Enhancing Histology Analysis with Interpretable Graph Neural Networks
Encadrants : Mohamed Mounib Benimam, Vannary Meas-Yedid
Email : mbenimam@pasteur.fr
Disponible : NON
Nombre d'étudiants : 1
Description :
Histological image analysis plays a crucial role in cancer diagnosis and prognosis, yet pixel-based analysis methods often lack interpretability and struggle to capture complex relationships among entities within tissues. This project focuses on introducing a novel approach that shifts from pixel-based analysis to entity-based analysis in histology, leveraging Graph Neural Networks (GNNs) to provide interpretable insights and enhanced control over biological priors
Pré-requis : Deep learning (Pytorch), Basic knowledge of graph theory
Travail demandé : After conducting a state-of-the-art review on graph neural networks, the student will develop and train a pipeline for classifying tumor malignancy using extracted biological entities to create a knowledge graph. The performance of this approach will be evaluated against baseline pixel-based methods. Additionally, the student will explore visualization and interpretability techniques.
Liens complémentaires : https://www.kaggle.com/datasets/ambarish/breakhis
https://github.com/vqdang/hover_net
https://github.com/vqdang/hover_net