A new study published in Translational Psychiatry has confirmed something that could fundamentally reshape how doctors diagnose, treat, and even talk about Alzheimer’s disease: the brains of people with Alzheimer’s and mild cognitive impairment don’t deteriorate in the same way.
Not even close.
Using a powerful analytical method called normative modeling, researchers mapped distinct structural subtypes in the brains of Alzheimer’s and MCI patients — comparing each person’s brain not to a group average, but to a personalized healthy baseline built from thousands of healthy people.
The headline finding is striking.
Rather than one uniform pattern of brain degeneration, the study identified multiple distinct subtypes, each with its own signature of which regions shrink, how fast, and in what sequence.
This matters enormously because right now, most clinical approaches treat Alzheimer’s as if it plays out the same way in every person.
It doesn’t.
And the implications of that simple fact stretch from the neuroscience lab all the way to the bedside, the clinical trial, and the doctor’s office.
What “Normative Modeling” Actually Means — In Plain English
Think about it this way.
If you want to know whether someone is unusually tall, you don’t compare them to the average height of people already diagnosed with a bone condition.
You compare them to the full, healthy range of human heights, adjusted for age, sex, and background.
Normative modeling does exactly that for the brain.
Researchers build a detailed statistical picture of what a healthy brain looks like across thousands of individuals, carefully accounting for age, sex, and other biological variables.
Then, when they scan the brain of someone with Alzheimer’s or MCI, they can pinpoint precisely which regions deviate from that healthy norm, by how much, and in which direction.
Each patient ends up with a kind of individualized brain deviation map — a record of exactly where their brain diverges from what would be expected for a healthy person of the same age and background.
As demonstrated in a key study from the Alzheimer’s Disease Neuroimaging Initiative (ADNI), this approach revealed that in people with Alzheimer’s, the highest proportion of outliers in any single brain region was less than 50%.
That means even in the brain areas most commonly associated with Alzheimer’s damage, like the hippocampus, fewer than half of patients were actually showing abnormal readings.
That is not what the “average patient” picture looks like.
It’s far messier, far more individual, and ultimately far more honest.
The Assumption That Has Been Holding Research Back
Here’s what most people get wrong about decades of Alzheimer’s neuroscience.
For a long time, researchers have studied the disease using what’s called a case-control approach: scan a group of Alzheimer’s patients, scan a group of healthy people, compare the averages, and report what’s statistically different between them.
It sounds reasonable.
But there’s a serious flaw buried in that logic.
When you average out the brains of hundreds of patients, the individual differences cancel each other out.
One patient might show severe shrinkage in memory-related regions like the hippocampus and entorhinal cortex.
Another might show loss primarily in regions that govern language or visual processing.
A third might have a completely different pattern concentrated in frontal areas responsible for planning and executive function.
Average them together, and what you get is a blurry composite that doesn’t accurately represent any real person walking through a clinic door.
It’s a bit like averaging the face of a thousand different people and then trying to use the result to identify someone in a crowd.
The Translational Psychiatry study was designed specifically to move beyond this limitation.
By using individual deviation analysis instead of group comparisons, the researchers could see the disease as it actually presents in each unique brain, not as a statistical fiction created by averaging.
The results painted a much more complex and clinically meaningful picture.
What the Subtypes Actually Look Like
So what are these distinct subtypes, and why should anyone outside a brain imaging lab care about them?
While exact subtype definitions vary across research groups and datasets, several consistent patterns have emerged from the normative modeling literature.
Some patients show predominantly medial temporal atrophy, meaning the hippocampus and surrounding memory regions shrink first and fastest.
These are the patients that most closely match the classic textbook description of Alzheimer’s — the person who forgets names, misplaces keys, and loses track of recent events.
Other patients show a more diffuse cortical pattern, where thinning spreads simultaneously across multiple brain regions rather than following a clear sequence.
Their symptom profile tends to be broader and harder to categorize neatly.
A third group shows primarily posterior cortical atrophy, with prominent changes in areas governing visual processing and spatial awareness.
These patients may first struggle with reading, judging distances, or recognizing faces — symptoms that can lead to years of misdiagnosis because they look less like “typical” Alzheimer’s.
And then there is a subtype that shows minimal structural deviation even with confirmed Alzheimer’s pathology in the brain.
