Correlation of cognitive dysfunction and diffusion tensor MRI measures in patients with mild and moderate multiple sclerosis

Authors


Abstract

Purpose

To compare the diffusion tensor imaging (DTI) measures of multiple sclerosis (MS) patients and healthy subjects in every brain voxel and to correlate them with Paced Auditory Serial Addition Test (PASAT) scores.

Materials and Methods

Fractional anisotropy (FA), and mean, longitudinal, and transverse diffusivity are compared between control subjects and MS patients, which were subdivided as mildly and moderately impaired. In addition, PASAT scores are correlated for both MS groups with the diffusion measures. An optimized voxel based analysis (VBA) method, in terms of coregistration, atlas construction, and image smoothing, was thereby used.

Results:

Diffusion differences between the control subjects and the patients with MS were found in the corpus callosum, inferior longitudinal fasciculus, cortico spinal tracts, forceps major, superior longitudinal fasciculus, and cingulum. In addition, we observed significant correlations of the FA and PASAT scores in the left inferior longitudinal fasciculus, the forceps minor, the capsula interna and externa, the genu of the corpus callosum, the left cingulum, the superior longitudinal fasciculus, and the corona radiata.

Conclusion:

Diffusion differences were observed between the mildly impaired MS patients and control subjects. In addition, different diffusion measures correlated with PASAT scores for cognitive decline in parietal, frontal, as well as temporal white matter (WM) regions. J. Magn. Reson. Imaging 2010;31:1492–1498. © 2010 Wiley-Liss, Inc.

ALTHOUGH IT HAS BEEN demonstrated that conventional magnetic resonance (MR) images are sensitive for detecting multiple sclerosis (MS) plaques, the T2 lesions reflect the clinical outcome only to a limited extent (1). One of the major drawbacks of conventional T2 lesion imaging in MS is the lack of pathophysiological specificity if this technique. T1 hypointense lesions are more specific, and correlate with axonal density in MS. Newer MRI techniques, such as diffusion tensor imaging (DTI), may show more subtle pathological changes in brain tissue that is normal-appearing on conventional images, which do show white matter (WM) lesions and atrophy (1). DTI provides in vivo and non-invasive information about the orientation and integrity of WM fiber bundles. Diffusion measures that are often used to quantify WM damage include the fractional anisotropy (FA), which is a normalized measure of the diffusion anisotropy, the mean diffusivity (MD), which is the average diffusion in a voxel and the longitudinal and the transverse diffusivities L1 and L23, which measure the diffusion along and perpendicular to the WM bundle (2).

It has been reported in other studies that cognitive dysfunction in MS patients might be related to WM lesions, normal appearing brain tissue on conventional MRI, and cortical as well as deep gray matter (3). Rao et al (3) suggested that cognitive impairment is induced by a disruption of the cortico-subcortical circuits, connecting the frontal cortices to thalamus and basal ganglia. However, other studies reported that posterior brain regions and the corpus callosum might also play a role in the cognitive malfunctioning (4). In addition, it has been suggested that a slowing of information processing speed might be related to sensory-motor damage. Since DTI provides measures of WM integrity, it is a relevant technique to investigate the correlation of WM microstructural properties with cognitive function in MS patients (5, 6). Rovaris et al (7) examined the relationship between DTI and cognition in relapsing-remitting MS patients using a whole brain histogram analysis. They observed moderate correlations between the MD and neuro-psychological test scores that measured memory, speed of information processing, and verbal fluency. Lin et al (8) found that the MD of the corpus callosum correlated with the Paced Auditory Serial Addition Test (PASAT) score. PASAT is a commonly used experimental paradigm to evaluate sustained attention, working memory and speed of information processing in MS. More recently, Dineen et al (5) examined the whole brain for correlations of FA and cognitive dysfunction. They detected significant correlations of PASAT and FA in the body and splenium of the corpus callosum, the forceps major, the left cingulum, the right inferior longitudinal fasciculus, the left superior longitudinal fasciculus, and the arcuate fasciculus. In their study, Mesaros et al (4) observed correlations of PASAT and FA in the corpus callosum.

The aim of this study was to examine differences of diffusion measures between healthy subjects on the one hand and mildly and moderately affected MS patients on the other hand. In addition, the relationship between the PASAT tests of cognitive decline and micro-structural WM breakdown, as assessed by the DTI measures, was studied in an automated whole brain analysis. To this end, an optimized voxel based analysis (VBA) approach, in terms of coregistration, atlas construction, and smoothing, was used to compare the diffusion properties of all subjects in every brain voxel and to correlate them with PASAT scores.

