Search tips
Search criteria 


Logo of nihpaAbout Author manuscriptsSubmit a manuscriptHHS Public Access; Author Manuscript; Accepted for publication in peer reviewed journal;
Stroke. Author manuscript; available in PMC 2014 January 1.
Published in final edited form as:
PMCID: PMC3558033



Background and Purpose

Objective imaging methods to identify optimal candidates for late recanalization therapies are needed. The study goals were 1) to develop MRI and CT multiparametric, voxel-based predictive models of infarct core and penumbra in acute ischemic stroke patients, and 2) to develop patient-level imaging criteria for favorable penumbral pattern based on good clinical outcome in response to successful recanalization.


An analysis of imaging and clinical data was performed on two cohorts of patients (one screened with CT, the other with MRI) who underwent successful treatment for large vessel, anterior circulation stroke. Subjects were divided 2:1 into derivation and validation cohorts. Pretreatment imaging parameters independently predicting final tissue infarct and final clinical outcome were identified.


The MRI and CT models were developed and validated from 34 and 32 patients, employing 943,320 and 1,236,917 voxels respectively. The derivation MRI and two-branch CT models had an overall accuracy of 74% and 80% respectively, and were independently validated with an accuracy of 71% and 79% respectively. The imaging criteria of 1) predicted infarct core ≤ 90 mL and 2) ratio of predicted infarct tissue within the at-risk region ≤ 70% identified patients as having a favorable penumbral pattern with 78–100% accuracy.


Multiparametric voxel-based MRI and CT models were developed to predict the extent of infarct core and overall penumbral pattern status in patients with acute ischemic stroke who may be candidates for late recanalization therapies. These models provide an alternative approach to mismatch in predicting ultimate tissue fate.

Keywords: Ischemic stroke, Penumbra, Magnetic resonance imaging, Computed tomography, Infarction, Recanalization, Mechanical embolectomy, Diffusion-weighted imaging, Perfusion-weighted imaging

Vessel recanalization with intravenous thrombolytic therapy improves functional outcomes in patients presenting within 3 to 4.5 hours of ischemic stroke onset.14 However, treatment is limited by risk of symptomatic intracerebral hemorrhage and the narrow therapeutic time window beyond which there is no proven benefit.5, 6 There is a recognized need for advanced imaging methods to select patients most likely to benefit from recanalization therapies, and exclude patients who would fare worse or not benefit, particularly in later time windows.7, 8 Important goals of multimodal imaging in acute stroke have, therefore, included identification of the infarct core (irreversibly injured tissue that cannot be salvaged even with tissue reperfusion), the ischemic penumbra (tissue at risk of infarction but salvageable with early restoration of blood flow), and regions of benign oligemia (tissue with reduced blood flow but not at risk of infarction).911 Patients most likely to benefit from treatment, particularly in late time windows, are postulated to be those with smaller regions of infarct core, substantial regions of ischemic penumbra, and a documented vessel occlusion.

While positron emission tomography (PET) has been considered the gold standard for defining the ischemic core, penumbra and benign oligemia12, it is not a practical imaging modality in the routine, clinical, acute stroke setting. As such, attention has focused on the role of multimodal MRI and multimodal CT for defining the infarct core and the penumbra. Infarct core has been estimated with MRI sequences including diffusion-weighted imaging (DWI), and with CT measures of cerebral blood volume (CBV) or cerebral blood flow (CBF).1319 A number of approaches to identification of the ischemic penumbra with both MRI and CT have also been proposed, including MRI diffusion-perfusion mismatch, and CBV or CBF thresholds; however, these approaches have some limitations.9, 2025 Of note, there has been a growing appreciation that the extent of the infarct core rather than the penumbra is the strongest predictor of outcome in the setting of a large vessel occlusion.1519 Since the above approaches (typically based on single parameter thresholds) for distinguishing core, penumbra and benign oligemia have shown only modest success, multiparametric models incorporating information from various sequences have been proposed as offering the potential to improve the prediction accuracy of these compartments.2628