Their symptoms appear to be driven more by disruptions in how brain networks communicate than by visible tissue loss on a standard MRI.
This last group is particularly important.
As highlighted in a 2025 study in Biological Psychiatry, both Alzheimer’s and MCI patients cluster into subtypes based on functional network deviation patterns, and these subtypes carry meaningfully different implications for cognitive trajectory and disease progression.
The MCI Window: Where This Research Matters Most Urgently
Mild cognitive impairment is where these findings carry their greatest urgency.
MCI sits at a crossroads.
It’s the phase between normal aging and Alzheimer’s dementia — and critically, not everyone with MCI goes on to develop Alzheimer’s.
Some people improve.
Some remain stable for years.
Others deteriorate rapidly.
Right now, doctors have limited tools to reliably tell these groups apart at the individual level.
Normative modeling offers a promising path forward.
By identifying which structural subtype a person with MCI belongs to, researchers may be able to predict who is at highest risk of conversion to dementia early enough to actually intervene.
One particularly useful metric to emerge from this work is the total outlier count (tOC): a simple score representing how many brain regions deviate significantly from a healthy norm for that individual.
Higher tOC scores have been consistently linked to worse cognitive test performance, abnormal protein levels in cerebrospinal fluid, and a higher likelihood of progressing from MCI to full Alzheimer’s dementia within three years.
This is exactly the kind of concrete, patient-specific number that clinicians could realistically use.
Think of it like a blood pressure reading, but for brain health.
A 2022 real-world study from a clinical cohort applied normative modeling to 86 patients who had confirmed amyloid pathology detected by PET scan.
The total outlier count in those patients ranged from just 1 to 120 out of 148 assessed brain regions, with a median of 21.5.
That extraordinary range in a group all diagnosed with the same disease says everything about how individual the disease truly is.
Why the Brain Gets More Different, Not More Similar, Over Time
One of the more counterintuitive findings from this body of research is that heterogeneity actually increases as Alzheimer’s progresses.
You might reasonably expect that as the disease advances and more brain tissue is affected, patients would converge toward a common pattern of widespread deterioration.
The data say the opposite.
A 2025 study in Alzheimer’s & Dementia by Kumar and colleagues applied a deep learning-based multimodal normative framework across three types of brain imaging — structural MRI, amyloid PET, and tau PET — in over 1,000 participants.
The results showed that patients at more advanced stages of Alzheimer’s dementia were significantly more dissimilar from each other than patients at the MCI stage.
In other words, the disease doesn’t funnel everyone toward the same endpoint.
Instead, individual factors including genetics (particularly variants of the APOE gene), vascular health, co-existing conditions, and potentially even life history, create increasingly divergent brain trajectories as the disease matures.
Interestingly, the research also found that tau protein deposits showed more heterogeneity than amyloid across patients.
Amyloid, which tends to build up early and broadly, produced fewer distinct subtypes in imaging analysis.
Tau, which spreads in patterns that appear to be more individual, produced more.
This might partially explain why amyloid-targeting drugs have had such a difficult road in clinical trials, even when they successfully cleared amyloid from the brain.
Clearing amyloid may not be enough on its own, because the damage being done by tau and neurodegeneration is playing out differently in each person.
The Personalized Disease Severity Index: A New Way to Score Brain Health
One of the most exciting practical developments to emerge from normative modeling research is the concept of a personalized disease severity index (DSI).
Rather than relying on a single scan type or comparing a patient to a group average, the DSI combines information across multiple imaging modalities into a single, patient-specific score of brain health deviation.
The 2025 Alzheimer’s & Dementia study that developed the DSI metric showed it was able to do several things that simpler measures can’t.
It tracked progression over time.
It correlated meaningfully with cognitive test performance.
It flagged patients at higher risk of deterioration before symptoms necessarily worsened on clinical measures.
And crucially, it was reproducible across two independent patient cohorts — which is far from guaranteed in neuroscience research, where findings often fail to hold up when tested on new populations.
This kind of multimodal score moves the needle from population-level statistics to individual-level prediction, which is where clinical medicine needs to be if it’s going to catch Alzheimer’s early enough to make a difference.
Brain Electrical Patterns Tell a Similar Story
The heterogeneity findings aren’t limited to structural brain scans.
