For generations, medicine has hoped that the brain's architecture might reveal its disorders as plainly as a broken bone reveals itself on an X-ray. A large new analysis, published in Nature Neuroscience and led by researchers at Monash University, quietly dismantles that hope for autism, depression, and bipolar disorder — finding that structural MRI scans produce no consistent, reproducible signatures across independent studies for these conditions. Where Alzheimer's disease speaks clearly in brain images, psychiatric illness, it seems, speaks in many voices at once, and no single picture cap
Brain imaging fails to find consistent signatures for autism, depression, bipolar disorder
The signal will not emerge from the noise, no matter how much data we collect.
Mark
So the study looked at thousands of scans and found that brain changes don't show up consistently across different research groups. Why does that matter so much?
Mimi
Because for forty years, the implicit promise has been that if we just scan enough brains, we'll find the biological fingerprint of each disorder. This study says that fingerprint might not exist—at least not in the way we've been looking for it. It's the difference between a problem with our tools and a problem with our assumptions.
Mark
But Alzheimer's disease did show consistent signatures. So it's not that MRI doesn't work at all.
Mimi
Right. Alzheimer's shows robust, reproducible changes across studies. The difference is that Alzheimer's is a disease of accumulation—plaques and tangles that damage the brain in fairly predictable ways. Autism or depression might not work that way. They might be fundamentally heterogeneous. Two people with depression could have completely different brain architectures and still meet the same diagnostic criteria.
Mark
The study found that schizophrenia improved with larger sample sizes, but autism didn't. Why the difference?
Mimi
Schizophrenia has narrower diagnostic criteria—you need a consistent pattern of symptoms for at least six months. That specificity might mean the people who get that diagnosis share more in common biologically. Autism and depression are diagnosed across much wider ranges of presentation. You could have two autistic people with almost nothing in common neurologically.
Mark
So what do researchers do now? Stop doing MRI studies?
Mimi
No. But stop expecting MRI alone to solve the problem. The researchers suggest combining brain imaging with genetics, molecular data, and clinical tracking over time. And some are proposing we stop using categorical diagnoses altogether—instead grouping people by trait profiles that might cut across traditional disorder boundaries.
Mark
Is that realistic? Can psychiatry really move away from the diagnostic categories it's built on?
Mimi
It's already starting. Large consortiums like ENIGMA are pooling data across countries and integrating multiple data types. It's slower and messier than the old model, but it might actually work.
The Pulse
- Decades of structural MRI research in psychiatry may have been chasing a signal that does not exist in the form scientists assumed — a finding that challenges the foundation of neuroimaging-based biomarker research.
- Unlike Alzheimer's disease, which yields consistent brain signatures across studies, autism, depression, and bipolar disorder showed persistent, irreconcilable inconsistencies even after controlling for age, sex, and scanner variation.
- Simulating ever-larger sample sizes offered no rescue for autism or mood disorders — reproducibility simply did not improve with scale, suggesting the problem is biological heterogeneity, not insufficient data.
- Schizophrenia offered a partial exception, with reproducibility rising alongside sample size, possibly because its narrower diagnostic criteria capture a more biologically coherent group of patients.
- The field is now being pushed toward integration — combining MRI with genetics, molecular biology, and longitudinal clinical data through large consortia like ENIGMA — and some researchers are calling for abandoning categorical diagnoses altogether.
- The reckoning is not just methodological but conceptual: if two people share a diagnosis yet have entirely different symptom profiles and brain patterns, the category itself may be obscuring the biology.
For generations, medicine has hoped that the brain's architecture might reveal its disorders as plainly as a broken bone reveals itself on an X-ray. A large new analysis, published in Nature Neuroscience and led by researchers at Monash University, quietly dismantles that hope for autism, depression, and bipolar disorder — finding that structural MRI scans produce no consistent, reproducible signatures across independent studies for these conditions. Where Alzheimer's disease speaks clearly in brain images, psychiatric illness, it seems, speaks in many voices at once, and no single picture captures them all.
For decades, neuroscience has pursued a seductive promise: scan the brain and find the disorder written in its folds. A new analysis, led by Alex Fornito at Monash University and published in Nature Neuroscience, suggests that promise has been built on unstable ground.
Fornito's team examined thousands of structural MRI scans from people diagnosed with autism, depression, bipolar disorder, schizophrenia, and schizoaffective disorder, comparing them against scans from people with Alzheimer's disease. The central question was simple: when independent teams scan different patients with the same diagnosis, do they see the same brain changes? For Alzheimer's disease, the answer was yes — consistent, reproducible signatures emerged across studies. For the psychiatric conditions, the answer was mostly no. Cortical thickness varied. Grey matter patterns diverged. And crucially, this was not a matter of poor methodology or incompatible equipment.
The team then tested whether the problem would dissolve with larger samples, simulating what would happen as cohort sizes grew. For schizophrenia, reproducibility did improve — eventually approaching Alzheimer's-level consistency when simulated groups exceeded 200 participants. But for autism, depression, bipolar disorder, and schizoaffective disorder, more data changed nothing. The inconsistency held.
