Séminaire du CERAMATHS - DMATHS : exposé d'Hernan F. Garcia-Arias

Le séminaire du département de mathématiques du CERAMATHS accueillera à 15h Hernan F. Garcia-Arias (Universidad de Antioquia, Colombie), jeudi 12 décembre 2024, pour l'exposé suivant :

Probabilistic Correspondence Analysis in Pediatric Health using Variational Mixture Models

Analyzing brain structures in medical imaging presents significant challenges, mainly due to the heterogeneity of pediatric diseases. Quantitative measurement of brain changes is essential for evaluating clinical outcomes linked to anatomical factors. From an artificial intelligence perspective, establishing correspondences between nonrigid brain structures often relies on similarity-based measures, which are frequently unavailable. In this talk, I will present an unsupervised probabilistic framework for shape matching in brain structures using variational unsupervised learning. This approach effectively captures the underlying representation of surface descriptors associated with brain anatomy. By leveraging a variational Gaussian process latent variable model, we learn group-wise latent space representations of these descriptors, enabling the establishment of unsupervised correspondences. Experimental results demonstrate the model's ability to capture non-linearities in nonrigid brain structures from real-world neurodevelopmental data, highlighting its potential for monitoring anatomical changes in both healthy and abnormal brain shapes.