Diffusion tensor imaging detects age related white matter change over a 2 year follow-up which is associated with working memory decline.
ABSTRACT Diffusion tensor imaging (DTI) is a sensitive method for detecting white matter damage, and in cross sectional studies DTI measures correlate with age related cognitive decline. However, there are few data on whether DTI can detect age related changes over short time periods and whether such change correlates with cognitive function.
In a community sample of 84 middle-aged and elderly adults, MRI and cognitive testing were performed at baseline and after 2 years. Changes in DTI white matter histograms, white matter hyperintensity (WMH) volume and brain volume were determined. Change over time in performance on tests of executive function, working memory and information processing speed were also assessed.
Significant change in all MRI measures was detected. For cognition, change was detected for working memory and this correlated with change in DTI only. In a stepwise regression, with change in working memory as the dependent variable, a DTI histogram measure explained 10.8% of the variance in working memory. Change in WMH or brain volume did not contribute to the model.
DTI is sensitive to age related change in white matter ultrastructure and appears useful for monitoring age related white matter change even over short time periods.
Article: Brain structure and function in chronic obstructive pulmonary disease: a multimodal cranial magnetic resonance imaging study.[show abstract] [hide abstract]
ABSTRACT: Brain pathology is a poorly understood systemic manifestation of chronic obstructive pulmonary disease (COPD). Imaging techniques using magnetic resonance (MR) diffusion tensor imaging (DTI) and resting state functional MR imaging (rfMRI) provide measures of white matter microstructure and gray functional activation, respectively. We hypothesized that patients with COPD would have reduced white matter integrity and that functional communication between gray matter resting-state networks would be significantly different to control subjects. In addition, we tested whether observed differences related to disease severity, cerebrovascular comorbidity, and cognitive dysfunction. DTI and rfMRI were acquired in stable nonhypoxemic patients with COPD (n = 25) and compared with age-matched control subjects (n = 25). Demographic, disease severity, stroke risk, and neuropsychologic assessments were made. Patients with COPD (mean age, 68; FEV(1) 53 ± 21% predicted) had widespread reduction in white matter integrity (46% of white matter tracts; P < 0.01). Six of the seven resting-state networks showed increased functional gray matter activation in COPD (P < 0.01). Differences in DTI, but not rfMRI, remained significant after controlling for stroke risk and smoking (P < 0.05). White matter integrity and gray matter activation seemed to account for difference in cognitive performance between patients with COPD and control subjects. In stable nonhypoxemic COPD there is reduced white matter integrity throughout the brain and widespread disturbance in functional activation of gray matter, which may contribute to cognitive dysfunction. White matter microstructural integrity but not gray matter functional activation is independent of smoking and cerebrovascular comorbidity. The mechanisms remain unclear, but may include cerebral small vessel disease caused by COPD.American Journal of Respiratory and Critical Care Medicine 05/2012; 186(3):240-5. · 11.08 Impact Factor
[show abstract] [hide abstract]
ABSTRACT: The aging brain's structural development constitutes a spatiotemporal process that is accessible by MR-based computational morphometry. Here we introduce basic concepts and analytical approaches to quantify age-related differences and changes in neuroanatomical images of the human brain. The presented models first address the estimation of age trajectories, then we consider inter-individual variations of structural decline, using a repeated measures design. We concentrate our overview on preprocessed neuroanatomical images of the human brain to facilitate practical applications to diverse voxel- and surface-based structural markers. Together these methods afford analysis of aging brain structure in relation to behavioral, health, or cognitive parameters.Frontiers in Neuroinformatics 01/2012; 6:3.