The two primary aims of this study were: 1) to create multiparametric, MRI and CT-based models for distinguishing infarct core from non-core hypoperfused tissue on a voxel-by-voxel basis by employing cohorts of patients undergoing successful vessel recanalization and studied pretreatment with multimodal MRI or CT; and 2) to derive patient-level volume thresholds for predicted infarct core and infarct fraction (volume of core/volume of at-risk tissue X 100) to identify a favorable penumbral pattern (small core, large penumbra; likely to benefit from reperfusion therapy) and non-penumbral pattern (large core, small penumbra; unlikely to benefit from reperfusion therapy). The ultimate goal was to use these categorizations to stratify enrollment in a clinical trial designed to test the imaging selection hypothesis for endovascular recanalization therapies for stroke. According to this hypothesis, patients with a favorable penumbral pattern and successful recanalization would be most likely to have a good clinical outcome and therefore this group would be the appropriate target for treatment.29



The two predictive models, one for MRI and one for CT, were developed independently by analysis of two datasets (1 for MRI, 1 for CT) obtained from patients meeting the following criteria: 1) acute ischemic stroke involving the anterior circulation; 2) recanalization therapy (any combination of IV tPA, intra-arterial thrombolysis, and mechanical embolectomy) initiated within 8 hours of symptom onset; 3) multimodal CT or MRI performed prior to treatment; 4) large vessel intracranial anterior circulation occlusion documented prior to treatment (distal internal carotid artery, M1 or M2 middle cerebral artery); 5) recanalization (Thrombolysis In Myocardial Infarction [TIMI] score of 2 or 3)30 achieved and documented post-treatment; and 6) follow-up imaging obtained within 7 days to document final infarct size. Functional outcome was assessed on days 30–90 employing the modified Rankin Scale (mRS), with good functional outcome defined as a mRS score of 0–2. For the MRI model, subjects were identified from a prospective study of diffusion-perfusion MRI changes in patients receiving endovascular recanalization therapy at UCLA. For the CT model, subjects were provided from two institutions (UCSF, University of Pittsburgh) routinely performing multimodal CT prior to recanalization treatments for acute ischemic stroke. The study was approved by the Institutional Review Boards of all sites involved.

For both models, both patient-based (e.g. NIHSS, age) and imaging voxel-based variables were considered for model development. Detailed methodology regarding MRI and CT image acquisition and processing is provided in the supplemental methods section online.

Statistical Methods: Voxel-Based Predictive Models

For each model (MRI or CT), the cohort was randomly divided into a derivation group (2/3 of the cohort) for model development, and a validation group (1/3 of the cohort) for model assessment.

To develop the voxel-level predictive models, a multivariate random coefficient logistic regression analysis was performed on data from the derivation sample (SAS procedure GENMOD and custom programming in SAS, similar to SAS Macro GLIMMIX) using backward stepwise regression with a liberal p < 0.15 variable retention criterion for the derivation data. This model allows infarct observations among voxels from the same person to be correlated, with observations from different subjects treated as independent. Model fit was assessed by examining three statistics: 1) maximum rescaled R2 statistic; 2) receiver operating characteristic (ROC) curve area (C statistic); and 3) unweighted model accuracy defined as [sensitivity + specificity] / 2. For the CT cohort, a second branch to the model was derived to handle voxels that were identified as having less reliable perfusion parameter estimates due to the numeric instability of the deconvolution process under conditions of extremely poor signal-to-noise ratio. The logit score was obtained for each voxel from the logistic model and a cutoff (threshold) score was found such that the unweighted accuracy above was maximized, thus simultaneously maximizing both sensitivity and specificity. The supplemental methods section provides greater detail on model development and the variables considered in both of the models.

Statistical Methods: Patient-Level Categorical Assignments

After the voxel-level models to predict infarct core were completed, an algorithm was developed to apply the results of the voxel model at the level of individual patients to distinguish patients with a favorable penumbral pattern (good candidate for treatment) vs. non-penumbral pattern (poor candidate for treatment). Heuristically-derived thresholds were identified that best correlated with good functional outcome (mRS 0–2) based on 1) total volume of tissue at-risk that was already irreversibly infarcted (predicted infarct core volume), and 2) the proportion of tissue at-risk that was infarcted (infarct fraction). The threshold was chosen to maximize the accuracy, defined as the unweighted average of the sensitivity and specificity. Infarct fraction was defined as: (predicted infarct volume×100) / at-risk volume. For this purpose, a more conservative Tmax threshold of ≥ 4 seconds (compared to the threshold of ≥ 2 seconds to develop the core prediction model) was chosen to define the tissue at-risk/penumbral region for MRI and an MTT threshold ≥ 6 seconds was chosen for CT.


