Can Autism Be Seen on a Brain Scan?

No brain scan can currently diagnose autism. Despite decades of neuroimaging research revealing real, measurable differences in brain structure and function between autistic and non-autistic individuals, none of these findings are reliable enough at the individual level to replace behavioral assessment. A 2024 systematic review and meta-analysis put it plainly: MRI-based diagnosis of autism is “potent and desired, but not ready for clinical application yet.” The gap between group-level patterns that show up clearly in studies and the kind of person-by-person accuracy a clinician needs turns out to be wide, and the reasons for that gap are themselves revealing about what autism is.

What Brain Scans Actually Show in Autism Research

Researchers have been scanning the brains of autistic individuals since the 1980s, and they have found a lot. The differences are not imaginary or trivial. They span brain size, cortical thickness, the wiring between regions, chemical signaling, and how different brain areas coordinate during rest and tasks. The problem is not that nothing shows up. The problem is that too many things show up, they vary widely from person to person, and they overlap with patterns seen in other conditions.

Structural MRI studies consistently find that young children who go on to receive an autism diagnosis tend to have larger-than-expected brain volumes in the first few years of life. One study found that two-year-olds with autism had roughly 9% larger cerebral cortex volume compared to controls, and this enlargement persisted to ages four and five without further acceleration in growth rate during that interval.¹ A separate study of high-risk infants found that the accelerated growth in total brain volume became apparent in the second year of life, between 12 and 24 months, with large effect sizes distinguishing infants later diagnosed with autism from those who were not.²

This early overgrowth pattern, however, does not persist uniformly across the lifespan. The cortex appears to follow a different thinning trajectory in autism: faster-than-typical expansion in early childhood, then accelerated thinning during later childhood and adolescence, followed by a slower rate of thinning in early adulthood.³ By early adulthood, cortical thickness in autistic individuals may converge with that of non-autistic peers.⁴ The ENIGMA consortium, which pooled brain imaging data from thousands of people, confirmed that the largest cortical thickness differences between autistic and non-autistic individuals appear around adolescence.⁵

So if you scanned a six-year-old with autism and a 25-year-old with autism, the structural differences you might detect would look quite different, and might not be detectable at all in the older person. That age dependence is one reason a single scan cannot serve as a diagnostic tool.

Differences in Brain Wiring and Connectivity

Beyond the shape and size of brain structures, researchers have examined how different brain regions communicate with each other. Functional MRI measures blood flow patterns to infer which regions are active at the same time, and diffusion tensor imaging tracks the white matter tracts that physically connect distant brain areas.

On the functional side, studies have repeatedly found altered connectivity in networks that handle social cognition, attention, and self-referential thought. The default mode network, the salience network, and cerebellar regions have been identified as areas where connectivity patterns differ most between autistic and non-autistic individuals.⁶ Some of these differences correlate with sensory sensitivities that many autistic people experience. Youth with autism showed greater activation in primary sensory areas as well as the amygdala and orbitofrontal cortex in response to sensory stimuli, and the degree of this over-reactivity tracked with parent-reported sensory sensitivity, even after accounting for anxiety.⁷

White matter studies tell a complementary story. A review of 48 diffusion tensor imaging studies found that autistic individuals tended to show reduced structural integrity across many white matter tracts, most consistently in the corpus callosum (the thick bundle connecting the two hemispheres), the cingulum, and temporal lobe pathways.⁸ Another study found widespread white matter compromise across many tracts without finding a single location where the autism group showed stronger structural organization than the control group.⁹

One proposed explanation for this pattern involves chronic neuroinflammation. Evidence points to sustained microglial activation in the brains of autistic individuals, and when these immune cells stay active too long, they can damage synaptic connections and contribute to the underconnectivity seen on scans.¹⁰ This is still an area of active investigation, not a settled explanation, but it illustrates how structural and functional findings can point toward underlying biological processes.

Can a Computer Detect Autism From a Brain Scan?

