Description
Date depot: 18 juillet 2024
Titre: Pseudo-healthy image synthesis for the detection of anomalies in the brain, a multi-modal approach
Directrice de thèse:
Ninon BURGOS (ICM)
Domaine scientifique: Sciences et technologies de l'information et de la communication
Thématique CNRS : Sciences de l’information et sciences du vivant
Resumé: Neuroimaging is essential for diagnosing and treating neurological disorders, but integrating various imaging techniques is challenging due to large data volumes. Current frameworks lack effective multi-modal integration, necessitating new analytical tools for better data processing and decision-making. Unsupervised anomaly detection creates models of healthy brain images to identify anomalies in patient images. Variational autoencoders have shown promise but produce blurry images, potentially missing subtle lesions. The project aims to improve pseudo-healthy image quality using multi-modal approaches and will involve exploring generative adversarial networks and diffusion models. Additionally, an evaluation framework for multi-modal anomaly detection will be developed, ensuring lesion consistency across imaging modalities. The methods will be validated with large datasets and applied to neuroimaging for diagnosing disorders like Alzheimer's. These developments will be integrated into ClinicaDL, an open-source platform for reproducible neuroimaging processing with deep learning.
Doctorant.e: Roy Hugues