Key words
perivascular spaces - glymphatic system - MR imaging
Introduction and historical context
Introduction and historical context
In recent years the significance of perivascular spaces (PVS) has increasingly become
a topic of clinical and scientific interest. However, the first descriptions of PVS
date back over 150 years. They were first observed and mentioned for the first time
in 1843 by the French physician Durand-Fardel in his paper “Traité du ramollissement
du cerveau”. In an autopsy report he described a “état criblé” consisting of sieve-like
small holes in the brain, primarily in the white matter and in the corpus striatum
[1] ([Fig. 1]). These spaces were initially called Virchow-Robin spaces in reference to the two
pathologists Virchow and Robin. In 1851, in the article “Ueber die Erweiterung kleinerer
Gefäße”, the German pathologist Virchow described a space in the vascular wall between
the tunica intima/media and the tunica adventitia and called it “dissecting ectasia”
[2]. In 1859, in his article “recherches sur quelques particularités de la structure
des capillaires de l’encéphale”, the French anatomist Robin described a closed space
within the tunica adventitia [3]. These spaces were described as non-pathological findings in both publications.
Today, PVS are defined as fluid-filled spaces surrounding small penetrating cerebral
vessels.
Fig. 1 “état criblé”: axial T2-weighted sequence with multiple enlarged perivascular spaces
at the level of the basal ganglia.
Particularly in connection with the model of cerebral clearance by the glial-lymphatic
(“glymphatic”) system [4], PVS increasingly became the focus of scientific interest. It is hypothesized that
these spaces play a central role in the drainage system of the central nervous system
and are thus involved in the removal of metabolic waste products. It is therefore
also postulated that they could serve as biomarkers for brain health and disease.
A connection with neurodegenerative diseases like Alzheimer’s or Parkinson’s disease
is suspected. Magnetic resonance imaging (MRI) as a noninvasive imaging method allows
detailed in-vivo visualization of components of the glymphatic system like the perivascular
spaces. The goal of this review is therefore to provide an overview of MR imaging
of perivascular spaces and to give a summary of the various assessment methods in
order to determine relevant points for the clinical routine.
MRI and characteristics of perivascular spaces in various sequences
MRI and characteristics of perivascular spaces in various sequences
PVS can be visualized in almost every cranial MRI examination in routine sequences
due to the constant improvements in the spatial resolution of modern scanners [5].
In particular, T2-weighted (T2w) TSE sequences provide a high contrast-to-noise ratio
(CNR) for the visualization of perivascular spaces [6]. Additional sequences can further facilitate the differentiation between perivascular
spaces and their differential diagnoses ([Tab. 1]). In the region of the basal ganglia (BG), the PVS run along the lenticulostriate
vessels. Continuity to the basal cisterns can be observed. In the centrum semiovale
(CSO), they begin several millimeters below the cortex [6]
[7]. Tractography with diffusion tensor imaging shows that the PVS and their central
vessels run parallel to the axons of the white matter [8]. In T2w and balanced steady-state free precession (bSSFP) sequences, ependymal openings
and thus direct connections to the ventricular system could be detected in the periventricular
PVS. The openings are largely located in the basal lateral wall of the anterior horn
of the lateral ventricle [9]. In the midbrain, they are seen on highly T2w images at the pontomesencephalic junction
mainly between the cerebral peduncles and the substantia nigra. At the diencephalic-mesencephalic
junction, PVS are mainly located dorsal to the cerebral peduncles and are significantly
smaller [10] ([Fig. 2a]). To date, studies have only been able to examine some PVS within the cortex ex
vivo using 7 T devices [11]. They may become dilated in histological specimens during the fixation process and
be significantly smaller in vivo than in white matter so that the spatial resolution
of MRI or the CNR is too low.
Table 1
Characteristics of perivascular spaces in different MRI-sequences.
|
Sequence
|
Characteristics of perivascular spaces
|
Comment
|
|
T2w
|
Hyperintense, clear margins
Oblong when the course is parallel to the slice/round-oval when the course is perpendicular
to the slice
|
Standard sequence for visual assessment in axial view
|
|
T1w
|
Hypointense, clear margins
|
|
|
EPC
|
Hypointense
|
Delimitation from vessels without surrounding perivascular spaces; these are hyperintense
here
|
|
FLAIR
|
Hypointense, clear margins
|
Delimitation from differential diagnoses (e. g., WMH, gliosis, and small lacunae)
|
|
TOF, SWI/T2*
|
No signal in the TOF sequence
No SWI changes
|
Relation to arterial and venous vessels
|
|
DWI
|
No diffusion restrictions
|
Calculation of the DWI-ALPS index
|
|
DTI
|
|
Tractography for correlation to the course of the perivascular spaces
Calculation of the DTI-ALPS index
|
Fig. 2 a axial T2-weighted sequence at the level of the midbrain. The arrows indicate perivascular
spaces on both sides. b coronal T2-weighted sequence at the level of the basal ganglia. The arrow indicates
focal dilatation of a perivascular space. c axial T2-weighted sequence at the level of the basal ganglia. The white arrow shows
a round-ovoid course. The black arrow shows a linear course. The white arrowhead indicates
a central vessel in a perivascular space.
