White matter hyperintensities (WMH) of presumed vascular origin, also referred to
as leukoaraiosis, are a very common finding on brain magnetic resonance imaging (MRI)
or computed tomography (CT) in older subjects and in patients with stroke and dementia.
They are associated with cognitive impairment, triple the risk of stroke and double
the risk of dementia. Knowledge of their pathology derives mostly from post mortem
studies, many from some years ago. These, by their nature, were generally small, sampled
from selected brain regions and probably reflect late-stage disease. They focus on
features of demyelination and axonal degeneration, which may be easier to detect histopathologically
than changes in extracellular fluid. Here we review advances in brain magnetic resonance
imaging (MRI) that are revealing white matter hyperintensities at earlier stages,
and changes in normal-appearing white matter that indicate tissue pathology, less
marked than those found in WMH. These “pre-visible” changes show that altered interstitial
fluid mobility and water content, which may be reversible, probably predate demyelination
and axonal damage, which are less likely to be reversible and are probably a late-stage
phenomenon. Neuroimaging is also revealing the dynamic nature of WMH, their interactions
with other pathological features such as secondary cortical and long tract damage,
and contribution to accumulating brain damage. These insights provide opportunities
to improve understanding the etiology and pathogenesis of small vessel disease, and
represents an enormous unfinished agenda for preventing accumulation of brain damage,
and its associated cognitive and physical problems, from mid to later life. Recognizing
the earliest stages leading to WMH development will provide important opportunities
to prevent (or even reverse) brain damage due to small vessel disease at the earliest
stages, and ameliorate its cognitive, physical, stroke and dementia consequences.
Historical Perspective
Worldwide, about 15 million people have a stroke, from which 6 million die and 5 million
are left permanently disabled each year1; 35.6 million people worldwide are estimated
to be living with dementia, and this is expected to triple by 2050. Although Alzheimer’s
disease is the most commonly diagnosed form of cognitive impairment in older people,2
cognitive impairment due to vascular disease alone, or in addition to Alzheimer’s
disease, is very common and contributes to substantial worsening of cognitive function
for a given burden of Alzheimer’s disease pathology.3 The most common form of vascular
cognitive impairment typically results in lesions visible on brain scanning known
collectively as small vessel disease (SVD) and described in more detail below. Although
stroke and dementia are priorities for many governments, the cause of the 20% of strokes
and about 40% of dementias attributed to SVD4 remains unclear, with limited options
for prevention and treatment.
The finding of altered areas of white matter attenuation on computed tomography (CT)
was first brought to prominence in the late 1980s by Hachinski and colleagues. They
described patchy low attenuation in the periventricular and deep white matter, which
they referred to as leukoaraiosis (Figure 1).5 These patchy white matter changes are
more obvious as abnormal areas of signal intensity on magnetic resonance imaging (MRI)
due to the latter’s better sensitivity to soft tissue changes than CT, particularly
for subtle alterations in water content. The signal changes predominate in the periventricular
and deep white matter, so were commonly referred to originally as “white matter lesions”
although they are also recognized to occur in the deep gray matter.6 These areas are
hyperintense on T2-weighted (T2), proton density-weighted (PD), and fluid attenuated
inversion recovery (FLAIR) MRI sequences, so are now by consensus referred to as “white
matter hyperintensities” (WMH), or “subcortical hyperintensities” where deep gray
matter is also involved.6 They are also hypointense on T1-weighted and hyperintense
on T2*-weighted sequences, but the most sensitive structural sequence for visualizing
WMH on MRI is generally the FLAIR sequence (Figure 1).
Figure 1
Examples of WMH on (A) CT, (B) MR FLAIR, (C and D) MR FLAIR and T2-weighted imaging.
A, Three adjacent CT images from 1 patient with severe WMH. B, Four different subjects
showing, L to R, normal to severe WMH. C, FLAIR and T2-w, same subject, show WMH and
a lacune (arrow). D, top FLAIR, bottom T2-w images showing that when subtle, WMH are
more easily seen on FLAIR. CT indicates computed tomography; FLAIR, fluid attenuated
inversion recovery; MR, magnetic resonance; WMH, white matter hyperintensities.
MRI came into wider use in clinical practice and for research in the late 1980s and
early 1990s. Early MRI scanners of the 1980s–1990s were generally of lower field strength
than are available routinely today. This affected their sensitivity. For example,
the earliest scanners in clinical use were 0.2 or 0.5 Tesla field strength, and inevitably
were less sensitive to subtle tissue changes than are the 1.5 or 3 T MR scanners that
are in widespread use today. The early studies also had fewer sequences available
and these were less sensitive to brain soft tissue changes than those available now.
For example, it was common to use T1-weighted (T1) or T2 sequences, T1 being of similar
sensitivity to subcortical changes as is modern CT scanning, while T2 is relatively
insensitive to subtle white matter abnormalities and is less good at identifying changes
adjacent to a cerebrospinal fluid (CSF)-containing space than is FLAIR (eg, in the
immediate periventricular tissues). Hence, when considering imaging-pathological correlations,
imaging-clinical, or imaging-cognitive correlations, it is important to bear in mind
that early studies of MRI-pathology correlations may have lacked sensitivity to detect
the more subtle changes that are visible today, and thus may have influenced our understanding
of the pathophysiology towards what are probably more established, permanent changes.
