Extent and Distribution of White Matter Hyperintensities in Stroke Patients
The Sydney Stroke Study
Background and Purpose— White matter hyperintensities (WMHs) on T2-weighted MRI are common in stroke patients and healthy elderly individuals. The detailed anatomical distribution of these lesions in stroke patients has not been examined.
Methods— A total of 112 stroke or transient ischemic attack patients and 87 matched control subjects from the Sydney Stroke Study underwent MRI scans that included a T2-weighted fluid-attenuated inversion recovery (FLAIR) sequence. WMHs were delineated from each FLAIR MRI by an automated method. Region of interest and voxel-wise statistical parametric mapping approaches were applied to examine the volume, distribution, and severity of WMHs of the patient and control groups, and subgroups with large or lacunar infarcts.
Results— Stroke subjects had significantly more WMHs than controls in all brain regions except the occipital lobe and in all arterial territories except the anterior callosal and anterior medial lenticulostriate. In the frontotemporal regions, average WMH volumes in patients were >3.5× those in controls. The total number of discrete WMHs was not different in the 2 groups, but stroke patients had more large (>20 mm) and high-intensity lesions. Subjects with lacunar infarcts had more WMHs than those with large infarcts, who, in turn, had more WMHs than control subjects. Lacunar infarction subjects had more WMHs than subjects with large thromboembolic or cardioembolic strokes. Those with anterior arterial territory infarction had more WMHs in the frontal regions. Subjects with single or multiple lacunes did not differ in volumes of WMHs.
Conclusions— Stroke patients have significantly more WMHs in nearly every brain region than healthy controls. Those with lacunar infarcts are particularly affected. WMHs represent a significant proportion of the ischemic lesion burden in stroke and transient ischemic attack patients.
White matter hyperintensities (WMHs) and basal ganglia hyperintensities are common on T2-weighted MRI in stroke patients1 as well as in healthy older individuals,2 with their extent being much greater in patients with cerebrovascular disease.1 The study of these hyperintensities is important for many reasons. First, they represent a significant proportion of the burden of pathology in the brains of stroke patients. Second, there is evidence that they are an independent predictor of future stroke.3 Third, they are associated with cognitive impairment over and above what can be accounted for by the infarction.4 In fact, vascular cognitive impairment (VCI) correlates better with WMHs than stroke volume in many of these patients.5
In spite of their obvious importance in patients with cerebrovascular disease, WMHs in this group have received inadequate attention. The topographic distribution of WMHs in the brains of stroke patients has not been well studied compared with healthy control subjects. Relevant questions in this regard are: Are there more WMHs in periventricular as well as deep white matter regions? Is the frontal white matter preferentially affected, as would be predicted from the nature of the cognitive impairment? Further, it is not known whether the excess is only in subjects with lacunar infarcts or if it also occurs in those with large cortical infarcts. Are patients with multiple lacunes different in their burden of WMHs? The latter is important because it has been argued that the etiology of single lacunes is often different from that of multiple lacunes.6 It is also not known whether stroke patients have more focal WMHs or an excess of large confluent WMHs, or both.
In this study, we attempt to address some of these questions in a sample of subjects with cerebrovascular disease and healthy comparison subjects. We used a newly developed automated method of quantitation of WMHs and mapped them topographically in 3D brain space.
The study sample was drawn from the ongoing longitudinal Sydney Stroke Study of consecutive admissions for ischemic stroke or transient ischemic attacks (TIAs) to 2 teaching hospitals affiliated with the University of New South Wales, recruited between May 1997 and June 2000. An ischemic stroke was defined as “rapidly developing clinical signs of focal (or global) disturbance of cerebral function, with symptoms lasting 24 hours or longer, with no apparent cause other than of vascular origin” in which a brain computed tomography (CT) or MRI scan does not show intracranial hemorrhage. TIA was defined as sudden focal neurological deficits lasting <24 hours and not associated with cerebral infarction on CT scan. The diagnosis of stroke or TIA was independently made by 2 neurologists. Other inclusion criteria were age between 50 and 87 years, good knowledge of English, absence of severe aphasia (<3 on the Aphasia Severity Rating Scale of the Boston Diagnostic Aphasia Examination), and ability to give informed consent. Exclusion criteria included hemorrhagic stroke, persistent impairment of consciousness (>7 days) after stroke, concomitant central nervous system disease known to affect cognition, concomitant medical disease that is judged to possibly limit life expectancy or affect cognition secondarily, mental retardation, alcohol dependence, or contraindications to MRI.
