Variability of Vascular Territory in Stroke
Pitfalls and Failure of Stroke Pattern Interpretation
Background and Purpose We investigated the efficacy and feasibility of determining infarction mechanisms and underlying embolic or hemodynamic pathologies from topographical patterns of ischemic damage seen on computed tomography or magnetic resonance imaging.
Methods Infarction patterns from 22 patients with ipsilateral severe, hemodynamically relevant carotid stenosis (n=6) or occlusions (n=16) were superimposed, using two matching algorithms, onto maps showing the variability of the cerebral vascular territories as determined from recent cadaver studies. These images were used to classify the infarctions as border-zone or territorial for the two conditions of minimal and maximal middle cerebral artery distribution.
Results Classification of infarction patterns resulting from carotid stenosis was independent of the territorial extension map chosen: 83% were classified as territorial. Classification of patterns due to carotid occlusion, however, varied highly; 81% of infarctions were considered territorial when the maximal middle cerebral artery distribution map was used, whereas only 19% were when the minimal territorial extension map was used.
Conclusions The current concept that stroke mechanisms can be inferred from the interpretation of stroke patterns seen on computed tomography scans or magnetic resonance imaging is significantly confounded by the demonstrated variability in intracranial vascular distributions. Stroke pattern interpretation appears to be highly dependent on the in vivo vascular tree of the individual, which is unknown to the examiner. This calls into question the reliability of classifying infarction patterns as border-zone or territorial. Determination of true underlying stroke mechanisms requires a comprehensive approach and cannot be based solely on stroke pattern interpretation.
Microcirculatory and macroscopic vessel disease, as well as emboli of various origins, can impair normal cerebral circulation, resulting in infarction. The workup of stroke patients now includes neuroimaging to infer the pathological basis of the infarction from anatomic landmarks and topographical features evident on imaging studies. It has been suggested that CT and MR imaging are useful in differentiating between hemodynamically induced stroke and stroke caused by embolism.1
Investigation of this suggestion is particularly useful when one considers that in the case of high-grade carotid stenosis and carotid occlusion, from the earliest days and even in the autopsy literature, it has been clear that several mechanisms may be at work, even in the same patient.
In clinical practice, severe carotid stenosis is commonly considered to result in ipsilateral infarction of brain tissue at the distal margins of the associated vascular distribution.1 2 3 4 5 6 Such an infarction pattern has been termed watershed or border-zone,1 2 3 4 5 6 boundary zone,4 extraterritorial,1 low-flow,3 and equal perfusion-pressure boundary infarction.7 The inclusion of such border-zone regions in the infarcted territory demonstrated by CT or MR imaging is therefore characteristically interpreted as implying a hemodynamic mechanism.
In contrast, infarction of cerebral territory solely perfused by one cerebral artery is thought to be caused by occlusion and subsequent thrombosis of that artery, generally as a result of an embolism by either a carotid plaque or a thrombus of cardiac origin.3 8 According to the assumption that occlusion of an intracranial vessel affects only its particular distribution, infarcts on CT scans or MR images thought to be isolated purely within a single vascular distribution are called territorial infarcts and are generally felt to be embolic in origin.3
In their recently published maps of vascular variability based on an autopsy series, van der Zwan and colleagues7 9 10 demonstrated that the variability of the cerebral vascular territories (Fig 1⇓) is significantly greater than is generally assumed. One of the most vital and clinically relevant implications of these reports is that the interpretation of individual infarction patterns may be confounded by such vascular variability. If such confounding is significant, one must question the practice of managing a diagnostic workup to determine the cause of stroke on the basis of interpretations of CT or MR images as border-zone (eg, hemodynamic disease) versus territorial (eg, embolic disease) infarction patterns. We investigated these questions by analyzing individual CT or MR images of infarction patterns of known carotid origin using the vascular variability maps of van der Zwan and coworkers.7
Subjects and Methods
We retrospectively reviewed the records of 87 patients who were either admitted with acute stroke to the Department of Neurology, Klinikum Mannheim, or seen in our clinic because of a previous history of symptomatic stroke caused by severe (>80%) extracranial stenotic carotid disease and carotid occlusions during 20 months from 1990 to 1992. Twenty-two patients (16 with carotid occlusions and 6 with carotid stenoses >80%) had CT or MR images that closely matched those shown by van der Zwan et al.7 Using two digitized representations of the images at the basal ganglia and ventricular levels presented by van der Zwan et al (Fig 1⇑) as the standards for brain sectioning, we digitized CT scans from 18 individuals and MR images from 4. We then analyzed the images off-line on a computer workstation using the following protocol.
