Article Text

Download PDFPDF
Original research
Topographic distribution of cerebral infarct probability in patients with acute ischemic stroke: mapping of intra-arterial treatment effect
  1. A M M Boers1,2,3,
  2. O A Berkhemer1,4,
  3. C H Slump2,
  4. W H van Zwam5,
  5. Y B W E M Roos6,
  6. A van der Lugt7,
  7. R J van Oostenbrugge8,
  8. A J Yoo9,
  9. D W J Dippel4,
  10. H A Marquering1,3,
  11. C B L M Majoie1
  12. on behalf of the MR CLEAN trial investigators
    1. 1Department of Radiology, Academic Medical Center, Amsterdam, The Netherlands
    2. 2Department of Robotics and Mechatronics, University of Twente, Enschede, The Netherlands
    3. 3Department of Biomedical Engineering and Physics, Academic Medical Center, Amsterdam, The Netherlands
    4. 4Department of Neurology, Erasmus MC University Medical Center, Rotterdam, The Netherlands
    5. 5Department of Radiology, Maastricht University Medical Center, Maastricht, The Netherlands
    6. 6Department of Neurology, Academic Medical Center, Amsterdam, The Netherlands
    7. 7Department of Radiology, Erasmus MC University Medical Center, Rotterdam, The Netherlands
    8. 8Department of Neurology, Cardiovascular Research Institute Maastricht (CARIM), Maastricht University Medical Center, Maastricht, The Netherlands
    9. 9Division of Neurointervention, Texas Stroke Institute, Plano, Texas, USA
    1. Correspondence to Anna M M Boers, Department of Biomedical Engineering and Physics, Academic Medical Center, PO Box 22660, Amsterdam 1100 DD, The Netherlands; a.m.boers{at}amc.uva.nl

    Abstract

    Background Since proof emerged that IA treatment (IAT) is beneficial for patients with acute ischemic stroke, it has become the standard method of care. Despite these positive results, recovery to functional independence is established in only about one-third of treated patients. The effect of IAT is commonly assessed by functional outcome, whereas its effect on brain tissue salvage is considered a secondary outcome measure (at most). Because patient and treatment selection needs to be improved, understanding the treatment effect on brain tissue salvage is of utmost importance.

    Objective To introduce infarct probability maps to estimate the location and extent of tissue damage based on patient baseline characteristics and treatment type.

    Methods Cerebral infarct probability maps were created by combining automatically segmented infarct distributions using follow-up CT images of 281 patients from the MR CLEAN trial. Comparison of infarct probability maps allows visualization and quantification of probable treatment effects. Treatment impact was calculated for 10 Alberta Stroke Program Early CT Score (ASPECTS) and 27 anatomical regions.

    Results The insular cortex had the highest infarct probability in both control and IAT populations (47.2% and 42.6%, respectively). Comparison showed significant lower infarct probability in 4 ASPECTS and 17 anatomical regions in favor of IAT. Most salvaged tissue was found within the ASPECTS M2 region, which was 8.5% less likely to infarct.

    Conclusions Probability maps intuitively visualize the topographic distribution of infarct probability due to treatment, which makes it a promising tool for estimating the effect of treatment.

    • Brain
    • CT
    • Stroke
    • Technology
    • Thrombectomy
    View Full Text