The control group comprised volunteers (n=110) from the same neighborhood, recruited from community organizations, and matched group-wise for age and gender with stroke patients. They were screened for absence of stroke, cognitive impairment, and psychiatric disorder during history and examination, with the same exclusion criteria as the stroke patients.
Subjects received a medical, neurological, and psychiatric assessment by a physician and a detailed neuropsychological assessment by a clinical psychologist. A diagnosis of vascular dementia, VCI, or no cognitive impairment was made by consensus by the research team at a meeting attended by a neuropsychiatrist, an old age psychiatrist, a neurologist, and 2 or more clinical psychologists.5 The stroke diagnoses according to the Oxfordshire Community Stroke Project (OCSP) classification7 and National Institute of Neurological Disorders and Stroke (NINDS) clinical classification8 were also finalized at this meeting.
Magnetic Resonance Imaging
MRI was performed in 112 patients 3 months after the stroke/TIA and 87 controls. After excluding those with incomplete MRI data or a poor quality MRI, 107 stroke patients (TIA 17) and 82 control subjects were included.
All MRI examinations were performed on a 1.5-T GE Signa scanner (GE Medical Systems) using the following protocol: a scout midsagittal cut for anterior commissure–posterior commissure plane alignment; a T1-weighted fast spoiled gradient-recalled sequence 3D anatomical MRI, with 1.5-mm-thick contiguous coronal (repetition time [TR]/echo time [TE] 14.3/5.4 ms) slices; matrix size 256×256; field of view [FOV] 250 mm; in-plane spatial resolution 0.976×0.976 mm/pixel; and a T2-weighted fluid-attenuated inversion recovery (FLAIR) sequence MRI, 4-mm-thick coronal slices with no gap and matrix size 256×256 (TR/TE/inversion time 8900/145/2200 ms). There was an FOV of 250 mm and in-plane spatial resolution of 0.976×0.976 mm per pixel.
There were 2 main components to image analysis. The first was the automated detection, delineation, severity rating, and volumetric measurement of WMHs from FLAIR images. Details of the algorithm and correlations between the results from the automated method and visual rating are described previously.9 In brief, we constructed an age-specific FLAIR template in Montreal Neurological Institute (MNI)-space.10 Spatial normalization of the coregistrated FLAIR and T1-weighted MRIs was then performed using FLAIR template as the target. Then detection and grading of WMH from each normalized FLAIR image with T1-weighted image as reference were carried out. We visually inspected each WMH map generated by the computer algorithm and manually removed false classification of WMH from the map.
WMH maps thus generated were binary images, the voxel values of which indicated either the presence or absence of WMH on that location. Linear and nonlinear transforms were applied onto each individual MRI in warping them into MNI-space. The WMH thus measured is the relative WMH lesion load rather than the absolute volume. Removal of stroke infarcts that appeared to have similar signal intensities with WMH in FLAIR scans was done manually on the WMH map by referencing it to its corresponding 3D T1-weighted anatomical image to confirm the stroke infarct site.
The second main component was region of interest (ROI) and voxel-wise analysis of 189 WMH brain maps. Quantitative ROI analysis was used in investigating the WMH volume differences between the stroke and control groups in anatomical regions and arterial territorial partitions.9 Voxel-wise analysis was performed to generate statistical parametric maps in detecting white matter anatomical structures with significant differences between 2 groups. To prepare the images, we applied a Gaussian smoothing kernel (full width at half maximum, 10 mm) on the individual WMH map to increase the signal-to-noise ratio. The resulting blurred WMH map can be thought of as an estimate of the probability that the subject has a WMH at that location.
Because the abnormal white matter signal varied in its intensity, we categorized it into “low” and “high” intensity lesions.9 The former usually appears as a milky fuzziness, whereas the latter as a white opacity to the naked eye. Because the neuropathological validity of this distinction has not been demonstrated, we present the total hyperintense lesions for most analyses and additionally present the analysis for high-intensity lesions for the whole brain only. The number of nonconnected discrete WMHs was calculated automatically by a computer program that also estimated the diameter of each WMH, assuming it to be a sphere. Because FLAIR slice thickness was 4 mm, small WMHs may not appear bright enough to be detected by our algorithm. As the result, we found that small hyperintensities, such as those in the brain stem, were underestimated.