WMH had been overlooked somewhat by pathologists up until CT and MRI became available,7,8
much of the focus of pathology examinations in the last century being on lacunes (small
CSF-containing cavities)9,10 that are easier to detect pathologically than subtle
WMH.11,12 The diffuse and often subtle changes of WMH may be hard to see macroscopically
on brain sections until they are advanced13; when subtle, the full extent of WMH may
be difficult to appreciate histologically unless specifically sought in aging-related
changes14 or in other white matter diseases such as multiple sclerosis.15 In contrast,
lacunes, ie, small CSF-containing cavities, are more obvious pathologically and had
been described in numerous detailed pathology studies (summarized in Bailey et al16)
although there were not many imaging-pathology correlations for lacunes either.
Clinical Importance
Until relatively recently, WMH were generally dismissed as inevitable consequences
of “normal” advancing age. This is clearly not true. Although WMH do become more common
with advancing age,17 their prevalence is highly variable. Furthermore, numerous studies
indicate that they have important clinical and risk factor associations, emphasizing
that they should not simply be overlooked as inevitable “silent” consequences of the
aging brain.
In a meta-analysis of 22 longitudinal studies, WMH were clearly associated with progressive
cognitive impairment, a 2-fold increase in the risk of dementia and a 3-fold increase
in risk of stroke.18 WMH also affect physical function, resulting in abnormal gait19
and disturbed balance.20 WMH increase the risk of late onset depression.21 Furthermore,
WMH are highly heritable,22 and they vary with familial longevity being less frequent
in subjects with long-lived parents.23 WMH are inversely associated with intelligence
in youth,24 and with educational attainment.25 Whether these latter associations indicate
relationships between intelligence and risk factor exposure or reflect brain resilience
to damage,26 are as yet unknown.
The prevalence of WMH increases with increasing vascular risk factors, including hypertension,27–29
diabetes,30 smoking,31,32 as well as with many other as yet undetermined risk factors.33
Risk factor exposure seems to be particularly important if it occurs in middle age,29,34
but the relative importance of different risk factors may also vary in different age
groups. A large multicenter study of WMH in 2699 stroke patients in 11 stroke centers
in China suggested that raised cholesterol was a more important risk factor for WMH
in older hypertensive patients whereas age alone was the major risk factor in older
non-hypertensive patients. Diabetes mellitus was an important risk factor in younger
non-hypertensives.35 It is unclear as yet if hypertension has most of its effect directly
on the brain, or if it results in systemic vascular stiffening, which in turn affects
the brain white matter36,37 and cognition.38 Accumulating evidence, including associations
with left ventricular hypertrophy, tend to point to the former explanation,35,37 but
more longitudinal studies to determine the sequence of development of WMH in relation
to risk factors are required. Additionally, there are as yet mixed results from randomized
trials on whether risk factor reduction can prevent WMH progression or cognitive decline
in older subjects,39–41 and some evidence that too aggressive blood pressure reduction
in old age could actually worsen white matter damage42 and cognitive decline, perhaps
if autoregulation is impaired through vessel stiffening and loss of vasoreactivity.
Further trials of risk factor management are needed and specifically at different
ages because different vascular risk factors may have more impact at different ages
(eg, hypertension in middle age, cholesterol at older ages35). Physical activity is
suggested to protect against WMH in cross-sectional studies,43–45 but whether exercise
can prevent progression or development of WMH remains to be tested in randomized clinical
trials.
WMH are part of the spectrum of small vessel disease (SVD) which includes lacunar
(or small subcortical) ischemic and hemorrhagic stroke, lacunes, microbleeds, perivascular
spaces and brain atrophy (Figure 2).6,46 All of these features, individually, are
associated with cognitive impairment,47–51 even when subtle.
Figure 2
STRIVE examples of different features of small vessel disease, including white matter
hyperintensities. Reproduced with permission from Wardlaw et al.6 DWI indicates diffusion-weighted
imaging; FLAIR, fluid attenuated inversion recovery; GRE, gradient recalled echo;
STRIVE, standards for reporting vascular changes on neuroimaging; SWI, susceptibility-weighted
imaging.
The dynamic nature of these inter-related lesions is illustrated by studying patterns
of lesion evolution (Figure 3): acute small subcortical infarcts can disappear (10%),
remain looking like a WMH (60% to 70%), or cavitate to form a lacune (20% to 30%)52–54;
lacunes may form at the edge55 or (in our experience) in the middle of a WMH; WMH
can “grow” at the edges of small subcortical infarcts56; incident lacunes57 and WMH58
are associated with cortical thinning and cerebral atrophy59; all of which indicate
progressive and accumulating brain damage. Furthermore, WMH increase risk of brain
damage in the presence of other pathologies, for example, they associate with infarct
growth and worse outcome after large artery stroke,60,61 and they predict poor functional
stroke outcome.62–64
Figure 3
Diagram of dynamic mechanisms by which WMH and SVD lead to brain damage. See separate
file. SVD indicates small vessel disease; WMH, white matter hyperintensities.