    Statistics from Altmetric.com

    Introduction

    IA treatment (IAT) for patients with acute ischemic stroke is rapidly becoming the standard method of care in comprehensive stroke centers after the emergence of new evidence showing that IAT is beneficial for patients with a proximal large vessel occlusion.1 This was first reported by the Multicenter Randomized Clinical Trial of Endovascular Treatment for Acute Ischemic Stroke in the Netherlands (MR CLEAN) trial, and later confirmed by four other randomized clinical trials.2–5 Although positive results have accumulated, a large population still remain functionally disabled after IAT. Despite high recanalization rates with retrievable stents, functional independence at 90 days is achieved in only 33–72%, depending on the patient selection methods used. It is of great importance to understand this treatment effect and to treat only those who will potentially benefit, sparing them treatment risks.6 Pretreatment imaging-derived factors may play an important role in improving patient selection and maximizing the clinical utility of IAT.7 ,8 Proof is accumulating that baseline factors, such as occlusion location,9 Alberta Stroke Program Early CT Score (ASPECTS),10–12 and collateral circulation capacity,13 ,14 are related with infarct volume and patient outcome at 90 days, underlining the importance of image-based patient selection for IAT.15 ,16 However, surprisingly little is known about the effect of these image-based factors and treatment on the extent and topographic distribution of the resulting cerebral infarction.

    This study aimed to provide more insight into the effects of IAT by presenting the infarct probability at anatomical locations for specific stroke subpopulations. This approach allows the generation of infarct probability maps for specific patient subpopulations who share common features, such as occlusion location and treatment allocation.

    Methods

    We have generated probability maps by combining infarct distributions of specific patient subpopulations from the MR CLEAN trial1 using registration of patient-specific follow-up non-contrast CT (FU-NCCT) images with a reference CT image. The frequency of infarction at a specific location of this combination is here presented as the infarct probability for a similar patient population. Details of this method are given below.

    Patient selection

    Patient eligibility and methods of the MR CLEAN trial have been reported previously.17 All patients or their legal representatives provided written informed consent before randomization. The study protocol was approved by a central medical ethics committee and the research board of each participating center. Briefly, MR CLEAN was a randomized clinical trial of IAT (intervention group) versus no IAT (control group) in patients with a proximal arterial occlusion in the anterior circulation demonstrated on vessel imaging, treatable within 6 h after symptom onset. Eligible patients had an occlusion of the distal intracranial carotid artery (ICA), middle cerebral artery (M1 or M2), or anterior cerebral artery (A1 or A2). The primary outcome of the MR CLEAN trial was the score on the modified Rankin Scale at 90 days.

    We included all patients with 3–9-day FU-NCCT images. For this substudy, patients with hemicraniectomy, NCCT images with extreme motion or beam hardening artifacts leading to technical errors precluding automatic registration or segmentation and those with insufficient scan quality were additionally excluded.

    Infarct segmentation

    Whole-brain imaging was performed in 16 medical centers in the Netherlands with thin-section acquisition by 28 different CT scanners. Segmentation of the infarcts was performed on FU-NCCT images before its registration with the reference image because segmentation on the transformed images (as a result of the registration) is more error prone due to interpolation of the image data. Infarcts were segmented using a previously presented and validated automatic method,18 which resulted in binary masks of the infarcted areas. All segmentations were inspected by an experienced neuroradiologist (CBM) who was blinded to treatment allocation. The final infarct volume (FIV) was calculated by multiplying the number of voxels of the segmented infarct with its voxel size.

    Registration with a reference image

    To combine the topographic distributions of the infarcts, all binary masks of a predetermined subpopulation were transformed to a reference CT image, which aligned all patient-specific images with each other. Registration was performed by a subsequent rigid and affine transformation to adjust for major differences in position, orientation, and size by using the open-source software Elastix (V.4.6; http://elastix.isi.uu.nl). Subsequently, the aligned binary infarct masks were inspected for registration inaccuracies by an experienced observer (AMMB).

    Anatomic labeling of the topographic distribution of the infarcts was facilitated by registering the reference CT image with an in-house developed atlas. The atlas is based on the Laboratory of Neuro Imaging Probabilistic Brain Atlas (LPBA40),19 as described by Boers et al.20 This atlas consists of 27 labeled brain structures, most of which are within the cortex. Furthermore, labeling was also performed according to the ASPECTS regions by the registration of an ASPECTS atlas with the reference CT image. The 10 ASPECTS regions were manually delineated using ITK-Snap21 and were verified for anatomical correctness by an experienced neuroradiologist (CBM).

