École Nationale Supérieure de
Techniques Avancées

Abstract

In recent years, artificial intelligence (AI) has aroused real enthusiasm in research and it is now an omnipresent notion when it comes to medical imaging. Deep learning computer methods are in full development to provide tools for the analysis of histological sections to pathologists. In this report we present the different steps leading to the classification of these images, from the creation of a dataset, thanks to cell segmentation, to the evaluation of the performance of several graph neural networks (GNN), through adjustment of the training parameters of these networks. We then show that GPU usage and training time are not limiting factors, unlike dataset size.