing the emotionally-active images to those obtained when the women watched films or recalled experiences that were “emotionally-neutral.” When the resulting film- and recall-stimulated areas were compared, Reiman found an unexpected difference in activity in a part of the brain called the limbic system. The limbic system is a series of structures in the middle of the brain that folds around and above the brain’s oldest evolutionary region, the brain stem.

“Traditionally, for the last 50 years, researchers have proposed that the limbic structures mediate emotion,” Reiman says. “But if you think about it, you should be able to divide emotion into its component parts.”

In Reiman’s study, certain structures in the limbic system lit up when the women became happy, sad, or disgusted in response to watching the film (an “external” source of emotion). But these same structures did not light up when the women felt the same emotions while reliving a previous experience (an “internal” source).

Reiman suggests his results indicate a biological distinction among what psychiatrists call “the three Es of emotion:” evaluation, experience, and expression. Evaluation is the cognitive process that labels events as emotionally significant.

Results showed that the women’s experience and expression of emotion were the same in both situations. That is, the women felt the same feelings and showed them in the same ways. Reiman thinks the limbic system is probably involved in evaluating external experiences and labeling them with emotional significance.

By contrast, two other areas of the brain, the medial frontal lobe and the thalamus, were stimulated by both film- and recall-generated emotion. The finding challenges traditional ideas about the seat of human emotion. It suggests that these areas—and not the limbic structures—are those common to all emotional responses. Reiman presented his findings to the Society of Neurosciences last November.

Reiman’s next step is to see if there are any functional differences among the three emotions he monitored. No matter what his results, they will add to an “explosion” in excitement and knowledge arising from studies using PET and other functional brain imaging techniques.

“Such techniques are helping to revolutionize our understanding of the human mind and brain,” he says. “They largely depend on the continued development of more precise methods of image analysis.”

It’s a reciprocal relationship that also drives advances on the CAGD side. The joint effort is seen as so important that the work received a major grant from the Flinn Foundation last year. It was only the second time that ASU scientists have received such an award for biomedical research, and one of the first times such an extensive collaboration has extended between ASU and the U of A.

The ASU group would like to automate their techniques enough so that researchers, doctors, and technologists could use them at the hospital. Rockwood and Sen look forward to developing “real-time” applications that animate brain function. They would like to show the sequence of an emotion or a feeling—where it starts and how it progresses—as it takes place.

“Eventually, we may be able to make models of the brain and do nonintrusive surgery,” Barnhill says. “This is really important for people to understand. We’re not experimenting on anybody; we’re going to model the surgery on the computer instead.”—Alana Mikkelsen

Brain-imaging research at ASU is supported by the Flinn Foundation, the National Science Foundation, and Silicon Graphics Inc. For more information, contact Robert E. Barnhill, Ph.D, Alyn Rockwood, Ph.D., or Thomas Foley, Ph.D., Department of Computer Science and Engineering, (602) 965-3190; or contact Eric Reiman, M.D., Good Samaritan Regional Medical Center, (602) 239-2373.

Go to: Engineering and Technology: Computer Science

Home Page | Arts | Business | Education | Engineering | Health | Life Sciences | Physical Sciences | Social Sciences

Who we are | Contact Info