A total of 34 patient datasets were employed for the MRI cohort and 32 datasets for the CT cohort. Mean age for both groups was 72, and median baseline NIHSS score was 16 (range 5–26) and 15 (range 4–25) for the MRI and CT cohorts respectively. Treatment categories for the MRI cohort were: 12 pure intra-arterial thrombolysis, 11 bridging IV tPA to intra-arterial thrombolysis, and 11 mechanical thrombectomy. Treatment categories for the CT cohort were: 3 pure IV tPA, 1 only mechanical thrombectomy, 12 bridging IV tPA to endovascular and 16 with combined endovascular mechanical thrombectomy and intra-arterial thombolytic. There were no significant differences in characteristics comparing the MRI vs. CT cohorts, nor were there differences between the derivation vs. validation groups within each cohort (data not shown).

MRI Voxel-Level Model

A total of 943,320 voxels were included in the MRI model development and validation dataset. In the derivation dataset, there were 636,967 voxel observations; of these, 24% (152,603 voxels) were identified as proceeding to infarct on the follow-up scan. The final predictive model of infarct core is provided in Online Supplemental Table 1. The final MRI model included the following measures: the apparent diffusion coefficient value relative to the contralateral hemisphere (rADC), the cerebral blood flow value relative to the contralateral hemisphere (rCBF), mean transit time value relative to the contralateral hemisphere (rMTT) and the Tmax value minus the mean Tmax value of the contralateral hemisphere (sTmax). This model flags a voxel as likely to proceed as infarct if the equation incorporating these variables is greater than -1.13. Table 1 summarizes the performance characteristics for this model. Accuracy of this model was 75%, sensitivity 68%, and specificity 82%, with an ROC C=0.81 and max R2=0.34. In the validation dataset, there were 306,353 voxel observations; of these, 24% were identified as proceeding to infarct. Accuracy of the above multiparametric MRI model applied to the validation group was 71%, sensitivity 62%, and specificity 80% with an ROC C=0.76 and max R2= 0.24. Case examples of the MRI model are shown in Figures 1 and and22.

Figure 1
Case example of the voxel-based model and predictive algorithm from the MRI cohort in a patient with a non-penumbral pattern. The pretreatment color-coded Tmax perfusion image is shown in the upper left panel and the baseline diffusion image in the upper ...
Figure 2
Case example of the predictive MRI model applied in a patient with a penumbral pattern. Following recanalization, the patient developed a relatively small infarct and had a good outcome with a day 90 mRS of 1.
Table 1
Model Characteristics

CT Voxel-Level Model

A total of 1,236,917 voxels were included in the CT model development and validation dataset. A two-branch combined model was developed. The final predictive models for each branch are provided in Online Supplemental Table 1. Branch one of the predictive model uses continuous variables derived from the deconvolution process (CBF, CBV, MTT and Tmax) along with the patient-level variable of the NIHSS score at baseline, and flags a voxel as likely to proceed to infarction if the sum of the equation is greater than -0.503. There were 567,397 voxel observations in the derivation dataset; of these, 47% (267,049 voxels) were identified as proceeding to infarct on the follow-up scan. Accuracy of this model in the derivation dataset was 77%, sensitivity 77%, and specificity 77%, with an ROC C=0.84 and max R2=0.35 (Table 1). In the validation dataset, there were 208,395 voxel observations; of these, 24% were identified as proceeding to infarct. Accuracy of the above model applied to this group was 79%, sensitivity 88%, and specificity 69% with an ROC C=0.85 and max R2= 0.28.

Branch two utilizes non-continuous (boolean) variables derived from the fundamental signal characteristics (arrival time, area under curve) of voxels where perfusion parameters cannot be reliably determined using deconvolution. Non-continuous voxels were identified as having low signal-to-noise and/or non-physiologically relevant values (e.g. values out of expected range) as might occur, for example, in the region of a large vessel. The final model incorporates 12 imaging variables along with the pretreatment NIHSS score and flags a voxel as infarct if the sum of the equation is greater than -0.324. There were 389,550 voxel observations in the derivation dataset; of these, 42% (163,512 voxels) were identified as proceeding to infarct on the follow-up scan. Accuracy of this model in the derivation dataset was 85%, sensitivity 88%, and specificity 81%, with an ROC C=0.93 and max R2=0.54. In the validation dataset, there were 71,575 voxel observations; of these, 29% were identified as proceeding to infarct. Accuracy of the above model applied to this group was 78%, sensitivity 89%, and specificity 67% with an ROC C=0.86 and max R2=0.31.

