Analysis of spatio-temporal brain imaging patterns by hidden markov models and serial MRI images

Authors

  • Ying Wang,

    Corresponding author
    1. Section of Biomedical Image Analysis, Department of Radiology, University of Pennsylvania, Philadelphia, Pennsylvania
    • Correspondence to: Dr. Ying Wang; Section of Biomedical Image Analysis, Department of Radiology, University of Pennsylvania, Philadelphia, PA 19104. E-mail: ying.wang@uphs.upenn.edu

    Search for more papers by this author
  • Susan M. Resnick,

    1. Laboratory of Behavioral Neuroscience, National Institute on Aging, Baltimore, Maryland
    Search for more papers by this author
  • Christos Davatzikos,

    1. Section of Biomedical Image Analysis, Department of Radiology, University of Pennsylvania, Philadelphia, Pennsylvania
    Search for more papers by this author
  • the Baltimore Longitudinal Study of Aging and the Alzheimer's Disease Neuroimaging Initiative


  • Part data used in the preparation of this article was obtained from the Baltimore Longitudinal Study of Aging (BLSA), and part data were from the Alzheimer's Disease Neuroimaging Initiative (ADNI) database. The investigators within the ADNI contributed to the design and implementation of ADNI and/or provided data but did not participate in analysis or writing of this article.

  • We have no conflict of interest to declare.

Abstract

Brain changes due to development and maturation, normal aging, or degenerative disease are continuous, gradual, and variable across individuals. To quantify the individual progression of brain changes, we propose a spatio-temporal methodology based on Hidden Markov Models (HMM), and apply it on four-dimensional structural brain magnetic resonance imaging series of older individuals. First, regional brain features are extracted in order to reduce image dimensionality. This process is guided by the objective of the study or the specific imaging patterns whose progression is of interest, for example, the evaluation of Alzheimer-like patterns of brain change in normal individuals. These regional features are used in conjunction with HMMs, which aim to measure the dynamic association between brain structure changes and progressive stages of disease over time. A bagging framework is used to obtain models with good generalization capability, since in practice the number of serial scans is limited. An application of the proposed methodology was to detect individuals with the risk of developing MCI, and therefore it was tested on modeling the progression of brain atrophy patterns in older adults. With HMM models, the state-transition paths corresponding to longitudinal brain changes were constructed from two completely independent datasets, the Alzheimer Disease Neuroimaging Initiative and the Baltimore Longitudinal Study of Aging. The statistical analysis of HMM-state paths among the normal, progressive MCI, and MCI groups indicates that, HMM-state index 1 is likely to be a predictor of the conversion from cognitively normal to MCI, potentially many years before clinical symptoms become measurable. Hum Brain Mapp 35:4777–4794, 2014. © 2014 Wiley Periodicals, Inc.

Ancillary