Summary: Different behavioral features of autism spectrum disorder (ASD) map to distinct neuroanatomical differences. Using artificial intelligence on MRI data, researchers identified individualized brain variations linked to ASD symptoms.
Source: Boston College
Boston College neuroscientists report that variability in behavior among people with autism spectrum disorder (ASD) corresponds closely to differences in brain anatomy. Their findings, published in the journal Science, show that AI-driven analysis of MRI scans can reveal which specific brain regions differ in individuals with ASD and how those differences relate to symptoms.
This advance offers a promising route to better understand the biological basis of ASD and to support development of personalized approaches to diagnosis and treatment.
The research team applied artificial intelligence to structural magnetic resonance imaging (MRI) data from more than 1,100 people with ASD. They used a novel contrastive machine-learning method to generate individualized simulations of what each person’s brain might look like without ASD, enabling direct comparison between actual and simulated anatomy.
“We found that different individuals with ASD show differences in distinct brain areas,” said Aidas Aglinskas, a Boston College postdoctoral researcher and co-author of the study. “By creating AI-simulated versions of each brain, we could isolate the ASD-specific anatomical features and identify which regions vary across individuals.”
A key challenge in previous MRI studies has been separating brain differences related to ASD from unrelated variability caused by genetics, age, measurement noise, or other factors. The new contrastive approach disentangles ASD-specific neuroanatomical variation from shared variation present in typical control participants, making it possible to link anatomy to individual symptom profiles.
The research team, including Assistant Professors of Neuroscience Joshua Hartshorne and Stefano Anzellotti, used pattern-detection techniques similar in concept to how deep generative models create realistic but synthetic images. Instead of fabricating photos, they used these models to simulate a neurotypical counterpart for each participant with ASD. Comparing real and simulated brains revealed the anatomical differences most likely to reflect ASD.
“ASD-related differences in brain anatomy can be masked by unrelated sources of variation,” Aglinskas explained. “Our AI method teases apart those sources, revealing relationships between individual anatomical differences and specific symptoms that were previously hidden.”
Among the notable findings, the researchers observed substantial neuroanatomical variation across ASD individuals, but these variations did not form clear, discrete subtypes. Instead, differences were better represented as continuous dimensions that affect particular sets of brain regions. In other words, at the level of brain structure, ASD appears to be organized along graded neuroanatomical axes rather than as distinct categorical brain subtypes.
This continuous, multi-dimensional structure of ASD-specific variation helps reconcile conflicting hypotheses about whether ASD should be divided into discrete neuroanatomical subtypes. The new evidence suggests that while behavior and symptoms vary markedly across individuals, the underlying anatomical differences form gradients across brain regions rather than distinct clusters.
Looking ahead, the authors emphasize the need to link these anatomical findings to brain function and behavior. “Two brains may look similar structurally yet operate differently,” said Stefano Anzellotti. The team plans to extend their AI tools to examine functional connectivity and other aspects of brain organization to better understand how structural variation translates into differences in cognition, social interaction, and other ASD-related behaviors.
A major goal of this line of research is to enable the use of brain imaging and AI to inform tailored clinical care. By identifying the specific brain systems affected in each individual, clinicians may eventually be able to design more targeted interventions and supports that address the particular neural and behavioral profile of a person with ASD.
About this AI and ASD research news
Author: Press Office
Source: Boston College
Contact: Press Office – Boston College
Image: The image is in the public domain
Original Research: Closed access.
“Contrastive machine learning reveals the structure of neuroanatomical variation within Autism” by Aidas Aglinskas et al. Science
Abstract
Contrastive machine learning reveals the structure of neuroanatomical variation within Autism
Autism spectrum disorder (ASD) is highly heterogeneous. Identifying systematic individual differences in neuroanatomy could inform diagnosis and personalized interventions.
The challenge is that these differences are entangled with variation due to other causes: individual differences unrelated to ASD and measurement artifacts.
We used contrastive deep learning to disentangle ASD-specific neuroanatomical variation from variation shared with typical control participants. ASD-specific variation correlated with individual differences in symptoms.
The structure of this ASD-specific variation also addresses a long-standing debate about the nature of ASD: At least in terms of neuroanatomy, individuals do not cluster into distinct subtypes; instead, they are organized along continuous dimensions that affect distinct sets of regions.