Article: Risk and Determinants of Dementia in Patients with Mild Cognitive Impairment and Brain Subcortical Vascular Changes: A Study of Clinical, Neuroimaging, and Biological Markers-The VMCI-Tuscany Study: Rationale, Design, and Methodology.[show abstract] [hide abstract]
ABSTRACT: Dementia is one of the most disabling conditions. Alzheimer's disease and vascular dementia (VaD) are the most frequent causes. Subcortical VaD is consequent to deep-brain small vessel disease (SVD) and is the most frequent form of VaD. Its pathological hallmarks are ischemic white matter changes and lacunar infarcts. Degenerative and vascular changes often coexist, but mechanisms of interaction are incompletely understood. The term mild cognitive impairment defines a transitional state between normal ageing and dementia. Pre-dementia stages of VaD are also acknowledged (vascular mild cognitive impairment, VMCI). Progression relates mostly to the subcortical VaD type, but determinants of such transition are unknown. Variability of phenotypic expression is not fully explained by severity grade of lesions, as depicted by conventional MRI that is not sensitive to microstructural and metabolic alterations. Advanced neuroimaging techniques seem able to achieve this. Beside hypoperfusion, blood-brain-barrier dysfunction has been also demonstrated in subcortical VaD. The aim of the Vascular Mild Cognitive Impairment Tuscany Study is to expand knowledge about determinants of transition from mild cognitive impairment to dementia in patients with cerebral SVD. This paper summarizes the main aims and methodological aspects of this multicenter, ongoing, observational study enrolling patients affected by VMCI with SVD.International journal of Alzheimer's disease. 01/2012; 2012:608013.
Diffusion Tensor Imaging detects age-related white matter change over a two-year
follow-up which is associated with working memory decline
R.A. Charlton PhD1, F. Schiavone MSc1, T.R. Barrick PhD1, R.G. Morris PhD2, and H.S.
Markus DM FRCP1.
1 Clinical Neuroscience, St George’s University of London, UK.
2 Department of Psychology, King’s College, Institute of Psychiatry, University of London,
Corresponding Author: Rebecca Charlton
St George’s University of London,
London, SW17 0RE
Tel: ++ 44 (0) 20 8725 5385
Fax: ++ 44 (0) 20 8725 2950
Keywords: cognitive decline, diffusion tensor imaging, longitudinal, normal aging, working
Word count: 2,688
Background: Diffusion tensor imaging (DTI) is a sensitive method for detecting white matter
damage, and in cross-sectional studies DTI measures correlate with age-related cognitive
decline. However, there is little data on whether DTI can detect age-related changes over
short time-periods, and whether such change correlates with cognitive function.
Methods: In a community sample of 84 middle-aged and elderly adults MRI and cognitive
testing was performed at baseline and after two-years. Changes in DTI white matter
histograms, white matter hyperintensity (WMH) volume and brain volume were determined.
Change over time in performance on tests of executive function, working memory and
information processing speed were also assessed.
Results: Significant change in all MRI measures was detected. For cognition, change was
detected for working memory and this correlated with change in DTI only. In a stepwise
regression, with change in working memory as the dependent variable, a DTI histogram
measure explained 10.8% of the variance in working memory. Change in WMH or brain
volume did not contribute to the model.
Conclusions: DTI is sensitive to age-related change in white matter ultrastructure and appears
useful for monitoring age-related white matter change even over short time periods.
There is increasing evidence that white matter damage plays a prominent role in age-related
cognitive decline. Post-mortem studies show evidence of white matter damage including loss
of myelinated white matter fibres 1, age-related activation of microglia 2 and microglial cell
inclusions believed to represent phagocytosis of myelin 3. MRI white matter hyperintensities
(WMH), best seen on T2-weighted and FLAIR sequences, become more common with age 4,
and correlate with cognitive impairment 5;6. Correlations are strongest with those cognitive
domains, such as executive function, believed to be more sensitive to changes in white matter
integrity and deterioration in cortical-subcortical connectivity 7. However, correlations
between WMH and cognition are only moderate and some studies have failed to find any
Diffusion tensor imaging (DTI) is a MRI technique that may measure white matter functional
integrity more directly. Even where white matter appears normal on T2-weighted or FLAIR
images, DTI values have indicated reduced white matter integrity 9, suggesting that subtle
damage is present even in regions outside WMH. Histopathology studies have also
demonstrated the presence of white matter damage outside WMH in Alzheimer’s disease 10.
Moreover, cross sectional studies in normal aging 11 and cerebral small vessel disease (SVD)
12;13 have shown that DTI measures correlate more strongly with cognition (and particularly
executive function) than WMH volume . In these SVD studies WMH no longer remained
independently related to cognition once DTI was entered in multivariate analysis 12;13.