The effects of all these SVD features on cognition are cumulative,65 (in submission)
providing further indication that together these SVD features are closely related
pathologically,52 and represent cumulative brain damage,32 the prevention of which
should help ameliorate brain tissue damage, reduce loss of normal brain cognitive
and physical function and preserve independent survival in old age.43
What Do Pathology Studies Suggest That WMH Are Due To?
Pathology studies are, unfortunately, infrequent12,66,67 compared with the number
of WMH captured in imaging studies. There are particularly few pathology studies that
have linked individual lesions seen on MRI with their pathological examination.12,68,69
Pathology studies have been hampered by difficulty in matching up individual small
lesions on imaging with their pathological counterpart,11,68 with limitations of sampling,13
of fixation,70 of definitions,71–73 they provide “snapshots” of disease evolution,
and because end-stage damage may obliterate the earliest stages of disease.
As described above, the earlier pathology reports focused on demyelination and axonal
loss in WMH7 and described the changes as “ischemic”.74,75 Demyelination and axonal
destruction imply that the changes are permanent. Indeed, extensive WMH were associated
with reduced density of glia and vacuolation.8 More subtle WMH on MRI were associated
with microglial and endothelial activation.13 Some studies differentiate periventricular
from deep WMH12,76,77 although imaging studies indicate that periventricular and deep
WMH are probably mostly part of a continuous pathology.76,78 A recent very large study
in 2699 patients with stroke from 11 stroke centers in China created statistical maps
of WMH distribution and provided more evidence to support treating periventricular
and deep WMH as a continuous pathology, any apparent difference in distribution in
some patients simply reflecting an earlier stage in disease.35 Notwithstanding, the
available pathology describes periventricular WMH as having discontinuous ependyma,
gliosis, loosening of the white matter fibers, and myelin loss around tortuous venules
in perivascular spaces,11,12,79 the gliosis, demyelination, and fiber loss worsening
as the periventricular WMH worsened. In deep WMH, there was demyelination, gliosis,
and axonal loss around perivascular spaces, with vacuolation and increased tissue
loss as the lesions became more severe.12 The changes were heterogenous.77 WMH in
patients with AD showed more microglial activation than in WMH of age-matched controls.80
The variation in pathological severity might help explain some of the variation in
the association between WMH severity and cognitive change in old age80 although this
needs to be verified in larger and more heterogeneous populations with good cognitive,
dementia and SVD phenotyping.
Some reports indicate the presence of proteins in the perivascular tissues in human
WMH81 and suggest that the arteriolar wall thickening seen in SVD is also a consequence
of the movement of plasma proteins into the vessel walls.82 The proteins include fibrinogen,
immunoglobulins, thrombomodulin,83 and occur in subcortical grey and white matter
and are also seen in normal appearing white matter and are associated with microglia.81
The microglia and proteins were more frequent in areas of WMH with more advanced tissue
loss.81 Increased albumin was found in the interstitial tissues of brains of older
subjects with WMH84 although it was uncertain as to where the albumin had come from
in this study as the endothelial tight junction proteins were said to be intact. Others
have not identified direct evidence of blood-brain barrier (BBB) impairment despite
identifying proteins associated with endothelial activation83 and that had extravasated
into the perivascular tissues.85 However, the cerebral blood flow and the cerebrovascular
endothelial surface area are both so large that there would not have to be much breach
of the BBB for proteins and fluid to accumulate in the interstitial spaces. It is
likely also that such tissue fluid shifts are difficult to identify pathologically
or overlooked as fixation artifact, where fixation methods involve freezing or dehydration,
which may affect tissue content.70
Numerous in vivo studies have found raised CSF:plasma albumin ratios with advancing
age, in patients with Alzheimer’s disease, in vascular dementia, and in patients with
WMH,86 and suggested a role for chronic brain edema in forming WMH and more advanced
damage in the form of lacunes.87–89 Undoubtedly the pathological picture is mixed
and complex90: despite the risk factor association, immunohistochemical and gene expression
microarray studies suggesting a role for ischemia, hypoxia, or hypoperfusion, studies
also show immune activation, BBB dysfunction, altered cell metabolic pathways, and
glial injury.90 The abnormalities are now being recognized in normal-appearing white
matter as well as in WMH. Therefore, regardless of where the albumin or other plasma
proteins came from, or how they got into the interstitial tissues, there does appear
to be a role for loss of normal BBB function and fluid shifts into the brain leading
to secondary brain damage91,92; if arrested or reversed early, the interstitial fluid
shifts may be more reversible than demyelination and axonal loss.
What Does Imaging Suggest?