Accuracy of the combined CT model incorporating both branches for the derivation dataset was 80%, sensitivity 82%, specificity 78%. Accuracy of this combined model applied to the validation group was 79%, sensitivity 88%, and specificity 69%. Case examples of the CT model are shown in Figure 3.

Figure 3
Case examples of the voxel-based model and predictive algorithm from the CT cohort. The images on the left of each pair show the predicted infarction in red and the predicted penumbra in green. The images on the right of each pair show the final infarct ...

Patient-Level Penumbral Pattern Classification

For the per-patient classification of favorable penumbral vs. non-penumbral pattern for the MRI cohort, patients were accurately predicted as having an independent final clinical outcome with reperfusion if 1) the irreversibly infarcted region identified by the MRI voxel-based predictive model was ≤ 90 ml and, 2) the infarct fraction (ratio of predicted infarct tissue within the at-risk region) was ≤ 70%. These criteria classified good final outcome with 85% (95% CI: 69%–95%) accuracy across all 34 MRI patients, and with 100% (95% CI: 88%–100%) accuracy in the 29 MRI patients who did not develop a parenchymal hematoma (Figure 4). For the per-patient classification of penumbral vs. non-penumbral pattern for the CT cohort, the same volume thresholds for core and infarct fraction employed in the MRI model were applied.

Figure 4
Scatter plot of patient outcomes based on model-predicted volume of irreversibly infarcted tissue (Predicted Infarct Volume) and proportion of at-risk tissue that is penumbral (Predicted Infarct Fraction): MRI cohort is shown in panel A, and CT cohort ...

Follow-up mRS scores were not available for 5 subjects and therefore these cases were not included in the per-patient pattern analysis. These criteria classified good final outcome with 78% (95% CI: 58%–91%) accuracy overall, and with 85% (95% CI: 62%–97%) accuracy in those patients who did not develop a parenchymal hematoma (Figure 4). The incorrect classifications in the CT cohort were due to motion in one case, causing a false prediction of large regions of infarct, and inadequate brain coverage for the three cases with a poor outcome but labeled as penumbral. If these latter cases had full brain coverage, they would have met the volume criteria for a non-penumbral pattern.


Employing a similar approach, we developed multiparametric models for both MRI and CT to predict infarct core on a voxel-by-voxel basis in patients undergoing successful vessel recanalization for acute ischemic stroke. We then developed an algorithm to identify patients as having a favorable penumbral pattern (good candidate for treatment) or a non-penumbral pattern (unlikely to benefit from treatment). The infarct core models had overall accuracies of 71–80% in identifying tissue fated to be infarcted despite reperfusion. Across both models, the favorable penumbral versus non-penumbral pattern achieved 85–100% accuracy in predicting good outcome in response to successful recanalization therapy in patients who did not develop a subsequent hematoma.

The voxel-level models predicting infarct core were derived from tissue within the at-risk region that evolved to infarction on follow-up imaging despite successful vessel recanalization. This approach is based on the premise that the distinction between infarct core and penumbra is optimally developed in patients who achieve successful early recanalization, whereas the distinction between penumbra and benign oligemia is optimally developed in patients who do not achieve early vessel recanalization. Reperfusion typically follows recanalization, resulting in rescue of salvageable tissue and in non-rescue of already infarcted tissue, permitting discrimination of penumbra from core. In contrast, when reperfusion does not occur, tissue injury continues, resulting in death of salvageable as well as already infarcted tissue, and survival only of non-threatened tissue, permitting discrimination of benign oligemia from penumbra, but not penumbra from core.

