Artificial intelligence to advance the study of autism

Autism spectrum disorders (ASD) are a group of neurodevelopmental disorders that affect the ability to interact and communicate socially and occur, among other things, along with repetitive, restricted and stereotyped behaviors; limited interests and unusual reactions to sensory stimuli such as light, noise, textures or temperature. Although there is no cure, there are treatments and these are aimed at enhancing strengths and providing supports that favor personal development, social inclusion and quality of life of people with ASD and their families. However, because ASDs are diverse, complex, and sometimes occur alongside other diseases or learning disorders, treatment (as well as its diagnosis) may sometimes not be easy to find. In this sense, in order to contribute to the design of new therapies, Argentine scientists used a computational tool to simulate the functioning of the cerebral cortex and to better understand the relationship between physiological and perceptual processes that would take place in people with ASD.

Children with ASD commonly have problems with nonverbal language, as they have difficulty understanding, and therefore using, hand gestures, eye contact, and facial expressions.

As detailed on the website of CONICET, the organization to which the researchers of this study belong, the balance between sensory information from the outside world and their own expectations occurs, in people with autism, in a less convenient way than in people without autism, since for the former, sensory perception intensifies and expectations are attenuated. The group of experts then set out to try to understand why that balance in people with ASD is different. They analyzed observations about the physiology of autism using a neural network trained from artificial intelligence techniques to process visual stimuli, while mimicking the behavior of the human primary visual cortex. Specifically, the team wanted to find out if perceptual differences between people with and without autism could be explained by differences in physiological functioning. This would allow, according to Rodrigo Echeveste, of the Institute for Research in Signals, Systems and Computational Intelligence of CONICET and the National University of the Littoral, to advance in the design of new therapies. The work was published in the journal Network Neuroscience and in it they describe how modifying the physiology in the artificial neural network “began to weigh less previous expectations and more the perception of stimuli.” Therefore, at least in the model, both aspects would constitute “two sides of the same coin.” Most models represent the neurotypical functioning of the brain. With this work we show that this approach is also very useful to understand the sensory processing of people with autism,” said Echeveste. In the future, he considers that “it would be interesting to be able to scale these models to be able to capture more complex behavioral phenomena, which will require new technical developments in the machine learning tools used to train these networks.”

Original source in Spanish