DTI provides quantitative measures which are suitable for incorporation into large-scale
clinical trials. It may provide a useful surrogate marker to evaluate the effect of therapeutic
interventions on age-related cognitive decline. Before adopting DTI as a surrogate marker it is
first necessary to show that changes in DTI parameters can be detected over short time
periods, of similar duration to those used in clinical trials. Secondly it is necessary to show
that it is more sensitive to change than other conventional MRI markers such as T2-WMH,
and than repeated neuropsychological testing which is currently the usual method of
assessment. Thirdly change in DTI should correlate with change in clinical parameters,
including cognition. Currently there are no published longitudinal ageing studies using this
technique which can answer these questions. Therefore we performed serial DTI imaging at
baseline and two-year follow-up in a community population of middle-aged and elderly
individuals to determine whether changes in DTI parameters could be detected over this time
period. Cognitive testing was also performed at both time-points.
The population sample of 106 people (55 males, 51 females; age range 50-90 years; mean
69) was recruited via a local family practice by random sampling as part of the prospective
GENIE study; this sample has been described in detail previously 14. Briefly, participants had
no contraindications to MRI, and no prior psychiatric or neurological disorder, and had
English as their first language. Out of this sample 84 participants re-attended at follow-up (48
males, 36 females; mean age [range] 71 [55-91]). Of those who did not attend follow-up; 3
were deceased, 2 declined to participate due to ill-health, 1 had moved away, 16 chose not to
provide a reason for withdrawal (an option required by our ethical approval). Of the 84
follow-ups, 82 completed all cognitive assessments, 81 attended for MRI, with 2 having MRI
only. Five did not complete the MRI and had missing data; 3 had poor quality or missing
baseline DTI and were therefore not included in longitudinal analysis. In summary,
longitudinal data was available for: cognition =82, brain volume =80, WMH =79, DTI =73
(69 people completed the study without any missing variables).
Participants underwent a battery of standardized neuropsychological tests. Mean scores were
calculated for executive function, working memory and information processing speed, using
methods as described previously 14. Briefly executive function comprised: the Stroop test total
correct 15, letter fluency (FAS) total correct, category fluency (animals and boys names) total
correct, and the following sub-tests from the Delis-Kaplan Executive Function System (D-
KEFS) 16: the Trails test - number-letter switching minus motor speed, and Towers total
achievement score. Working memory comprised the digit span backwards and letter-number
sequencing subtests from the Wechsler Memory Scale III (WMS-III) 17. Information
processing speed included sub-test A (total completed) from the Attention, Memory and
Information Processing Battery (AMIPB) 18, digit symbol number completed from Wechsler
Adult Intelligence Scale Revised (WAIS-R) 19, and the time to complete the grooved
pegboard with the dominant hand 20.
At baseline, raw scores for each test were transformed into z-scores (using means and
standard deviations for the whole group); follow-up data was transformed into z-scores using
the means and standard deviations from baseline data. Z-scores were re-coded so a high score
reflected good performance. Mean scores were calculated from the z-scores for each ability at
baseline and follow-up.
Magnetic Resonance Imaging
All MRI was performed on a General Electric 1.5T(22 mT/m) Signa scanner. Whole brain
DTI (acquisition matrix=96x96, FOV=240mm x 240mm; TE=80ms; TR=7s, maximum b-
value=1000 s mm-2) was acquired in two interleaved series of 4 repeats, each containing
twenty-five 2.8mm slices, with a gap of 2.8mm, providing contiguous whole brain coverage.
Images were acquired with no diffusion weighting (b= 0 s mm-2) and in 6 directions and the
negative of those 6 to eliminate diffusion-imaging gradient cross terms. Also acquired were:
FLAIR (slices=28; thickness=5mm; acquisition matrix=256x256, FOV=240mm x 240mm;
TE=120ms; TR=9000ms); and T1-weighted whole brain volume scan (slices=96;
thickness=1.5mm; acquisition matrix=256x256x92, FOV=240mm x 240mm; TE=3ms;
TR=17ms; Flip Angle=10º).
All data were analyzed on an independent workstation (Sun Blade 1500; Sun Microsystems,
Mountain View, CA).
DTI: The analysis derived measures of mean diffusivity (MD) and fractional anisotropy (FA),
two main descriptors of white matter ultrastructural integrity. Images were realigned to
remove linear eddy current distortions using automated image registration (AIR) software
21;22. Diffusion tensor elements were computed at each voxel 23 and diagonalized to determine
eigenvalues and eigenvectors, from which MD and FA values were generated.
