This article comes from my weekly Monday livestream, Neurofeedback & Chill. I clean up the talk for the page, anonymize the audience questions, and keep the science.
Is Autism One Brain Type or Many?
If you have spent years inside the autism world as a parent, an autistic adult, or someone who works with autistic people, you already know the category is doing a lot of work. The same label covers a child with profound sensory flooding and no verbal prosody, and a high-functioning adult with intact social skills, an obsessive special interest, and a touch of OCD. Those two brains are running different architecture. The diagnostic label tells you very little about what is happening underneath.
For decades, autism research lumped everyone with an ASD diagnosis into one group. When you cluster different phenomena and measure them together, you get a weighted average that describes nobody. That is part of why the autism literature has looked contradictory for so long. Studies disagreed because they were measuring different brains and calling them the same thing.
Over the last two years, several teams flipped the logic. Instead of treating autism as one population, they asked: if you cluster within the autism group, do distinct subtypes fall out? The answer is yes. Multiple methods, multiple research teams, converging on the same conclusion. Autism is an umbrella over a handful of recurring brain architectures.
What Do the MRI and Imaging Subtypes Show?
The imaging work is where this started to get solid.
MRI work pulling joint patterns of gray matter and resting-state connectivity has found that classic autism carries a different brain signature than high-cognitive-function presentations: more subcortical connections, more disruption in the default mode network. The old categories like Asperger's and PDD-NOS mapped onto something real, but the boundaries were crude and the fit to the data was poor.
Functional connectivity work has reported distinct subtypes. One showed widespread hyper-centrality, heavy connectivity and activation down the midline. One showed localized, modular abnormalities. One looked closer to a typical brain but with specific network deviations rather than regional ones. Each lined up with a different symptom profile.
Other work clustered on language, cognition, and motor skills and found robust early-childhood subtypes: one with higher language and cognitive skills, one with more global delays. These did not just score differently. They showed different fMRI patterns and different relationships between brain connectivity and gene expression. Different biology under different behavior.
Add transcriptomics and the picture sharpens further. Studies combining structural MRI, functional MRI, behavioral measures, and gene transcription have found subtypes with distinct neuroanatomical and behavioral signatures. These look like separate developmental stopping places, like distinct nodules on a string rather than points on a severity line getting progressively worse. The precision-psychiatry argument is the same: biologically anchored subtypes outperform the lumped category.
Autism consists of several different developmental configurations that can show up on the brain's developmental tree, each a distinct arrangement of resources.
Do These Subtypes Show Up on EEG?
They do. Resting-state EEG meta-analyses show distinct power differences in autistic groups. One pattern carries low alpha power and elevated fast frequencies, beta running up into the gamma range. Low alpha power tracks with detail-focused perception that does not assemble into a whole. The flooded fast frequencies track with difficulty in social reciprocity. A separate pattern shows globally elevated theta with hypercoherent fronto-central connectivity, producing front-midline theta features: obsessiveness, tics, repetitive behavior.
For the technically inclined: EEG microstates also differ in autism. Microstates are brief whole-brain timing patterns that roll through the brain continuously. Microstate C runs in fast temporal loops, roughly 40 to 100 milliseconds, the loops that support processing and thinking.
A short detour on why that timing matters. You do not perceive something the moment light hits your retina. The retina transduces light into neural signals, the information splits between the left and right visual fields into opposite hemispheres, then routes through the thalamus. It takes about two synapses to reach the thalamus, where awareness starts to creep in around 60 to 90 milliseconds. Decision-making adds another 40 to 50. Then action. A full perceive-and-act loop sits around 180 milliseconds at the fast end. Below that, you are reacting rather than acting, favoring speed over accuracy. When microstate C timing is disrupted, this deep processing-speed loop is affected, and you see it in resting delta and alpha as well.
What Are the Three QEEG Phenotypes of Autism?
Taking that imaging and EEG research and lining it up against what I see on QEEG brain maps, three recurring phenotypes emerge. These are plausible EEG phenotypes, the same way EEG phenotypes generally describe patterns that show up across populations. QEEG compares you to population data and looks for patterns that usually mean something. It produces a description of resources and possibilities, not a diagnostic verdict.
The Default-Mode-Network-Light Phenotype
The DMN, the self-referential network, runs underconnected. Posterior sensory and social regions are not checking in cleanly. Alpha power is low, connectivity is weak across long ranges, and individual modules run hot in isolation. You see a lot of cingulate activation, perseveration, rumination, and sensory difficulty.
This person has trouble shifting between inner and outer focus and gets absorbed internally. When language is intact, they can process plenty of detail but struggle to assemble the social gestalt, the implicit meaning. They can feel like they are watching the world from far away. When language is not online, this looks like stimming, scripting, and difficulty breaking repetitive sensory habits.
The Front-Midline-Theta-Heavy Phenotype
Front-midline and global theta are very high. There is impulsivity and sensory flooding. This phenotype overlaps heavily with the ADHD presentation, and in kids it often draws both diagnoses or an argument about which one it is. The brain here is overwhelmed because the filter runs slow, not because the incoming signal is too loud.
