Figure 1.
Histograms of EDSS scores, age (in years), sex, and duration from disease onset (in years) for the 98 patients in the study.
Table 1.
Descriptive statistics of the demographic information on the patients.
Figure 2.
Review of the steps behind the five algorithms.
Figure 3.
One slice from five different registration methods for three subjects (one subject on each row).
The MNI template brain is shown on the first row (left).
Figure 4.
The histogram of brain lesions for 98 patients based on a rigid registration of the images to a template brain indicating the number of patients out of 98 having lesions at each voxel.
Figure 5.
The lesion histograms for 98 patients (showing the number of patients out of 98 having lesions at each voxel) based on “ANTS affine” (top left), “ANTS diffeo” (top right), “FSL nonlinear” (bottom left) and “DARTEL” (bottom right) spatial registration algorithms.
Red: voxels where more patients have lesions; Blue and light blue: voxels where fewer patients have lesions.
Figure 6.
GLSR: the proportion of lesion voxels for each registration algorithm that are not in the white matter in the template space.
Each dot (left) is the GLSR for one subject for a particular registration algorithm: “rigid” (black), “ANTS affine” (red), “ANTS diffeo” (blue), “FSL nonlinear” (green) and “DARTEL” (purple). Higher and more variable across subjects is worse. The size of the dots is proportional to the TLV for each patient. Larger dots correspond to higher TLV. The beanplots (right) show the distribution of GLSR for each registration algorithm.
Table 2.
Properties of the GLSR statistic computed for each of the registration algorithms (higher is worse).
Figure 7.
P-values (uncorrected) for testing using models
–
.
From left to right: spatial registrations “rigid”, “ANTS affine”, “ANTS diffeo”, “FSL nonlinear”, and “DARTEL”. Bright red: p-values close to 0 to black: p-values close to 1. The p-value maps are overlaid on a grayscale template.
Figure 8.
P-values for testing using models
–
after applying Bonferroni and FDR corrections.
From left to right: spatial registrations “rigid”, “ANTS affine”, “ANTS diffeo”, “FSL nonlinear”, and “DARTEL”. Red: small p-values ().
Figure 9.
P-values (uncorrected) for testing using ordinal regression models with independent variables as in models
–
.
From left to right: spatial registrations “rigid”, “ANTS affine”, “ANTS diffeo”, “FSL nonlinear”, and “DARTEL”. Bright red: p-values close to 0 to black: p-values close to 1. The p-value maps are overlaid on a grayscale template.