Diagnosing schizophrenia has long depended on conversations, clinical observations, and careful assessments of a person’s symptoms. But what if artificial intelligence could help psychiatrists identify measurable changes in the brain that are difficult to recognise through conventional examinations?
Researchers in Taiwan are exploring that possibility with BrainProbe, an AI-powered brain imaging platform developed at Taipei Veterans General Hospital (TVGH).
In July 2025, the hospital announced that its system reached a reported 91.7% accuracy in helping assess schizophrenia. It uses magnetic resonance imaging (MRI) and deep learning to analyze patterns of neurological abnormalities.
The platform also offers three-dimensional brain visualizations and estimates of biological brain aging. These features may help doctors understand and explain neurological changes linked to psychiatric disorders.
This achievement is promising and raises an important question: Can AI-based brain imaging improve how we diagnose schizophrenia without taking away the clinical judgment that remains essential in psychiatric care?
As of October 2026, researchers are moving beyond just predicting diagnoses and are exploring personalized brain stimulation therapies. Still, there are big questions about external validation, clinical usefulness, and availability that have not been answered.
What Is Schizophrenia, and Why Is Diagnosis Difficult?
Schizophrenia is a complicated mental health condition that can affect how people perceive reality, organise their thoughts, express emotions, and interact with others.
Its symptoms vary substantially between individuals and may include:
- Hallucinations: Hearing, seeing, or experiencing things that others do not perceive.
- Delusions: Strongly held beliefs that remain despite convincing evidence to the contrary.
- Disorganised thinking: Difficulty organising thoughts or communicating coherently.
- Negative symptoms: Reduced motivation, emotional expression, or interest in social activities.
- Cognitive difficulties: Problems with attention, memory, and making decisions.
According to the World Health Organisation’s schizophrenia fact sheet, updated in September 2026, approximately 27 million people worldwide, or about 1 in 300 individuals, live with schizophrenia.
Although schizophrenia can cause significant disability, effective treatment and support are available, and recovery results vary considerably between individuals.
Why Isn’t There a Simple Medical Test?
Unlike conditions that can be identified through a specific laboratory marker, schizophrenia has no single definitive blood test or brain scan.
Psychiatrists generally rely on a complete analysis that considers symptoms, medical and family history, functioning, and other potential explanations for psychosis.
According to the UK National Health Service, diagnosis usually requires assessment by a mental health specialist.
Brain imaging may be used to investigate other possible medical causes, but routine MRI scans are not currently an established standalone diagnostic test for schizophrenia.
This gives researchers an important opportunity to use AI to find helpful brain patterns while still keeping professional psychiatric assessment central to the process.
What Is BrainProbe?
BrainProbe is an artificial intelligence platform created to analyse brain MRI images and offer quantitative information about structural and functional abnormalities associated with neuropsychiatric conditions.
It was developed by a research team led by Professor Albert Chih-Chieh Yang, associated with Taipei Veterans General Hospital and National Yang Ming Chiao Tung University in Taiwan.
In a July 3, 2025 announcement, Taipei Veterans General Hospital reported that the platform could assist schizophrenia assessment with 91.7% accuracy.
The developers described a system built from MRI data collected from more than 1,500 Taiwanese participants, including healthy individuals and people with psychiatric conditions.
Instead of relying only on what clinicians can see in an MRI scan, BrainProbe uses deep learning algorithms to find subtle brain patterns and create quantitative assessments.
The hospital describes several capabilities:
- AI-assisted analysis of brain MRI scans.
- Detection and localisation of abnormalities in specific brain regions.
- Estimation of biological brain age.
- Monitoring of changes associated with brain ageing and disease progression.
- Three-dimensional visualisation to support clinical interpretation.
- MRI-guided planning for research into non-invasive brain stimulation treatments.
These features could give clinicians additional biological information when evaluating patients with complex or unclear psychiatric symptoms.
Original announcement: Taipei Veterans General Hospital, Smart Brain Imaging Platform, July 2025.
How Does BrainProbe Use AI to Analyse the Brain?
BrainProbe combines MRI technology with machine learning to evaluate patterns that may distinguish groups of people with schizophrenia from comparison populations.
Its basic process includes several stages.