The scans were also rated visually by a trained rater with good inter-rater (intraclass correlation coefficients from 0.7 to 0.9 on various measures) and intrarater (intraclass correlation coefficients 0.8 to 0.9) reliability determined on 10 scans each. Periventricular hyperintensity and deep WMH (DWMH) hyperintensity were rated on a 0 to 3 scale modified from Fazekas et al,11 with the images displayed on a computer console. Brain infarctions were identified on T1-weighted images and categorized into large infarcts (cortical or subcortical, diameter >15 mm) and lacunar infarctions (single or multiple, 5 mm<diameter<15 mm). Infarction volume was determined for each infarct by manual tracing. Inter-rater reliability for these measurements was established on 10 scans (intraclass correlation coefficients >0.7).
Statistical tests were performed with SPSS version 12 (SPSS). General linear model (multivariate) was used in group comparisons of WMH volumes. Age, gender, and hypertension were controlled for as covariates in the models. Bonferroni adjustments were made in post hoc multiple comparisons.
During the duration of recruitment, 1050 patients with possible stroke or TIA were screened for suitability, and 252 who met inclusion and exclusion criteria entered the study. The major reasons for exclusion, in order of frequency, were: (1) critical medical condition; (2) refusal by subject or family member; (3) lack of competence in English; (4) severe aphasia; and (5) age >87 years. By the time of detailed assessment at 3 to 6 months after the cerebrovascular event, 42 were lost to follow-up (33 from withdrawal, 3 were deceased, 4 relocated outside Sydney, 1 was not contactable, and 1 had probable dementia), yielding 210 patients as eligible subjects. Additionally, 110 healthy controls were recruited. MRI scans were obtained for 199 subjects (112 stroke patients plus 87 control subjects). The sociodemographic and clinical characteristics of the subjects are presented in Table 1. Predictably, the stroke subjects had more risk factors for cerebrovascular disease. Subjects with (n=199) and without (n=163) MRI scans were compared regarding age, years of education, gender and Mini Mental State Examination (MMSE) scores,18 and no significant differences were noted, suggesting a lack of systematic bias in subject inclusion. The criteria for the subject exclusion of MRI scans were mainly refusal by the subjects, followed by implant such as pacemaker, claustrophobia, illness, nursing home residents, and small number of missing MRI scans from the MRI center.
The categorization of the strokes according to the OCSP7 and NINDS8 classifications is presented in Table 2. We found that 69.2% patients had brain infarcts visible on T1- and T2-weighted MRI scans 3 to 6 months after stroke, which were large infarcts in 36.4% (right-sided 25, left-sided 11, both sides 3), including 9.3% with large and lacunar infarcts, and only lacunar infarcts in 32.7%. Incidentally, 7 (8.5%) control subjects had MRI evidence of infarction, lacunar in 6, and large infarct in 1. In patients with lacunar infarcts only, the mean number of infarcts was 2 (range 1 to 6), and in those with large infarcts, the mean was 1.8 (range 1–5). The mean volume of lacunar infarcts was 0.59 mL, and that of large infarcts was 6.4 mL.
The mean (SD) volume of total WMH (mild and severe) for a stroke/TIA patient was 37.53 (38.44) mL, more than twice the average WMH found in a control subject, which was 17.65 (16.02) mL per subject, representing 6.35% and 2.98% of the total white matter for the 2 groups, respectively. Mean volumes of severe WMH were 13.50 (21.97) mL and 4.20 (5.99) mL per subject, or 2.28% and 0.71% of the total white matter, for stroke and control groups, respectively. The severity of WMHs in TIA subjects was in between controls and stroke patients, but because of small sample size, the TIA group was not significantly different from the other 2 groups.
