AIR4Health
THUNDER: Fostering Trust in AI-Driven Healthcare
Funded by the European UnionHorizon Europe · MSCA Staff ExchangesGrant No. 101236749Trustworthy Generative AI · Healthcare
THUNDER: Fostering Trust in AI-Driven Healthcare
SecUre and uNbiased knowleDge-guided gEneRative AI — Generative AI, particularly large language models (LLMs), can synthesise realistic healthcare datasets to address data scarcity, but its adoption is held back by concerns about semantic validity, fairness, bias, fidelity, privacy, and real-world utility. THUNDER aims to forge a comprehensive framework for trustworthy and responsible generative AI in healthcare, with sepsis as its target clinical application.
Funding Support
European Union · Marie Skłodowska-Curie Actions
THUNDER has received funding from the European Union’s Horizon Europe Research and Innovation Programme under the Marie Skłodowska-Curie Actions (MSCA), Grant Agreement No. 101236749, with an EU contribution of €771,540. The project involves 154 staff secondments across the consortium.
The Challenge
Machine learning offers transformative opportunities for healthcare, from precision medicine to operational optimisation, but progress is constrained by limited access to diverse, high-quality data, made worse by fragmentation, data scarcity, and strict privacy regulations. Off-the-shelf generative AI can create synthetic data, but it is not ready for clinical use. THUNDER addresses four key gaps:
- No standardised frameworks for evaluating synthetic data
- Trust deficits in healthcare generative models
- The resource intensiveness of current approaches
- Design-induced opaqueness in AI decision-making
The THUNDER Framework
Anchored in expert medical knowledge, the framework rests on three Research Initiatives that together support safe and reliable generative AI:
RI-1. Standardised Evaluation
The Ruler. Comprehensive metrics for evaluating synthetic healthcare data, covering fidelity, utility, and privacy, as well as semantic validity and fairness.
RI-2. Knowledge-Guided Generation
The Engine. Generative models that use RAG, LLMs, and expert-defined domain knowledge to produce trustworthy, clinically relevant synthetic data.
RI-3. Frugal & Interpretable Design
The Vehicle. Resource-efficient models that are interpretable by design, using approaches such as neural architecture search (NAS), spiking neural networks (SNNs), and explainability.
Target application: sepsis. THUNDER focuses on sepsis, a WHO-designated global health priority, to demonstrate and validate the framework’s real-world clinical impact.
Work Packages
Research
- WP3: Foundations of metrics for synthetic data evaluation
- WP4: Pioneering knowledge-guided generative models for trustworthy synthetic data
- WP5: Advanced frugal-by-design ML for actionable insights
- WP6: Iterative clinical validation and refinement of AI models
Management & Training
- WP1: Management
- WP2: Training, communication, and dissemination
Edinburgh Napier University Team
ENU Team Lead
Prof. Hongnian Yu
Fellow of the IET, Prof. Yu leads ENU’s THUNDER team, with expertise in digital health, robotics, and neural networks.
Supporting Experts
Multidisciplinary Team
Dr Mary Abboah-Offei (public health), Prof. Lai Xu (AI-driven predictive models), Dr Shuang Cang (machine learning), Dr Paul de Vrieze (digital twin simulations), and Dr Luigi La Spada (AI-driven signal processing), along with 10 aligned PhD students.
Consortium
A pan-European and international collaboration of 11 participants and 13 associated partners, coordinated by the Université de Lille (France). Roles below follow the EU CORDIS project record.
Participants
- Université de Lille (France): coordinator
- Université de Versailles Saint-Quentin-en-Yvelines (France)
- Edinburgh Napier University (UK)
- University of Surrey (UK)
- Universidade da Coruña (Spain)
- Universidad de Sevilla (Spain)
- Vicomtech (Spain)
- KU Leuven (Belgium)
- Tomas Bata University in Zlín (Czechia)
- National Engineering School of Sfax (Tunisia)
- Business & AI (Tunisia)
Associated Partners
- Assistance Publique – Hôpitaux de Paris (AP-HP) (France)
- Predict SAS (France)
- LMU Munich (Germany)
- University of Vienna (Austria)
- Canary Islands Health Research Institute Foundation (FIISC) (Spain)
- FlyThings Technologies (Spain)
- PharmaDataHub (PDH) (Portugal)
- Revelia (Czechia)
- AINIGMA Technologies (Belgium and Greece)
- The Education University of Hong Kong (Hong Kong, China)
- VinUniversity (Vietnam)
- École de technologie supérieure (ÉTS) (Canada)
Early Results
Since the January 2026 kick-off, consortium members have published in the Journal of Biomedical Informatics and had papers accepted at conferences including IJCAI-ECAI 2026, ECML PKDD 2026, ICONIP 2026, ESANN 2026, and WCCI (IJCNN) 2026. Secondments between partners are under way.