Prediction of stroke outcome with echoplanar perfusion- and diffusion-weighted MRI

Ischemic stroke is a common and heterogeneous condition. Complex and variable factors, particularly the degree and duration of hypoperfusion, determine the final infarct size.¹ Intense interest has focused on the ischemic penumbra, defined as functionally impaired but potentially salvageable ischemic brain tissue surrounding an irreversibly damaged infarct core,^2,3 because identification of the penumbra may allow rational selection of therapeutic agents in acute stroke.

Our group has shown previously that hypoperfusion volumes measured using SPECT predict acute stroke outcome,⁴ as have others.^5,6 Although concomitant demonstration of the ischemic core is possible with PET,^7,8 this is only available in a small number of research centers, with practical difficulties and safety considerations limiting the potential for serial studies. Significant advantages are offered by rapid MRI techniques, such as echoplanar magnetic resonance imaging (EPI),⁹ which can detect lesions with impaired perfusion (perfusion imaging) and cellular failure (diffusion-weighted imaging) concomitantly.^10-16 This report utilizes these techniques in patients with acute cerebral infarction.

Perfusion imaging (PI) makes use of the signal loss that occurs during the dynamic tracking of the first pass of an IV paramagnetic contrast agent. A concentration-time curve is then derived. Different hemodynamic measurements such as bolus arrival time, relative mean transit time, and relative cerebral blood volume and flow can then be calculated and displayed as perfusion maps,¹⁷ although as yet no consensus exists on the ideal parameter for measurement of absolute perfusion.¹⁸ To date, little attention has been given to the predictive value of PI in acute stroke.

Diffusion-weighted imaging (DWI) measures the diffusional movement of water within the brain and can identify acute ischemic injury within minutes of stroke onset. Increased DWI signal intensity is a function of a restriction in this diffusional movement.^10,19,20 Although diffusion abnormalities typically evolve into infarction, very early reperfusion in experimental animal models may lead to lesion reversal.^20,21 Such reversibility has not been shown convincingly in humans, leading to speculation that DWI could be used to distinguish irreversibly damaged tissue.^22-24 DWI lesions may predict clinical severity and outcome in acute stroke.²⁵

Combined PI and DWI has the potential to be more powerful than either study alone in predicting infarct outcome.^14,26 To date, however, there have been few reports of the simultaneous use of these techniques in acute stroke. In one study,¹³ the combination of PI and DWI was superior to conventional MRI in predicting improvement following severe acute stroke. Comparison of PI and DWI lesions has led to different patterns being described, with a large PI deficit including a smaller DWI abnormality postulated to predict ischemic lesion enlargement.^14,16 Further characterization of these patterns and their potential for evolution is required.

In this prospective, serial observational study we examine the predictive utility of PI and DWI in determining evolution and outcome in acute stroke. We hypothesized that acute regional perfusion deficits would correlate with clinical severity, clinical outcome, and final infarct size. We also hypothesized that the combined use of PI and DWI would be more powerful than either alone at identifying patterns in which ischemic regions have the potential to expand.

Methods. Patients. Patients with sudden onset of focal neurologic deficit consistent with hemispheric ischemic stroke were recruited prospectively from the Stroke Service of the Royal Melbourne Hospital. Stroke onset was defined as the last time the patient was known to be without neurologic deficit. We aimed to describe patterns of acute and subacute PI and DWI abnormalities, identify potentially viable ischemic tissue, and evaluate stroke outcome at a time when neurologic recovery had largely plateaued. We therefore required acute EPI studies within 24 hours(preferably within 6 hours) of stroke onset, subacutely (days 3 to 5), and at outcome (day 90).

Patients were excluded if they had cerebral hemorrhage, preexisting significant nonischemic neurologic deficits (including dementia or extrapyramidal disease), or a history of prior stroke that would hamper interpretation of clinical and radiologic data. There were no age, gender, or handedness exclusions. Patients subsequently enrolled in acute stroke therapy trials were not excluded. We postulated that any potential benefit of these therapies would not weaken correlations between acute imaging and outcome measurements, and inclusion of these patients would not therefore confound the examination of our hypotheses. The study was performed with the approval of the ethics committee at our institution, and written informed consent was obtained from the patient or next of kin.

