Home Computer Science Computational Diffusion MRI: MICCAI Workshop, Athens, Greece, October 2016
Co-localization ofNeurite Density and Functional Activity
We used the same approach to map the z-scores stemming from BOLD fMRI data of each individual onto its cortical surface. The z-scores were computed voxel-wise and integrated at the scale of the cortical region. So, at the end of the two pipelines, we get for each subject and for each vertex of the pial mesh, the integral of the neurite density over the cortical thickness, as well as, for each functional contrast, the integral of its z-scores over the same cortical thickness. A Gaussian smoothing was then applied onto the pial surface in order to enhance the peaks of neurite density and the peaks of functional activity. All the individual surface maps were finally averaged together to better visualize the colocalization of those peaks at the group level (see Fig. 3).
In order to characterize the overlap between the fi„tra and z-score maps, we chose the S0rensen-Dice coefficient defined by the Eq. (2).
Fig. 3 Group level average z-scores and fintra maps
For the computation of the binary textures A and B in the Eq. (2), we took the individual smoothed maps offi„tra and z-scores for each contrast. For the fi„tra, we set the threshold to half of the standard deviation (ie, threshold = mean — 0.5 * stddev) and for the z-scores, as it can be negative, we limited the threshold to one standard deviation (ie, threshold = mean + stddev). The two maps have their own range of values and we had to adapt each threshold independently: it has been done empirically but would benefit from an automatized process in the future. Then, we used Eq. (2) to compute the Dice index for each cortical area defined by the Destrieux’s parcellation at the subject level.
|< Prev||CONTENTS||Next >|