To date, the most commonly employed method of identifying the ischemic core and penumbra using MRI has been visual assessment of diffusion-perfusion mismatch. This approach is based on assumptions that a hyperintense diffusion lesion represents the irreversibly injured core and that the full extent of the surrounding perfusion abnormality identifies the ischemic penumbra.31 However, this model has been shown to have some limitations: some of the milder diffusion abnormal tissue is salvageable with early reperfusion, and much of the milder perfusion abnormal tissue is experiencing benign oligemia, insufficient to produce tissue infarction.9, 32, 33 Moreover, single parameter thresholds have only shown modest accuracy.32, 33 Due to the recognized limitations of mismatch models and single parameter thresholds, we, as others, hypothesized that models incorporating information from multiple sequences or image-based measurements would have the potential to improve prediction accuracy. These models provide an alternative approach to predicting tissue fate with the advantage of incorporating pathophysiologic information from multiple sequences and the possibility of improving model accuracy as technology improves. Future studies in larger datasets will allow comparison of our models to other approaches to predicting infarct core and penumbra, including mismatch.

There have been at least three other published studies on multiparametric MRI prediction of tissue outcome following stroke. In all three, baseline imaging was performed relatively late – within 12 to 24 hours after onset – and recanalization/reperfusion was not a prerequisite.27, 34, 35 None of these studies focused purely on differentiating core from penumbra. For CT, one study reported that an interaction term including both CBV and CBF had the highest accuracy for distinguishing core from penumbra.26, 36

Comparability of the CT and MRI models is an important consideration, particularly in light of the negative Desmoteplase in Acute Ischemic Stroke 2 (DIAS 2) clinical trial results.37 In DIAS 2, patients could be enrolled with either multimodal MRI or CT based on visual appearance of core-perfusion mismatch. There has been speculation that the two approaches for defining a penumbral pattern were not comparable in DIAS 2, since MR patients had subsequent growth of the core lesion, whereas the CT patients had growth attenuation. There was also further concern that visual assessment of mismatch may not be an optimal approach to identifying penumbra.

A number of studies comparing MRI and CT approaches to defining core and penumbra have suggested that consistent results for predicting core and penumbra can be achieved. By employing similar voxel-based model development and validation approaches in CT and MRI cohorts of patients with a confirmed proximal large vessel occlusion and subsequent recanalization, we believe our models are comparable. While the point estimates suggest that the CT infarct core model showed greater accuracy in core identification in imaged slices, the MR infarct core model may have had a countervailing advantage of whole brain coverage and greater accuracy in predicting patient outcome; however these differences should be interpreted with caution due to the overlap in confidence intervals. Both CT and MRI patient outcome algorithms used the same core and infarct fraction tissue volume thresholds to identify penumbral and non-penumbral patterns. It is important to note some differences in the model development and results. First, the MRI model was developed prior to the CT model on an older dataset; lessons learned from development of the MRI model improved our approach to development of the CT model, including a two-branch approach. Second, perfusion CT provides more accurate absolute CBF and CBV measures compared to MRI. Third, the CT models predicted a greater fraction of at-risk voxels would proceed to infarction. This may reflect differences in the two cohorts overall, or inherent differences in the models or imaging modalities themselves. An additional advantage to the voxel-based approach employed here is that all calculations have been fully automated using the same approach to AIF selection and deconvolution, as well as automated lesion volume determinations of per-patient outcome status. Fully automated image analysis avoids the potential pitfalls of variability in visual assessment of penumbral-core mismatch status.

We derived an algorithm for favorable penumbral status using our predictive model which differentiated patients with good versus poor clinical outcome after successful recanalization. This approach was based on the premise that all patients had a large vessel occlusion with a disabling clinical deficit prior to treatment, and therefore good functional outcome reflected salvage of clinically-meaningful penumbral tissue. These voxel-based tissue fate and patient-level favorable penumbral pattern predictive models are currently being employed in the NIH-funded Mechanical Retrieval and Recanalization of Stroke Clots Using Embolectomy (MR RESCUE) multicenter trial.29