For baseline data, images were segmented (using the EPI-B0 [b=0 s mm-2 images]) into grey
matter, white matter, and CSF concentration maps in SPM2
(http://www.fil.ion.ucl.ac.uk/spm/), incorporating a correction for image intensity non-
uniformity 24. For each participant, hard segmentations of each tissue type (grey, white, CSF)
were generated to include all voxels whose probability was greater than the combined
probability of the remaining tissue types. Each segmentation was checked to assure accuracy.
To ensure DTI histogram measurements for both time-points were taken from the same brain
region, follow-up EPI-B0 images were co-registered to the baseline EPI-B0 images using an
affine transformation in SPM2. This realigned data was used to generate follow-up
histograms using the segmentation computed from baseline data.
DTI Histograms: Histograms were calculated for MD and FA maps of all brain white matter,
using the segmentation described above. MD histograms bin width was set to 4x10-5, between
0.0 and 0.004 (4x10-3). FA histograms bin width was 0.01, between 0.0 and 1.0. To correct for
individual differences in brain volume each histogram was normalized over the number of
voxels in the segmented images. Histograms do not necessarily demonstrate Gaussian profiles
in MD and FA data (see Figure 1), so mean and standard deviation measures may be less
relevant than other descriptors of the histogram, such as median, peak height frequency or
peak height intensity, skewness (i.e. how asymmetric is the histogram?) and kurtosis (i.e. how
peaked is the histogram?) which were used in the analysis. Distributions with kurtosis of less
than zero are referred to as having negative kurtosis or being platykurtic (i.e. having a flatter
peak, and longer, fatter tails than the normal distribution) and those with a kurtosis of greater
than zero are referred to as having positive kurtosis or being leptokurtic (i.e. distributions with
narrower, well-defined peaks and fatter, shorter tails than the normal distribution). It is also
helpful to consider the histogram parameters in terms of their difference in symmetry from the
normal distribution, termed skewness. Here zero indicates a symmetrical distribution, whereas
a distribution with larger tails to the left or right is referred to as possessing negative or
positive skewness, respectively. Furthermore, a greater skewness magnitude indicates greater
asymmetry of the histogram distribution. Figure 1 clearly shows the MD distribution to be
positively skewed and leptokurtic, whereas the FA distribution is positively skewed and less
leptokurtic than the MD distribution. Furthermore, there is an interaction between the
measured histogram parameters, for example, the MD distribution, which is highly leptokurtic
has a greater peak height frequency than the FA distribution, while the MD distribution also
has a greater positive skew (i.e. it is more asymmetrical) than the FA.
In order to assess the reproducibility of DTI measurements, scans were repeated within a
week in 8 healthy volunteers. Test-retest errors for DTI histogram measures indicated good
reproducibility. For example, normalized peak height frequency and median are given below,
and other parameters demonstrated similar error levels. Test-retest errors: MD normalized
peak height frequency (mean = 4 x 10 -6, standard deviation (SD) = .00003), MD median
(mean = 4 x 10 -6, SD = .00002), FA normalized peak height frequency (mean = -.003, SD =
.009), FA median (mean = .004, SD = .015).
White Matter Hyperintensities (WMH): FLAIR images were loaded into dispunc (David
Plummer, University College London, UK), and WMH were outlined using the contour
function. All areas of increased signal were included unless: i) the area was less than 10
pixels, ii) the area was a narrow band (less than one pixel wide) along the ventricles, iii)
hyperintensities were due to the presence of blood vessels. Baseline and follow-up images
were blinded for date of acquisition, and WMH on both scans were outlined concurrently.
Lacunar infarcts were counted on T1 images by an experienced observer. Although lacunar
infarcts were rare (4 in 1 subject, 2 in 2 subjects, and 1 in 8 subjects, with none in the
remainder), where WMH were present adjacent to a lacunar infarct, the lacunar infarct
volume was excluded from the WMH volume.
