USP AI Breakthrough: 90% Accuracy in Mental Disorder Diagnosis from Brain Scans

Revolutionizing Mental Health Research at Brazil's Top University

  • research-publication-news
  • usp-ai
  • mental-health-diagnosis
  • graph-neural-networks
  • francisco-rodrigues
New0 comments

Be one of the first to share your thoughts!

Add your comments now!

Have your say

Engagement level
a black and white photo of the word mental health
Photo by Marcel Strauß on Unsplash

The Groundbreaking AI Development at USP

A remarkable advancement in mental health diagnostics has emerged from the University of São Paulo (USP), Brazil's premier higher education institution. Professor Francisco Aparecido Rodrigues, a leading expert in complex systems and data science at USP's Institute of Mathematical and Computer Sciences (ICMC), has pioneered an artificial intelligence (AI) tool that identifies mental disorders with over 90% accuracy. This innovation leverages machine learning algorithms trained on brain imaging data, such as magnetic resonance imaging (MRI) and electroencephalogram (EEG), to detect subtle neural patterns associated with conditions like autism spectrum disorder (ASD), schizophrenia, and epilepsy. 70 71

The tool's potential to revolutionize early diagnosis is particularly timely, given the escalating mental health challenges faced by university students and faculty across Brazil. Traditional diagnostic methods rely heavily on clinical interviews and subjective assessments, often delaying intervention. Rodrigues' AI offers an objective, data-driven alternative, analyzing functional brain networks to pinpoint alterations before symptoms fully manifest.

Brain scan analysis using AI at USP for mental disorder detection

Profile of Innovator Francisco Rodrigues

Francisco Rodrigues is a Full Professor at ICMC-USP in São Carlos, heading the Complex Systems Laboratory. With a PhD in Physics and expertise in network science, machine learning, and data science, he has authored over 200 publications, garnering more than 12,000 citations and an h-index of 43. His work bridges mathematics, computer science, and neuroscience, applying graph theory to model brain connectivity. 146

Rodrigues' interdisciplinary approach stems from collaborations with neuroscientists and clinicians, including partnerships with German researchers. Recently recognized with the prestigious Friedrich Wilhelm Bessel Research Award from the Alexander von Humboldt Foundation—valued at €60,000—he will spend time in Germany advancing his work, underscoring USP's global stature in AI-driven research. 103

For aspiring researchers, Rodrigues exemplifies the career paths available in Brazil's vibrant higher education sector. Opportunities abound in research jobs at institutions like USP, where faculty can lead cutting-edge projects funded by FAPESP and CNPq.

How the AI Works: Graph Neural Networks in Brain Analysis

At the core of Rodrigues' AI is graph neural network (GNN) technology, a type of deep learning model designed for data structured as graphs—networks of nodes (brain regions) connected by edges (functional connectivity). Functional brain networks are constructed from fMRI or EEG time series, where pairwise metrics like Spearman correlation or transfer entropy quantify interactions between regions of interest (ROIs), such as those from the Bootstrap Analysis of Stable Cluster (BASC) atlas. 157

The process unfolds step-by-step:

  • Data Preprocessing: Raw MRI/EEG signals are filtered and segmented into ROIs (e.g., 122 regions).
  • Connectivity Matrix Generation: Compute statistical dependencies to form adjacency matrices representing brain graphs.
  • Graph Features Extraction: Apply complex network measures like degree centrality, clustering coefficient, k-core decomposition, and community structure.
  • Machine Learning Classification: Feed features into classifiers (e.g., logistic regression, random forest) or GNNs for diagnosis, using techniques like SHAP for interpretability.
  • Validation: Cross-validation on datasets like ABIDE (500 subjects for ASD) yields AUC scores up to 0.99.

This methodology not only diagnoses but reveals insights, such as reduced connectivity in the left ventral posterior cingulate cortex to the cerebellum in ASD patients.

Targeted Disorders and Impressive Validation Results

The AI excels in multiclass classification, distinguishing ASD, attention deficit hyperactivity disorder (ADHD), schizophrenia, Alzheimer's, and healthy controls. In a landmark study using ABIDE dataset (242 ASD, 258 typical development), it achieved 99% accuracy, precision, recall, and F1-score with logistic regression on connectivity matrices. 157

Other validations include EEG-based diagnosis of schizophrenia and Alzheimer's (Journal of Physics: Complexity, 2022) and DMT-induced brain changes (PLOS One, 2022). For smaller datasets, sliding window augmentation maintains ~81% accuracy with just 30 patients, making it practical for Brazilian clinics. 135

These results surpass traditional methods, positioning USP at the forefront of neuroinformatics.

