Kereval is one of 17 successful applicants to a call for expressions of interest issued by the Health Data Hub (HDH), which is implementing the government’s strategy for the National Health Data System (SNDS).

The project led by Kereval, known as Santé2Fair, aims to develop a methodological framework for assessing the fairness of artificial intelligence (AI) models and associated health data, with an initial application in oncology, using clinical and imaging data from the SNDS.

The project is structured around two main components: the design of a detailed – but generic – methodology for analysing statistical and algorithmic biases in AI systems within the healthcare sector, and a multi-level strategy to mitigate these biases.

The work will be conducted on real-world use cases in oncology, within the HDH’s secure infrastructure.

Test models (specialised AI systems) will be developed and trained on data from two projects led by the HDH.

The rapid development of AI in healthcare brings major challenges in terms of the quality, representativeness and fairness of the data used to train models. These issues are particularly critical in oncology, a field with high clinical stakes where data is rich but heterogeneous, and where AI models are increasingly used for medical image analysis.

In particular, biases present in datasets – demographic imbalances, inconsistent coding practices or disparities in access to healthcare – can be amplified by machine learning algorithms, leading to medical decisions that systematically disadvantage certain population groups.

Expertise in AI testing

Kereval has a multidisciplinary team specialising in software engineering and AI, working in areas such as trustworthy AI, MLOps and robustness testing.

The company supervised a Cifre PhD thesis, in partnership with ENSTA Bretagne, which resulted in several scientific publications on the definition of testing methodologies for AI systems. Furthermore, Kereval collaborates with the Signal and Image Processing Laboratory (LTSI) at Rennes University Hospital, particularly on the structuring and use of clinical and medical-administrative data.

The company is also conducting specific research on the evaluation and formalisation of fairness in machine learning algorithms. These topics are central to European strategies on AI and health data, particularly in the development of the European Health Data Space (EHDS). Kereval’s teams contribute to this effort through the European projects i2X and myHealth@myHands, supported by the European Commission.