There are several limitations to the present study. The total number of patients available for model development was modest; even so, the patient cohorts contributed data from approximately 2 million voxels for model development. Tissue swelling at day 7 was partially addressed by coregistration of images, but likely added noise to voxel fate mapping analyses. The area of hypoperfusion we included in our analysis region when deriving the MRI infarct core model is known to include regions of benign oligemia. However, the inclusion of benign oligemia regions among candidate voxels would not be expected to alter substantially derivation of parameters that distinguish core from penumbra. We employed multiparametric techniques to model tissue core but single parameter techniques to model tissue at-risk, and excluded voxels with a Tmax less than 2 seconds. The single-parameter thresholds for distinguishing penumbra from benign oligemia have only performed modestly well in the literature and have not been completely validated. It is also possible that voxels without a perfusion deficit at the time of imaging could have already proceeded to infarction (e.g. infarction of lenticulostriates followed by distal clot migration) or could do so ultimately. This may have led to an overestimation of model sensitivity. While use of a random coefficient logistic regression analysis accounts for correlation among voxels from the same person, it does not take account of the 3-dimensional spatial distribution of the voxels. Pooled data from other studies allowing a larger sample size will be needed to appropriately address and understand correlation patterns. Future refinements of the models are likely to achieve greater accuracy by improved definitions of the at-risk region employing multiparametric techniques (including an ADC or CBV measure), incorporating larger datasets, and employing delay-insensitive deconvolution for perfusion processing. A two-branch approach for the MRI model may also achieve improved accuracy by addressing perfusion voxels having abnormal signal characteristics due to insufficient curve fit. Of note, our model is only applicable to anterior circulation strokes and its utility in the posterior circulation needs to be independently assessed.

In summary, we have developed and validated multiparametric MRI and CT models that predict infarct core on a voxel-by-voxel basis as well as presence of favorable penumbral and non-penumbral patterns on a per-patient basis. These models show comparable sensitivity and accuracy. These approaches are being employed to stratify patients to ensure equal representation of penumbral and non-penumbral subjects among treatment groups in the MR RESCUE clinical trial.

Supplementary Material


The authors wish to thank Philips Healthcare who provided a software analysis program for processing CT perfusion imaging: Advanced Brain Perfusion (Philips Healthcare, Cleveland, OH).

Funding Source: This study was supported by a grant from the National Institutes of Neurological Disorders and Stroke (NINDS) / National Institutes of Health (NIH), P50 NS044378, and the American Heart Association Bugher Foundation Award for Stroke (Kidwell, CS) 2001–2004.


Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

Conflict of Interest / Disclosures: Max Wintermark, MD has received research grants from Philips Healthcare and GE Healthcare. Dr. Jahan has served on the speakers’ bureau for Concentric Medical, Inc. and serves as a consultant for Covidien, Inc. The University of California, Regents receive funding for Dr. Saver’s services as a scientific consultant regarding trial design and conduct to Covidien and Grifols. Drs. Jahan, Alger, Starkman, Liebeskind, Gornbein, and Saver, are employees of the University of California, which holds a patent on Merci Retriever devices for stroke.


