Information-Theoretic Clustering of Neuroimaging Metrics Related to Cognitive Decline in the Elderly

• Published in: Medical Computer Vision: Algorithms for Big Data pp. 13 - 23
• Main Authors: Daianu, Madelaine, Ver Steeg, Greg, Mezher, Adam, Jahanshad, Neda, Nir, Talia M., Yan, Xiaoran, Prasad, Gautam, Lerman, Kristina, Galstyan, Aram, Thompson, Paul M.
• Format: Book Chapter
• Language: English
• Published: Cham Springer International Publishing 2016
• Series: Lecture Notes in Computer Science
• Subjects: Brain connectivity; Diffusion weighted imaging; Gray matter; Machine learning; Spectral graph theory
• ISBN: 3319420151; 9783319420158
• ISSN: 0302-9743; 1611-3349
• DOI: 10.1007/978-3-319-42016-5_2

As Alzheimer’s disease progresses, there are changes in metrics of brain atrophy and network breakdown derived from anatomical or diffusion MRI. Neuroimaging biomarkers of cognitive decline are crucial to identify, but few studies have investigated how sets of biomarkers cluster in terms of the information they provide. Here, we evaluated more than 700 frequently studied diffusion and anatomical measures in 247 elderly participants from the Alzheimer’s Disease Neuroimaging Initiative (ADNI). We used a novel unsupervised machine learning technique - CorEx - to identify groups of measures with high multivariate mutual information; we computed latent factors to explain correlations among them. We visualized groups of measures discovered by CorEx in a hierarchical structure and determined how well they predict cognitive decline. Clusters of variables significantly predicted cognitive decline, including measures of cortical gray matter, and correlated measures of brain networks derived from graph theory and spectral graph theory.