METHODS

Subjects

A total of 20 patients with definite MS according to the McDonald criteria were included in this study (9). Enrolled subjects did not have a relapse for at least 30 days before entry into the study, did not use sedatives, and had a visual acuity above 20–40, as measured on a Snellen chart. Ten patients with an expanded disability status scale (EDSS) between 0 and 3, referred to as MS group 1, and 10 patients with an EDSS between 4 and 7, referred to as MS group 2, were selected (10). MS patient group 1 contained nine relapse-remitting (RR) MS patients and one secondary-progressive (SP) patient, whereas four RR and six SP MS patients were included in group 2. A control group of 10 healthy volunteers was matched to both patient groups for age, gender and educational level. Volunteers using medication, having a first or second degree relative with MS, or having visual impairment were excluded. The demographics, educational level, EDSS, and PASAT scores for the three subject groups are presented in Table 1. All subjects were right handed. The study was approved by the hospital ethics committee and all subjects gave written informed consent before entering the study.

Table 1. Study Information
 ControlsMS1MS2
  1. MS = Multiple Sclerosis, RR = Relapse-Remitting, SP = Secondary-Progressive, f = female, m = male, PASAT = Paced Auditory Serial Addition Test, EDSS: Expanded Disability Status Scale.

Number of subjects101010
Gender (f/m)5/55/54/6
MS type (RR/SP)na9/14/6
Age (year, mean ± sd)42 ± 1043 ± 941 ± 7
Education (year, mean ± sd)14 ± 214 ± 213 ± 2
PASAT (mean ± sd)53 ± 549 ± 946 ± 13
EDSS (mean ± sd)na2 ± 16 ± 1
Disease duration (mean ± sd)na12 ± 711 ± 5

Image Acquisition

DTI data sets were obtained on a 1.5 T MR scanner using an spin echo (SE) echo planar imaging (EPI), sequence with the following acquisition parameters: TR = 10.4 seconds; TE = 100 msec; diffusion gradient = 40 mT.m−1; field of view (FOV) = 256 × 256 mm2; number of slices = 60; voxel size = 2 × 2 × 2 mm3; b = 700 seconds.mm−2; acquisition time = 12 minutes 18 seconds. Diffusion measurements were performed along 60 directions with 10 b0-images.

Image Processing for VBA

DTI data sets were processed as follows:

  • All DTI data sets are transformed to a FA template in Montreal Neurological Institute (MNI) space with an affine transformation using Multimodality Image Registration using Information Theory (MIRIT) based on the FA maps (11).

  • A population specific DTI atlas was constructed from these affinely aligned data sets (12).

  • The affinely coregistered data sets were transformed to the population specific atlas using a high-dimensional non-rigid coregistration algorithm that was adopted to include all tensor information during the iterative alignment procedure (13, 14). The preservation of principal direction (PPD) tensor reorientation strategy was thereby incorporated.

  • The resulting images were smoothed with an adaptive, anisotropic smoothing kernel (full-width at half-maximum [FWHM] = 3 mm) (15, 16). This spatially dependent, anisotropic kernel was estimated from the FA maps and subsequently applied to the FA, L1, L23, and MD images.

After this pre-processing of the data sets, two analyses were performed:

  • Analysis 1: the diffusion properties of the different subject groups, i.e., the control group, MS patient group 1, and MS patient group 2, are compared using a non-parametric Kruskall-Wallis analysis in each voxel. A correction for multiple comparisons based on the false discovery rate (FDR) (FDR threshold of 0.05) was thereby applied to account for the multiple testing. In order to verify the specific differences between the various groups, Mann-Whitney U-tests were subsequently applied between the control group and the MS patient group 1, the control group and the MS patient group 2, and between both MS patient groups, including the aforementioned FDR based correction for multiple comparisons.

  • Analysis 2: Non-parametric Spearman correlation tests are performed in each voxel to quantify the relation between the different diffusion properties and the PASAT score. Again, an FDR based correction for multiple comparisons was performed.

RESULTS

The three subject groups did not significantly differ in age, gender distribution, disease duration, or educational level (Table 1). MS group 2 contained significantly more SP MS patients than RR MS patients, compared to MS group 1 (Chi square test, P < 0.05). PASAT scores were significantly lower in the MS group 2 compared to the control group (see Fig. 1, P < 0.05 in a Kruskall-Wallis test). A post hoc comparison showed a significantly lower number of correct answers for MS patient group 2, compared to controls (P < 0.05). There was no significant difference between the control group and the MS patient group 1 (P = 0.484), nor between both patient groups (P = 0.239) in this post hoc comparison.