For the purposes of our analysis, digitized individual CT or MR imaging slices had to be congruent with the corresponding van der Zwan7 brain sections.11 12 This required a normalization procedure that was performed in two steps. First, we chose the two slices most closely matching the van der Zwan slices at the caudate level and 12 mm below. The cortical surface, basal ganglia, and ventricular structures of those individual CT scans or MR images were then segmented to serve as reference structures. Using an initial elliptical matching procedure for the cortical surface and a second planar elastic matching algorithm for the basal ganglia and ventricular structures, the individual CT or MR imaging slices were fitted to be congruent to the corresponding van der Zwan slices.7 11 12
Every individual infarction area attributable to the known carotid disease and consistent with the patient’s clinical symptoms and signs was then identified, outlined, and pasted automatically onto the corresponding cut in the standard brain sections.
For each patient, every infarction area (now displayed on our two standard brain sections) was superimposed over the others in a “sandwich-like” fashion, creating a composite image of all infarcted areas for each level, respectively. The graphic composition was programmed to darken the regions where overlapping of infarcted areas occurred. This resulted in a summation image comparable with a histogram in which those areas with a higher incidence of overlap were displayed as darker gray on a spectrum going from white to black.
We then superimposed this summation image onto the minimal and maximal territorial extension maps of the middle cerebral artery (MCA), anterior cerebral artery (ACA), and posterior cerebral artery (PCA) shown by van der Zwan et al.7 9 10 Fig 2⇓ shows the summation images for all infarcted areas for the images at the levels of the basal ganglia and lateral ventricular system.
The final infarction summation images, superimposed onto the territorial extension maps, were then classified as territorial or border-zone infarction patterns. Analysis was performed separately for the minimal and the maximal territorial extensions at each of the two levels.
The classification of the 22 infarction areas as territorial or border-zone is presented in Table 1⇓. Infarctions assumed to be caused by hemodynamically relevant carotid disease were located within the entire hemisphere. They were distributed in territorial MCA locations as well as in border-zone areas between the MCA, ACA, and PCA (Fig 2⇑).
When we used the van der Zwan7 small MCA territorial variant with wide border-zone margins, 13 of 16 stroke patients (81%) with complete carotid occlusion showed infarction involving border-zone areas of either or both PCA or ACA territories (Fig 3⇓, top). The remaining 3 patients (19%), however, showed typical territorial MCA ischemia without any border-zone involvement (Fig 3⇓, bottom). As shown in Table 2⇓, when the analysis was repeated using the van der Zwan territorial extension map of the large MCA territory with narrow ACA/MCA and PCA/MCA border-zone margins, only 3 of 16 stroke patients (19%) with carotid occlusion showed MCA infarction involving either border zone (Fig 3⇓, bottom). Therefore, confirmation of the association between carotid occlusion and involvement of borderline zones varied between 19% and 81%, depending on the choice of territorial borders. For the 6 patients with carotid stenosis, the interpretation was independent of the territorial extension map chosen. Five patients (83%) displayed pure autonomous MCA territory infarction (Fig 4⇓, top). Only 1 patient (17%) showed junctional MCA/PCA involvement (Fig 4⇓, bottom).
The common belief that carotid stenosis or occlusion results in border-zone infarction held true in 64% of our patients when the small territorial variant with maximal MCA border-zone extension was applied. However, when the wide territorial variant with the minimal MCA border zones was used, this tenet held true in only 19% of patients (Table 3⇓).
