CYBER-HEALTH (Cybersecurity of externalized and mutualized healthcare data and processing) research activities are transversal to the LaTIM project. They are organized into two axes to address key issues centered on data and their uses in the era of big health data and artificial intelligence applications in healthcare:
- Continuous protection of medical data – fighting against information leaks and traceability by design – The aim of this axis is to develop data protection tools (e.g., watermarking and crypto-watermarking of data and AI models, radioactive data, AI based digital content forensics) in order to protect data all along their life cycle and to give back the control to the data producer onto the data he or she outsources.
- Secure processing of secured medical data – the objective of which is to allow the secure processing of data (e.g. statistical and machine learning algorithms) in partially/fully externalized/distributed/federated open environments; data secured with the help of the mechanisms developed in the previous axis or with more classic security tools.
These activities are developed in different contexts (medical imaging, genetics, public health, data storage on DNA molecules and connected medical devices) and in several national and international projects.
CYBER HEALTH leads CYBAILE, an industrial chair in Cybersecurity and Trusted AI in Healthcare, in partnership with THALES, AiiNtense, and SOPHIA GENETICS.
Continuous protection of medical data – fighting against information leaks and traceability by design
The aim of this axis is to develop protection tools for data all along their life cycle, from their acquisition to their reuse, and to give back the control to the data producer onto his data he or she outsources. These tools encompass the development of nvel approaches of:
- Watermarking and crypto-watermarking of data (images and genetic or public health databases)
- Radioactive data for machine learning
- AI based digital content forensics.
More recently, CYBER HEALTH investigates the security of the storage of data on DNA molecules, extending the above solutions to these context.
Some projects
- PAROMA-MED, Privacy Aware and Privacy Preserving Distributed and Robust Machine Learning for Medical Applications – crypto-watermarking of data
- TracIA, France 2030 PEPR Digital Health – watermarking based traceability of data and AI models
- CYBAILE, Inserm industrial Chair (Thales, AiiNTENSE, SOPHIA GENETICS) – Data and model watermarking, fight against posionning attacks
- TRUSTINClouds, France 2030 PEPR CLOUD – Data watermarking
- CARBURE, DGA RAPID project – Image digital forensic
- PC4, France 2030 PEPR MolecularXiv – watermarking and encryption of data
- DNASEC, ANR project – Watermarking of data and of the support
Secure processing of secured medical data
The aim of this axis is to develop secure processing of data, these ones being statistical or machine learning based, in partially/fully externalized/distributed/federated but open environments while considering data secured with classic security tools or with the help of the mechanisms developed in axis 1. It is at the origin of different solutions in the fields of:
- AI model watermarking
- Attacks and defenses against backdooring, poisoning and byzantine attack
- Secure processing of data (e.g. homomorphic encryption, multiparty computation).
Some projects
- CYBAILE, Inserm industrial Chair (Thales, AiiNTENSE, SOPHIA GENETICS) – Data and model watermarking, fight against posionning attacks
- SSF-ML-DH, France 2030 PEPR Digital Health – watermarking based traceability of data and AI models
- PAROMA-MED, Privacy Aware and Privacy Preserving Distributed and Robust Machine Learning for Medical Applications – watermarking of models
- TracIA, France 2030 PEPR Digital Health – watermarking based traceability of data and AI models
TEAM LEAder
Gouenou Coatrieux, Full Professor, IMT Atlantique
Permanent members
Reda Bellafqira, Associate Professor, IMT Atlantique
Mehdi Ben-Ghali, Research Engineer, Inserm
Kassem Kallas, Inserm Researcher, Inserm
Technician
Bilelle Triki