Brain Volume Measurements: These were performed using SIENAX and SIENA software
(http://www.fmrib.ox.ac.uk/fsl/siena/index.html) 25. Using SIENAX, T1-weighted volume
images were used to calculate baseline normalized brain volume, controlling for individual
differences in the skull size. Percentage change in brain volume between the two time-points
was automatically calculated using SIENA.
Analysis was performed on the participants who attended at both baseline and follow-up.
Measures were normally distributed at baseline, follow-up, and on the measure of change for
all cognitive data, DTI parameters and brain volume. WMH volumes were not normally
distributed at baseline or follow-up, and were therefore log transformed. The change
measurement for WMH volume was normally distributed.
Paired sample t-tests assessed the difference between baseline and follow-up measures.
Change between the time-points was calculated as a ratio using the following equation
b S ; follow-up=
f S ):
This equation was used to take into account baseline levels of functioning, and reduces
distortion due to scalar differences in change. In order to investigate whether changes in
imaging measures predicted change in cognitive decline, only variables that demonstrated
significant change over the two-year delay were included in subsequent analysis. Pearson’s
correlations were used to assess the relationships between change in imaging and change in
cognitive variables. Stepwise regression analyses were performed with change in imaging
measures as independent variables, and change in cognitive abilities as the dependent
There were no differences between the individuals who did and did not attend for follow-up
in baseline measures of age, blood pressure, WMH-volume, DTI parameters, or information
processing speed. The drop-out group had smaller baseline normalized brain volume (t= -
2.518, p= .013) and lower baseline executive function (t= -2.70, p= .008) and working
memory (t= -2.00, p= .048) scores. Demographic details, baseline scores and follow-up scores
are shown in table 1.
Page left intentionally blank
See Table 1 on next page
Table 1: Cognitive ability scores and MRI measures for baseline and 2 year follow-up data.
Mean (and standard deviations)
Paired sample t-test
of data at baseline
Executive Function .076 (.684) .119 (.697) t= -.979, p= .330
Working Memory .082 (.844) -.073 (.810)
t= -2.57, p= .012
Information Processing Speed .062 (.757) .010 (.467) t= .640, p= .524
Normalized Whole Brain Volume
1589.11 (75.16) 1569.71 (81.71)
t = 4.38, p < .001
WMH (in mm3)
t = -5.03, p < .001
MD - Normalized peak height
.19547 (.03144) .19110 (.03426)
t = -9.83, p < .001
MD - Peak height intensity .00074 (.00004) .00075 (.00004)
t = 2.55, p = .013
MD – Skewness 5.07 (.989) 4.36 (.821)
t = -.486, p < .001
MD – Kurtosis 34.33 (15.574) 28.04 (10.776)
t = -3.49, p = .001
MD – Median .00080 (.00004) .00082 (.00004)
t = 5.73, p < .001
FA - Normalized peak height
.03112 (.00386) .03342 (.00437)
t = 3.17, p = .002
FA - Peak height intensity .22288 (.04020) .21493 (.03485)
t = -5.70, p < .001
FA – Skewness .811 (.153) .849 (.159)
t = 11.70, p < .001
FA – Kurtosis .570 (.417) .659 (.436)
t = 6.51, p < .001
FA – Median .282 (.028) .275 (.025)
t = -7.66, p < .001
All DTI measures changed significantly between the two time-points with an increase in
median MD and a decrease in median FA (table 1 for data and t-tests). Figure 1 shows the
baseline and follow-up MD and FA histograms for the two time-points. Change in the MD
distribution between time-points was characterized by a significant increase in median MD
and a significant decrease in positive kurtosis. These parameter changes caused the
distribution to be significantly more symmetrical (i.e. have smaller positive skew) and
significantly less peaked (i.e. have a decreased normalized peak height frequency). Change in
the FA distribution between time-points was characterized by a significant decrease in median
FA and a significant increase in positive kurtosis. In turn these parameter changes caused the
distribution to be more significantly asymmetrical (i.e. have greater positive skew) and
significantly more peaked (i.e. have an increased normalized peak height frequency). Figure 2
illustrates the change in MD and FA normalized peak height frequency between baseline and
WMH, expressed as a percentage of brain volume, increased from 7.21% (SD 9.79) at
baseline to 9.00% (SD 12.39) at follow-up (t = -5.03, p< .001), see table 1. Normalized whole
brain volume decreased from 1589.63 cm3 (SD 76.76) at baseline to 1569.71 cm3 (SD 80.71)
at follow-up (t = 4.38, p< .001), a mean reduction of 0.87% (SD 0.993), approximately 0.44%
per year, see table 1.
