Multiscale Adaptive Marginal Analysis of Longitudinal Neuroimaging Data with Time-Varying Covariates

Authors

  • Martha Skup,

    Corresponding author
    1. Division of Biostatistics, Yale University, New Haven, Connecticut, U.S.A.
      email: martha.skup@yale.edu
    Search for more papers by this author
  • Hongtu Zhu,

    Corresponding author
    1. Department of Biostatistics, University of North Carolina at Chapel Hill, Chapel Hill, North Carolina, U.S.A.
    2. Biomedical Research Imaging Center, University of North Carolina at Chapel Hill, Chapel Hill, North Carolina, U.S.A.
      email: hzhu@bios.unc.edu
    Search for more papers by this author
  • Heping Zhang

    Corresponding author
    1. Division of Biostatistics, Yale University, New Haven, Connecticut, U.S.A.
      email: heping.zhang@yale.edu
    Search for more papers by this author

email:martha.skup@yale.edu

email:hzhu@bios.unc.edu

email:heping.zhang@yale.edu

Abstract

Summary Neuroimaging data collected at repeated occasions are gaining increasing attention in the neuroimaging community due to their potential in answering questions regarding brain development, aging, and neurodegeneration. These datasets are large and complicated, characterized by the intricate spatial dependence structure of each response image, multiple response images per subject, and covariates that may vary with time. We propose a multiscale adaptive generalized method of moments (MA-GMM) approach to estimate marginal regression models for imaging datasets that contain time-varying, spatially related responses and some time-varying covariates. Our method categorizes covariates into types to determine the valid moment conditions to combine during estimation. Further, instead of assuming independence of voxels (the components that make up each subject’s response image at each time point) as many current neuroimaging analysis techniques do, this method “adaptively smoothes” neuroimaging response data, computing parameter estimates by iteratively building spheres around each voxel and combining observations within the spheres with weights. MA-GMM’s development adds to the few available modeling approaches intended for longitudinal imaging data analysis. Simulation studies and an analysis of a real longitudinal imaging dataset from the Alzheimer’s Disease Neuroimaging Initiative are used to assess the performance of MA-GMM. Martha Skup, Hongtu Zhu, and Heping Zhang for the Alzheimer’s Disease Neuroimaging Initiative.

Ancillary