Figure 1.

Results of PASAT test for the control group (green), the MS patient group with an EDSS score between 0 and 3 (orange) and the MS patient group with an EDSS score between 4 and 7 (red).

In Fig. 2, the significant voxels of the Kruskall-Wallis test are colored in white and superimposed on different axial slices of the atlas FA maps that were color-encoded for the diffusion direction. As can be seen in Fig. 2a, a lower FA in the MS patients compared to the control subjects is observed in the inferior longitudinal fasciculus, the capsula externa, and the forceps major. L1 decreases in the MS subjects are found in the inferior longitudinal fasciculus, the capsula interna, the body of the corpus callosum, and the corona radiata (see Fig. 2b). As shown in Fig. 2c and d, the L23 and MD are significantly increased in the MS patients compared to the control subjects in the inferior longitudinal fasciculus, the capsula interna and externa, genu, body, and splenium of the corpus callosum, the forceps major, and the corona radiata. After applying Mann-Whithney U tests to examine the specific group differences, no significant voxels were found when comparing the control group and the MS patient group 1 or when comparing both MS patient groups. The differences that were observed in the Kruskall-Wallis analysis thus originate from differences between the control group and MS patient group 2, as can be observed in Fig. 3a, b, c, and d.

Figure 2.

VBA results of the Kruskall-Wallis group analysis in which different diffusion properties, i.e., FA (a), L1 (b), L23 (c), and MD (d) are compared between the control group and both MS patient groups. A false discovery rate threshold of 0.05 was applied to select the significant voxels. [Color figure can be viewed in the online issue, which is available at www.interscience.wiley.com.]

Figure 3.

VBA of Mann-Whitney U tests that compare the FA (a), L1 (b), L23 (c), and MD (d) between the healthy subjects and the moderately impaired MS patients (fasle discovery rate threshold of 0.05). [Color figure can be viewed in the online issue, which is available at www.interscience.wiley.com.]

The voxels in which the cognitive test scores are significantly correlated with the diffusion measures are depicted in white and superimposed on axial slices of the color encoded FA maps in Fig. 4. The correlation results of the PASAT score with the FA, L23, and MD are displayed in Fig. 4a, b, and c, respectively. Significant correlations between PASAT and FA are found in the left inferior longitudinal fasciculus, the forceps minor, the capsula interna and externa, the genu of the corpus callosum, the left cingulum, the superior longitudinal fasciculus, and the corona radiata. As can be seen in Fig. 4b, similar regions contain significant correlations between the PASAT and the L23. Correlations between the PASAT score and the MD were observed in the capsula interna and externa, the superior longitudinal fasciculus, and the corona radiata (see Fig. 4c).

Figure 4.

VBA results of Spearman correlation tests between the FA (a), L23 (b), and MD (c) on the one hand and the PASAT scores on the other hand. A false discovery rate threshold of 0.05 was applied to select the significant voxels. [Color figure can be viewed in the online issue, which is available at www.interscience.wiley.com.]

In Fig. 5, the Spearman correlation coefficients ρ are depicted in the voxels that contain statistically significant correlations. The results are thereby superimposed on the atlas FA map. The correlation coefficients ρ of the PASAT scores with the diffusion measures are shown in Fig. 5a, b, and c.

DISCUSSION

In previous studies, diffusion differences between control subjects and MS patients were detected in the corpus callosum (4–6), the corticospinal or pyramidal tracts (4–6), the frontal WM (17), the forceps major (5, 6), the forceps minor (18), the inferior longitudinal fasciculus (5, 6), the fornix (14, 19), the cingulum (19), the superior longitudinal fasciculus (19), and the uncinate fasciculus (19). Our findings overlap to a certain degree with these results, since differences in the corpus callosum, inferior longitudinal fasciculus, cortico spinal tracts, forceps major, superior longitudinal fasciculus, and cingulum were observed between the control subjects and the patients with MS in the Kruskall-Wallis analysis. Our results furthermore indicated that these group differences were mainly caused by differences between the control subjects and the moderately impaired MS patients. In contrast to the study of Mesaros et al (4), no differences were found between the control subjects and the mildly affected MS patients after the applied post-hoc correction. This might be explained by the restricted statistical power, due to a relatively low number of mildly affected MS patients that were included in this study.