1. NINDS rt-PA Stroke Group. Tissue plasminogen activator for acute ischemic stroke. New England Journal of Medicine. 1995;333:1581–1587. [PubMed]
2. Hacke W, Kaste M, Bluhmki E, Brozman M, Davalos A, Guidetti D, et al. Thrombolysis with alteplase 3 to 4.5 hours after acute ischemic stroke. N Engl J Med. 2008;359:1317–1329. [PubMed]
3. Hacke W, Donnan G, Fieschi C, Kaste M, von Kummer R, Broderick JP, et al. Association of outcome with early stroke treatment: Pooled analysis of atlantis, ecass, and ninds rt-pa stroke trials. Lancet. 2004;363:768–774. [PubMed]
4. Lees KR, Bluhmki E, von Kummer R, Brott TG, Toni D, Grotta JC, et al. Time to treatment with intravenous alteplase and outcome in stroke: An updated pooled analysis of ecass, atlantis, ninds, and epithet trials. Lancet. 2010;375:1695–1703. [PubMed]
5. NINDS t-PA Stroke Study Group. Intracerebral hemorrhage after intravenous t-pa for ischemic stroke. Stroke. 1997;28:2109–2118. [PubMed]
6. Wahlgren N, Ahmed N, Eriksson N, Aichner F, Bluhmki E, Davalos A, et al. Multivariable analysis of outcome predictors and adjustment of main outcome results to baseline data profile in randomized controlled trials: Safe implementation of thrombolysis in stroke-monitoring study (sits-most) Stroke. 2008;39:3316–3322. [PubMed]
7. Fisher M. Characterizing the target of acute stroke therapy. Stroke. 1997;28:866–872. [PubMed]
8. Albers GW. Expanding the window for thrombolytic therapy in acute stroke. The potential role of acute mri for patient selection. Stroke. 1999;30:2230–2237. [PubMed]
9. Kidwell CS, Alger JR, Saver JL. Beyond mismatch: Evolving paradigms in imaging the ischemic penumbra with multimodal magnetic resonance imaging. Stroke. 2003;34:2729–2735. [PubMed]
10. Albers GW, Thijs VN, Wechsler L, Kemp S, Schlaug G, Skalabrin E, et al. Magnetic resonance imaging profiles predict clinical response to early reperfusion: The diffusion and perfusion imaging evaluation for understanding stroke evolution (defuse) study. Ann Neurol. 2006;60:508–517. [PubMed]
11. Davis SM, Donnan GA, Parsons MW, Levi C, Butcher KS, Peeters A, et al. Effects of alteplase beyond 3 h after stroke in the echoplanar imaging thrombolytic evaluation trial (epithet): A placebo-controlled randomised trial. Lancet Neurol. 2008;7:299–309. [PubMed]
12. Marchal G, Benali K, Iglesias S, Viader F, Derlon JM, Baron JC. Voxel-based mapping of irreversible ischaemic damage with pet in acute stroke. Brain. 1999;122:2387–2400. [PubMed]
13. Fisher M, Albers GW. Applications of diffusion-perfusion magnetic resonance imaging in acute ischemic stroke. Neurology. 1999;52:1750–1756. [PubMed]
14. Wintermark M, Flanders AE, Velthuis B, Meuli R, van Leeuwen M, Goldsher D, et al. Perfusion-ct assessment of infarct core and penumbra: Receiver operating characteristic curve analysis in 130 patients suspected of acute hemispheric stroke. Stroke. 2006;37:979–985. [PubMed]
15. Jovin TG, Yonas H, Gebel JM, Kanal E, Chang YF, Grahovac SZ, et al. The cortical ischemic core and not the consistently present penumbra is a determinant of clinical outcome in acute middle cerebral artery occlusion. Stroke. 2003;34:2426–2433. [PubMed]
16. Lev MH, Segal AZ, Farkas J, Hossain ST, Putman C, Hunter GJ, et al. Utility of perfusion-weighted ct imaging in acute middle cerebral artery stroke treated with intra-arterial thrombolysis: Prediction of final infarct volume and clinical outcome. Stroke. 2001;32:2021–2028. [PubMed]
17. Gasparotti R, Grassi M, Mardighian D, Frigerio M, Pavia M, Liserre R, et al. Perfusion ct in patients with acute ischemic stroke treated with intra-arterial thrombolysis: Predictive value of infarct core size on clinical outcome. AJNR Am J Neuroradiol. 2009;30:722–727. [PubMed]
18. Nighoghossian N, Hermier M, Adeleine P, Derex L, Dugor JF, Philippeau F, et al. Baseline magnetic resonance imaging parameters and stroke outcome in patients treated by intravenous tissue plasminogen activator. Stroke. 2003;34:458–463. [PubMed]
19. Yoo AJ, Verduzco LA, Schaefer PW, Hirsch JA, Rabinov JD, Gonzalez RG. Mri-based selection for intra-arterial stroke therapy: Value of pretreatment diffusion-weighted imaging lesion volume in selecting patients with acute stroke who will benefit from early recanalization. Stroke. 2009;40:2046–2054. [PMC free article] [PubMed]
