School of Information Technology, Engineering, Mathematics and Physics

Biomarker-Based Mental Health Estimation Using Machine Learning

USP’s Groundbreaking AI and Biosensor Research for Mental Health Advancement

Drone-Based AI Monitoring for Climate Change & Wildlife Conservation

USP’s Groundbreaking AI and Biosensor Research for Mental Health Advancement

The University of the South Pacific is leading a cutting-edge project titled “Estimation of Human Mental States Using Biomarkers and Machine Learning Models.” This interdisciplinary initiative explores how physiological biomarkers, wearable technology, and artificial intelligence can be combined to better understand, monitor, and ultimately improve mental health.

🧬 Project Overview

Traditional mental health diagnosis often relies on subjective assessments, fixed diagnostic categories, and infrequent clinical visits. This project aims to change that by developing a real-time, data-driven approach to estimating mental states using non-invasive biosensors and machine learning algorithms.

Project Acronym: Biomarker-Based Mental Health States (BBMHS)
Principal Investigator: Levente Orban (USP)
Partners:

  • Voicu Groza – University of Ottawa (Canada)
  • Co-funding provided by the University of Ottawa

Collaborators:

  • Sefanaia Qaloewai and Violet Erasito (FNU)
  • Bibhya Sharma and Mansour Assaf (USP)

Location: Suva, Fiji
Duration: 2 years
Funding: FJD $30,000 (PIURN)

What Makes This Project Unique?

  • Utilizes wearable biosensors to continuously collect physiological data such as heart rate variability, pupil dilation, and cortisol levels
  • Applies machine learning (e.g., Random Forest, SVM, Neural Networks) to classify mental states (healthy/unhealthy)
  • Follows the Research Domain Criteria (RDoC) framework developed by the U.S. National Institute of Mental Health, shifting away from outdated, rigid diagnostic labels
  • Merges clinical data, digital biomarkers, and AI for a truly integrative mental health assessment

🔬 Methodology

  1. Participant Selection: Adults aged 18–65, including both clinical and subclinical populations. Ethical approval and informed consent protocols will follow international standards.
  2. Biometric Data Collection:
    • HRV (Heart Rate Variability) via wearable ECG
    • Pupillary Response using eye-tracking systems
    • Cortisol Levels through salivary assays collected across the day
  3. Mental Health Assessment:
    • Validated scales: Hamilton Anxiety Rating Scale, Beck Depression Inventory
    • Computer-based cognitive and emotional tasks targeting RDoC domains
  4. AI Model Development:
    • Feature engineering from multimodal data
    • Dimensionality reduction (PCA)
    • Model training using supervised learning (SVM, RF, Neural Nets)
    • Evaluation using ROC AUC, precision, recall, and nested cross-validation
  5. Data Analysis & Computation:
    • Python 3.8 with Scikit-learn and NumPy
    • Statistical validation using the DeLong test
  • Develop a prototype system that uses wearable devices and AI to estimate mental health states
  • Validate mental state predictions against established clinical criteria
  • Publish and disseminate findings in peer-reviewed journals and international AI and psychology conferences
  • Contribute to future public health tools capable of early mental health screening and personalized intervention

This project is in collaboration with Professor Voicu Groza from the University of Ottawa, who brings technical expertise in biomedical computing and sensor development. This cross-institutional effort strengthens USP’s research leadership in AI and mental health innovation in the Pacific.

Project Timeline (2024–2025)

  • Presentation at the 2024 Symposium on Applied Computational Intelligence and Informatics, Hungary
  • Submissions to international journals in AI, Neuroscience, and Computational Psychology
  • Contribution to future mental health policies and tools in the Pacific

For more information or collaboration opportunities, contact:
📧 levente.orban@usp.ac.fj