Statistical probability maps for WMHs in both groups are presented in Figure 1. The value for each voxel in the 3D composite images represents the probability of being a WMH voxel in the cohort. Note that the maxima (of the voxels) are projected on to the orthogonal planes. The differences between the 2 groups are obvious visually. Although hyperintense signals on FLAIR images were present in all 189 subjects regardless of their group membership, the extent and distribution in the 2 groups varied considerably. The average WMH volumes were significantly greater in stroke/TIA than control subjects in nearly every ROI except occipital lobe, cerebellum, and brain stem. Because very few WMHs were detected in either cerebellum or brain stem, we will not discuss these 2 regions further. We divided periventricular region into the anterior horn, the posterior horn, and the periventricular body. The periventricular region was greatly affected in both groups, whereas the greatest differences between stroke/TIA and control groups were found in deep white matter regions, with relative increases of 358% in frontal, 362% in temporal, and 242% in parietal regions in the former. Because of the presumed ischemic origin of WMHs, we measured WMH volumes in different arterial territories. Except for the regions supplied by anterior callosal and anterior medial lenticulostriate arteries, all arterial regions were more affected in patients than control subjects. The boundaries of the ROIs are described previously.9
The relationship of each voxel between the stroke/TIA and control subjects’ WMH maps was examined with a general linear model with group membership as the independent variable and age and gender as covariates using SPM99. Of the 182 036 voxels (1.456 L) in the MNI-space thus examined, 26 843 voxels (0.215 L) appeared to have significantly greater probabilities of being a WMH for stroke patient than control subject (t>3.13; P<0.001 uncorrected), with no voxel showing the reverse. Considering the possible false activation rate of 0.001 resulting from multiple observations of 182 036 voxels, we may have ≈200 voxels (of 26 843 voxels) that were incorrectly highlighted. Figure 2 shows that significant WMH differences between the groups were mostly found in periventricular white matter, frontal lobe white matter, cingulate gyrus, and subgyral parietal and temporal lobes. The voxel with highest t value (t=5.96) was located in right frontal lobe.
In size, WMHs may appear as small punctate to large confluent areas. The total numbers of WMHs were 19.45 (8.23) and 19.69 (8.81) for patient and control groups, respectively. The numbers of punctate (<3 mm diameter) WMHs for stroke/TIA and control groups were 12.40 (6.28) and 12.46 (6.29), focal (3 to 12 mm) WMHs 4.63 (2.89) and 4.62 (3.12), and large (>12 mm) WMHs 2.43 (1.08) and 2.61 (1.28), respectively. None of the above showed any significant difference. Only 3 patient and 6 control subjects did not have large WMHs, and 87.8% of stroke/TIA and 85.3% of control subjects had at least 2 large WMHs. The discrepancy between the large group differences in WMH volumes and the group similarities in the number of discrete WMHs is explained by the more frequent observations of very large WMHs in the stroke/TIA group. There were a mean (SD) 1.31 (0.895) discrete WMHs, the diameter of which was >20 mm (523 voxels) in the stroke/TIA group, and 0.88 (0.935) in controls (t=3.22; df=187; P<0.002).
We examined the WMH volumes of stroke/TIA subcategories grouped according to OCSP classification and NINDS clinical classification. Results are shown in Table 3. Subjects with lacunar infarction had more extensive WMHs than those with thromboembolic or cardioembolic strokes. After Bonferroni adjustment for multiple comparisons, the WMH volumes of atheroembolic and lacunar subgroups were significantly different in the following regions: frontal (P=0.003); parietal (P=0.027); whole-brain mild and severe (P=0.016) and whole-brain severe only (P=0.018). There was no significant difference between atheroembolic and cardioembolic subgroups in any ROI. Subjects with stroke/TIA affecting the anterior circulation had more WMHs in the frontal and parietal white matter and had large “caps” bordering the anterior horns of the lateral ventricles. In those with posterior circulation strokes, the mean volumes of WMH in the cerebellum and brain stem were higher than those in the other subgroups but did not reach statistical significance.
The relationship of WMH with type of infarction was further examined by first comparing subjects with single lacune and those with multiple lacunes, and no significant differences in whole-brain WMH volumes were found (t=0.200; df=33; P=0.843). These subjects were treated as 1 group. General linear model was used in contrasting the WMH volumes in the groups of control with no infarction (CNI), control with lacunar infarction (CL), stroke with no infarction (SNI), stroke with large infarction (SS), stroke with lacunar infarction (SL), and stroke with large and lacunar infarction (SSL). All stroke/TIA groups (SS, SL, and SSL) had more WMH volumes than the CNI. The lacunar infarction groups (SL and SSL) had more periventricular WMH volumes than the SS group (P=0.033). There was no significant difference between the stroke/TIA patients and controls when only those who did not have infarction on MRI were considered. The linear regression analysis showed that the trend from the least WMHs to the most was in the order of CNI, SNI, CL, SS, SL, and SSL for whole-brain WMH (r=0.395; t=5.856; P<0.000), DWMH (r=0.384; t=5.670; P<0.000), and periventricular WMH (r=0.384; t=5.64; P<0.000; Table 4).