    Generation of topographic infarct probability maps

    The voxel-wise infarct probability maps were generated by averaging the aligned masks of infarcted tissue on a scale from 0% to 100% for a subpopulation. The infarct probability maps assume that the frequency of infarcted tissue in a subpopulation equals the infarct probability for a patient with similar characteristics in this subpopulation. To increase statistical power, all infarctions on the right hemisphere of the brain were mirrored over the midline to the left hemisphere in the generation of the probability maps.

    We generated probability maps of topographic infarct distribution for two treatment groups: patients who underwent IAT and patients who underwent standard care alone (control group). Comparison of these maps allows the visualization and quantification of the topographic distribution of tissue that may be saved with IAT. Furthermore, we generated different infarct probability maps for occlusions located in the ICA, M1, M2, A1 and A2 artery by combining infarct segmentations in the subpopulations characterized by the location of occlusion.

    By calculating the average voxel-wise infarct probability of all voxels within a labeled region, the regional impact is determined. Regional impact of IAT in 27 anatomical and 10 ASPECTS areas was determined by calculating the infarct probability difference between the IAT and control groups in those areas.

    Statistical analysis

    Binary variables are presented as a proportion of the population. Continuous variables are presented as mean±SD for normally distributed variables, and as median and IQR otherwise.

    Baseline characteristics were compared between the IAT and control groups, and with the MR CLEAN patients who were excluded for this substudy. Treatment impact for each anatomical and ASPECTS region was determined by comparing infarct probability maps between the treatment groups. In addition, FIVs were compared between the IAT group and control group for occlusion location. Comparisons were performed using Student t tests for normally distributed variables and Wilcoxon rank-sum tests (Mann–Whitney U tests) otherwise. A two-sided p value of <0.05 was considered statistically significant. Statistical analysis was performed using SPSS V.21.0 (IBM Corp, Armonk, New York, USA).

    Because we excluded a large number of patients who had hemicraniectomy and patients who died before follow-up imaging, the patient population of our study largely differs from the full MR CLEAN population. Therefore, we also performed a post hoc analysis comparing the MR CLEAN primary and secondary outcome measures and serious adverse events between the included and excluded patients.

    Results

    Of the 500 patients included in the MR CLEAN trial, 338 had 3–9-day FU-NCCT scans. Fifty-seven patients were subsequently excluded because of hemicraniectomy (21 patients), technical errors precluding automatic registration or segmentation (19 patients), and insufficient scan quality (17 patients), resulting in the inclusion of 281 patients for analysis.

    Online online supplementary table S1 displays the baseline characteristics of the included IAT and control populations, and of the MR CLEAN participants who were excluded for this substudy. Median age was 68 and 66 years for the IAT and control group, respectively; 41% were female. Baseline characteristics were evenly distributed across the three groups, apart from a significant difference in pre-stroke modified Rankin Scale score and treatment with IV alteplase.

    IAT versus standard care

    The topographic infarct probability maps for IAT and control group are shown in figure 1A, B. Topographic infarct probability maps for subpopulations grouped by occlusion location are shown in figure 1C–E. Table 1 shows infarct probabilities for each anatomical and ASPECTS region related to type of treatment. Treatment impact, defined as absolute change in infarct probability for these regions, is shown in figure 2 and table 1.

    Table 1

    Treatment impact on 27 anatomical regions and 10 ASPECTS regions

    Figure 1

    Topographic distribution of infarct probability for the MR CLEAN control (A) and IA treatment (IAT) (B) subpopulations, and the subpopulation with an occlusion at the intracranial carotid artery (ICA) (C), M1 (D), and M2 location (E). Colors indicate the probability of infarct for each voxel.

    Figure 2

    Treatment impact for anatomical regions (left) and ASPECTS regions (right). Treatment impact is defined as the difference in infarct probability for patients receiving IA treatment and those receiving standard treatment. Colors indicate absolute change in infarct probabilities for all occlusion locations combined. ASPECTS, Alberta Stroke Program Early CT Score.