There were no differences between the mean z-score for the two time-points for executive
function (mean= .042, SD= .393; t= -.979, p= .330) or information processing speed (mean= -
.054, SD= .759; t= .640, p= .524). In contrast working memory scores had declined at follow-
up compared to baseline (mean= -.155, SD= .547; t= -2.57, p= .012). See table 1.
Change Measures and Age
Correlations were performed to assess whether changes in working memory or imaging
measures were associated with increasing age. Age did not correlate significantly with change
in working memory (r = .034, p = .763). There were no significant correlations between age at
baseline and WMH (r = -.056, p = .624) or normalized brain volume (r = -.205, p = .070).
There were no significant correlations between FA histogram measures and age, and only one
significant correlation with change in the MD histogram measures and age, with MD
normalized peak height frequency (r = .244, p = .037).
Correlations between change in MRI parameters and change in working memory
The association between imaging and longitudinal cognitive data was only considered for
working memory, the measure that showed significant change at follow-up.
The working memory change ratio correlated with change in MD normalized peak height
frequency, MD skewness, and MD kurtosis. There were no correlations with change in FA
parameters (table 2). There were no correlations between either change in WMH (r= .098, p=
.396) or normalized brain volume (r= -.075, p= .517) and change in working memory.
To check that age-effects were not influencing the results the correlations between change in
working memory and change in the imaging measures were repeated, controlling for age at
baseline. The results remained largely unchanged with the working memory change ratio
remaining significantly correlated with change in MD normalized peak height frequency (r = -
.346, p = .004), MD skewness (r = -.270, p = .026), and MD kurtosis (r = -.306, p = .011), see
table 2. Secondary analyses were performed to assess the relationship between change in
imaging measures and change in both executive function and information processing speed.
No significant correlations were observed in these analyses.
Table 2: Correlations for associations between the change ratio for working memory and
change in imaging measures, and partial correlations controlling for age. Results for a number
of different parameters derived from the DTI histogram are shown.
Correlations Partial correlations
controlling for age
Normalized whole brain
WMH volume r= .098
MD FA MD FA
Normalized peak height
r = -.329
p = .005
r = .077
p = .523
r = -.346
p = .004
r = .085
p = .490
Peak height intensity r = -.125
p = .298
r = -.183
p = .126
r = -.118
p = .336
r = -.180
p = .142
Skewness r = -.252
p = .034
r = .019
p = .872
r = -.270
p = .026
r = .019
p = .879
Kurtosis r = -.292
p = .013
r = -.034
p = .777
r = -.306
p = .011
r = -.036
p = .771
Median r = .025
p = .839
r = -.118
p = .327
r = .035
p = .775
r = -.113
p = .359
Additional partial correlations were performed to investigate whether the relationship between
change in DTI parameters and change in working memory was independent of change in the
volume of WMH. The pattern of correlations remained the same: change in working memory
correlated significantly with change in MD normalized peak height frequency (r = -.320, p =
.007), MD skewness (r = -.251, p = .038), and MD kurtosis (r = -.294, p = .014).
Multivariate analysis of all MRI and working memory data
A stepwise regression analysis was computed with change in working memory as the
dependent variable. MD normalized peak height frequency was selected to represent the DTI
histogram as it demonstrated the most robust correlation. Independent variables were change
in MD normalized peak height frequency, WMH, and normalized brain volume. Only MD
normalized peak height frequency remained in the model (beta = .329, p= .005), explaining
10.8% of the variance in working memory (F(1,68) = 8.27, p= .005).
Correlations between baseline MRI parameters and change in working memory
No significant correlations were observed between baseline imaging measures and change in
Previous cross sectional studies have shown that DTI parameters correlate strongly with both
age and aspects of cognition including executive function and working memory 14;26;27.
Furthermore they have shown that correlations of DTI with cognition are generally stronger
than between WMH and cognition 12;13. The current study, describing longitudinal data from
the GENIE cohort, extends these findings and demonstrates that DTI is sensitive to short term
age-related white matter change.