Tensions et convergences dans la conception de nouveaux outils d’intelligence artificielle pour l’oncologie : le cas de la radiomique

The emerging field of radiomics aims to extract quantitative information from medical images. This advanced analysis technique seeks to identify specific biomarkers that can improve patient categorization and care. In oncology, clinicians collaborate with experts in image processing to design new predictive models based on artificial intelligence. This article demonstrates how technicizing diagnostic and prognostic tools has led both to a convergence of interests in this new field – that allows for an accumulation of scientific capital – and to tensions affecting, among other things, the criteria involved in the validation of technologies. In particular, the performance metrics used by researchers do not allow clinicians to measure their clinical usefulness, which is judged by the context of use. Therefore, various standards are applied to evaluate these new imaging biomarkers and their success depends on the articulation between medical and computational knowledge.