    The highest probability to infarct was observed in the insular cortex and putamen with, respectively, 47.2% and 43.2% for the control group and 42.6% and 41.5% for the IAT group. A significant decrease in infarct probability was seen in most anatomical regions in favor of IAT. With IAT, most tissue was saved in the middle temporal gyrus with a relative reduction in infarct probability of 8.2% compared with standard care (25.4% vs 17.2% mean infarct probability), whereas the only significant increase in infarct probability was found in the superior frontal gyrus (mean probability from 1.2% to 1.8%).

    A significant change in infarct probability between the treatment groups was seen in four of the 10 ASPECTS regions (M2, M3, M5 and M6) as shown in table 1. These changes were all in favor of the IAT group and located in the middle cerebral artery (MCA) cortex, with M2 being the region with the highest probability of tissue salvage of 8.5%.

    Final infarct volumes

    Median FIV was 61.3 mL (IQR 29.5–117.5 mL). Differences in FIV by occlusion location and treatment allocation are shown in table 2. A significant reduction in FIV was in favor of the IAT group. The largest FIV was observed in patients treated with standard care with an occlusion of the ICA (median 94.9 mL; IQR 47.9–153.6 mL). Patients with an M1 occlusion and treated with IAT demonstrated the smallest FIV (median 38.5 mL; IQR 16.9–87.7 mL). FIV differences between the IAT and standard care group could not be calculated for the A1 and A2 occlusions owing to the small number of cases. Table 2 shows a reduction in FIV associated with IAT for all other occlusion locations.

    Table 2

    Final infarct volume by occlusion location and treatment allocation

    Population comparison

    As expected, there was a large difference of the MR CLEAN outcome measures (see online supplementary table S2) and serious adverse events (see online supplementary table S3) between the included group of patients and the whole MR CLEAN population, since only patients with FU-NCCT scans were included in our study. Moreover, patients with hemicraniectomy were also excluded. Most notably, the adjusted common OR was 2.2 (95% CI 1.5 to 3.5) for the included group compared with 1.1 (95% CI 0.7 to 1.9) for the excluded patients.

    Discussion

    We have presented a new method to estimate the topographic distribution of infarct probability for specific stroke subpopulations grouped by occlusion location and treatment allocation. This approach has been applied to determine the effect of treatment in anatomical regions by comparing the probability maps of patients allocated to IAT with those allocated to usual care.

    The probability maps show that in our population of patients with acute ischemic stroke caused by a proximal occlusion of the anterior circulation the benefit from IAT is reflected in a lower infarct probability in the M2, M3, M5 and M6 ASPECTS regions. With IAT, the highest probability of tissue salvage was in the M2 ASPECTS territory, whereas the deeper located subcortical structures did not show IAT-related benefit. The last finding reflects the end-arterial supply to the deep structures, which are often irreversibly injured by the time of imaging when a very proximal occlusion obstructs blood flow in the lenticulostriate arteries.

    This study is not the first to present probability maps of patients with acute ischemic stroke. Previously, Phan et al22 ,23 presented digital maps of patients with middle and posterior cerebral artery branch occlusions, in which the boundaries of the arterial territories were studied. However, in their study the effect of treatment allocation was not examined and treatment effect was lacking. Alternatively, maps have been presented in which the location of infarction was linked with clinical outcomes.24

    In our comparisons, only patients with an occlusion of the M1 segment showed a statistically significant decrease in FIV when treated with IAT. However, a reduction was seen in all occlusion locations (except for the A1 and A2 artery owing to a lack of cases). This suggests that patients with an infarction due to an occlusion in the ICA or M2 segment of the MCA benefit less from IAT, which is in contradiction to the functional outcome of this same population.1 These differences can be explained by the exclusion of patients who died or who were discharged before follow-up imaging. The infarct probability maps represent the probability of infarct for patients who survive stroke (and who did not have hemicraniectomy). Furthermore, the FIV does not take the side of the occlusion into account. Since infarcts on the left hemispheric side are usually associated with worse functional outcome, this may also contribute to differences in significance.