To understand more about the pathophysiology of WMH, we must turn to imaging methods
that allow us to detect and quantify subvisible tissue changes on a per voxel basis
in vivo. Structural changes in the integrity of the brain’s white matter are commonly
observed through MRI methods such as diffusion tensor imaging (DTI), a widely available
imaging technique that provides quantitative measures of the mobility of water molecules
in vivo. Parameters obtained from the water diffusion tensor are used as diagnostic
markers for clinical applications. The most commonly used are fractional anisotropy
(FA), which indicates the deviation from pure isotropic diffusion of water mobility
in vivo, and mean diffusivity (MD), which measures the magnitude of diffusion of the
water molecules.93 Both the magnitude and directionality of the water diffusion are
affected by the underlying tissue architecture and can demonstrate, for example, alterations
in axonal microstructure or interstitial fluid.94–98
Further potential MRI biomarkers of white matter damage are the magnetization transfer
ratio (MTR) and the longitudinal relaxation time (T1). MTR is obtained from magnetization
transfer MRI99 and can show pathological changes in the white matter structure that
involve macromolecules in the cell membrane, such as inflammation or demyelination.100
T1 relaxation time is a weighted average of the free and bound water phases101 providing
quantitative information on brain water content, and is therefore a potential marker
for edematous brain tissue.94
Other methods of assessing alterations in normal brain tissue function use dynamic
imaging following intravenous injection of a contrast agent. In dynamic contrast-enhanced
MRI102 a series of T1-weighted images are acquired dynamically after injection, enabling
the measurement of BBB permeability.103,104 Similarly, dynamic susceptibility contrast
MRI is used to measure cerebral perfusion and estimate regional cerebral blood volume
and cerebral blood flow (CBF), again by tracking a contrast agent bolus using a T2*-weighted
sequence.105,106
The measurement of these MR imaging biomarkers in addition to structural “visible”
brain changes, can help to identify the pathophysiological changes within normal-appearing
white matter and WMH in vivo and inform our understanding of their pathophysiology.
Studies in humans have shown that MD and T1 increase while FA and MTR decrease significantly
in areas of WMH when compared to normal appearing white matter,98,107–109 indicating
pathological processes involving increased water content and mobility, demyelination,
and axonal loss. MD was reported as the parameter that discriminated best between
normal-appearing white matter and WMH (Figure 4).109 Perfusion MRI studies also showed
decreased CBF in regions of WMH,110,111 although it remains uncertain as to how much
the blood flow reduction represents reduced flow due to having less tissue to supply
but is not the primary cause of the damage,37,112 or is the primary cause of worsening
of tissue damage:113,114 there is some evidence for both explanations.115 The dynamics
of WMH progression and associations with CBF are complex and regionally specific.116
Figure 4
Images for FA, MD, T1, and MTR and corresponding values in normal-appearing white
matter and WMH. And corresponding ROC curves for each parameter’s ability to differentiate
normal white matter from WMH. MD shows a near perfect ROC curve. Top 2 images show
the original FLAIR image and the tissue segmentation into normal white matter and
WMH. FA indicates fractional anisotropy; FLAIR, fluid attenuated inversion recovery;
MD, mean diffusivity; MTR, magnetisation transfer ratio; NAWM, normal-appearing white
matter; ROC, receiver operator characteristic; WMH, white matter hyperintensities.
Imaging demonstrates that WMH are heterogeneous, ie, they represent different amounts
of tissue damage, as reflected in their degree of “whiteness” and in features seen
on more recent MR scanners such as “dirty white matter” (Figure 5). Moreover, studies
using MTR show that MTR differs between frontal and occipital WMH,117 either indicating
different stages of tissue damage or possibly different underlying tissue constructs.
Figure 5
FLAIR image through the lateral ventricles showing severe (ie, intense) WMH and less-intense
white matter damage (A), represented in blue and red, respectively in (B) and which
correspond with the intensities arrowed in the histogram (C). Less-intense damage
can be also observed in the T1-weighted modality as the yellow arrows show (D). FLAIR
indicates fluid-attenuation inversion recovery; WMH, white matter hyperintensities.
The relationships between different parameters observed in WMH differed from those
relationships in normal-appearing white matter (Figure 6). Significant correlations
appeared between MTR and both FA and MD in WMH, whereas these parameters were not
correlated in normal-appearing white matter, indicating that cellular breakdown in
WMH allows quantitative MRI parameters to take a much wider range of values than in
healthy tissue, where they are kept within tight bounds and are likely to be independent.98
Figure 6
Correlations between FA, MTR, MD and T1 in NAWM and WMH. In general, the T1 and FA,
MD and MTR show stronger correlations in WMH than in NAWM. FA indicates fractional
anisotropy; MD, mean diffusivity; MTR, magnetization transfer ratio; NAWM, normal-appearing
white matter; WMH, white matter hyperintensities.
Several reports indicate that the tissue structural and vascular changes spread further
than the visible area of the WMH, rather than being confined to the visible WMH109,111,118–120
consistent with recent pathological reports.90 The changes radiate into normal-appearing
white matter, particularly in the immediate peri-WMH tissue, indicating that the underlying
pathology is a diffuse process affecting much of the white matter and even other parts
of the brain, and that visible lesions, ie, the WMH, are probably only the “tip of
the iceberg.”
