Project Overview
Advancing Medical AI through Privacy-Preserving Federated Learning for Osteoporosis Prediction
About This Project
Osteoporosis is highly prevalent in older people, but detection is mostly ignored until bone fractures occur, which then affects daily lives and can be life-threatening. The diagnosis of osteoporosis requires the gold standard imaging modality, dual energy X-ray absorptiometry (DXA), which is typically available only in larger hospitals in Thailand; hence, limited access.
Traditional Machine Learning systems require data to be fused in a centralised server for model training. Collecting and aggregating patient data dispersed over various isolated sources, and then securely processing and managing the data are non-trivial tasks, particularly in Thailand where data protection legislation (Thai PDPA) came into force in June 2022.
This project initiates a cornerstone for transforming traditional osteoporosis diagnosis using DXA scans by a convenient, economical and privacy-preserving ML method leveraging the emerging Federated Learning (FL) technique over readily available datasets such as routine X-ray and abdominal CT scans. The FL system is then integrated with a virtual hospital in a metaverse with digital twin technology demonstrating its convenience, interactive nature, and effectiveness using healthcare services.
Project Aim
To develop and validate a comprehensive privacy-preserving federated learning system that can accurately predict osteoporosis risk from routine X-ray and CT scans, making bone health diagnosis more accessible and economical while maintaining the highest standards of data privacy under Thai PDPA compliance.
Key Objectives
Our research is structured around four primary objectives that will drive innovation in medical AI and healthcare data security.
Develop Accessible Diagnostic Methods
Create ML models that can predict osteoporosis from routine X-ray and CT scans available in smaller hospitals, reducing dependency on expensive DXA equipment limited to larger facilities.
Implement Thai PDPA Compliant Framework
Develop Privacy-Preserving Federated Learning systems that comply with Thailand's Personal Data Protection Act (PDPA) while enabling collaborative model training across healthcare institutions.
Create Metaverse Healthcare Integration
Integrate the federated learning system with a virtual hospital in metaverse using digital twin technology to demonstrate convenience, interactivity, and effectiveness of healthcare services.
Foster International Collaboration
Establish research partnerships with University of Glasgow and promote knowledge exchange in AI healthcare applications.
Expected Outcomes
This project will deliver tangible outcomes that advance both scientific knowledge and practical healthcare applications.
Lab Scale Prototype (TRL 5)
Functional AI system for osteoporosis prediction from X-ray images
International Publications
Q1 journal articles and international conference presentations
Enhanced Healthcare Security
Privacy-preserving medical AI framework for sensitive data protection
Skilled Workforce Development
Trained researchers in advanced AI and cybersecurity techniques
Project Impact
This research addresses the critical healthcare challenge of osteoporosis detection in Thailand by developing accessible, privacy-compliant AI solutions that transform routine medical imaging into powerful diagnostic tools. By leveraging federated learning and metaverse technologies, we aim to democratize bone health diagnosis while maintaining strict data protection standards under Thai PDPA legislation.
Healthcare Accessibility
Making osteoporosis diagnosis available in smaller hospitals across Thailand
PDPA Compliance
Setting standards for privacy-preserving medical AI under Thai legislation
Metaverse Integration
Pioneering virtual healthcare delivery through digital twin technology