A standard clinical MRI scan cannot diagnose autism. If you or your child had a brain MRI tomorrow, the radiologist would not be able to look at the images and say “this person is autistic.” The American Academy of Pediatrics and the American Academy of Neurology do not recommend routine brain MRI as part of an autism assessment, reserving it only for cases where something unusual in a person’s medical history or neurological exam warrants a closer look.1PubMed Central. Yield of brain MRI in children with autism spectrum disorder That said, research-grade MRI has uncovered real, measurable differences in how autistic brains develop and connect, and those findings are fueling efforts to eventually turn brain imaging into a diagnostic tool.
What Researchers Actually See on Brain Scans
Decades of MRI research have identified a handful of structural patterns that tend to show up more often in autistic individuals than in non-autistic controls. The most consistent finding involves brain size during early childhood. Studies show that abnormal brain overgrowth occurs during the first two years of life in children later diagnosed with autism, with the most pronounced enlargement in regions tied to social, emotional, and language processing.2PubMed. Brain development in autism: early overgrowth followed by premature arrest of growth That early burst of growth then slows or stalls, so the gap between autistic and non-autistic brain size narrows with age. By age two, brain volume enlargement is already present, and it persists at roughly the same degree into the preschool years without further acceleration.3PubMed Central. Early Brain Overgrowth in Autism Associated with an Increase in Cortical Surface Area Before Age 2
The overgrowth appears to be driven more by the brain’s surface area expanding than by the cortex getting thicker. In young boys with autism, cortical thickness was comparable to that of non-autistic peers, but surface area was significantly greater, especially in a subgroup with disproportionately large heads.4PubMed Central. Increased Surface Area, but not Cortical Thickness, in a Subset of Young Boys With Autism Spectrum Disorder In that subgroup, the overgrowth persisted throughout childhood without shrinking back toward typical levels.5PubMed Central. Longitudinal Evaluation of Cerebral Growth Across Childhood in Boys and Girls With Autism Spectrum Disorder Later in life, large-scale studies comparing autistic adults to non-autistic adults and people with other conditions have found thicker frontal cortex in the autism group.6PubMed Central. Subcortical Brain Volume, Regional Cortical Thickness, and Cortical Surface Area Across Disorders: Findings From the ENIGMA ADHD, ASD, and OCD Working Groups
These are group-level statistical trends, not landmarks a radiologist can spot on any individual scan. Many autistic people have brain volumes well within the typical range, and some non-autistic people have unusually large brains. The overlap between groups is enormous, which is precisely why a clinical MRI is uninformative for diagnosis.
How Brain Regions Talk to Each Other
Structural MRI shows the shape and size of brain regions. Functional MRI, or fMRI, captures which regions are active at the same time and how tightly they coordinate. This coordination, called functional connectivity, has become one of the most studied aspects of the autistic brain, and it tells a different story than anatomy alone.