Depending on the image plane, PVS appear either linear or oval since they follow the
course of the penetrating vessels [12], with the possibility of focal dilation and fluctuations in caliber [6]
[7] ([Fig. 2b]).
It has been reported that PVS usually have a spatial relationship to the perforating
arteries. Depending on the proportion between vascular diameter and the fluid filling
the PVS, the central vessels within the PVS can sometimes be visualized ([Fig. 2c]), e. g., on T1-weighted images after contrast administration [13] or in TOF angiography after contrast administration [14]. However, PVS are also seen without a corresponding artery and arteries are seen
without corresponding PVS [7]. Some studies with an additional T2*-weighted or gradient echo/susceptibility-weighted
(SWI) sequence also show that they do not or only minimally spatially correlate with
the course of venous vessels [7]
[15]. It is not yet clear whether this is due to a smaller size or different structure
of the perivenous PVS. It is also conceivable that these are collapsed due to the
lower pressure so that they are usually not visible. However, complete absence of
perivenous PVS is also conceivable. Though, other studies with coregistered T2w and
SWI sequences were able to show a perivenous location of at least a small portion
of perivascular spaces [16].
Visually the signal of PVS is isointense with respect to CSF and is therefore completely
suppressed in FLAIR sequences. In contrast to white matter hyperintensities (WMH),
they are hypointense and do not have a hyperintense margin unless they are not within
such a lesion [17]. However, the measured signal intensity of PVS has a low hyperintensity compared
to CSF which indicates different compositions [18] that have not yet been sufficiently examined. Further studies also show signal differences
between perivascular spaces of different regions thus also indicating variations in
fluid composition [19].
After intravenous administration of gadolinium-based contrast agent, PVS along the
perforating arteries and in the BG show significant enhancement on highly T2w FLAIR
images after approximately three or four hours [20]
[21]. In contrast, no comparable effect was observed in the PVS of the white matter in
the subinsular region [19]. One possible reason for this could be different drainage routes and functions between
PVS in various regions.
The combination of MRI sequences with different weightings could further increase
the visibility of PVS in the future. The enhanced perivascular space contrast (EPC),
a combination of T1w and T2w sequences, was established for this. Due to the inverse
signal behavior of fluid here, a combination of these sequences results in greater
contrast between PVS and tissue. A further advantage of this method is the differentiation
of vessels with and without surrounding PVS. Without PVS, vessels are hypointense
in the T2w sequence and hyperintense in the EPC. With surrounding PVS, vessels are
hyperintense in the T2w sequence and hypointense in the EPC [22].
To determine the activity of the glymphatic system, Taoka et al. evaluated the diffusion
tensor image analysis along the perivascular space (DTI-ALPS). The ALPS index is the
ratio of the diffusivity along the PVS to the diffusivity along the projection and
association fibers of a slice [23].
The same group also used their method for simple diffusion-weighted imaging. ADC maps
can be generated from DWI sequences on the axes described above. The DWI-ALPS index
can then be calculated in the same slice [24].
Perivascular spaces over the course of a lifetime and in health and disease
Perivascular spaces over the course of a lifetime and in health and disease
Studies show that PVS become more visible with increasing age [7]. Whether there are also changes in diameter is still a topic of discussion. The
underlying mechanisms are debated in the literature. Studies show that there is a
hereditary component [25]. It seems that men have a higher number of perivascular spaces [26]. Prospective studies were able to show that PVS are associated with age, lacunar
infarcts, and white matter lesions. Therefore, they are increasingly being seen as
further markers of cerebral small vessel disease [27]. A positive correlation between the number of PVS and white matter lesions is also
seen in cerebral autosomal dominant arteriopathy with subcortical infarcts and leukoencephalopathy (CADASIL) as a monogenetic and early-onset form of cerebral small vessel disease
[28]. Various PVS patterns could be detected in the case of spontaneous intracerebral
bleeding. Thus the prevalence of PVS in the BG appears to be associated with hypertensive
bleeding and the presence of PVS in the CSO appears to be associated with bleeding
in cerebral amyloid angiopathy [29]. A further genetic disease involving dilation of the PVS is, for example, mucopolysaccharidosis,
which as a lysosomal storage disease is associated with manifestations in the CNS
among other things. PVS are seen here primarily in unusual locations, e. g., along
the junction zone between gray and white matter. However, there is no correlation
with cognitive performance [30]. Even if the available data is still contradictory, the burden on PVS seems to correlate
with neurodegenerative diseases like multiple sclerosis, Alzheimer’s disease, and
Parkinson’s disease. The underlying mechanisms continue to be the subject of research.