    In this study, the location with the highest infarct probability was the insular cortex. There was no statistically significant difference found in this area for treatment allocation, which may suggest that the insular cortex is not salvaged by IAT. A possible explanation is that the blood supply of the insula is mainly provided by arteries that originate from the M2 segment of the MCA.25 In M1 occlusions, these areas are the most remote from potential leptomeningeal collateral flow and would be the first to deteriorate.

    Our study has several limitations. First, our population was biased by the use of images of a 3–9-day follow-up period, excluding both the very favorable and very unfavorable outcome cases that did not have a follow-up scan. Exclusion of these cases from our infarct probability maps might have resulted in a bias towards a homogeneous distribution of infarctions. The number of patients could be increased by including 1-day follow-up images of the patients who are missing from this study. However, the automated method was not yet considered sufficiently accurate for these images.

    Second, we chose to examine only two predictors: treatment allocation and occlusion location. In future work, the effect of other potentially important parameters, such as onset to treatment time, collateral capacity, and variations in the circle of Willis, on the distribution of infarct probability will be investigated.

    Third, we assumed both hemispheres to have the same vascularity, allowing us to combine all infarcts in one hemisphere and therefore improving the statistical power. However, this assumption might be an oversimplification when assessing different perfusion situations. Moreover, hemispheric lateralization correlates with stroke symptoms, management, and outcome.26–28 The correlation with hemisphere lateralization of additional baseline characteristics, such as gender, will further be evaluated in a planned pooled analysis of patient data from five randomized trials to ensure sufficient statistical power to distinguish effects from left- and right-sided hemispheric strokes.29

    Conclusions

    We have presented infarct probability maps which provide insight into stroke evolution and rescue, enabling visualization and assessment of effects of treatment allocation and occlusion location within affected anatomical areas. These maps can be generated for individual patients by generating probability maps for a subpopulation with similar baseline characteristics to visualize the probable effect of treatment. In our population of patients with acute ischemic stroke due to a proximal arterial occlusion of the anterior circulation, the insular cortex and the putamen had the highest infarct probability independent of treatment and occlusion location. Moreover, we found that the ASPECTS M2 region benefits most from IAT, with the highest probability of tissue salvage. With further data and model refinement, the presented methodology has the potential to become a valuable tool in estimating treatment effect, which may be used to support treatment selection in the management of patients with ischemic stroke.

    References

    View Abstract

    Footnotes

    • HAM and CBM contributed equally.

    • Collaborators The MR CLEAN investigators: DWJ Dippel1; A van der Lugt2; CBLM Majoie3; YBWEM Roos4; RJ van Oostenbrugge5; WH van Zwam6; OA Berkhemer1,3; PSS Fransen1,2; D Beumer1,5; LA van den Berg4; WJ Schonewille7; JA Vos8; CBLM Majoie3; YBWEM Roos4; PJ Nederkoorn4; MJH Wermer9; MAA van Walderveen10; RJ van Oostenbrugge5; WH van Zwam6; J Staals5; J Hofmeijer11; JA van Oostayen12; GJ Lycklama à Nijeholt13; J Boiten14; DWJ Dippel1; PA Brouwer2; BJ Emmer2; SF de Bruijn15; LC van Dijk16; L JKappelle17; RH Lo18; EJ van Dijk19; J de Vries20; PLM de Kort21; JSP van den Berg22; BAAM van Hasselt23; LAM Aerden24; RJ Dallinga25; MC Visser26; JCJ Bot27; PC Vroomen28; O Eshghi29; THCML Schreuder30; RJJ Heijboer31; K Keizer32; AV Tielbeek33; HM den Hertog34; DG Gerrits35; RM van den Berg-Vos36; GB Karas37; HA Marquering38; LF Beenen3; MES Sprengers3; SFM Jenniskens39; R van den Berg3; AJ Yoo40; PJ Koudstaal1; H Zwenneke Flach23; EW Steyerberg41; HF Lingsma41; MM Brown42; T Liebig43; T Stijnen44; T Andersson45; HP Mattle46; N Wahlgren47; Esther van der Heijden1; Naziha Ghannouti1; Nadine Fleitour4; Imke Hooijenga4; Annemieke Lindl-Velema5; Corina Puppels7; Wilma Pellikaan7; Kirsten Janssen9; Nicole Aaldering11; Marjan Elfrink11; Joke de Meris14; Annet Geerlings19; Gina van Vemde22; Ans de Ridder17; Paut Greebe17; José de Bont-Stikkelbroeck21; Willy Struijk15; Tiny Simons30; Gert Messchendorp28; Friedus van der Minne28; Hester Bongenaar32; Karin Bodde.26

