Artificial Intelligence to reveal new data on autism. A team from Boston College, USA, published in the journal Science a study based on neuroimaging of people with an autism spectrum disorder. The researchers observed that behavioral differences between individuals with this disorder are related to variations in brain structure.
Understanding the heterogeneity of the brains of people with autism spectrum disorder (ASD) could be critical to improving their quality of life, as it would enable specific diagnoses and more targeted behavioral interventions.
Now, researchers at Boston College (USA) have used machine learning for a detailed analysis of brain images of people with autism and have revealed that behavioral differences between people with this disorder are related to variations in brain structure. The results of the study have been published in the journal Science.
The team used this artificial intelligence (AI) technique to study the magnetic resonance imaging data of more than 1,000 individuals with ASD and compared these images with those provided by computational simulations of what their brains would look like if they did not have this disorder.
As Aidas Aglinskas, a neuroscientist at the US institution and co-author of the study, explains to SINC, “the variations studied are differences in neuroanatomy that indicate altered development in certain regions of the brain”.
Expanded and compressed brain areas
In this study,” he continues, “we have investigated the volumetric changes associated with autism spectrum disorder, identifying the brain areas that are expanded or compressed compared to what would be expected if that person did not have it”.
The expert indicates that they observed that the brains of individuals with autism “differ from each other in many brain regions, including those associated with known ASD symptoms, such as those involved in social cognition, language, and motor cortexes.”
He also points out that the fact that different people with ASD “may have different regions affected could help explain the large individual differences in symptoms: those affected by this disorder often present different symptoms of varying severity,” he stresses.
Autism differs, both in symptoms and neuroanatomy, from one individual to another. Previous research had already hypothesized that there might not be a single set of neuroanatomical correlates common to all individuals with ASD.
“Confirming these proposals has been difficult because identifying neural alterations specific to ASD is a complicated task,” Aglinskas says. “Brains are different due to many factors, including genetic variation not due to the disorder, which is difficult to control for in a research study.”
The team overcame that barrier by employing machine learning to identify patterns of neural variability that are specific to ASD, which made it possible to identify the neural pathways specifically affected, says Aglinskas, who conducted the research with Boston College assistant professors of neuroscience Joshua Hartshorne and Stefano Anzellotti.
Hidden neuroanatomical variations
“ASD-related differences in brain anatomy may be ‘hidden’ among differences that are unrelated to ASD,” notes Aglinskas. “As a result, it has been difficult to identify variations in brain anatomy that are related to different symptoms. Therefore, we used AI to separate the differences related to the disorder from those that were not.”
Using MRI data from 1,103 participants, the authors used an analysis method to detect images created from the visual data of study participants.
In this case, the researchers used computer-detected patterns to create a simulation of what the brain of each individual with ASD would look like if he or she did not have ASD. This was made possible by machine learning techniques, which separate individual differences in brain anatomy into ASD-specific and unrelated features.
“We were surprised to find that, despite observing large variation in brain anatomy among individuals with autism across multiple dimensions, subjects did not cluster into distinct, categorical subtypes as previously thought,” Aglinskas notes.
“At the level of brain anatomy, individual differences within ASD may be better captured by continuous dimensions than by categorical subtypes,” according to the coauthor, “but it is important to note that this does not rule out the possibility that categorical subtypes can be found with other types of brain measurements, such as functional imaging.”
Looking ahead, the authors point to the need to understand in more detail how these neuroanatomical differences affect behavior. Anzellotti stresses that they plan to use AI tools to look beyond brain structure for ways to better understand ASD diagnoses and the behavior of affected individuals.
Individual differences within ASD.
“Two brains can be very similar in shape and still function differently,” Anzellotti comments. “There are a number of other aspects of the brain that we will have to look at to get a complete picture. Right now, we’re focusing on functional connectivity, a measure of how ‘wired’ the brain is.”
A big question is whether that will show us anything new about individual differences within ASD. The goal of this work is to be able to use brain imaging data to help develop personalized healthcare approaches for people with autism, the author’s stress.
Aidas Aglinskas highlights that machine learning tools, which allow finding subtle patterns in large data sets such as those provided by neuroimaging, “we were able to unravel individual-level neuroanatomical differences specific to ASD; and we observed that these differences in the structure were related to ASD symptoms, bringing us closer to precision medicine approaches in these cases.”
The power of neuroimaging and data
For his part, the Spanish researcher Santiago Canals, from the Institute of Neurosciences (CSIC- UMH), who did not participate in the study, comments that “for those who still have doubts, this study shows the power of neuroimaging to advance our knowledge of the brain, when the right techniques are applied correctly”.
Canals add that the work “also highlights the extraordinary value of large data repositories and their open use for science”.
In this research, which has employed a group of control subjects and another with autism spectrum disorder about 1,000 in total the authors “have been able to disentangle the individual variability associated with autism from other sources of irrelevant variability, including the type of scanner on which the image data are acquired. Beyond the specific advance on our knowledge of the structure of the autism spectrum, the study offers an interesting analysis strategy with general applications,” he concludes.
¡Stay tuned! Go and visit our blog! New articles every week.