The most ambitious efforts to make brain scans diagnostically useful have involved training machine learning algorithms to classify individuals as autistic or non-autistic based on imaging data. The results so far are encouraging in small, well-controlled studies but have not crossed the threshold needed for clinical use. A systematic review and meta-analysis of these algorithms found that accuracy was “considered acceptable” only with structural MRI data, and even then, the authors cautioned that substantial limitations meant the results could not yet be trusted in a real clinical setting.¹¹

A central challenge is what happens when you try to combine data from multiple hospitals and scanners. Different MRI machines, scanning protocols, and patient populations introduce variation that has nothing to do with autism. This data heterogeneity makes it genuinely difficult for algorithms to separate signal from noise.¹² An algorithm trained at one research center may perform impressively on data from that center and poorly on data from another, which is exactly the opposite of what you need for a diagnostic tool that works everywhere.

There is also a less obvious technical problem: head motion. People with autism, especially children, tend to move more during scans. Even small head movements create artifacts in the data that can mimic or mask real connectivity differences. Research has shown that standard motion-correction methods may not fully address this, and that different correction approaches can change the apparent brain differences identified.¹³ If the thing your algorithm is detecting is partly “this person moved more during the scan,” you do not have a biomarker for autism.

The Infant Prediction Studies

Some of the most striking results in this field come from studies that scanned babies at high familial risk for autism before they had any diagnosable symptoms. In one study, functional connectivity MRI of 59 six-month-old infants with an older autistic sibling correctly predicted which babies would receive an autism diagnosis at 24 months. The algorithm achieved 100% positive predictive value, meaning every baby it flagged as likely to develop autism did receive the diagnosis. It correctly identified 9 of 11 babies who were later diagnosed, and none of the 48 babies who were not diagnosed were incorrectly flagged.¹⁴

Those numbers are remarkable, and they generated significant excitement when published. But the sample was small (59 infants, 11 diagnosed), the infants were pre-selected for high risk, and the results have not been replicated at a scale that would justify routine screening. A separate study of infant siblings found that brain volume and cortical surface area in the first two years of life correlated with the severity of the older sibling’s autism symptoms, suggesting a genetic component to early brain overgrowth patterns.¹⁵

These infant studies demonstrate that brain differences associated with autism can precede behavioral symptoms by months, which is scientifically significant. Whether scanning six-month-old babies will ever become a practical screening strategy is a separate question with significant cost, access, and ethical dimensions that go well beyond technical accuracy.

Why Autism Is Especially Hard to See on a Scan

Autism is not one thing. The heterogeneity of the condition is probably the single biggest obstacle to scan-based diagnosis, and it is worth understanding why. Two autistic people may share a diagnostic label and have strikingly different neurobiological profiles. Recent research has used computational approaches to identify distinct “biotypes” of autism, subgroups that share neurobiological features beneath their clinical similarity.¹⁶ Other researchers have attempted to stratify autistic individuals using brain connectivity patterns from EEG, arguing that behavioral subtyping alone is blind to the underlying neurobiology and has limited ability to predict developmental outcomes.¹⁷

Animal model research reinforces this picture. A study that performed brain-wide connectivity mapping across 16 different mouse models of autism-associated genetic mutations found a broad spectrum of connectivity abnormalities, with diverse and sometimes opposite patterns across different mutations.¹⁸ If the genetic causes of autism push the brain in many different directions, searching for one universal brain signature is inherently problematic. A recent study tried a different approach, grouping autistic individuals by how their brain imaging phenotypes correlated with gene expression profiles, and found three subgroups with different clinical severities and different underlying gene sets.¹⁷

Comorbidities add another layer. Autism frequently co-occurs with ADHD, and people with both conditions show a brain profile that is not simply a combination of the two individual patterns. One study found that the comorbid group had unique and more severe brain-behavior associations in prefrontal, striatal, and anterior cingulate regions, suggesting the overlap creates its own neurobiology rather than being additive.¹⁹ A large brain-charting study confirmed this, finding that the co-occurring autism-ADHD group showed a distinctive pattern of cortical thickness increases and surface area decreases that differed from either condition alone.²⁰

Sex Differences in Autism Brain Patterns

Most autism neuroimaging research has been conducted primarily with male participants, which creates a blind spot. When researchers have specifically compared brain connectivity in autistic males and autistic females, they have found strikingly different patterns. Males with autism tend to show reduced connectivity compared to non-autistic males, while females with autism often show increased connectivity compared to non-autistic females. Intriguingly, the pattern in autistic females resembled a shift toward the connectivity levels typically seen in non-autistic males, while autistic males shifted toward the pattern typically seen in non-autistic females.²¹

A systematic review of structural, functional, and white matter imaging studies confirmed that females with autism appear to follow distinct neurodevelopmental trajectories, and that collapsing data across age ranges in studies may mask these differences.²² The review also noted accumulating evidence for a “female protective effect,” the idea that the female brain may require a higher genetic burden before autism manifests, though very few studies had directly examined the brain circuits involved in this proposed protection.

