Quantitative diffusion tensor imaging in amyotrophic lateral sclerosis: revisited

Abstract

Voxel-based analyses (VBA) are increasingly being used to detect white matter abnormalities with diffusion tensor imaging (DTI) in different types of pathologies. However, the validity, specificity, and sensitivity of statistical inferences of group differences to a large extent depend on the quality of the spatial normalization of the DTI images. Using high-dimensional nonrigid coregistration techniques that are able to align both the spatial and orientational diffusion information and incorporate appropriate templates that contain this complete DT information may improve this quality. Alternatively, a hybrid technique such as tract-based spatial statistics (TBSS) may improve the reliability of the statistical results by generating voxel-wise statistics without the need for perfect image alignment and spatial smoothing. In this study, we have used (1) a coregistration algorithm that was optimized for coregistration of DTI data and (2) a population-based DTI atlas to reanalyze our previously published VBA, which compared the fractional anisotropy and mean diffusivity maps of patients with amyotrophic lateral sclerosis (ALS) with those of healthy controls. Additionally, we performed a complementary TBSS analysis to improve our understanding and interpretation of the VBA results. We demonstrate that, as the overall variance of the diffusion properties is lowered after normalizing the DTI data with such recently developed techniques (VBA using our own optimized high-dimensional nonrigid coregistration and TBSS), more reliable voxel-wise statistical results can be obtained than had previously been possible, with our VBA and TBSS yielding very similar results. This study provides support for the view of ALS as a multisystem disease, in which the entire frontotemporal lobe is implicated.

Figure 1

Flow chart of the different steps of the data processing for the original and the new voxel‐based analysis and the tract‐based spatial statistics analysis. FA, fractional anisotropy; MD, mean diffusivity; FSL, FMRIB Software Library; FNIRT, FMRIB's Nonlinear Registration Tool; S(n)PM, Statistical (non‐) Parametric Mapping; MNI, Montreal Neurological Institute; VBA, voxel‐based analysis; TBSS, tract‐based spatial statistics.

Figure 2

Two templates were used for spatial normalization of the FA and MD maps. In our original approach, we registered the subject's high‐resolution anatomical T1‐weighted image to the T1‐weighted Montreal Neurological Institute template (MNI) (A), of which a midcoronal (A, upper left), midsagittal (A, upper right), and an axial slice (A, lower left) are shown. The parameters of this transformation were subsequently applied to the FA and MD maps to bring these into MNI space. For this study, we built a population‐based diffusion tensor atlas from the DTI data of all subjects included in the study (B). This atlas was used as a template for registration of the FA and MD maps for both the new voxel‐based and the tract‐based spatial statistics analysis. A midcoronal (B, upper left), midsagittal (B, upper right), and an axial (B, lower left) slice of the DTI‐based color maps are shown (colors are according to the eigenvector associated with the largest eigenvalue, with red = left–right, green = anterior–posterior, and blue = inferior–superior and intensities are scaled in proportion to the magnitude of FA). As this atlas contains full diffusion tensor information, this information can be used to drive the coregistration process and it even allows performing whole‐brain fiber tractography [Jones et al., 2002] (B, lower right).

Figure 3

Maps of the coefficient of variance (COV; COV = standard deviation/mean) for FA (A) and MD (B), in which the color‐coded COV values within the applied WM mask were overlaid on coronal, axial and sagittal slices of the mean FA map for the old analysis (A/B, upper part of figure) and on the FA map of the population‐based DT atlas for the new analysis (A/B, lower part of figure). The color bar indicates the colors corresponding to values of the COV between 0 and 1.

Figure 4

Empirical cumulative distribution functions of the coefficient of variance (COV; COV = standard deviation/mean), determined on the warped FA (A), MD (B), L1 (C), L2 (E), L3 (F), and transverse diffusivity (LTR) (D) maps used for the original (full lines) and new (dotted lines) voxel‐based analysis. The cumulative distribution function represents the value of COV to the percentage of the distribution with values less than or equal to that value. For the new analysis, the CDFs of the COV are shifted leftward for all variables, which indicates reduction of residual variance after coregistration to the population‐based diffusion tensor atlas.