A network that researchers focus on heavily is the default mode network, a set of brain areas that are most active when a person is daydreaming, thinking about themselves, or reflecting on social situations. A meta-analysis pooling data from multiple fMRI studies found that autistic individuals show decreased connectivity in key parts of this network, including the medial prefrontal cortex, the precuneus, and the angular gyrus.7PubMed. Resting-state abnormalities in functional connectivity of the default mode network in autism spectrum disorder: a meta-analysis In plainer terms, the front and back of this network do not communicate as strongly in autistic individuals as they do in non-autistic controls.8PubMed Central. Resting state fMRI reveals a default mode dissociation between retrosplenial and medial prefrontal subnetworks in ASD despite motion scrubbing There is also evidence that the typical boundary between the default mode network and task-focused networks is blurrier in autism, with some regions flipping from negative to positive correlation.9PubMed Central. Integration and Segregation of Default Mode Network Resting-State Functional Connectivity in Transition-Age Males with High-Functioning Autism Spectrum Disorder: A Proof-of-Concept Study
Another consistent finding involves the “social brain,” the collection of regions that handle face recognition, emotional expression reading, and eye-gaze tracking. Reviews of fMRI studies have found that autistic individuals tend to show lower activity in the fusiform gyrus, the posterior superior temporal sulcus, the amygdala, and parts of the prefrontal cortex during tasks that involve processing faces and social cues.10PubMed Central. Functional magnetic resonance imaging of autism spectrum disorders More detailed work within the fusiform gyrus specifically has shown that the structural and functional deviations in this area are lateralized: the right hemisphere shows increased deviations in face-selective subregions, while the left hemisphere shows decreased deviations across both face-selective and retinotopic zones.11Nature Mental Health. A multimodal neural signature of face processing in autism within the fusiform gyrus The pattern is not as simple as “less activation equals autism,” though. Task design matters: when autistic participants match emotional expressions to word labels or passively view faces, the group differences often disappear.12PubMed Central. Face processing in autism spectrum disorders: from brain regions to brain networks
White Matter and the Brain’s Wiring
Beyond the gray matter where neurons cluster, the brain’s long-range wiring runs through bundles of white matter tracts. A type of MRI called diffusion tensor imaging can map these pathways and measure how organized the fibers are. A review of 48 diffusion imaging studies found that autistic individuals tend to have less organized white matter across many brain regions, most consistently in the corpus callosum (which connects the two hemispheres), the cingulum (involved in emotion and memory), and tracts running through the temporal lobe.13PubMed Central. Diffusion tensor imaging in autism spectrum disorder: a review More recent work mapping tracts in fine-grained detail confirmed these differences across nearly every major connecting pathway between frontal, temporal, parietal, and occipital regions, though the differences tended to be localized to specific portions of each tract rather than affecting them uniformly.14PubMed Central. Tract-specific analyses of diffusion tensor imaging show widespread white matter compromise in autism spectrum disorder
What this means functionally is that the autistic brain may have subtle differences in how efficiently signals travel between distant regions. This dovetails with the connectivity findings from fMRI: if the physical cables connecting regions are less tightly bundled, it makes sense that the functional conversations between those regions would look different too.
Can Machine Learning Close the Gap?
If the human eye cannot reliably distinguish an autistic brain from a non-autistic one, perhaps algorithms can. Researchers have been feeding MRI data into machine-learning classifiers for over a decade, and the accuracy numbers illustrate both the promise and the problem. Early efforts using structural MRI features on large, multi-site datasets topped out around 60% accuracy, barely better than a coin flip.15PubMed Central. Brain imaging-based machine learning in autism spectrum disorder: methods and applications Functional connectivity data pushed that to about 67%, and more sophisticated deep-learning architectures have nudged accuracy into the mid-70s on the same multi-site dataset.15PubMed Central. Brain imaging-based machine learning in autism spectrum disorder: methods and applications Some models using carefully selected features and neural networks have reported accuracy as high as 82%.16Frontiers in Neuroinformatics. Machine learning for autism spectrum disorder diagnosis using structural magnetic resonance imaging: Promising but challenging
Those numbers might sound encouraging, but context matters. Most of these models are trained and tested on the same large open dataset, called ABIDE, which pools scans from multiple countries and scanning sites. That introduces noise: different scanners, different imaging protocols, different demographic profiles. The highest accuracies tend to come from studies that either use smaller, more homogeneous samples or employ more aggressive feature selection, which risks overfitting to the peculiarities of one dataset. When models are tested on completely independent data from a new site, accuracy usually drops. Newer efforts are using explainable AI to identify which brain regions drive the classification, working with hundreds of participants across many sites, but the field has yet to produce a model robust enough for individual-level clinical use.17PubMed. Identification of critical brain regions for autism diagnosis from fMRI data using explainable AI: an observational analysis of the ABIDE dataset
Research combining MRI with other modalities like EEG has shown improved classification accuracy compared to either modality alone, which suggests that the future of brain-based autism detection may lie in fusing multiple types of data rather than relying on MRI in isolation.182nd International Conference on Cognitive, Green and Ubiquitous Computing, IC-CGU 2025. Synergizing EEG and MRI for Autism Classification: Exploring the Feasibility of Machine Learning Models
Why Autism Is So Hard to See on a Scan
One of the biggest obstacles to a brain-scan-based diagnosis is that autism is not one thing neurologically. The condition is highly heterogeneous at both the biological and behavioral level, meaning two autistic people may have very different brain profiles.19PubMed. Parsing Autism Heterogeneity: Transcriptomic Subgrouping of Imaging-Derived Phenotypes in Autism When you average across hundreds of autistic brains to find a “signal,” the individual variation within the autistic group can be as large as the difference between autistic and non-autistic groups. Researchers are increasingly trying to parse that variation into meaningful subgroups, hoping that brain imaging differences will become clearer once the autism umbrella is split into biologically coherent subtypes.