1. Collecting MRI Brain Images
MRI creates high-resolution images of brain structures without using ionising radiation.
Researchers can analyse differences in brain tissue, regional anatomy, and other measurable characteristics.
For BrainProbe, the hospital reported using a database containing information from more than 1,500 participants.
It’s important to note that just because the imaging database is large, it doesn’t mean all participants were included in the calculation of the reported 91.7% accuracy.
2. Identifying Patterns Through Deep Learning
Deep learning algorithms can analyse complex combinations of image features that may be difficult for human observers to evaluate consistently.
In schizophrenia research, investigators have examined differences involving cortical regions, white matter pathways, and networks associated with perception, emotion, and cognition.
BrainProbe reportedly focuses on abnormal patterns inside regions including the insular cortex and temporal lobes.
However, neurological abnormalities linked to schizophrenia are not unique to the condition, and findings can differ substantially among patients.
3. Producing a Visual Brain Assessment
One of BrainProbe’s distinctive features includes its ability to present quantitative findings via visual reports.
These may help clinicians explain patterns of brain differences to patients and their families.
Three-dimensional imaging could also make findings easier to interpret than conventional numerical reports alone.
However, an AI-generated visualization does not prove that someone has schizophrenia. It is just the result of a model and needs to be considered along with the patient’s symptoms, history, and clinical assessment.
4. Estimating Biological Brain Age
BrainProbe also incorporates models designed to estimate how an individual’s brain compares with patterns commonly observed at different ages.
Researchers have investigated whether some psychiatric disorders are associated with accelerated biological brain ageing.
Such estimates may eventually help monitor disease-associated changes or guide research into treatment responses.
But a higher brain-age estimate is not a diagnosis, and what it means for an individual patient is still unclear.
What Does BrainProbe’s 91.7% Accuracy Actually Mean?
The most widely reported finding about BrainProbe is its 91.7% accuracy in assessing schizophrenia.
That figure originated from announcements by Taipei Veterans General Hospital in July 2025 and was subsequently reported by Taiwanese news organisations.
Basically, 91.7% accuracy means the model correctly labels about 92 out of every 100 cases in the dataset it was tested on.
But this interpretation requires several qualifications.
Accuracy does not mean that every patient receiving a positive result has a 91.7% probability of having schizophrenia.
The significance of any diagnostic model depends on the population being tested and how researchers evaluated its performance.
For example, a model tested on carefully selected schizophrenia patients and healthy controls may perform differently when used among people with depression, bipolar disorder, substance-related psychosis, or other overlapping conditions.
Several details are essential when evaluating a medical AI system:
- Sensitivity: How effectively the model detects patients who have schizophrenia.
- Specificity: How effectively it identifies people who do not have schizophrenia.
- External validation: Whether the model performs reliably in independent hospitals and populations.
- Prospective evaluation: Whether it works when tested on new patients in real clinical settings.
- Clinical utility: Whether using the model improves diagnosis, treatment decisions, or clinical outcomes.
The public hospital announcements reviewed for this article did not provide enough detailed methodological information to independently establish all these performance measures for the headline 91.7% figure.
So, this result should be seen as developer-reported performance, not as proof that BrainProbe will have the same accuracy for every patient or hospital.
Is There Peer-Reviewed Research Supporting AI-Based Schizophrenia Classification?
Yes. Scientific research provides evidence that machine learning can distinguish certain brain imaging patterns linked to schizophrenia.
For example, a peer-reviewed study published in European Psychiatry evaluated machine learning models for schizophrenia classification using a single-site dataset of 440 participants.
The study, titled Evaluating the performance of machine learning models for automatic diagnosis of patients with schizophrenia based on a single-site dataset of 440 participants, demonstrates the academic interest in MRI-based classification methods.
Research paper: European Psychiatry, DOI: 10.1192/j.eurpsy.2021.2248.
This is related research involving Taiwanese institutions; it should not be treated as independent verification of BrainProbe’s specific 91.7% performance claim.
The main point is that promising experimental results are not the same as having a clinically validated diagnostic test.
A Reported Clinical Case: Detecting Neurological Abnormalities
During the platform’s public introduction, the research team described a patient in his thirties experiencing auditory hallucinations and persecutory delusions that disrupted everyday life.