The likely explanation lies in how these conditions are defined. Schizophrenia requires a specific, sustained pattern of symptoms — a relatively narrow diagnostic gate. Autism, depression, and bipolar disorder encompass far broader ranges of clinical presentation, meaning two people sharing a diagnosis may have different symptoms, different genetics, and different brain profiles entirely. If the underlying biology is that heterogeneous, no single structural signature will ever surface, regardless of how many scans are collected.
Experts outside the study point toward a different path forward. Rather than relying on structural MRI alone, researchers argue for integrating brain imaging with genetic data, molecular markers, and longitudinal clinical tracking. Large collaborative efforts like the ENIGMA consortium — pooling imaging data from 43 countries alongside genetic and epigenetic information — are already moving in this direction. Some scientists go further, proposing that traditional diagnostic categories be set aside in favor of trait-based classification systems that may better reflect the brain's actual biology.
Fornito's team plans to extend the work with genuinely larger datasets and whole-brain analyses rather than regional measures. The message to the field is unambiguous: the brain signatures psychiatry has long sought may not exist in the form researchers have been searching for them — and the search itself must change.
For decades, neuroscientists have chased a simple promise: scan the brain, find the disorder. Look at the images and you will see autism, or depression, or bipolar illness written in the architecture of grey matter and cortical folds. A new analysis suggests that promise has been built on sand.
Researchers led by Alex Fornito at Monash University examined thousands of structural MRI scans from people diagnosed with autism, depression, bipolar disorder, schizophrenia, and schizoaffective disorder, alongside scans from people with Alzheimer's disease. They applied the same analysis method to all the data, then asked a straightforward question: when independent research teams scan different groups of people with the same diagnosis, do they see the same brain changes? The answer, published in Nature Neuroscience, was mostly no.
Unlike Alzheimer's disease—which produced consistent, reproducible signatures across studies—the psychiatric conditions showed little agreement. Cortical thickness varied. Grey matter volume patterns diverged. The inconsistency persisted even after researchers accounted for age, sex, scanner differences, and other technical factors that might explain the noise. This was not a problem of sloppy methodology or incompatible equipment. Something deeper was wrong.
The finding lands as a reckoning. Structural MRI studies have produced conflicting results for years. Some research showed people with autism had thicker cortex in certain regions; other studies found the opposite. Until now, the field could assume these contradictions would eventually resolve—that with enough studies, enough data, enough computational power, the signal would emerge from the noise. Fornito's team tested that assumption mathematically, simulating what would happen if researchers simply kept collecting larger and larger samples. For schizophrenia, reproducibility did improve with scale, eventually matching Alzheimer's disease when simulated cohorts exceeded 200 participants. But for autism, depression, bipolar disorder, and schizoaffective disorder, larger sample sizes made no difference. The inconsistency persisted.
The reason may lie in how these conditions are defined. Schizophrenia requires a consistent pattern of symptoms lasting at least six months—a relatively narrow diagnostic gate. Autism, depression, and bipolar disorder encompass much broader ranges of clinical presentation. Two people with the same diagnosis might have entirely different symptom profiles, different genetic backgrounds, different life histories. If the brain changes associated with these conditions are heterogeneous—if they vary widely from person to person—then no single structural signature will ever emerge, no matter how many scans you collect.
Experts outside the study acknowledge the implications. Joshua Roffman at Harvard Medical School notes that cortical thickness measurements carry inherent measurement error, even under ideal conditions. Maria Di Biase at the University of Melbourne suggests that structural MRI alone cannot bear the weight of diagnosis. The future, she argues, lies in integration: combining brain imaging with genetic data, molecular markers, and longitudinal clinical tracking to build a more complete picture.
Large collaborative efforts are already moving in that direction. The ENIGMA consortium pools brain imaging data from 43 countries and layers in genetic and epigenetic information, searching for biological signatures too subtle to detect in smaller studies. Some researchers propose abandoning traditional diagnostic categories altogether, instead classifying people by trait profiles using frameworks like the Hierarchical Taxonomy of Psychopathology. These approaches might reveal more consistent brain patterns than the current system of categorical diagnoses.
Fornito's team plans to continue the work, testing whether reproducibility improves with genuinely larger datasets rather than simulations, and investigating whole-brain anatomical patterns instead of focusing on regional measures. The message to the field is clear: the old business-as-usual approach will not work. The brain signatures psychiatry has been searching for may not exist in the form researchers have been looking for them. The search itself may need to change.
Notable Quotes
If we keep doing business as usual, that's not going to happen—the noise will not wash out and we will not converge on a consensus of brain changes.— Alex Fornito, Monash University
Across this study and many others, we are not seeing evidence of discrete categories of mental disorders.— Louise Mewton, University of Sydney
Want the full story?
Read the original at The Transmitter ↗
Science & Space