    • Department of Neurology1, Radiology2, Public Health41, Erasmus MC; Department of Radiology3, Neurology4, Biomedical Engineering and Physics38, AMC, Amsterdam; Department of Neurology5, Radiology6, MUMC and CARIM; Department of Neurology7, Radiology8, Sint Antonius Hospital, Nieuwegein; Department of Neurology9, Radiology10, Medical Statistics and Bioinformatics44, LUMC; Department of Neurology11, Radiology12, Rijnstate Hospital, Arnhem; Department of Radiology13, Neurology14, MC Haaglanden, the Hague; Department of Neurology15, Radiology16, HAGA Hospital, the Hague; Department of Neurology17, Radiology18, UMCU; Department of Neurology19, Neurosurgery20, Radiology39, RadboudUMC, Nijmegen; Department of Neurology21, Sint Elisabeth Hospital, Tilburg; Department of Neurology22, Radiology23, Isala Klinieken, Zwolle; Department of Neurology24, Radiology25, Reinier de Graaf Gasthuis, Delft; Department of Neurology26, Radiology27, VUMC, Amsterdam; Department of Neurology28, Radiology29, UMCG, the Netherlands; Department of Neurology30, Radiology31, Atrium Medical Center, Heerlen; Department of Neurology32, Radiology33, Catharina Hospital, Eindhoven; Department of Neurology34, Radiology35, MST, Enschede; Department of Neurology36, Radiology37, Sint Lucas Andreas Hospital, Amsterdam; all in the Netherlands. Department of Radiology40, Texas Stroke Institute, Texas, USA; UCL Institute of Neurology42, National Hospital for Neurology and Neurosurgery, London, UK; Med. Fakultät43, Uniklinik Köln, Germany; Department of Radiology45, Neurology47, Karolinska Univeristy Hospital, Stockholm, Sweden; Department of Neurology46, University Hospital of Bern, Switzerland.

    • Contributors All authors read and approved the submitted manuscript. The manuscript has not been submitted elsewhere nor published elsewhere in whole or in part. All authors of this work met International Committee of Medical Journal Editors criteria for authorship and made substantial contributions to the conception and design, acquisition of data, analysis and interpretation of data, drafting, critical revision, and final approval of this manuscript.

    • Funding (1) AMMB is supported by a personal grant from the Stichting Toegepast Wetenschappelijk Instituut voor Neuromodulatie (TWIN). (2) The MR CLEAN trial was funded by the Dutch Heart Foundation and through unrestricted grants from AngioCare BV, Covidien/EV3, MEDAC Gmbh/LAMEPRO, and Penumbra Inc.

    • Competing interests None declared.

    • Ethics approval Medical ethics committee and resarch board of Erasmus MC University Medical Center.

    • Provenance and peer review Not commissioned; externally peer reviewed.

    Request Permissions

    If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.