According to the hospital’s account, BrainProbe analysis identified abnormalities in regions including the insula and temporal lobes.
These data contributed additional information to the clinical assessment, and the patient subsequently received a schizophrenia diagnosis and individualised treatment recommendations.
This example shows how AI-assisted imaging could help with clinical decision-making.
However, one reported clinical case cannot establish diagnostic reliability or demonstrate that using BrainProbe improves long-term clinical outcomes.
Controlled, independently validated studies are needed to answer those questions.
Beyond Diagnosis: Could BrainProbe Support Personalised Treatment?
The research team’s ambitions go beyond identifying brain differences.
BrainProbe’s MRI analysis and visualisation capabilities may potentially help researchers identify brain regions for focused treatments.
One area of interest is non-invasive brain stimulation.
Transcranial Magnetic Stimulation
Repetitive transcranial magnetic stimulation (rTMS) uses magnetic pulses to influence activity in selected brain regions.
It is already used in certain clinical contexts, particularly for some forms of depression.
Researchers are also investigating brain stimulation techniques for specific symptoms and conditions linked to schizophrenia.
Still, using MRI-guided targeting with BrainProbe does not prove that these treatments work for everyone with schizophrenia.
Clinical trials are needed to determine which patients might benefit, the magnitude of those benefits, and possible risks.
Time Interference Brain Stimulation
The team is also researching time-based interference brain stimulation, or TIBS.
This experimental approach seeks to influence deeper brain regions via patterns of electrical stimulation without requiring surgery.
AI-based imaging and computational modelling may help determine which brain regions should be targeted and how stimulation should be delivered.
For now, the technique remains investigational for schizophrenia and other psychiatric conditions.
October 2026 Update: What Has Changed?
An important development emerged on October 29, 2026, when Nanyang Technological University’s Lee Kong Chian School of Medicine in Singapore published an interview with Professor Albert Yang about BrainProbe and the future of AI-assisted psychiatry.
Yang explained that BrainProbe uses explainable AI to compare patient MRI findings with healthy brain data and produces visual reports to support clinical interpretation.
He also emphasised that the technology is meant to complement, not replace, professional clinical judgment.
The interview highlighted three developments.
Research Is Expanding Toward Individualised Brain Stimulation
The team is investigating how AI-generated MRI findings can help personalise time interference brain stimulation.
According to Yang, researchers have completed a Phase I safety study and are progressing toward clinical trials involving depression, schizophrenia, bipolar disorder, and Alzheimer’s disease.
Finishing an early safety study does not mean the treatment is effective.
Researchers Are Investigating Earlier Detection
Brain-ageing algorithms have identified differences linked with schizophrenia and other psychiatric disorders.
The researchers hope these patterns should eventually help identify individuals who need closer monitoring.
However, Yang acknowledged that further studies are needed to determine which markers meaningfully predict future illness.
Independent Validation Continues to be Essential
The interview also identified major barriers to routine clinical integration, including the need to evaluate across different populations, hospitals, and MRI systems.
Researchers need to show that AI-assisted assessments actually improve clinical decisions and outcomes, not just produce good classification scores.
2026 source: Nanyang Technological University, Treating Mental Illness More Precisely With AI, October 29, 2026.
The Limitations and Dangers of AI-Assisted Schizophrenia Diagnosis
Although BrainProbe has potential, it faces several challenges before its findings can be used in everyday psychiatric practice.
Differences Between Patient Populations
Brain structure and imaging measurements can vary with age, demographic characteristics, medical history, scanner settings, and other factors.
An AI model developed primarily using Taiwanese imaging data may not perform identically in patients from other countries.
Independent testing is necessary to establish generalizability.
Overlapping Psychiatric Conditions
Schizophrenia shares some clinical and biological characteristics with other psychiatric disorders.
An algorithm that distinguishes schizophrenia from healthy controls may not be equally effective at differentiating schizophrenia from other causes of psychosis.
So, real-world assessments need tougher comparisons than just comparing patients to healthy controls.
Risks of Incorrect Results
A false-positive result could contribute to unnecessary distress, inappropriate labelling, or additional medical investigations.
A false-negative result could create misplaced reassurance and potentially delay needed assessment.
These risks show why AI predictions should never be the only factor in making a psychiatric diagnosis.