The correlation between PVS and cognitive performance is also controversial in the
current literature.
Visual assessment
In relation to the visual scoring of PVS, the increasing number of publications has
resulted in a corresponding increase in the number of assessment methods.
A classification based on the distribution of PVS includes three characteristic locations:
Type I along the perforating lenticulostriate arteries in the BG, type II along the
perforating subcortical white matter arteries above the convexity, and type III in
the midbrain at the pontomesencephalic junction [31]. A type IV was recently suggested for the subcortical white matter of the anterior
superior temporal gyrus [32]. However, PVS can occur anywhere in the brain parenchyma.
Grading of the PVS based on size is sometimes used. To date, there is no standardized
definition here. In an early study, grade 1 was defined for a diameter of < 2 mm,
grade 2 for a diameter of 2–3 mm, and grade 3 for a diameter of > 3 mm [33]. Other studies define a lower limit of 1 mm [34] or 5 mm [31]. However, it is largely agreed that PVS < 3 mm are not considered dilated. In the
case of larger PVS, differential diagnoses like lacunar infarcts must be taken into
consideration [34]
[35]
[36]. In addition, the literature includes cases with so-called giant tumefactive perivascular
spaces with dimensions of greater than 1.5 cm ([Fig. 3a, b]). However, in spite of their size, they can be asymptomatic and an incidental finding
[38]. They can also be associated with numerous symptoms and cause focal mass effects
and even obstructive hydrocephalus [37]. In rare cases dilated PVS can be reversible, either in association with the resection
or regression of cerebral lesions or spontaneously [13].
Fig. 3 giant tumefactive perivascular spaces a axial T2-weighted sequence. b sagittal T1-weigted sequence.
More recent studies define dilation of the PVS in the supratentorial white matter
not based on diameter but rather on shape. The only differentiation is “not dilated”
when they have an even and linear course and “dilated” in the case of visible irregularities
or focal ectasia. According to the authors, this classification should have the advantage
that it is easier to use and the size does not need to be normalized to the individual
brain volume. Therefore, grading is less dependent on technical parameters [36].
To evaluate the quantity of dilated perivascular spaces, they can be counted in the
various regions and applied to the rating.
The most established scale in studies with visual assessment of PVS in BG and CSO
ranges from 0 to 4 with 0 = no PVS, 1 = 1–10, 2 = 11–20, 3 = 21–40, and 4 = > 40.
This scale is used for both hemispheres. In the case of asymmetry, the higher score
is used ([Fig. 4]) [39]. For the mesencephalon, scores of 0 = not visible and 1 = visible are used [39].
Fig. 4 Visual score for rating of perivascular spaces at the level of the basal ganglia
(upper row) and at the level of the centrum semiovale (bottom row) on axial T2-weighted
images. All images were acquired with a 3T-MRI scanner (MAGNETOM Vida 3 T, Siemens
Healthcare, Erlangen, Germany).
The unclear number of assessment methods results in limited comparability between
the studies to date. The goal of the “Uniform Neuro-Imaging of Virchow-Robin Spaces
Enlargement” (UNIVERSE) consortium therefore is to harmonize the visual assessment
of PVS and to make it more time efficient. Four relevant brain regions were defined:
the CSO, the BG, the hippocampus, and the mesencephalon. For the first two regions,
the perivascular spaces are counted in a predefined plane, while all perivascular
spaces are counted in the last two perivascular spaces [4].
As described above, there is increasing evidence that both PVS and WMH are biomarkers
for cerebral microvascular integrity and they are correlated with each other [26]. Initial approaches consider the two parameters together with a combined score.
The ratings for perivascular spaces of the BG, the CSO, and the midbrain and the Fazekas
score [41] are added [42].
Even if studies report good interobserver reliability of their visual scores, manual
assessment is very time-intensive and subjectivity of the observers cannot be eliminated.
The scores are often difficult to compare and are thus not suitable for longitudinal
studies. It must also be clarified whether the continuously improving technology with
higher field strengths and better resolution changes the detectability of PVS. Therefore,
it is necessary to question whether the visual rating scales used to date also need
to be adapted.