This compromised normal-appearing white matter integrity in the presence of WMH has
been observed through different quantitative imaging techniques. For example, CBF121
and cerebrovascular reactivity were reduced and BBB permeability was increased in
the normal appearing white matter of healthy elderly subjects and all were associated
with the presence of WMH.122 BBB permeability increases with advancing age during
normal aging, but is further increased in dementia particularly in patients with vascular
dementia and in patients with WMH.86 Others have found increases in BBB permeability
in NAWM in patients with WMH,123 in patients with lacunar versus cortical ischemic
stroke, with increasing age and numbers of enlarged perivascular spaces,124 and in
WHM in vascular dementia.125 The MD and FA of NAWM have been also associated with
WMH volume.120,126–128
However, studies typically include subjects of widely varying age and the changes
observed in NAWM integrity relative to WMH load could also be a consequence of the
older age of those subjects with more WMH.17,129,130 A more recent study in a large
cohort of older people of near-homogeneous age reported that FA, MD, MTR, and T1 were
all associated with the severity of WMH (highest Fazekas scores) – even after accounting
for age, gender, exposure to common vascular risk factors and the proximity to WMH
– proving that the changes observed were not just a function of age.109 Moreover,
although all 4 parameters measured in NAWM were related to WMH severity in those with
most WMH, only the MD of NAWM showed changes within the mildest range of WMH. Additionally,
of the 4 quantitative parameters of MD, FA, MTR and T1, it was MD that showed the
best differentiation of WMH from normal-appearing white matter, performing substantially
better on received operator characteristic (ROC) curves than did FA, MTR, or T1. This
combination of findings suggests that the earliest pathological processes responsible
for WMH involve changes in interstitial fluid mobility.109
Although observed by others,98 the concept of “WMH penumbra” was introduced by Maillard
et al118 after they observed that the effects on FA are not only globally influenced
by the WMH load, but also locally influenced by the distance into normal-appearing
white matter from the edge of the WMH. This idea of the more abnormal the tissue,
the closer the proximity to the WMH, was recently corroborated by measuring MRI quantitative
parameters in the normal-appearing white matter in “contours” set at increasing distance
from the visible edge of the WMH on FLAIR. In this contour-based analysis, MD decreased
and MTR increased with increasing distance from the WMH over all distances assessed,
whereas FA and T1 mainly showed only slight changes in the closest contours to the
WMH (Figure 7).109
Figure 7
Effect on MD, FA, MTR, and T1 of increasing burden of WMH, in the WMH and at 2 mm
distance increments from the WMH edge into the NAWM (2 mm-NAWMrem). Top left, illustrates
WMH (red) and contours (different color bands), right, 1 to 6, WMH Fazekas scores.
Lower left graphs show change in the 4 MR parameters from WMH and at increasing distance
into the NAWM; lower right graphs show the parameter changes at increasing distance
into NAWM split by total WMH burden (Fazekas score). Adapted from Muñoz-Maniega et al.109
FA indicates fractional anisotropy; MD, mean diffusivity; MTR, magnetisation transfer
ratio; NAWM, normal-appearing white matter; WMH, white matter hyperintensities.
Longitudinal studies suggest that people with lower white matter integrity at baseline
are more likely to progress to cognitive impairment or frank Alzheimer’s disease than
are those with higher white matter integrity.131 Having a high WMH burden increases
the risk of worsening WMH several years later,132 as well as of stroke, dementia,
and death.18 New lacunes form at the edges of WMH55 and WMH form at the edges of small
subcortical infarcts or lacunes,56 ie, damage begets damage (Figure 3).
The precise sequence of pathologic processes underpinning the microstructural changes
in white matter integrity within and around WMH are yet to be fully described. Pathology
studies have described rarefaction, demyelination, and axonal loss in WMH as outlined
above, which are compatible with the observed increases in MD and T1, and the decreases
in FA, MTR, and CBF detected using neuroimaging. Some changes such as minor fluxes
in fluid content may be relatively easier to detect with MR imaging, which is highly
sensitive to water shifts. Hence, neuroimaging in vivo is highly complementary to
pathology by providing tools to aid the identification of the earliest stages in the
pathological processes that end in the development of visible WMH.46 The increase
in MD in NAWM seen even in the presence of the mildest WMH burden, and the fact that
this parameter also provides excellent discrimination between both tissue types on
ROC curves, suggests that altered mobility of interstitial fluid occurs earlier in
the developing pathology of WMH in the aging brain than do changes in myelin (MTR),
axonal integrity (FA), or total water content (T1). However, further work relating
these imaging biomarkers to histopathological findings, especially at early stages
in disease, and in longitudinal studies, is required to understand fully the pathological
processes that are responsible for white matter damage within and around the WMH.
Limitations of Imaging Approaches to Determining WMH Pathophysiology
MRI processing methods for WMH detection and differentiation from artifacts, co-registration
of different types of images, and image processing methods such as bias field correction,
affect measurements of lesions and brain tissue parameters. The full analysis of these
limitations is beyond the remit of this paper, but these limitations are important
but rarely mentioned in imaging papers so are discussed briefly here for completeness.
Indeed the full scope of the effects of these limitations is only just beginning to
be understood by the clinical research community.