Sex differences compound the challenge. A systematic review of structural, functional, and diffusion MRI studies found that autism interacts with sex-related brain differences in complex ways. In regions where non-autistic males typically have larger volumes than non-autistic females, autistic females often showed patterns closer to non-autistic males, while autistic males sometimes resembled non-autistic females.20PubMed Central. Brain-based sex differences in autism spectrum disorder across the lifespan: A systematic review of structural MRI, fMRI, and DTI findings Since the overwhelming majority of autism neuroimaging research has been conducted on male participants, these sex-related differences mean that findings from male-heavy studies may not generalize to autistic women and girls at all.
Overlapping conditions also muddy the picture. Autism frequently co-occurs with ADHD, and both conditions are associated with brain differences that partly overlap. A large brain-charting study found that while ADHD was linked to broadly smaller cortical and white matter volumes, autism was associated with increased cortical thickness and volume in more specific areas like the superior temporal gyrus, along with greater ventricular volume.21PubMed Central. Brain-charting autism and attention deficit hyperactivity disorder reveals distinct and overlapping neurobiology In a person who has both conditions, these opposing patterns could partially cancel each other out on a scan, making the autism-specific signal even harder to isolate.
The Infant Brain and Early Detection
The most striking MRI results so far come not from trying to diagnose autism in older children or adults, but from scanning the brains of infants at high familial risk before any behavioral signs appear. A study scanning 59 six-month-old infants who had older siblings with autism used functional connectivity MRI to predict which babies would receive an autism diagnosis at age two. The algorithm correctly identified 9 of 11 infants later diagnosed, with no false positives among the 48 babies who did not go on to develop autism.22PubMed Central. Functional neuroimaging of high-risk 6-month-old infants predicts a diagnosis of autism at 24 months of age
A separate study tracked brain growth in 106 high-risk and 42 low-risk infants and found that rapid expansion of the brain’s surface area between six and twelve months preceded the brain volume overgrowth seen between twelve and twenty-four months in the 15 babies later diagnosed. A deep learning algorithm using surface area data from six- and twelve-month scans predicted the autism diagnosis at 24 months with a positive predictive value of about 81% and sensitivity of 88%.23PubMed Central. Early brain development in infants at high risk for autism spectrum disorder
These are small studies of a very specific population (infants already known to be at elevated risk because of family history), and they have not been replicated at the scale needed for clinical adoption. But they represent the strongest evidence that brain imaging might someday contribute to early identification, possibly before behavioral screening tools can pick up anything unusual. Whether this approach will work for children without a family history of autism remains an open question.
Brain Chemistry Through the MRI Lens
Standard MRI looks at structure. Functional MRI tracks blood flow as a proxy for neural activity. A third MRI technique, called magnetic resonance spectroscopy, measures the concentrations of specific brain chemicals. This has been used to probe a long-standing theory about autism: that the balance between excitatory and inhibitory signaling in the brain is disrupted.
A recent systematic review and meta-analysis of spectroscopy studies found that autistic individuals have lower concentrations of GABA (the brain’s primary inhibitory chemical) and lower levels of NAA (a marker of neural health), with these reductions most pronounced in children and in limbic brain regions relevant to core autism traits.24PubMed Central. Neurometabolite differences in Autism as assessed with Magnetic Resonance Spectroscopy: A systematic review and meta-analysis Smaller studies have found that while average GABA levels in the anterior cingulate cortex did not differ between autistic and non-autistic boys, lower GABA correlated with more severe autism symptoms within the autistic group.25PubMed Central. Brain MR spectroscopy in autism spectrum disorder-the GABA excitatory/inhibitory imbalance theory revisited This is a good example of how neuroimaging findings in autism tend to play out: a signal emerges at the group level or correlates with symptom severity, but the between-group overlap is too large for the measurement to work as a diagnostic marker on its own.