Our study supports the finding that WMHs are common in stroke and TIA patients as well as in otherwise healthy elderly comparison subjects, and their extent is greater in patients.1,12 This group difference was not accounted for by an overall increased number of discrete WMHs but by an excess of large WMHs with diameter >20 mm in the patient group. Patients also had more “severe” hyperintense signals; whereas the volumes of WMHs were increased ≈2-fold in patients compared with controls, WMHs with high intensity signal were 3-fold in the patients. It is reasonable to assume that the more intense signals are indicative of a greater degree of damage to the white matter. Assuming that patients had more severe cerebral ischemia, it is the large and high-intensity WMHs that characteristically reflect such damage. The large WMHs have been shown to indicate loss of myelin staining, variable extent of astrogliosis, rarefaction of the neuropil, and loss of oligodendrocytes and axons on neuropathology.13 Some of them even contain small central lacunes that are not resolved by MRI but are visible microscopically.14
The hyperintensities were distributed throughout the white matter in both groups. The periventricular region is particularly affected by this change, but the regions with the greatest difference between patients and controls were the frontal, temporal, and parietal lobes. The frontotemporal white matter had a 3.5-fold difference between the groups. Although WMHs were frequent in the occipital white matter, they did not differ in the 2 groups. Hyperintense lesions were uncommon in the white matter of the cerebellum and the brain stem in both groups. These findings suggest that all brain regions are not equally affected by white matter ischemia, and the anterior brain regions bear a larger burden. Furthermore, the deep white matter is differentially affected in the patient group than the periventricular white matter. This may reflect differences in the pathogenesis of deep and periventricular WMHs.
A not unexpected finding of the study was the increased WMH load in subjects with lacunar infarction, irrespective of whether they had a history of stroke or not or whether they had an additional large stroke. Surprisingly, WMH volumes were not different in those with single or multiple lacunes. Lacunar infarction is mostly caused by segmental arterial disorganization (lipohyalinosis), usually secondary to hypertension,15 whereas incomplete infarction of the white matter, which is the likely basis of WMHs in this group, is attributable to arteriolosclerosis affecting medullary arteries.16 However, the 2 pathologies share many of the same risk factors, thereby explaining their co-occurrence.17 Large and lacunar infarcts are different in their pathogenesis, which explains the different WMHs in these groups. Nevertheless, some risk factors for cerebrovascular disease, such as hypertension and diabetes, are common to large and lacunar infarcts, and the 2 types of infarcts were coexistent in many of our subjects. These subjects were closer to the lacunar group in their WMH volumes. There was also a regional association of stroke site and the distribution of WMHs. Anterior circulation strokes were more associated with WMHs in the frontal and parietal regions and with larger anterior caps. However, we cannot rule out the possibility that the occurrence of a stroke in a brain region led to the development of noncontiguous hyperintensities in the same region. A longitudinal study is needed to decide whether increased WMHs predispose the individual to the development of strokes in a particular region.
Our stroke patients had more cerebrovascular risk factors than the controls. Because we were interested in the extent and topography of the WMHs in this study, we did not examine the role played by all the risk factors in the etiology of WMHs. Besides age and gender, we controlled for hypertension in our analysis, and the differences between lacunar and other strokes were still significant. Schmidt et al1 argued that the excess WMHs in stroke patients could be accounted for by the higher rates of cerebrovascular risk factors, and the ischemic attacks were not important per se. This is a reasonable conclusion because the WMHs are not caused by the infarction but by small vessel disease, some of which is also responsible for infarction. However, hypertension alone is unable to account for the difference, and other pathogenetic factors must be examined. The presence of these lesions indicates that strokes and TIAs occur in brains that are already affected by much ischemic pathology. Deficits such as cognitive impairment seen after a stroke should not automatically be attributed to the recent infarction but may be a result of accumulation of pathology, much of which may have been unrecognized previously.