Automatic evaluation
A growing number of studies are addressing the semi-automatic or fully automatic analysis
of PVS. The increasing number of studies has resulted in an increasing number of assessment
methods. Therefore, there are multiple filter and segmentation methods that can potentially
be applied to PVS. Initial approaches explicitly for segmenting PVS were seen in the
early 2000 s when Descombes et al. first used a “marked point process framework” optimized
via “reversible jump Markov chain Monte Carlo algorithm” to detect multiple small
lesions [43].
These methods can roughly be divided into image processing for increasing the visibility
of PVS and methods using artificial intelligence, deep learning algorithms in this
case, with some overlap. Preprocessing to create a mask by segmenting gray and white
matter and CSF and if necessary any subcortical structures is usually first performed.
For image processing, subgroups can be defined like intensity-based filter methods,
which assign voxels to the PVS on the basis of intensity differences, and the Vesselness
filter methods, which are used to identify tubular structures.
Early approaches were based on semi-automatic intensity-based segmentation in a predefined
slice through the BG [44]. The following studies performed edge detection and processing of voxel-specific,
high positive spatial gradients using the MATLAB “imgradient” function [8] in various regions [45].
A further study developed a method of Voxel-based multimodal autoidentification of
PVS that performs automatic analysis and allocation of Voxels based on normalized
signal intensities of the gray and white matter and the ventricular system in spatially
coregistered T1w, FLAIR, T2w and PDw sequences [46]. An approach based on this by the same work group used only T1w and FLAIR sequences
in order to reduce the acquisition time and to minimize the effect of movement between
coregistered sequences [47].
Other studies used Vesselness filter methods like the Frangi filter [48]. Since the goal of this method is to show tubular structures, it can be effectively
used for the visualization of PVS. PVS were defined here as tubular structures with
a length of 3–50 mm [49]. Further studies using the Frangi filter with slight changes have been performed
[6]. The filter can be used for T1w, T2w, and EPC images. The threshold values need
to be adapted to the applied image and they must have the same resolution. The PVS
can then be counted based on the resulting mask [22].
In addition to the Frangi filter, there are other filters like the Jerman filter and
the RORPO filter based on morphological path operators. Which filters are used, for
example in studies, depends on whether a high sensitivity or specificity or a high
positive or negative predictive value is targeted [50].
A combination of the two methods mentioned above is also possible. For instance, Vesselness
filters were applied in a study to highlight all vessel-like structures on T2w images.
To automatically detect the PVS in a predefined slice, probability maps based on a
binary kNN classification (k-nearest neighbor) were used [51].
With the further development of artificial intelligence, DL algorithms were also developed
for the segmentation of PVS. However, a large number of manually annotated data sets
are typically needed The visual PVS scores described above are ultimately usually
used as a ground truth. However, a certain amount of data sets are still needed for
training and validation. For example, “random forest” models [52]
[53] or “convolutional neural networks” (CNN) [11]
[54]
[55]
[56] are used as the basis for such approaches.
Since as described above the 4 main regions are the midbrain, hippocampus, BG, and
CSO, one study designed an automatic quantification method for these regions. The
approach used in this method is based on the CNN regression, which was designed for
the detection of small objects within a region of interest. Due to the potentially
different shapes and differential diagnoses of PVS in these areas, the CNN was trained
specifically for these regions [57].
The DL methods can also be combined, e. g., with Frangi filter and CNN-based classification
via 3 D deep CNN [25].
Automatic methods are capable of detecting a high number of even very small PVS thus
allowing deeper analyses. Therefore, quantitative parameters like an exact count or
the diameter of PVS can be determined [6]. Furthermore, the length and individual PVS volume can be calculated, which can
then be corrected as a volume fraction for the entire brain volume [8]
[45]. Morphological differences like linearity [46] and shape between regions can also be examined potentially more objectively.
Automatic PVS assessment is still in the initial phase and it is difficult to compare
the different approaches due to the various sequences and devices with different field
strengths. To actively promote research in this field, the Medical Image Computing
and Computer-Assisted Interventions Association (MICCAI) started the Vascular Lesions
Detection (VALDO) Challenge in 2021. The detection of PVS is a subcategory: Better
comparability is desired since proposals are evaluated based on the same guidelines
and data sets. (https://valdo.grand-challenge.org/).
Summary
PVS are visible in practically every cranial MRI examination even using routine clinical
sequences. They are a normal finding in people without neurological abnormalities.
However, their clinical relevance is not yet fully understood. To date, there is only
minimal data from healthy persons to determine a cut-off between physiological and
pathological. There are numerous scoring systems for visual assessment. Automatic
segmentation is a promising approach for standardizing findings and ensuring comparability
between studies.
