Description
Date depot: 2 avril 2024
Titre: AI-based biomarker from phenotypic images and genomic spatial transcriptomics of histological slides of cancerous tissues
Directeur de thèse:
Nicolas LOMENIE (LIPADE)
Directeur de thèse:
Julien CALDERARO (APHP (UP))
Domaine scientifique: Sciences et technologies de l'information et de la communication
Thématique CNRS : Images et vision
Resumé: We aim to develop new methodologies at the frontier of AI, computer vision and computational pathology, allowing the automatic discovery of new discriminating biomarkers for the study and medical/biological understanding of tumor heterogeneity, in the larger context of personalized medicine. Practically, the doctoral student, relying on the state of the art on multimodal foundation models, will aim to propose new models and algorithms to optimize / fine-tune giga-models from the literature for the different tasks of the doctoral project (e.g. tissue segmentation, analysis of spatial interactions between tumor regions of interest, prediction of pathologies at the WSI level, etc.). The starting point will be the implementation of pretext tasks implementing the biologist's knowledge for self-supervised training and the optimization of multi-modal foundation models. These models will be fed from multimodal data (1) visual from WSI and (2) genetic from spatial Transcriptomics. The self-supervised aspect will make it possible to limit as much as possible the use of annotated data, which is costly for doctors, and to limit the latter to the fine-tuning of models on the downstream tasks mentioned above and to their evaluation.