As can be observed in Figs. 2 and 3, especially differences in L23 and MD between the subject groups were detected, as was also observed in (20). Although further studies are needed, recent work suggests that demyelination and axonal degeneration cause an increase of the transverse diffusivity L23 and a decrease of the longitudinal diffusivity L1, respectively (2). However, since no ground-truth about the underlying microstructural damage is known in our population of subjects, it is impossible to correlate the observed changes in the DTI measures to the exact micro-structural pathophysiology. Here, it is important to stress that DTI can be considered as a highly sensitive technique with a low specificity of the underlying changes in WM tissue organization.

The results of our study can be subdivided in three parts: WM structures that contain differences based on EDSS and are correlated with PASAT, WM structures that contain differences based on EDSS but are not correlated with PASAT, and WM structures that contain differences not based on EDSS but are correlated with PASAT.

The first category, i.e., differences based on EDSS and PASAT, contains the inferior longitudinal fasciculus, capsula interna and externa, genu of the corpus callosum, and the corona radiate. This is in agreement with the literature, since Dineen et al (5) demonstrated that FA values in the inferior longitudinal fasciculus correlated with PASAT scores. Capsula interna abnormalities are important for pyramidal and gait function, which links these regions to EDSS score, but were also demonstrated in age associated memory impairment, linking them to cognitive function (21). The corona radiata shows reduced FA values in MS patients compared to normal controls (5). Since the corona radiata contains fibers linking capsula interna to cortical areas, this probably also accounts for a positive correlation between the DTI abnormalities in this structure and both EDSS and PASAT scores. The genu of the corpus callosum is also implicated in other cognitive pathologies. Alzheimer patients have lower FA values in this structure than normal controls (22).

Diffusion differences based on EDSS, but not on PASAT, were observed in the forceps major and the body of the corpus callosum. EDSS score is highly driven by ambulatory function. It has been shown that DTI abnormalities in the corpus callosum correlate with gait function in other study populations, such as the elderly (23). Visual dysfunction also leads to an increase in EDSS scores. Since the forceps major connects the occipital lobes, this explains the correlation between the forceps major and EDSS scores.

WM structures, in which the diffusion measures were correlated with PASAT, but not with EDSS, included the forceps minor, the cingulum, and the superior longitudinal fasciculus. PASAT performance relies heavily on attentional processes, which activate the anterior cingulate in a positron emission tomography (PET) paradigm (24). The cingulum is associated with memory functions in primates, traumatic brain injury and schizophrenia (25). Other aspects of the task necessitate executive control, working memory, and speed of information processing (24). This means that prefrontal and limbic pathways are important in the completion of the PSAt task. Limbic and prefrontal regions have also been implicated in MS-cognitive deterioration in a voxel-wise statistical analysis on 452 MS patients (26). The superior longitudinal fasciculus for instance, connects the dorsolateral prefrontal cortex to superior parietal regions, and is involved in verbal memory in MS (5).

In recent years, VBA is increasingly being applied to examine DTI data sets of subjects with various neurological or psychiatric disorders. Compared to the region of interest (ROI) post-processing approach, VBA is less laborious, is not observer-dependent, does not need to outline the complex three-dimensional (3D) WM architecture by 2D ROIs, and does not need an a priori hypothesis regarding the spatial location and extent of the expected pathology induced WM differences. However, VBA has also several drawbacks. For example, all the images need to be aligned perfectly to an atlas, which is not possible in practice due to the inter-subject variability. In addition, in order to enhance the sensitivity and specificity of the pathology detection, the data sets are filtered with a smoothing kernel width that should match the size and shape of the expected pathology, which are not known in advance, and can vary across the subjects. It is therefore virtually impossible to smooth the data sets with the appropriate filter. Furthermore, no consensus is reached about the optimal statistics for a given study, about the optimal post-hoc correction method, or about the post-hoc correction threshold that should be used. In this study, a post-hoc correction for multiple comparisons was included to account for the large number of statistical tests (i.e., for each voxel). However, the fact that multiple diffusion measures were investigated in this analysis (i.e., the “multiple hypothesis testing” correction) was not taken into account in the statistical analysis. Another limitation of this study is the relatively low number of subjects that are included, which can potentially affect the statistical power. In addition, the disease subtype, i.e., RR vs. SP patients, was not similarly distributed in both MS groups. No cardiac gating was applied during the acquisition or pre-processing of the DTI data sets.

In conclusion, in this work DTI data sets of MS patients were analyzed using a VBA method. In this VBA framework, an optimized image coregistration, atlas construction, and anisotropic smoothing method was implemented. Diffusion differences were observed between the mildly impaired MS patients and control subjects. In addition, different diffusion measures correlated with PASAT test scores for cognitive decline in parietal, frontal, as well as temporal WM regions.

Ancillary