Russian scientists use MRI scans to predict children's intelligence
Russians win fourth place in an NIH competition.
By Allen Laiche | Feb 11, 2020 | Print-friendly

Russian scientists took fourth place in an international competition of MRI-based methods for predicting adolescent intelligence. The scientists, from Russia's Skoltech Center for Computational and Data-Intensive Science and Engineering (CDISE), used new techniques that they said could gauge a child's "fluid intelligence," or brain power that the child is born with and that has little to do with acquired knowledge or skills.

Their model predicted children's fluid intelligence level and the "target variable"intelligence over time, which can be affected by learning and environmental nurturingindependently of age, gender, brain size, or the type of MRI scanner used, according to the researchers. They published their results in the journal Adolescent Brain Cognitive Development Neurocognitive Prediction.

The competition dates back to 2013, when the U.S. National Institutes of Health (NIH) launched a grand-scale study of adolescent brains to evaluate if and how teenagers' hobbies and habits affect their brain development. NIH scientists also wondered if MRI scans could predict a young person's intelligence level and compiled a massive database of brain scans of children ages 9-10.

The NIH subsequently jumpstarted the competition and made the database available to international research groups wishing to compete. To enter the competition, each group would have to build a predictive model based on brain images.

The Russian researchers accomplished the task by building a network architecture that applies several mathematical models to more accurately predict the outcomes, as well as an "ensemble method" for analyzing the MRI data.

"Our team develops deep learning methods for computer vision tasks in MRI data analysis, amongst other things," said CDISE Ph.D. student Ekaterina Kondratyeva. "With this approach, one can classify an image as it is, without first reducing its dimension and, therefore, without losing valuable information."