Online Program

Return to main conference page
Friday, May 31
Computational Statistics
Change Point Detection
Fri, May 31, 5:20 PM - 6:25 PM
Grand Ballroom K

Robust Graph Change-Point Detection for Brain Evolvement Study (306248)

Brain Caffo, Johns Hopkins University 
Xi Chen, New York University 
Fang Han, University of Washington 
Lexin Li, University of California, Berkeley 
*Honglang Wang, Indiana University-Purdue University Indianapolis 

Keywords: FMRI, change-point detection, Gaussian graphical model, least absolute deviation, splines

This paper studies brain structural evolvement from resting-state functional magnetic resonance imaging. The brain structure is characterized by a series of Gaussian graphical models, and we propose a robust data-driven method for inferring the structural changes of multiple graphs. The graphs correspond to different subjects, are aligned by, e.g., the ages of the subjects, and need to be estimated from the subject level data. We propose to estimate the structural changes of these graphs through a three-step procedure. First, we employ a kernel-smoothing approach to estimate multiple graphs at different ages simultaneously. Secondly, we summarize graphical information, such as the number of edges, global and local efficiency, for each estimated graph, and align them as a curve. Lastly, we propose a robust least-absolute-deviation (LAD) type penalization procedure with the fused Lasso (FL) penalty, named LAD-FL, to infer the change-points in those graph summary metrics. Our method is theoretically well understood, and results show that it could effectively capture the brain evolvement pattern.