Cost and Availability
MRI scanning requires specialised equipment, technical expertise, and healthcare resources.
Expanding MRI-based psychiatric assessment could create financial and management challenges, especially in settings where basic mental health services are already difficult to access.
A technology that works well in research must also prove its clinical benefits justify the resources it requires.
Data Privacy and Transparency
Brain imaging datasets contain sensitive medical information.
Clinical AI systems require safeguards for informed consent, data security, model governance, and monitoring for potential bias.
Explainable outputs can help clinicians understand what a model finds, but being transparent does not always mean the results are reliable.
Does BrainProbe Mean Schizophrenia Can Now Be Diagnosed With a Brain Scan?
No. BrainProbe does not establish that schizophrenia can now be diagnosed definitively from an MRI scan.
The platform provides an emerging way to add brain imaging information to psychiatric assessment.
The reported results are scientifically interesting, but clinical diagnosis still depends on a thorough evaluation by qualified healthcare professionals.
This distinction is particularly important because schizophrenia is heterogeneous: two people with the same diagnosis may have substantially different symptoms, brain imaging findings, treatment needs, and recovery experiences.
So, the most promising future use may be as one part of a broader assessment, together with clinical history, structured psychiatric interviews, and other validated information.
The Future of AI in Precision Psychiatry
BrainProbe illustrates a larger scientific effort to develop more objective methods for understanding mental health conditions.
Rather than relying on a single biological measurement, future systems may combine brain imaging, clinical information, mental evaluations, and other relevant data.
Researchers hope such approaches will eventually help clinicians make more well-informed decisions, distinguish among overlapping conditions, and investigate therapies customized to individual patients.
But progress should be judged by more than just an impressive accuracy percentage.
The key question is whether these systems can improve care in real-world settings, as shown by independent evaluations, without causing unacceptable risks.
For BrainProbe, future research needs to show strong diagnostic performance, clarify its role in clinical practice, and prove whether AI-guided interventions offer real benefits.
Conclusion
Taiwan’s BrainProbe represents a potentially valuable development in AI-assisted psychiatric assessment.
By combining MRI scans, deep learning, and brain visualization, researchers at Taipei Veterans General Hospital have shown an approach that could give clinicians extra biological information.
The team’s reported 91.7% accuracy has attracted attention, while research developments through October 2026 suggest growing interest in using brain imaging to guide individualised treatment.
However, major questions remain about independent validation, clinical effectiveness, accessibility, and safety.
BrainProbe is important not because it could replace psychiatrists with AI, but because it explores whether objective brain measurements can make psychiatric assessment more informative and personalized.
This possibility deserves more research, careful evaluation, and realistic expectations.
References
- Taipei Veterans General Hospital (July 3, 2025). Smart Brain Imaging Platform for Schizophrenia Assessment. Original institutional announcement.
https://www.vghtpe.gov.tw/vghtpe/ophopd/bmsc/ps/News!one.action?gcode=A05&nid=14804 - Taiwan Healthcare+ (July 3, 2025). Taipei Veterans General Hospital Reaches AI Diagnostic Milestone: Brain Imaging Identifies Schizophrenia With 91.7% Accuracy.
https://www.taiwan-healthcare.org/en/news-detail?id=0sysst69w74xp5da - Nanyang Technological University (October 29, 2026). Treating Mental Illness More Precisely With AI. Interview with Professor Albert Yang.
https://www.ntu.edu.sg/medicine/news-events/news/detail/treating-mental-illness-more-precisely-with-ai - Lee et al. (2022). Evaluating the Performance of Machine Learning Models for Automatic Diagnosis of Patients With Schizophrenia Based on a Single Site Dataset of 440 Participants. European Psychiatry.
https://doi.org/10.1192/j.eurpsy.2021.2248 - World Health Organisation (September 11, 2026). Schizophrenia: Fact Sheet.
https://www.who.int/en/news-room/fact-sheets/detail/schizophrenia - National Health Service. Schizophrenia: Diagnosis.
https://www.nhs.uk/mental-health/conditions/schizophrenia/diagnosis/ - National Institute of Mental Health. Understanding Psychosis.
https://www.nimh.nih.gov/health/publications/understanding-psychosis