In a cognitively intact or gifted version, you see enormous procrastination. The anterior cingulate gets stuck in fast alpha and struggles to climb into a beta mode. Functionally, the cingulate has trouble reconciling the left-side appraisal of approach with the right-side appraisal that things are worth avoiding, so behavior stalls. Scripting and repetitive behavior show up here too.
The Sensory-Gamma / High-Beta Phenotype
Beta and gamma run very high, with elevated local connectivity especially in posterior sensory and social regions, and very little coupling across networks. Alpha is chaotic, with poor coherence and high phase lag, usually dragged low in amplitude. Delta tends to run high amplitude.
This is the intense sensory experience: hearing fluorescent lights, feeling the tags in a shirt constantly, recognizing certain inputs with incredible precision but failing to integrate across channels. The resolution is cranked up and the bandwidth is narrow and flooded. The same low-connectivity, high-local-module configuration can produce savant abilities alongside real difficulty using those gifts outside a narrow context.
Why Does Subtyping Change Anything for Treatment?
Two reasons. First, research. If studies keep mixing physiology, behavior, and complaints into one group, we will never understand what autism is or what supports it. Separating the architectures lets us run precision research.
Second, precision care. Once you know which phenotype is in front of you, the neurofeedback protocol follows. For the DMN-light phenotype, you train network integration, alpha-theta work, and connectivity training at the midline, and you address cross-hemisphere sensory imbalance at the temporal sites (T3/T4, T5/T6). For the frontal-theta-heavy phenotype, you run a midline protocol with a theta inhibit instead of a beta inhibit, build response inhibition, and bring up some SMR along the way. For the sensory-gamma phenotype with flooded beta, slowed alpha, and sped-up delta, you work bottom-up sensory integration and cross-network coordination.
Subclinical seizure phenomena are common in this landscape. Across autism and related neurodevelopmental conditions, a substantial fraction of individuals show atypical subclinical epileptiform activity. SMR training is the original neurofeedback finding for seizure suppression (Sterman & Friar, 1972). That sensorimotor tone is probably part of why neurofeedback helps so broadly here, the same tone that helps you stay still, stay asleep when a car passes, and suppress seizure discharge.
Which Autism Features Respond Best to Neurofeedback?
For strength of impact, in terms of what the research and the maps point to: anxiety, sensory integration, and executive function. Sleep regulation and seizure-related instability are also highly tractable. Think of neurofeedback resting on three legs: sleep, stress, and attention. By stress I mean the whole cluster of overarousal and underarousal, the perseveration, rumination, social and sensory load. Those three are foundational, they overlap, and they are routinely dysregulated in autism. Get them under control and the results can be powerful.
Language is the hardest resource to move. Language has a critical period and does not retain the plasticity that attention, sensory, and social processing keep over time. If someone lacks receptive language, the right-parietal Wernicke's-area machinery that extracts meaning, productive language rarely develops from nothing. With decent receptive language and a little productive output from Broca's area on the left, more language usually develops with training. Those are the slower cases.
For the more typical presentations, the AuDHD and Asperger's-type profiles, you tend to see results on the sub-resources: perseveration, inflexibility, inattentiveness, sleep, sensory integration. You work those individually without centering the label. A QEEG brain map gives you the target, and where there is a target, the resource is generally tractable.
What Predicts a Good Neurofeedback Response?
A clear target on the map and a functional level high enough to support training. Above a certain functional level, you can get strong results on sensory and social processing, anxiety circuits, and executive function. Below it, some resources move much more slowly, language most of all.
People are also weirder than their maps. I have seen one person with profound autism, hot posterior beta, and no eye contact, and another with very similar hot beta over the right TPJ who seeks sensory input constantly, reaches for texture, and maintains eye contact readily. The map narrows the search. It does not finish it.
One concrete example of how local this gets: the fusiform face area. Atypical function there produces prosopagnosia, face blindness, where you recognize someone by their glasses, their voice, or their shirt rather than their face. That same region is involved in autism and in eye contact. Train it and people who do not give eye contact naturally tend to give more. If eye contact is overwhelming, training that area to relax can make it less so.
What This Reframes for Families and Autistic Adults
A QEEG brain map is a description of patterns and possibilities. That is where your agency lives. If your front-midline is generating a lot of theta and you find yourself with intrusive thoughts, nail-biting, or a tic, front-midline theta is the likely driver. Training it down is one thing to explore. If the map shows front-midline beta with intrusive thoughts, that runs more OCD-flavored, and N-acetylcysteine, methylation work, neurofeedback, and SSRIs are all things the research and clinicians point to. None of this is medical advice. It is a way of understanding your own architecture.
Across all three phenotypes, the individual pinch points matter more than the label. Two people in the same archetype can carry very different patterns of difficulty. The useful question is which resources are cramped and how that shows up day to day.
The research is converging from imaging, EEG, genetics, and transcriptomics on one conclusion: autism is many related brain types. For families and autistic adults, that means the path forward is to find the specific architecture, name its strengths and its cramps, and work the resources you can actually move. A brain map gives you the target, and the protocol follows from there.