Global Recognition: Humboldt Prize and Beyond

In February 2026, Rodrigues received the Humboldt Foundation's Bessel Prize, one of 20 awarded worldwide for transformative research. The €60k grant funds a year-long collaboration in Frankfurt, lecturing on complex systems and ML while expanding datasets with German partners. 71 103

This accolade highlights USP's excellence, attracting funding and talent. Rodrigues noted, "São resultados que realmente poderão transformar vidas," emphasizing objective diagnostics via MRI.FAPESP Agency

a close up of a typewriter with a paper on it

Photo by Markus Winkler on Unsplash

Mental Health Crisis in Brazilian Higher Education

Brazilian universities grapple with alarming mental health rates: 44-60% of students report common mental disorders (TMC), including depression (51%) and anxiety (42-98%). At USP and peers, factors like academic stress exacerbate issues, with women and low-income students most affected. 124 136

Post-pandemic, graduate students show 56% TMC prevalence. Rodrigues' AI could screen campus populations, aiding services like USP's mental health programs. For faculty, higher-ed career advice stresses work-life balance amid rising burnout.

Statistics on mental health prevalence among Brazilian university students

Challenges: Data Scarcity and Regulatory Hurdles

Key obstacles include limited Brazilian datasets—Rodrigues relies on US ABIDE data—and EEG/MRI acquisition difficulties (patients immobile 40+ mins). Anvisa approval for clinical use may take 10 years. Ethical data collection and bias mitigation are paramount. 71

Solutions: Minicerebros (organoids) simulate disorders in labs, testing drugs on desynchronized neurons. International collaborations address gaps.

Future Outlook: Minibrains and Predictive Diagnostics

Rodrigues envisions predictive tools flagging Alzheimer's risk 10 years early via brain changes. Minibrains from animal embryo cells will model networks, accelerating drug discovery. USP's CeMEAI center supports scaling.DW Article

This aligns with Brazil's AI health push, potentially integrating into SUS primary care.

AI's Role in Brazilian Higher Ed Research Boom

USP leads Brazil's AI surge, with FAPESP funding interdisciplinary hubs. Similar projects: voice-based depression detection (90% accuracy). Unis foster university jobs in AI, from postdocs to faculty. 24

Rate professors pioneering this at Rate My Professor.

Career Paths in AI for Health at Brazilian Unis

Rodrigues' success inspires: PhDs in complex systems lead to roles at Harvard, Google. Brazil needs AI experts; explore faculty positions or postdoc opportunities. Advice: Master GNNs via open courses.

A wooden table topped with scrabble tiles spelling news and deep seek

Photo by Markus Winkler on Unsplash

  • Benefits: Interdisciplinary impact, funding.
  • Risks: Data ethics, competition.
  • Comparisons: USP vs Unicamp AI hubs.

Stakeholder Perspectives and Implications

Psychiatrists praise objectivity; ethicists urge equity. Impacts: Reduced misdiagnosis, cost savings. For unis, enhanced research prestige attracts talent.

G1 Coverage | Autism Paper

Looking Ahead: Transforming Mental Health Care

USP's AI heralds precise, predictive diagnostics, vital for Brazil's 11.5% mental disorder rate. Aspiring pros, check higher-ed-jobs, rate-my-professor, career-advice, university-jobs. Share thoughts below.

Frequently Asked Questions

🧠What is the USP AI tool developed by Francisco Rodrigues?

The AI uses graph neural networks on MRI and EEG data to analyze brain functional connectivity, diagnosing disorders with >90% accuracy.

🔬Which mental disorders does the AI detect?

Primarily autism spectrum disorder (ASD), schizophrenia, epilepsy; expanding to Alzheimer's, depression, ADHD via multiclass classification.

📊How accurate is Rodrigues' AI?

Lab tests show 99% AUC/accuracy for ASD on ABIDE dataset; >90% overall for various disorders, outperforming traditional methods.Scientific Reports

🏆What prize did Francisco Rodrigues win?

Friedrich Wilhelm Bessel Research Award (€60k) from Alexander von Humboldt Foundation, enabling Germany collaboration.

🎓Why is this relevant for Brazilian universities?

With 44-60% students facing depression/anxiety, AI enables early screening. Explore Rate My Professor for USP experts.

🌐How does graph neural network (GNN) work here?

GNNs model brain regions as nodes, connections as edges; learn patterns from connectivity matrices for classification.

⚠️Challenges in implementing this AI?

Data scarcity, regulatory approval (Anvisa), MRI access. Solutions: organoids, international data sharing.

🔮Future plans for the research?

Predictive diagnostics (e.g., Alzheimer's 10yrs early), minibrains for drug testing, clinical trials.

📈Mental health stats in Brazil unis?

50%+ depression, 45%+ anxiety; higher in females, post-grads. AI aids campus wellness.

💼Career opportunities from this research?

AI health roles at USP-like unis. Check research jobs, career advice.

📚Where to learn more about USP AI research?

Rodrigues Lab, FAPESP CeMEAI.

Can this AI predict future disorders?

Yes, detects pre-symptomatic changes, e.g., schizophrenia risk years ahead.