The strengths of this study are the relatively large sample size, the breadth of pathology, and the detailed examination of the WMHs, which were delineated by a computerized quantitative method. Previous studies of WMHs have mostly relied on visual ratings within large brain regions. A limitation of the study is that subjects were examined after a stroke or TIA, and it could not be determined what impact the event might have had on the WMHs. We manually removed the hyperintense brain region surrounding the index infarction, but it is possible that there were remote regions that were affected by the ischemic event. Being a cross-sectional study, it does not describe the development of WMHs, and their ischemic origin is based on informed speculation. Moreover, a number of eligible subjects in our study did not undergo an MRI scan, making our sample not truly representative of a community stroke sample.7 However, when we compared the subjects in the study to those who were excluded on clinical variables such as age, gender, education, and MMSE scores, no significant differences were noted.
In summary, we have described in detail the severity and distribution of WMHs in stroke and TIA, supporting their prominence in these patients. Future studies should examine these lesions longitudinally using similar methodology to understand their evolution and pathogenesis.
- Received January 6, 2004.
- Revision received September 8, 2004.
- Accepted September 13, 2004.
Schmidt R, Fazekas F, Kleinert G, Offenbacher H, Gindl K, Payer F, Freidl W, Niederkorn K, Lechner H. Magnetic resonance imaging signal hyperintensities in the deep and subcortical white matter: a comparative study between stroke patients and normal volunteers. Arch Neurology. 1992; 49: 825–827.
Pantoni L, Garcia J. The significance of cerebral white matter abnormalities 100 years after Binswanger’s report: a review. Stroke. 1995; 26: 1293–1301.
Vermeer S, Hollander M, van Dijk E, Hofman A, Koudstaal PJ, Breteler M. Silent brain infarcts and white matter lesions increase stroke risk in the general population: the Rotterdam Scan Study. Stroke. 2003; 34: 1126–1129.
Sachdev PS, Brodaty H, Valenzuela MJ, Lorentz L, Looi JC, Wen W, Zagami AS. The neuropsychological profile of vascular cognitive impairment in stroke patients: the Sydney Stroke Study. Neurology. 2004; 62: 912–919.
Mantyla R, Aronen HJ, Salonen O, Pohjasvaara T, Korpelainen M, Peltonen T, Standertskjold-Nordenstam CG, Kaste M, Erkinjuntti T. Magnetic resonance imaging white matter hyperintensities and mechanism of ischemic stroke. Stroke. 1995; 30: 2053–2058.
Bamford J, Sandercock P, Dennis M, Warlow C, Jones L, McPherson K, Vessey M, Fowler G, Molyneux A, Hughes T, et al. A prospective study of acute cerebrovascular disease in the community: the Oxfordshire Community Stroke Project 1981–86. 1. Methodology, demography and incident cases of first-ever stroke. J Neurol Neurosurg Psychiatry. 1988; 51: 1373–1380.
Roman GC, Tatemichi TK, Erkinjuntti T, Cummings JL, Masdeu JC, Garcia JH, Amaducci L, Orgogozo JM, Brun A, Hofman A, et al. Vascular dementia: diagnostic criteria for research studies. Report of the NINDS-AIREN International Workshop. Neurology. 1993; 43: 250–260.
Evans AC, Collins DL, Mills, Brown ED, Kelly RL, Peters TM. 3D statistical neuroanatomical models from 305 MRI volumes. Proceedings of IEEE-Nuclear Science Symposium and Medical Imaging Conference. 1813–1817;1993.
Fazekas F, Chawluk JB, Alavi A, Hurtig HI, Zimmerman RA. MR signal abnormalities at 1.5 T in Alzheimer’s dementia and normal aging. Am J Neuroradiol. 1987; 149: 351–356.
Sullivan P, Pary R, Telang F, Rifai AH, Zubenko GS. Risk factors for white matter changes detected by magnetic resonance imaging in the elderly. Stroke. 1990: 21; 1424–1428.
Fazekas F, Kleinert R, Offenbacher H, Schmidt R, Kleinert G, Payer F, Radner H, Lechner H. Pathologic correlates of incidental MRI white matter signal hyperintensities. Neurology. 1993; 43: 1683–1689.
Fisher CM. Lacunes: small, deep cerebral infarcts. Neurology. 1965; 15: 774–784.
Lammie GA. Pathology of small vessel stroke. Br Med Bull. 2000; 56: 296–306.