Methods to quantify WMH volume are evolving rapidly. Many were developed for use in
one study, reflect the particular subject, image acquisition characteristics and WMH
burden of that study population. Few, if any, are in use in clinical practice. Many
research groups develop their scanning protocol and WMH segmentation approach and
apply it to subsequent studies, without further validation. The protocol may not be
changed or improved partly due to lack of availability of source codes and/or software
and because of a natural desire to minimize the effect of protocol changes on measured
variables. Most studies use a semi-automatic thresholding approach to identify WMH
on FLAIR images. These threshold selection methods may then be applied in different
studies assuming that the reliability will not change, which is unlikely to be correct
as FLAIR images are not quantitative and the signal to noise ratio may not be the
same. Thresholding on a single image such as FLAIR requires individualized thresholds
(“one size” does not “fit all cases”); multispectral approaches using several sequences
combined are more accurate.133 A full description of the WMH volume measurement methods
used in different studies to date is beyond the scope of this paper but is available
upon request.
WMH volume measurement is time consuming, hence larger studies seek automated approaches
but it is not known whether all studies check each case individually for the accuracy
of segmentation, or the rigor with which this is performed where done. However, there
is as yet no automated approach that can identify WMH accurately without any human
input – all require visual assessment and manual correction, particularly in populations
with advanced age or stroke where the brains are more likely to be abnormal (Figure
8) and features of similar signal characteristics like stroke lesions, if not excluded,
will distort the WMH volume measurements with subsequent alterations on the outcome
of the study.134
Figure 8
Some factors which confound the delineation of WMH. FLAIR axial images from a patient
4 days after a stroke (A) and a year later (B): while the boundary of the periventricular
WMH can be well perceived on the baseline scan, it coalesces with the large infarct
after a year (yellow arrow) making it impossible to quantitatively determine accurately
the increase in the stroke lesion separated from the increase in the WMH. Different
patient (C) has a recent small subcortical stroke lesion arrowed in yellow, and artifacts
arrowed in blue; (D) a lacune (yellow arrow) confluent with the periventricular WMH,
real thalamic hyperintensity arrowed in magenta and thalamic artifact arrowed in blue
create difficulties for measuring the WMH. FLAIR indicates fluid attenuation inversion
recovery; WMH, white matter hyperintensities.
Artifacts that mimic WMH (to a computer algorithm) are numerous (Table 1) and require
manual editing.135,136 Guidelines exist to differentiate WMH from artifacts,135,137
but these lack uniformity and consensus. Some class periventricular hyperintense thin
lines around the lateral ventricles as artifact,135,139 but others found a high correlation
between signal intensity levels and width of the periventricular WMH, total WMH volumes,
periventricular WMH severity, vascular risk factors, and stroke,76 indicating that
periventricular WMH are true tissue abnormalities and should not be considered artifacts.78
Table 1
Common Artifacts That Confound the Identification of WMH on FLAIR MRI
Artifact Type
Artifact Location
Characteristic Effect
References
CSF flow
Fourth ventricle, aqueduct, cistern ventral to mesencephalon
Hyperintense tissue surrounding these structures; periventricular hyperintensity with
constant gradient
136,138–145
CSF flow
Bilateral sylvian fissures and insular cortex
Uniform hyperintensity along the external capsule
136,146,147
CSF flow
Third ventricle and Fornix
Hyperintense tissue surrounding the third ventricle and fornix; diffuse periventricular
hyperintensity gradually decereasing in strength towards the antero-medial thalamic
nucleus and internal capsule
136,140,142,144,147
CSF flow
Lateral ventricles
Hyperintense cluster near the walls of the anterior horns of the lateral ventricles
136,140,142,144,147
CSF flow
Choroid plexus
Small hyperintense cluster above the temporal horns of the lateral ventricles
144
Corticospinal tract
Pathways of the corticospinal tracts
Bilateral and symmetric hyperintensities with similar signal level in the junction
between the internal capsule and the medial thalamic nucleus
147
Magnetic field susceptibility
Amygdaloid nucleus and anterior temporal poles
Wide hyperintense clusters covering the amygdaloid nuclus and the anterior temporal
poles
136,145,148
Magnetic field susceptibility and CSF flow
Cingulate gyrus
Hyperintense tissue along the longitudinal cerebral fissure on the anterior section
and on the territory of the anterior cerebral artery
136
Patient motion (eye movement)
Around the eyes
Ghosting effect on tissue around the eyes
136,143
Patient motion (Head movement)
Throughout the whole image
Concentric hyperintense rings that distort/mask real WMH
136,138,148,149
Image reconstruction (Gibbs ringing)
Throughout the brain accentuated near the skull borders
Concentric hyperintense rings that follow signal intensity borders and distort/mask
small WMH
145
Blood flow
Near the sinuses and main arteries
Small linear hyperintensities near the sinuses and carotid arteries
145
Note, only publications that were dedicated to analyse and describe these artifacts
on FLAIR MRI and/or propose methods to compensate them are referenced. CSF indicates
cerebrospinal fluid; FLAIR, Fluid Attenuated Inversion Recovery; MRI, magnetic resonance
imaging; WMH, white matter hyperintensities.
