Mapping How the Brain Matures
New analysis tools may eventually help doctors predict autism and other developmental disorders.
Using a new way of analyzing brain-imaging data, scientists have mapped out how the complex networks of connections in the brain evolve as children age. The researchers are now using the technology to examine how brain development in children with specific disorders, such as autism, veers off the norm. Ultimately, researchers aim to use the technology to predict, for example, whether a child at risk for autism will actually develop the disorder, or what treatments might work best for that individual.
Previous research by the same team at the Washington University School of Medicine had shown that between age five and 30, the short-range connections in the brain tend to weaken, while longer-range connections get stronger. In the new study, scientists plotted the trajectory for normal brain development and showed for the first time that they could determine a child’s development based on a brain scan.
Brain-imaging technologies that measure neural activity, such as functional magnetic resonance imaging (fMRI), have proven to be a valuable scientific tool over the last decade, but they’re still rarely used in medicine. That’s because most fMRI studies aggregate data from multiple people, making the technology inappropriate for diagnosing patients. “We realized that if the promise of fMRI is going to apply to understanding individual patients, we need to come up with a new approach, one that can take advantage of an individual’s data,” says Bradley Schlaggar, a pediatric neurologist at Washington University. Schlaggar led the new study.
The researchers developed a twist on functional connectivity–a relatively new approach to analyzing brain-imaging data. With this approach, fMRI is used to record spontaneous fluctuations in brain activity as an individual lies quietly in the brain scanner and does nothing in particular. Brain areas that are well-connected will fluctuate in synchrony, providing an indirect way of mapping the brain’s functional networks.
Schlaggar’s team studied more than 200 people age five to 30 and recorded brain activity in 160 brain areas. These areas had been identified in previous studies as important nodes of neural connectivity. The team then used a machine-learning algorithm to examine how activity in each of those areas fluctuated with every other area. The analysis revealed patterns in the data “that the naked eye wouldn’t be able to surmise,” Schlaggar says. He adds that the work requires such heavy computational processing that it probably would have been impossible on the computers available to the researchers five to 10 years ago.
Researchers can then look at the features that the algorithm deemed most important and learn what is changing in the brain’s functional architecture. Over the course of development, different brain regions became segregated from each other as the connections between them weakened. “It paints a picture of a brain that gradually organizes into distinct subnetworks,” says Olaf Sporns, a neuroscientist at Indiana University who was not involved in the study. “The networks appear to become more differentiated with respect to each other and more integrated internally.” It’s not yet clear exactly how this change affects the way children and teenagers think, but Sporns speculates that as the brain becomes more differentiated, with distinct neural networks, people get better at switching between tasks. Previous research has identified specific neural networks involved in memory and attention, or sensory and motor function.
The research is part of a growing trend to analyze how the brain’s networks interact, rather than focusing on specific brain areas. Schlaggar says this approach, as well as the ability to look at how multiple factors change simultaneously, allowed his team to analyze individual brain scans. The researchers could predict an individual’s age from just five minutes of brain-imaging data. Other scientists have applied a similar approach to other types of brain-imaging data, using it to determine what word someone is thinking of, for example. “I think this is going to become a new way of looking at the brain, and of finding new rules for how the brain develops,” says Martijn van den Heuvel, a neuroscientist at the Utrecht University Medical Center, who was not involved in the study.
Researchers are now using the same approach to study such developmental disorders as Tourette’s syndrome and autism. These and other developmental and psychiatric disorders have proven difficult to detect with traditional brain-imaging tools. “One of the questions I want to ask is, how can I predict the clinical course of an eight-year-old with Tourette’s? Will it be a transient tic disorder? Or full blown Tourette’s with [other psychiatric disorders]?” says Schlaggar. “We have little ability now to make predictions, and a tremendous need to make strong predictions about what’s going to happen to a patient over the next 10 years.”
Keep up with the latest in machine learning at EmTech MIT.
Discover where tech, business, and culture converge.
September 11-14, 2018
MIT Media Lab