20. Sobesky J, Zaro Weber O, Lehnhardt FG, Hesselmann V, Neveling M, Jacobs A, et al. Does the mismatch match the penumbra? Magnetic resonance imaging and positron emission tomography in early ischemic stroke. Stroke. 2005;36:980–985. [PubMed]
21. Kakuda W, Lansberg MG, Thijs VN, Kemp SM, Bammer R, Wechsler LR, et al. Optimal definition for pwi/dwi mismatch in acute ischemic stroke patients. J Cereb Blood Flow Metab. 2008;28:887–891. [PubMed]
22. Takasawa M, Jones PS, Guadagno JV, Christensen S, Fryer TD, Harding S, et al. How reliable is perfusion mr in acute stroke? Validation and determination of the penumbra threshold against quantitative pet. Stroke. 2008;39:870–877. [PubMed]
23. Shih LC, Saver JL, Alger JR, Starkman S, Leary MC, Vinuela F, et al. Perfusion-weighted magnetic resonance imaging thresholds identifying core, irreversibly infarcted tissue. Stroke. 2003;34:1425–1430. [PubMed]
24. Guadagno JV, Warburton EA, Jones PS, Fryer TD, Day DJ, Gillard JH, et al. The diffusion-weighted lesion in acute stroke: Heterogeneous patterns of flow/metabolism uncoupling as assessed by quantitative positron emission tomography. Cerebrovasc Dis. 2005;19:239–246. [PubMed]
25. Dani KA, Thomas RG, Chappell FM, Shuler K, MacLeod MJ, Muir KW, et al. Computed tomography and magnetic resonance perfusion imaging in ischemic stroke: Definitions and thresholds. Ann Neurol. 2011;70:384–401. [PubMed]
26. Murphy BD, Fox AJ, Lee DH, Sahlas DJ, Black SE, Hogan MJ, et al. Identification of penumbra and infarct in acute ischemic stroke using computed tomography perfusion-derived blood flow and blood volume measurements. Stroke. 2006;37:1771–1777. [PubMed]
27. Wu O, Koroshetz WJ, Ostergaard L, Buonanno FS, Copen WA, Gonzalez RG, et al. Predicting tissue outcome in acute human cerebral ischemia using combined diffusion- and perfusion-weighted mr imaging. Stroke. 2001;32:933–942. [PubMed]
28. Jacobs MA, Mitsias P, Soltanian-Zadeh H, Santhakumar S, Ghanei A, Hammond R, et al. Multiparametric mri tissue characterization in clinical stroke with correlation to clinical outcome: Part 2. Stroke. 2001;32:950–957. [PubMed]
29. Kidwell CS, Jahan R, Alger JR, Schaewe TJ, Guzy J, Starkman S, et al. Design and rationale of the mechanical retrieval and recanalization of stroke clots using embolectomy (mr rescue) trial. Int J Stroke. 2012 In Press. [PMC free article] [PubMed]
30. The thrombolysis in myocardial infarction (timi) trial. Phase i findings. Timi study group. New England Journal of Medicine. 1985;312:932–936. [PubMed]
31. Baird AE, Benfield A, Schlaug G, Siewert B, Lovblad KO, Edelman RR, et al. Enlargement of human cerebral ischemic lesion volumes measured by diffusion-weighted magnetic resonance imaging [see comments] Ann Neurol. 1997;41:581–589. [PubMed]
32. Kidwell CS, Saver JL, Mattiello J, Starkman S, Vinuela F, Duckwiler G, et al. Thrombolytic reversal of acute human cerebral ischemic injury shown by diffusion/perfusion magnetic resonance imaging. Ann Neurol. 2000;47:462–469. [PubMed]
33. Chalela JA, Ezzeddine MA, Calabrese TM, Latour LL, Baird AE, Luby ML, et al. Diffusion and perfsuion changes two hours after intravenous rt-pa therapy: A prelminary report. Stroke. 2002;33:356–357.
34. Schlaug G, Benfield A, Baird AE, Siewert B, Lovblad KO, Parker RA, et al. The ischemic penumbra: Operationally defined by diffusion and perfusion mri. Neurology. 1999;53:1528–1537. [PubMed]
35. Jacobs MA, Zhang ZG, Knight RA, Soltanian-Zadeh H, Goussev AV, Peck DJ, et al. A model for multiparametric mri tissue characterization in experimental cerebral ischemia with histological validation in rat: Part 1. Stroke. 2001;32:943–949. [PubMed]
36. Murphy BD, Fox AJ, Lee DH, Sahlas DJ, Black SE, Hogan MJ, et al. White matter thresholds for ischemic penumbra and infarct core in patients with acute stroke: Ct perfusion study. Radiology. 2008;247:818–825. [PubMed]
37. Hacke W, Furlan AJ, Al-Rawi Y, Davalos A, Fiebach JB, Gruber F, et al. Intravenous desmoteplase in patients with acute ischaemic stroke selected by mri perfusion-diffusion weighted imaging or perfusion ct (dias-2): A prospective, randomised, double-blind, placebo-controlled study. Lancet Neurol. 2009;8:141–150. [PMC free article] [PubMed]