Preclinical Alzheimer's Disease Prediction using Graph Neural Networks

Abstract

Deep learning has revolutionized many machine learning tasks in recent years, ranging from image classification and video processing to speech recognition and natural language understanding. The data in these tasks are typically represented in the Euclidean space. However, there is an increasing number of applications where data are generated from non-Euclidean domains and are represented as graphs with complex relationships and interdependency between objects. The complexity of graph data has imposed significant challenges on existing machine learning algorithms. Recently

Alzheimer?s disease (AD) is the most common form of dementia and it is considered as a biological continuum that can begin decades before the first cognitive symptoms. The detection of healthy but amyloid positive individuals is an opportunity for the prevention of the disease but non-invasive and cost-efficient amyloid detection techniques are needed to reduce the number of unnecessary, invasive, expensive PET/CSF tests. The aim of this project is to study the state of the art of Deep Learning on graphs or Geometric Deep Learning and its most known models: Graph Neural Networks in order to use them to predict the preclinical stage of Alzheimer?s disease with parcelled and processed MRI, which have been expressed as graphs using the regions of interest defined by the brain parcellation atlases as nodes and their volumes and other features as node signals. Two different datasets have been used and addressed as two independent graph classification tasks. Furthermore, the results have been interpreted carrying out a class activation mapping technique that determines what are the most relevant brain regions for the models to predict the preclinical stage.