Differentiating WMH from other SVD lesions such as perivascular spaces6,8,150 and
lacunes6 may be difficult. Differences in magnet strengths can lead to inconsistent
assessment of WMH, eg, subtle tissue changes in normal-appearing white matter could
go unnoticed at 3 T due to noise151,152 or vice versa (Figure 9).153 Reported mean
intra- and inter-scanner coefficients of variation in automatic volumetric measurements
of brain structures range from 0.87% to 15.1% (median 4.80%). Only limited information
is available for the range of values for direct comparisons of WMH assessed at 1.5
and 3 T.153,154
Figure 9
Three different patients, but with similar amounts of WMH, scanned with slightly different
slice thicknesses and at 1.5 or 3 T, illustrate the resulting differences in signal
of WMH and normal tissues and hence the need for changes in tissue and WMH volume
measurement methods. Slice thickness 1 mm2 (A), 0.479 mm2 (B), and 1 mm2 (C) and MR
scanner field strength—1.5 T (A and B) and 3 T (C) affect image appearance. Alterations
in imaging parameters influence the image appearance and hence the performance of
any software used to measure the WMH as all software ultimately rely on differences
in signal characteristics between the tissue and lesions of interest. MR indicates
magnetic resonance; WMH, white matter hyperintensities.
Image registration methods impose limitations, eg, native versus standard space, atlases,
and registering structural to DTI or other echoplanar images, all can distort lesion
location and size. In one study of recent small subcortical infarcts of <1.5 cm diameter,
we found that the volume of some acute lesions was more than doubled if measured after
registering the image in standard space against a routinely available atlas, rather
than performing the measurement in native space (ie, on the patient’s own brain scan)
prior to registration. Mapping the images of interest against brain atlases allows
analysis of lesion distribution (eg, lobar distribution, vascular territories, etc),
but relies on accurate registration of a brain template to the brains of the subjects
in the population of interest. While state-of-art non-linear registration methods155
have demonstrated very good performance, these have yet to be tested in SVD or aging
studies. This is important as it may influence perceived lesion or tract location
and represents a complex challenge that needs to be addressed. In general, atlases
should be used that are as representative as possible of the age range and population
of interest. Unfortunately, no atlases as yet include structures that are typically
affected by WMH, such as the major arterial “borderzones,”156 but variation in vascular
territories may account for some variation in WMH burden.157
Irrespective of the method used to quantify WMH, one factor that affects the assessment
of WMH is the subtle variation in the radio-frequency magnetic field known as magnetic
field inhomogeneity. Image processing methods that correct for magnetic field inhomogeneities
have been used in some studies.158–162 Some of these methods are scanner and protocol-specific159,162
while others are part of publicly available software where the algorithms assume that
the software is applied to brain images with only voxels containing CSF, gray matter,
and white matter, after all other tissue types are removed (http://brainsuite.org/processing/surfaceextraction/bfc/,
http://bioimagesuite.yale.edu/manual/guide/correction.aspx). If applied to FLAIR images
with WMH, these tools may try to remove the WMH or focal infarcts, which are interpreted
as inhomogeneities (Figure 10). Hence, these methods should be avoided until it is
clear that they do not simply remove the WMH or alter their apparent distribution.
Figure 10
Effect of incautious use of bias field correction on cortical infarct and WMH. The
left image shows the original MR FLAIR image with a right frontal cortical infarct
(arrow) and numerous WMH. Following bias field correction (left) the frontal infarct
appears much smaller, some of the WMH have become more (eg, posterior areas) and some
now appear split into smaller components (mid centrum semiovale) as the bias field
smoothing software tries to “even out” the distribution of signal intensities across
the image. FLAIR indicates fluid attenuation inversion recovery; MR, magnetic resonance;
WMH, white matter hyperintensities.
Two final features of WMH that have been overlooked to date are (1) the presence of
ill-defined subtle hyperintense white matter that are non-continuous and diffuse,
with varying erratic intensity patterns emerging from the lateral ventricle walls,
sometimes known as “dirty white matter”163 and which may be an indicator of pre-lesional
changes (Figure 5); and (2) the fact that WMH can get smaller or disappear rather
than continuously enlarging (Figure 1). “Dirty white matter” is probably a real feature
indicating subtle tissue damage, as suggested by the signal change in Figure 5. It
will influence where the boundary of the WMH is set and hence associations with other
MR parameters like FA and MD.
Figure 11
Example of disappearing white matter hyperintensities. Left, scan at presentation
with small subcortical infarct. Right, scan 1 year later. Inset image is diffusion
image from original presentation showing small subcortical infarct that precipitated
medical consultation. The images were obtained on the same scanner, using the same
sequence with careful quality assurance maintained throughout. Note the lesions have
become smaller and less white at 1 year compared with presentation.
