Read our COVID-19 research and news.

A noisy map of myelin in the brain (left) becomes more stable and sensitive (right) thanks to a new mathematical technique.

M. BOUHRARA ET AL., NEUROIMAGE, 127, 456 (2016)

Clever math enables MRI to map molecules implicated in multiple sclerosis, other diseases

BOSTON—MRI scanners can map a person's innards in exquisite detail, but they say little about composition. Now, physicists are pushing MRI to a new realm of sensitivity to trace specific biomolecules in tissues, a capability that could aid in diagnosing Alzheimer's and other diseases. The advance springs not from improved scanners, but from better methods to solve a notoriously difficult math problem and extract information already latent in MRI data.

The new techniques, described this month at a meeting of the American Physical Society here, could soon make the jump to the clinic, says Shannon Kolind, a physicist at the University of British Columbia (UBC) in Vancouver, Canada, who is using them to study multiple sclerosis (MS). "I don't think I'm being too optimistic to say that will happen in the next 5 years," she says. Sean Deoni, a physicist at Brown University, says that "any scanner on the planet can do this."

An MRI scanner uses magnetic fields and radio waves to tickle the nuclei of hydrogen atoms—protons—in molecules of water, which makes up more than half of soft tissue. The protons act like little magnets, and the scanner's strong magnetic field makes them all point in one direction. A pulse of radio waves then tips the protons away from the magnetic field, causing them to twirl en masse, like so many gyroscopes. The protons then radiate radio waves of their own.

The scanner tracks how that signal decays in time, which happens in two ways. The twirling protons "relax" back to the direction of the magnetic field, and they also interact with one another. Those two processes are described by a pair of time constants, T1 and T2, which are like half-lives in radioactive decay. Applying various sequences of radio pulses, the scanner measures the time constants, which depend on the water molecules' chemical environment and, hence, the tissue type. By tracking how the constants vary across the body, the scanner maps the structure of soft tissue.

However, a standard MRI scan cannot trace specific biomolecules because it treats each millimeter-size volume element of tissue as if it contains a single material. To look for a particular biomolecule, researchers must assume that each voxel contains at least two materials—the target biomolecule and everything else. They then have to measure at least two sets of time constants: one pair for the water surrounding a biomolecule of interest and another for the water farther away.

Disentangling those two sets is devilishly difficult, akin to listening to the same note played simultaneously on two xylophones and telling which note faded faster. To mathematicians, the problem is "ill-posed," meaning that a little noise will cause estimates of the overlapping time constants to vary wildly, Richard Spencer, a physicist and physician at the National Institute on Aging (NIA) in Baltimore, Maryland, explained at the meeting.

In 1994, Alex MacKay, a physicist at UBC, focused on teasing apart two values of the constant T2 to produce the first MRI brain map showing myelin, the fatty molecule that insulates nerve fibers. At first, he had to scan for 25 minutes to image a single slice through the brain. In 2006, Deoni published a more complex pulsing protocol that extracts two values for each of T1 and T2 and can scan the whole brain in 10 minutes.

In recent years, Spencer and his NIA colleague Mustapha Bouhrara developed a statistical approach that improves on MacKay's and Deoni's techniques. When disentangling two decaying signals, conventional statistical analysis yields estimates for time constants that are precise but buffeted by noise. So Spencer and Bouhrara employ a so-called Bayesian approach, which yields a probability distribution for each time constant.

Although less precise, the probability distributions are far more reproducible. "It's a huge contribution, absolutely," Deoni says.

The analysis takes up to 4 hours of computation, but the payoff could be big. In 2017, Spencer and Bouhrara reported they could map molecules called proteoglycans in cartilage in the knee, which could help physicians decipher the origins of osteoarthritis. Last year, they reported that adults with mild cognitive impairment and Alzheimer's disease have less myelin in their brains than healthy subjects, which jibes with other evidence that loss of myelin may play a role in the disease.

Similarly, neurologists have long known that MS patients develop brain lesions in which myelin is lost. Using the new scans, Kolind has found that some MS patients lose myelin elsewhere in the brain, too, suggesting the loss precedes the lesions. Deoni and colleagues are studying myelin in 1200 healthy children over time to see how it develops with age. The data suggest social factors such as wealth can also lead to differences in myelination, Deoni says.

Spencer says his team is eager to move into diagnostic applications. "Our hope always has been to turn this into a clinical tool rather than just a toy for research." To that end, Kolind and others at UBC are scanning healthy subjects to develop a map of normal myelin against which abnormal scans might stand out. Still, researchers need to compare the methods with clinical data and tissue samples for specific diseases, which could take years.