Education

New AI Study Links Learning Disabilities to Brain Hubs

University of Cambridge uses AI neural networks to study the biological brain.

Posted Apr 11, 2020

Geralt/Pixabay
Source: Geralt/Pixabay

What causes learning difficulties in children? Conventional neuroscience suggests that various regions of the brain serve as predictors for different disorders associated with learning disabilities. However, a new neuroscience study published in Current Biology on April 6, 2020 has findings that go against existing views on developmental disorders. University of Cambridge researchers used artificial intelligence (AI) machine learning to demonstrate how the brain’s neural hubs, rather than brain regions, are a strong predictor of learning-related cognitive problems in children.

Learning disabilities in children is not uncommon. Seven million students, or 14 percent of all public-school students in the U.S. between the ages of 3 and 21, received special education services under the Individuals with Disabilities Education Act (IDEA) in 2017-18, according to a National Center for Education Statistics 2019 report. In England, 14.9 percent of all pupils have special-education needs and 3.1 percent have an Education, Health and Care Plan in January 2019 according to the British Department for Education (DfE), a department of Her Majesty’s Government that is responsible for children’s services and education.

Learning disabilities may include specific diagnoses such as developmental language disorder (DLD), dyslexia (learning disability in reading), and dyscalculia (learning disability in math). Neurological disorders that can impact a child’s ability to learn include attention deficit and hyperactivity disorder (ADHD), autism spectrum disorder (ASD), and dyspraxia, a form of developmental coordination disorder (DCD) that affects motor coordination.

The diagnostic groups for learning disabilities may intersect and are highly wide-ranging. Additionally, even the symptoms are diverse within the diagnostic categories. Given these challenges, it made sense to the researchers to adopt a transdiagnostic strategy that spans diagnostic categories. The research team of senior author Duncan Astle, Joni Holmes, Joe Bathelt and Roma Siugzdaite from the MRC Cognition and Brain Sciences Unit at the University of Cambridge took a transdiagnostic approach to understand how brain differences relate to childhood learning difficulties using artificial neural networks to analyze data from brain images.

The researchers used data from cognitive, behavioral, and learning tests as well as magnetic resonance imaging (MRI) scans of 479 children for an artificial neural network that uses an unsupervised machine learning algorithm. Unsupervised machine learning can capture complex, non-linear relationships, which is useful in transdiagnostic studies.

The team used a Growing Hierarchical Self-Organizing Map (GHSOM), a type of artificial neural network that is a variant of a Self-Organizing Map (SOM) that is typically used for data visualization of high dimensional data space into a two-dimensional representation. GHSOMs have multiple layers arranged hierarchically, in which each layer has a number of independent SOMs.

The algorithm spotted different brain profiles in the data. The researchers then created whole-brain white-matter connectomes using diffusion-weighted neuroimaging in order to test what would happen when disconnecting hubs.

The researchers discovered that children with poorly connected brain hubs had severe and widespread cognitive impairments, and those with well-connected brain hubs either had no cognitive issues or had select cognitive deficits. The findings emphasize the importance of focusing on the areas of cognitive issues when it comes to targeted interventions and less on the diagnostic classification itself—opening the door for novel therapeutics that target the connectivity of the brain’s hub in the future.

Copyright © 2020 Cami Rosso All rights reserved.