Research using EEG (electroencephalography), which measures the brain’s electrical activity, has reached strikingly similar conclusions.
A 2025 study published in npj Parkinson’s Disease applied normative modeling to EEG data from Alzheimer’s and Parkinson’s patients across 14 clinical sites.
Around 30% of Alzheimer’s patients showed abnormal spectral deviations in brain wave patterns.
But when researchers looked at functional connectivity, meaning how different brain regions communicate electrically with each other, roughly 80% of patients showed deviations from the healthy norm.
The spatial overlap of those abnormal features across patients was less than 25%.
Said plainly: each brain was misfiring in its own way.
The pattern didn’t generalize across patients any more than the structural atrophy patterns did.
This convergence across two completely different types of brain measurement — structure and electrical function — makes a strong case that the heterogeneity seen in Alzheimer’s is real, deep, and consistent.
Why Clinical Trials Have Been Failing, and What This Could Change
The stakes of this research extend well beyond academic neuroscience.
Alzheimer’s clinical trials have a notoriously poor track record.
Dozens of promising drug candidates have been tested in large, expensive Phase III trials, and the vast majority have failed to show significant benefit in the treated group.
One leading explanation that has gained serious traction in recent years is patient heterogeneity.
When a trial enrolls participants who have different subtypes of the disease, the average treatment effect across the full group can look flat even when the drug is genuinely helping one subtype and doing nothing (or worse) for another.
The signal gets buried in the noise of biological diversity.
As reviewed in Biological Psychiatry, the field is increasingly recognizing that robust and reproducible subtyping frameworks will be essential for translating heterogeneity research into actual clinical practice, including better-designed trials.
If researchers can stratify trial participants by structural and functional subtype before the trial begins, they stand a much better chance of detecting true treatment effects.
The same logic applies to personalized care planning.
A patient whose brain shows primarily temporal lobe atrophy may respond differently to memory support strategies compared to a patient whose atrophy pattern is frontal or visual.
One-size-fits-all dementia care is becoming increasingly difficult to justify as the science makes clear just how different these brains actually are from one another.
The Challenges That Still Need Solving
To be fair, this research is not without its complications.
One persistent issue is reproducibility.
Different research groups using different datasets, different clustering algorithms, and different imaging protocols can arrive at subtypes that don’t map cleanly onto each other.
The number of identified subtypes varies across studies, ranging from two to four depending on the imaging modality and the population studied.
Another challenge is that most of the normative modeling studies published so far have used cross-sectional data, meaning a single snapshot in time rather than tracking the same patients over years.
That makes it harder to distinguish between subtypes that represent different stages of the same progression versus genuinely different disease variants.
Researchers at University College London have been working on deep normative models that incorporate multiple imaging types at early disease stages to address exactly this gap, but it’s a significant methodological challenge.
And finally, vascular health, meaning the health of blood vessels in and around the brain, is an important contributing factor to cognitive decline that has not yet been fully integrated into most normative modeling frameworks.
Future work will need to account for white matter damage, blood flow changes, and small vessel disease as part of a complete heterogeneity picture.
The Bigger Shift: From Average to Individual
There is a quiet revolution happening across neuroscience and psychiatry.
It’s a shift away from the assumption that any disease — whether Alzheimer’s, depression, schizophrenia, or ADHD — can be fully understood by studying average differences between groups.
The brains of people within any diagnostic category are profoundly individual.
For diseases as complex and devastating as Alzheimer’s, that individuality may be one of the most important facts we can possibly understand.
As laid out in an influential review in Brain: A Journal of Neurology, moving beyond the average patient isn’t merely a philosophical preference.
It is a scientific and clinical necessity.
The patterns hidden within that heterogeneity — the subtypes, the trajectories, the outlier brain regions, the individual deviation maps — may hold the clearest answers yet about who gets Alzheimer’s, when their symptoms will worsen, which treatments might help them, and why.
The next time you read a headline about Alzheimer’s research, it’s worth pausing to ask a simple question: whose brain are we actually talking about?
Because if the science is telling us anything right now, it’s that the answer is not as simple as any of us once assumed — and that complexity, uncomfortable as it is, may turn out to be one of the most valuable things we’ve ever discovered about this disease.