Watershed infarctions have been reported in over 40% of patients with carotid stenotic disease or occlusions by Wodarz and colleagues.5 6 They called this finding “typical” yet not “specific” and suggested that border-zone involvement is one of the neuroimaging features commonly seen in severe carotid vessel disease. In another report, Bogousslavsky and Regli13 found that 69% of 26 patients with ipsilateral internal carotid artery occlusions had watershed infarctions. Both of these reported percentages are in keeping with our range of 19% to 64%. However, embolic ischemia may be another mechanism of border-zone infarction.2 13 14 15 16 Pollanen and Deck14 and Torvik and Skullerud15 reported border-zone infarction patterns that were attributable to cerebral embolism rather than to low perfusion of adjacent distal cerebral vessels. A similar argument has been made recently for three cases by Graeber and coworkers.17 Although the nature of such cases is most probably embolic and appears to represent pure territorial infarction in individuals with wide MCA margins, these cases would not be recognized as embolic because of the prevailing territorial concept, which assumes a narrower territorial margin area and would place them in the border zone.
Our results suggest that interpatient variation in the topographic distributions of major cerebral vessels, as shown by van der Zwan et al,7 9 10 significantly confounds the classification of CT or MR image stroke patterns into border-zone or territorial categories because the variable vascular tree and vascular collateral network specific to an individual is unknown to the examiner. We conclude that the approach of determining stroke mechanisms from stroke pattern interpretation does not appear to be a useful plan in patients with carotid disease.
Given the uncertainty suggested by our data in reliably determining the presence or absence of true border-zone infarction, we believe that a workup prompted by a presumed hemodynamically induced infarction pattern (border-zone infarction) should always include a search for relevant vascular disease; more often than expected, the same pattern may be due to multiple embolism of arterial or cardiac origin.
This work was supported in part by an educational stipend from the Deutsche Forschungsgemeinschaft awarded to Erhard W. Lang (La 916 1/1) and in part by the Deutsche Forschungsgemeinschaft, Sonderforschungsbereich 258/K4 at the Universität Heidelberg. We are grateful to Dr A. van der Zwan, Department of Neurological Surgery, University Hospital, Utrecht, Netherlands, for supplying the anatomic reference material. We are also indebted to Dr H. Haken, Department of Physics, University of Stuttgart, Germany, for his helpful suggestions during the methodological development of pattern recognition procedures. We also wish to thank Pamela Derish for editing the manuscript and Cindy Huff for manuscript preparation.
Reprint requests to Michael Hennerici, MD, Department of Neurology, Universität Heidelberg, Klinikum Mannheim, Theodor Kutzer Ufer, D-68135 Mannheim, Germany.
- Received November 2, 1994.
- Revision received January 30, 1995.
- Accepted March 6, 1995.
- Copyright © 1995 by American Heart Association
Ringelstein EB, Zeumer H, Angelou D. The pathogenesis of strokes from internal artery occlusion: diagnostic and therapeutic implications. Stroke. 1983;14:867-875.
Rodda RA, Path FRC. The arterial patterns associated with internal carotid infarcts. Stroke. 1986;17:69-75.
Wodarz R, Ratzka M, Grobe D. Der Grenzzoneninfarkt als besondere Infarktkonstellation bei Karotisinsuffizienz. Rofo Fortschr Geb Röntgenstr Nuklearmed. 1981;134:128-131.
van der Zwan A, Hillen B. Review of the variability of the territories of the major cerebral arteries. Stroke. 1991;22:1078-1084.
van der Zwan A, Hillen B, Tulleken CAF, Dujovny M. A quantitative investigation of the variability of the major cerebral arterial territories. Stroke. 1993;24:1951-1959.
Bogousslavsky J, Regli F. Unilateral watershed cerebral infarcts. Neurology. 1986;36:373-377.
Torvik A. The pathogenesis of watershed infarcts in the brain. Stroke. 1984;15:221-223.