That WMH can reduce or disappear should not be a surprise if we accept that much of
the signal change on FLAIR or T2 is due to shifts in water content and not just permanent
myelin loss or axonal damage. WMH regression was noted in only one previous published
case.164 Most longitudinal studies report “no change” or “progression” of WMH. Three
studies mentioned “minor regression” in some patients, but classified these as “no
progression” without further exploration,165 attributed reductions in WMH volume to
measurement variability,166 or reported (small) negative volume changes without comment.167
We have noted reductions in WMH volume as well as increases, eg, about a third of
200 patients who presented with minor stroke had a small reduction in WMH volume on
repeat MRI a year after stroke (in submission). In a few cases, the changes were more
obvious (Figure 1). The reasons for reduction in WMH are unclear but might relate
to improved vascular risk factor control. If some WMH are areas of tissue edema, then
reduction in tissue edema would both reduce WMH and decrease brain volume. Indeed,
in patients with CADASIL new WMH were associated with subtle increased brain volume.168
Finally, fluctuations in WMH and white matter subvisible damage might account for
the return to normal cognition seen in a few studies in patients with mild cognitive
impairment.2 Although these fluctuations have been attributed to recovery from depression,
delirium, heart failure, or autoimmune disorders, the improving cognition could equally
be attributed to resolution of transient fluid-related white matter damage before
permanent axonal injury or demyelination has occurred.
Implications for Clinical Practice
There is strong evidence that WMH are clinically important markers of increased risk
of stroke, dementia, death, depression, impaired gait, and mobility, in cross-sectional
and in longitudinal studies. They associate with brain damage such as global atrophy59
and other features of small vessel brain damage,32 with focal progressive visible
brain damage,55,57 are markers of underlying subvisible diffuse brain damage,19,98,109,119
and predict infarct growth and worse outcome after large artery stroke.60,61 They
could be considered as the neuroimaging marker of “brain frailty.”169
However, they should not be viewed only as “untreatable” or “permanent”: in vivo imaging
indicates that water shifts and water content are prominent and represent early changes
in WMH. Given the associations of WMH with traditional vascular risk factors, it is
reasonable to manage risk factors according to current guidelines until further data
from randomized controlled trials are available. Future trials should assess the effect
of risk factor reduction by age group because common vascular risk factors may act
differently at different ages. For example, blood pressure reduction in older people
may need to work to less stringent target levels than in younger people because some
older people with less reactive vasculature may become more dependent on blood pressure
for cerebral perfusion than in subjects whose vasculature retains its elasticity and
responsiveness. Similarly, cholesterol reduction may be relatively more important
in older people to protect the brain than in younger people where its main action
may be on the heart. These speculations are based on emerging observations (eg, and
the PODCAST trial, paper in preparation) and require careful testing in future trials.
Given the lack of readily available computational analysis methods, and the simplicity
and durability of deriving visual WMH scores,170 it is reasonable to use visual scores
in clinical practice – at least these would provide some more precise quantification
than purely descriptive reports. The addition of imaging methods such as DTI to routine
clinical assessment of patients with minor stroke, cognitive, or aging would facilitate
quantification of mean diffusivity or fractional anisotropy in normal-appearing white
matter to indicate subvisible brain damage.
Implications for Research
Research should target the enormous unfinished agenda that constitutes brain damage
represented by WMH and diffuse small vessel disease. There should be more imaging-neuropathology
analyses of individual lesions in humans and in relevant experimental models. There
should be more awareness that WMH are heterogeneous, can diminish as well as increase,
that they represent a tip of the iceberg in terms of damage to the brain and lead
to progressive global brain damage through local and remote brain tissue damage. More
longitudinal multimodal imaging studies in well characterized subjects from middle
to old age, from different ethnic, socioeconomic, and clinical backgrounds, are required
to fully understand what influences the varying patterns of WMH and their associations
with cognition, gait, mood, and vulnerability to stroke. Multimodal imaging is needed
to assess parameters like MD, FA, MTR, T1, perfusion, etc, contemporaneously in normal
and abnormal white and gray matter. Better methods to measure WMH are needed that
are reliable and efficient, with minimal human input, for large population studies.
These need to be more sophisticated to detect subtleties of WMH, change over time
and response to treatment. These should be able to detect differences in intensity
of WMH, not just size, and to recognize underlying general changes in brain white
matter with increasing age and by exposure to risk factors. Better measurement methods
are essential for new trials to test risk factor modification, repurposed drugs, new
drugs, lifestyle modifications, etc, to prevent progressive brain damage from WMH.
Finally, rather than focusing clinical trials on “Alzheimer’s disease” or “Vascular
cognitive impairment”, clinical diagnoses which likely represent mixtures of pathologies,
perhaps it would be better to focus on intermediary markers of brain damage, such
as WMH, which likely reflect more specific pathologic mechanisms.
Sources of Funding
The authors acknowledge institutional research funding from The Row Fogo Charitable
Trust (salary support for Valdés Hernández) and Age UK (salary support for Muñoz-Maniega).
In addition, the concepts developed in the review derive from research funded by numerous
sources over many years but primarily including the Scottish Government Chief Scientist
Office, the Wellcome Trust, the UK Medical Research Council, the Scottish Funding
Council SINAPSE Initiative, The Stroke Association, and Chest Heart Stroke Scotland.
Disclosures
None.