FURI | Spring 2024

Geometric Deep Learning and Anatomical Landmarking for Improved Preclinical Alzheimer’s Disease Diagnosis

This study develops a preclinical Alzheimer’s disease screening method by applying geometric deep learning on hippocampal structures. Researchers will first extract the hippocampal structures from the MRI images and build volumetric meshes with existing tools. In the model, traditional graph convolution is enhanced by customized volumetric Laplace Beltrami Operators (LBOs). Moreover, the team will extract anatomical landmarks using hierarchical Bayesian networks, which are then integrated into model with a cross attention mechanism. As a result, model achieves higher classification accuracies among all diagnosis groups. Future work will focus on refining the model and expanding its applicability to additional disorders.