For any future diagnostic tool based on brain imaging, these sex differences are a practical problem. An algorithm trained primarily on male brains may systematically miss autistic females, perpetuating a diagnostic gap that already exists in behavioral assessment.

Neurochemical Signals

Magnetic resonance spectroscopy (MRS) is a less well-known scanning technique that can measure concentrations of specific brain chemicals without a blood draw or biopsy. In autism research, it has been used to investigate the balance between excitatory and inhibitory neurotransmitters, particularly glutamate and GABA. The theory that an imbalance between excitation and inhibition in the brain contributes to autism has been around for over two decades, and MRS provides a way to test it in living people.

One study measuring GABA levels in the anterior cingulate cortex of autistic boys found a significant negative correlation: the more severe the autism symptoms, the lower the GABA levels, consistent with the excitation-inhibition imbalance theory.²³ A review of MRS studies in autism concluded that differences in glutamate and GABA balance may serve as both a biomarker and a potential treatment target, though the field acknowledged that findings had been inconsistent across studies.²⁴

Neurochemical imaging illustrates a broader theme in autism brain research: individual findings are suggestive and sometimes exciting, but they have not converged into a single, robust biomarker. The brain chemistry differences are real but variable, region-specific, and measured in small samples.

What This Means If You or Your Child Are Seeking a Diagnosis

If a clinician suggests getting a brain scan to diagnose autism, that is not part of standard clinical practice. The current gold standard for autism diagnosis remains behavioral evaluation, typically using structured tools like the ADOS-2 (Autism Diagnostic Observation Schedule) and developmental history interviews. A comprehensive evaluation considers communication patterns, social interaction, sensory responses, and restricted or repetitive behaviors. Brain scans play no role in this process in mainstream medicine.

That said, a doctor might reasonably order a brain MRI for a child being evaluated for autism to rule out other conditions. Structural abnormalities, tumors, metabolic disorders, or evidence of prior brain injury can all cause behavioral symptoms that overlap with autism. In those cases, the scan is not looking for autism but looking for something else that might explain the symptoms.

Some private clinics market brain scans, including SPECT scans and quantitative EEG, as diagnostic tools for autism and other neurodevelopmental conditions. Major professional organizations have not endorsed these as diagnostic methods. The gap between what these scans can reveal in a research context and what they can tell you about a specific individual seeking a diagnosis remains wide. Paying for a scan that a clinician reads as “consistent with autism” gives you something that looks like precision but lacks the validation needed to mean what it appears to mean.

Nonhuman Primate and Animal Models

Some of the clearest evidence that brain differences in autism are biologically real, rather than artifacts of how humans behave in scanners, comes from animal studies. Researchers can create animal models that mimic specific genetic mutations or environmental exposures associated with autism and then scan their brains under controlled conditions, eliminating the behavioral and movement-related confounds that plague human imaging.

A study of 16 different autism-associated mouse mutations mapped brain-wide connectivity and found that the mutations produced a genuine spectrum of connectivity abnormalities, with some mutations causing increased connectivity in certain areas while others caused decreases.¹⁸ In a primate model, researchers triggered an immune response during pregnancy (maternal immune activation) and found that the offspring showed elevated free water in the brain and brain volume changes detectable on MRI, providing evidence that prenatal immune disruption can produce measurable brain changes visible on standard imaging.²⁵

These animal studies reinforce a key point: there are genuine, biologically driven brain differences associated with autism-relevant genetic and environmental factors. The challenge in human diagnosis is not that the differences are illusory but that they are many, variable, and entangled with measurement noise. Whether continued advances in scanning technology, larger datasets, and smarter algorithms will eventually close that gap is an open question. The 2024 meta-analysis called MRI-based autism diagnosis “potent and desired” for a reason: the research community believes the signal is real.²⁶ Extracting it reliably from any single person’s scan, on any scanner, at any age, is the part that has not been solved.