TY - JOUR
T1 - Joint modelling of diffusion MRI and microscopy
AU - Howard, Amy FD
AU - Mollink, Jeroen
AU - Kleinnijenhuis, Michiel
AU - Pallebage-Gamarallage, Menuka
AU - Bastiani, Matteo
AU - Cottaar, Michiel
AU - Miller, Karla L.
AU - Jbabdi, Saad
N1 - Publisher Copyright:
© 2019 The Authors
PY - 2019/11/1
Y1 - 2019/11/1
N2 - The combination of diffusion MRI (dMRI) with microscopy provides unique opportunities to study microstructural features of tissue, particularly when acquired in the same sample. Microscopy is frequently used to validate dMRI microstructure models, addressing the indirect nature of dMRI signals. Typically, these modalities are analysed separately, and microscopy is taken as a gold standard against which dMRI-derived parameters are validated. Here we propose an alternative approach in which we combine dMRI and microscopy data obtained from the same tissue sample to drive a single, joint model. This simultaneous analysis allows us to take advantage of the breadth of information provided by complementary data acquired from different modalities. By applying this framework to a spherical-deconvolution analysis, we are able to overcome a known degeneracy between fibre dispersion and radial diffusion. Spherical-deconvolution based approaches typically estimate a global fibre response function to determine the fibre orientation distribution in each voxel. However, the assumption of a ‘brain-wide’ fibre response function may be challenged if the diffusion characteristics of white matter vary across the brain. Using a generative joint dMRI-histology model, we demonstrate that the fibre response function is dependent on local anatomy, and that current spherical-deconvolution based models may be overestimating dispersion and underestimating the number of distinct fibre populations per voxel.
AB - The combination of diffusion MRI (dMRI) with microscopy provides unique opportunities to study microstructural features of tissue, particularly when acquired in the same sample. Microscopy is frequently used to validate dMRI microstructure models, addressing the indirect nature of dMRI signals. Typically, these modalities are analysed separately, and microscopy is taken as a gold standard against which dMRI-derived parameters are validated. Here we propose an alternative approach in which we combine dMRI and microscopy data obtained from the same tissue sample to drive a single, joint model. This simultaneous analysis allows us to take advantage of the breadth of information provided by complementary data acquired from different modalities. By applying this framework to a spherical-deconvolution analysis, we are able to overcome a known degeneracy between fibre dispersion and radial diffusion. Spherical-deconvolution based approaches typically estimate a global fibre response function to determine the fibre orientation distribution in each voxel. However, the assumption of a ‘brain-wide’ fibre response function may be challenged if the diffusion characteristics of white matter vary across the brain. Using a generative joint dMRI-histology model, we demonstrate that the fibre response function is dependent on local anatomy, and that current spherical-deconvolution based models may be overestimating dispersion and underestimating the number of distinct fibre populations per voxel.
KW - Diffusion MRI
KW - Fibre response function
KW - Histology
KW - Orientation dispersion
KW - White matter
UR - http://www.scopus.com/inward/record.url?scp=85070922780&partnerID=8YFLogxK
U2 - 10.1016/j.neuroimage.2019.116014
DO - 10.1016/j.neuroimage.2019.116014
M3 - Article
C2 - 31315062
AN - SCOPUS:85070922780
SN - 1053-8119
VL - 201
JO - NeuroImage
JF - NeuroImage
M1 - 116014
ER -