Network analysis of brain connectivity, or Connectomics, has emerged as an important interdisciplinary field, making strides in advancing both fundamental scientific knowledge on the structure of the brain, as well as providing insights into the pathology of neurological disorders. Advances in neuroimaging technologies have enabled acquisition of high-resolution datasets on brain activities in normal and diseased populations, while advanced machine learning methods hold promise to obtain data-driven insights into the functional architecture of the brain. Functional connectivity analysis is typically carried out in two steps. First, one estimates a network among different brain regions by studying strengths of associations among the time course of neurophysiological signals for different subjects (patients or healthy controls). Next, one compares the networks between different groups of subjects and seeks network features prevalent in specific groups of interest using statistical methods. Two emerging challenges in this field are the presence of heterogeneity amongst subjects in large study cohorts, and developing predictive models to construct robust and interpretable results. The central goal of this proposal is to address these challenges by developing machine learning methods equipped with uncertainty quantification measures, suitable for high-dimensional network data for heterogeneous populations. Upon completion, these methods are expected to provide automated, robust and more accurate discovery of connectivity patterns that is prevalent in heterogeneous populations of patients.
We aim to accomplish this goal by pursuing two specific aims: (1) develop estimation and inference methods for frequency domain measures of high-dimensional functional connectivity networks, (2) develop a framework of mixed effects model of high-dimensional functional connectivity networks that accounts for heterogeneity among subjects and enables discoveries more likely to generalize in large cohorts. For each aim, novel machine learning methods for integrative analysis of structural and functional connectivity will be developed using a mathematical model of network diffusion. We will also calibrate and validate our proposed methods on data from Human Connectome Project, and on multiple sclerosis (MS) patients. The proposed approach is innovative since it integrates machinery across diverse disciplines, including statistics, machine learning and network analysis to address important challenges learning large functional connectivity graphs. The proposed research is significant in that it is expected to have both scientific and translational impact.

Public Health Relevance

The proposed research is relevant to public health because understanding brain connectivity patterns using rigorous machine learning methods can help seek common patterns in large databases in an automated fashion, and can potentially lead to development of neuroimaging based diagnostic tools and sensitive prognostic measures. The proposed research will thus use fundamental scientific knowledge about brain to assist translational research in neurological disease.

National Institute of Health (NIH)
National Institute of Neurological Disorders and Stroke (NINDS)
Exploratory/Developmental Grants (R21)
Project #
Application #
Study Section
Biostatistical Methods and Research Design Study Section (BMRD)
Program Officer
Babcock, Debra J
Project Start
Project End
Budget Start
Budget End
Support Year
Fiscal Year
Total Cost
Indirect Cost
Cornell University
Biostatistics & Other Math Sci
Earth Sciences/Resources
United States
Zip Code