Figure 5

Results of the voxel‐based analysis (VBA), comparing the FA maps of ALS patients and controls that were generated for the original (A) and new (B) VBA and results of the comparison of the FA skeletons that were generated using tract‐based spatial statistics (TBSS) (C). WM regions in which a significant FA reduction could be demonstrated for ALS patients (n = 28) compared with controls (n = 26) using nonparametric statistics are shown in red on coronal/axial/sagittal slices of the mean FA map for the original VBA, of the FA map of the DTI atlas for the new VBA, and of the mean FA map of the TBSS analysis. Note that the results of the TBSS analysis have been enhanced for visualization purposes. A similar pattern of FA reductions in ALS patients that included both motor and extramotor WM areas was demonstrated. (E) In our original analysis, we assessed the WM integrity of the corticospinal tract (CST) by spatially interpolating the fiber tractography reconstructions of the corticospinal tracts, which allowed plotting the mean FA values (bold lines; thin lines indicate +/− standard error of the mean) over the entire cranio‐caudal course of the CST. This analysis showed that FA was significantly reduced in ALS patients (green) compared with controls (blue) at the level of the posterior limb of the internal capsule and the subcortical WM (indicated by double arrows and a red asterisk), which was also demonstrated in the original and new VBA, as well as in the TBSS analysis. A reconstructed mean CST (D) with DTI based color‐encoding overlaid on an anatomical image (flipped to match the x‐axes of the graphs) is shown for anatomical reference, together with a vertical bar in white indicating the anatomical structures belonging to the CST by following abbreviations: MO, medulla oblongata; P, pons; CP, cerebral peduncle; PLIC, posterior limb of the internal capsule; CR, corona radiata; SWM, subcortical WM.

Figure 6

Results of the voxel‐based analysis (VBA), comparing the MD maps of patients with ALS and controls, generated for the original (A) and new (B) analysis and results of the comparison of the MD skeletons that were generated using tract‐based spatial statistics (TBSS) (C). WM regions in which a significant increase of MD could be demonstrated in patients with ALS compared with controls are shown in blue on coronal/axial/sagittal slices of the mean FA map for the original VBA, of the FA map of the DTI atlas for the new VBA and of the mean FA map of the TBSS analysis. Note that the results of the TBSS analysis have been enhanced for visualization purposes. In the original VBA, only a very limited number of clusters showed this effect, whereas in the new VBA and TBSS analysis, a more extensive pattern of significant increase in MD could be demonstrated. (E) In our original study, interpolation of the tract data of the reconstructions of the corticospinal tract (CST) showed that MD was significantly increased in patients with ALS (green) compared with controls (blue) in the cranial parts of the CST. In contradiction, no significant changes in MD within the CST were found in our original VBA. The new VBA and TBSS analysis show more congruent results, as significant MD changes in patients with ALS were found from the level of the internal capsule up to the level of the corona radiata. A reconstructed mean CST (C) with DTI based color‐encoding overlaid on an anatomical image (flipped to match the x‐axes of the graphs) is shown for anatomical reference, together with a vertical bar in white indicating the anatomical structures belonging to the CST (same abbreviations as in Fig. 5).

Figure 7

Regions showing significant positive correlation between FA values and the patients' scores on the ALS functional rating scale are shown in yellow on coronal/axial/sagittal slices of the mean FA map for the old voxel‐based analysis (A), of the FA map of the DTI atlas for the new voxel‐based analysis (B), and of the mean FA map of the tract‐based spatial statistics analysis (C). Note that the results of the TBSS analysis have been enhanced for visualization purposes. In all analyses, this correlation was present throughout the WM.

Figure 8

Regions showing significant negative correlation between MD values and the patients' scores on the ALS functional rating scale are shown in green on coronal/axial/sagittal slices of the mean FA map for the old voxel‐based analysis (VBA) (A), of the FA map of the DTI atlas for the new VBA (B) and of the mean FA map of the tract‐based spatial statistics analysis (C). Note that the results of the TBSS analysis have been enhanced for visualization purposes. No such correlation could be demonstrated in our original VBA, whereas in the new VBA and TBSS analysis, this correlation was clearly present.