TY - JOUR
T1 - Two step Gaussian mixture model approach to characterize white matter disease based on distributional changes
AU - Kim, Namhee
AU - Heo, Moonseong
AU - Fleysher, Roman
AU - Branch, Craig A.
AU - Lipton, Michael L.
N1 - Publisher Copyright:
© 2016 Elsevier B.V.
PY - 2016/9/1
Y1 - 2016/9/1
N2 - Background Magnetic resonance imaging reveals macro- and microstructural correlates of neurodegeneration, which are often assessed using voxel-by-voxel t-tests for comparing mean image intensities measured by fractional anisotropy (FA) between cases and controls or regression analysis for associating mean intensity with putative risk factors. This analytic strategy focusing on mean intensity in individual voxels, however, fails to account for change in distribution of image intensities due to disease. New method We propose a method that aims to facilitate simple and clear characterization of underlying distribution. Our method consists of two steps: subject-level (Step 1) and group-level or a specific risk-level density function estimation across subjects (Step 2). Results The proposed method was demonstrated with a simulated data set and real FA data sets from two white matter tracts, where the proposed method successfully detected any departure of the FA distribution from the normal state by disease: p < 0.001 for simulated data; p = 0.047 for the posterior limb of internal capsule; p = 0.06 for the posterior thalamic radiation. Comparison with existing method(s) The proposed method found significant disease effect (p < 0.001) while conventional 2-group t-test focused only on mean intensity did not (p = 0.61) in a simulation study. While significant age effects were found for each white matter tract from conventional linear model analysis with real FA data, the proposed method further confirmed that aging also triggers distribution-wide change. Conclusion Our proposed method is powerful for detection of risk factors associated with any type of microstructural neurodegenerations with brain imaging data.
AB - Background Magnetic resonance imaging reveals macro- and microstructural correlates of neurodegeneration, which are often assessed using voxel-by-voxel t-tests for comparing mean image intensities measured by fractional anisotropy (FA) between cases and controls or regression analysis for associating mean intensity with putative risk factors. This analytic strategy focusing on mean intensity in individual voxels, however, fails to account for change in distribution of image intensities due to disease. New method We propose a method that aims to facilitate simple and clear characterization of underlying distribution. Our method consists of two steps: subject-level (Step 1) and group-level or a specific risk-level density function estimation across subjects (Step 2). Results The proposed method was demonstrated with a simulated data set and real FA data sets from two white matter tracts, where the proposed method successfully detected any departure of the FA distribution from the normal state by disease: p < 0.001 for simulated data; p = 0.047 for the posterior limb of internal capsule; p = 0.06 for the posterior thalamic radiation. Comparison with existing method(s) The proposed method found significant disease effect (p < 0.001) while conventional 2-group t-test focused only on mean intensity did not (p = 0.61) in a simulation study. While significant age effects were found for each white matter tract from conventional linear model analysis with real FA data, the proposed method further confirmed that aging also triggers distribution-wide change. Conclusion Our proposed method is powerful for detection of risk factors associated with any type of microstructural neurodegenerations with brain imaging data.
KW - Aging effects
KW - Density function estimation
KW - Diffusion tensor imaging
KW - Fractional anisotropy
KW - Gaussian mixture model
UR - http://www.scopus.com/inward/record.url?scp=84977109171&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=84977109171&partnerID=8YFLogxK
U2 - 10.1016/j.jneumeth.2016.04.024
DO - 10.1016/j.jneumeth.2016.04.024
M3 - Article
C2 - 27139737
AN - SCOPUS:84977109171
SN - 0165-0270
VL - 270
SP - 156
EP - 164
JO - Journal of Neuroscience Methods
JF - Journal of Neuroscience Methods
ER -