Table 1.
Anatomical changes in the sixteen subjects used as the multiple atlases.
Table 2.
Anatomical changes in 25 subjects for testing the multiple atlases application.
Figure 1.
Demonstration of the unique anatomical features revealed by multi-contrast images generated in DTI and GMM.
Histograms of the five contrasts, FA, MD, EV-x, EV-y, and EV-z, of five adjacent structures are shown, including two white matter structures (the ALIC and the PLIC), two gray matter structures (the caudate and the thalamus), and the ventricle. In each subplot, blue indicates the histogram of the corresponding contrast within that specific structure, green represents the probability density of each single Gaussian, and red shows the weighted sum of all Gaussians. Abbreviations are: ALIC: Anterior limb of internal capsule and PLIC: Posterior limb of internal capsule.
Figure 2.
A comparison of parcellating using a single-contrast image and multi-contrast images, in terms of overlap accuracy.
The mean Dice overlaps and the standard deviations of the eighteen ROIs obtained from automated parcellations based on five contrasts (red), the single FA contrast (green), the single MD contrast (blue), the vector x y z contrasts (yellow), as well as the inter-rate (patterned). The mean values are calculated across fourteen different subjects. Star marks indicate significant difference among the four sets of Dice results by ANOVA (p<<0.05). Abbreviations are: GCC – genu of corpus callosum; BCC – body of corpus callosum; Caud – caudate; Put – putamen; ALIC – anterior limb of internal capsule; PLIC – posterior limb of internal capsule; CG – cingulate gyrus; MCP – middle cerebellar peduncle; SLF – superior longitudinal fasciculus; CST – corticospinal tract.
Figure 3.
A correlation comparison of the automated caudate parcellation from using a single-contrast image and multi-contrast images.
A plot of the correlation between the automated and the manual measurements of the size of the caudate in both hemispheres in square millimeters. Results from the four automated parcellation methods are compared: 5-contrast (red), FA-only (green), MD-only (blue), and EV-only (yellow).
Figure 4.
A comparison of the CST correlation obtained from using a single-contrast image and multi-contrast images.
A correlation plot between the automated and manual measurements of the sizes of left and right corticospinal tracts (CST). Results from the four automated parcellation methods are compared: 5-contrast (red), FA-only (green), MD-only (blue), and EV-only (yellow).
Table 3.
Pearson correlations between manual and four different automated ROI area measures, with bold typesetting indicating that the correlation between the automated and the manual measures is statistically significantly stronger than other methods (p<0.025).
Figure 5.
Examples of CST parcellations from single- and multi-contrast approaches.
Demonstration of the parcellation accuracy of the CST in three representative cases with different degrees of anatomical abnormalities. Results from five different approaches are compared.
Table 4.
Pearson correlations between the mean FA value within the manual ROI and the automated parcellations, with bold typesetting suggesting that the correlation between the automated and the manual measures is statistically significantly stronger than other methods (p<0.025).
Table 5.
Pearson correlations between the mean diffusivity (MD) value within the manual ROI and the automated parcellations, with bold typesetting indicating that the correlation between the automated and the manual measures is statistically significantly stronger other methods (p<0.025).
Figure 6.
Examples of whole brain parcellations.
Results of the whole brain parcellations into 159 structures in three representative cases with large anatomical variability. The parcellation results are superimposed on color (upper row) and MD (bottom row) images.
Figure 7.
Depiction of the cross-subject variability in different whole brain structures of interest.
Demonstration of the cross-subject variability (std/mean) within the 16 healthy subjects for each of the 193 anatomical regions. The population variability in terms of ROI volume (top row), the mean MD of each ROI (middle row), and the mean FA of each ROI (bottom row) are shown. The results are presented for three difference axial slices using a colormap with the color scale ranging from 0 to 0.5.