Blood Flow Patterns in the Autistic Brain
Yet another MRI-based technique, arterial spin labeling, measures blood flow through the brain without requiring contrast dye. A study of autistic children found lower blood flow in the frontal lobe, hippocampus, temporal lobe, and caudate nucleus compared to non-autistic children, with the number of affected regions increasing with age.26PubMed Central. Application of Three-Dimensional Pseudocontinuous Arterial Spin Labeling Perfusion Imaging in the Brains of Children With Autism A separate study found a more nuanced pattern in children: some areas showed increased blood flow while others showed decreased flow, and the relationship between blood flow and functional connectivity that exists in non-autistic children was disrupted in the autistic group. Greater disruption in both measures was associated with more social difficulties.27PubMed Central. Altered resting perfusion and functional connectivity of default mode network in youth with autism spectrum disorder
Perfusion imaging is still a niche corner of autism research, but it offers something that structural and functional MRI do not: a physiological measurement that might reflect how efficiently the brain is being supplied rather than how big it is or which areas light up together. Whether this adds enough unique information to improve diagnostic classification remains to be seen.
What Getting an MRI Is Actually Like for Autistic People
There is a practical side to the question of MRI and autism that rarely makes it into research discussions. MRI scanners are loud, enclosed, and require the person to lie completely still for extended periods, all of which can be extremely difficult for autistic individuals who are sensitive to noise, confined spaces, or physical discomfort. A survey of parents and caregivers found that successful scans depended on staff making specific adjustments: changing the lighting, letting the person feel the scan bed beforehand, allowing a parent to stay in the room and hold their hand, offering mirror glasses so the person could maintain visual contact with a caregiver, and having a specialist explain the procedure using concrete tools like Lego models or pre-visit videos.28PubMed Central. Strategies to improve the magnetic resonance imaging experience for autistic individuals: a cross-sectional study exploring parents and carers’ experiences
A separate survey of autistic adults in the UK echoed these themes, identifying effective communication, an adjusted MRI environment, scans tailored to the individual’s specific needs and preferences, and well-trained staff as the key factors that made scans possible without sedation.29PubMed Central. Toward Autism-Friendly Magnetic Resonance Imaging: Exploring Autistic Individuals’ Experiences of Magnetic Resonance Imaging Scans in the United Kingdom, a Cross-Sectional Survey This matters both clinically and for research, because if the scanning experience itself is a barrier, the data we collect will be biased toward autistic individuals who can tolerate the procedure most easily, which skews our understanding of what autistic brains look like as a whole.
The Developmental Trajectory Problem
One reason brain-imaging findings in autism resist being turned into a simple diagnostic test is that the differences are not static. They change with age, and they change differently depending on the brain region. In early childhood, surface area in certain regions is larger in the autistic group. But longitudinal studies tracking cortical development show that some of these surface area differences narrow over time, while others actually widen, with autistic children showing a smaller decline in surface area with age than their non-autistic peers in regions like the banks of the superior temporal sulcus and the supramarginal gyrus.30PubMed Central. Development of cortical thickness and surface area in autism spectrum disorder A scan that looked for the early-overgrowth pattern might pick it up at age two but miss it entirely by age ten, when the pattern has been replaced by a different set of regional differences.
This moving target means that any diagnostic algorithm would need to be calibrated not just to “autism vs. no autism” but to a specific age range, possibly a specific sex, and potentially a specific autism subtype. The field is still working on defining those subtypes reliably, let alone building age- and sex-specific normative brain charts against which to compare an individual scan. Until that groundwork is more complete, brain MRI will remain a powerful research tool for understanding autism’s neurobiology while falling short of the precision needed for clinical diagnosis.