Clinical assessment. The Canadian Neurological Scale (CNS), which renders a validated neurologic impairment score,²⁷ was performed just before the acute and subacute EPI studies. Outcome clinical assessments were performed on the same day as the final MR study and consisted of a repeat CNS score, and scores derived from the Barthel Index (BI) and the Rankin Scale (RS).²⁸ The BI is a validated functional disability score, and the RS is a validated handicap scale. These clinical scales were used because they measure different aspects of recovery following stroke. All clinical assessments were performed by a neurologist or neurology resident trained in their administration, and were administered without knowledge of the PI or DWI results.

Imaging. All MR scans were obtained using a 1.5-T EPI-equipped whole-body scanner (Signa Horizon SR 120; General Electric, Milwaukee, WI) using a standard protocol optimized to obtain high-quality images as rapidly as possible in ill and potentially uncooperative patients. Sequences were always performed in the same order, with an initial T1-weighted sagittal localizer, the diffusion-weighted sequence, perfusion sequence, a proton density and T2-weighted fast-spin double-echo sequence (repetition time[TR]/echo time [TE]/TE, 3,000/10/60 msec), EPI spin echo sequence, phase contrast MR angiography, and finally a contrast enhanced T1-weighted sequence. Similar slice positions were used to facilitate comparisons. Only the DWI, PI, and T2-weighted image are reported here, with imaging times of 1 minute 23 seconds, 1 minute 21 seconds, and 1 minute 41 seconds, respectively, and a total "table time" for these sequences of 15 to 20 minutes.

Diffusion imaging. Diffusion-weighted imaging was obtained using a multislice, single-shot spin-echo EPI sequence. The rapid acquisition times made cardiac or respiratory gating and special head restraint unnecessary. Slice thickness was 6 mm with a 1-mm gap, with the number of slices set to include the whole brain (average of 15). Matrix size was 256× 128 (128 × 128 for the first seven patients), field of view was 40 × 20 cm, and TR/TE was 6,000/100 msec (TE was 75 msec for the first five patients). Diffusion gradient strength was varied between 0 and 22 mT/m, resulting in five b values of increasing magnitude from 0 up to 1,200 sec/mm², where b values are related to the strength, duration, and separation of the diffusion-sensitizing gradients as described previously.²⁹ Because the diffusion of water may be directionally dependent (anisotropic), measurement in only one direction can lead to incorrect image interpretation.³⁰ The diffusion gradient was therefore applied in each of three orthogonal directions (x, y, z), and an average of these measurements was calculated to give the trace of the diffusion tensor, which is reported to minimize the effects of diffusion anisotropy.³¹

Perfusion imaging. Perfusion images were obtained using dynamic first-pass bolus tracking of gadolinium diethylenetriamine penta-acetic acid(Gd-DTPA) using an EPI spin-echo sequence with a TR/TE of 2,000/70 msec. The Gd-DTPA bolus (0.2 mmol/kg) was administered by manual IV injection over approximately 5 seconds via a large-bore cannula in the antecubital fossa. The concentration-time curve obtained was processed on a voxel-by-voxel basis, allowing determination of an observed or relative mean transit time(rMTT) map, where the rMTT is related to the sum of the true MTT plus the injection time.¹⁷ We found that in addition to giving the most visually distinct perfusion deficit border, as previously reported,¹⁶ the rMTT map also resulted in perfusion deficits of greater volume than with other hemodynamic measurement maps, suggesting it was more representative of the maximum perfusion impairment.

Data analysis. Postprocessing of MR images was performed using customized software based on a commercial image analysis application(Advanced Visualization Systems, Waltham, MA) using an Indigo workstation(Silicon Graphics Inc., Mountain View, CA). DWI volumes were measured using the maximum diffusion sensitivity isotropic image because this showed the greatest contrast between the hyperintense infarct and the surrounding tissue. The edge of the infarct was identified visually and regions of interest (ROIs) were outlined using a manual pixel-wise method. The area of the ROI was multiplied by the slice thickness plus the interslice gap, and then summed. Known anatomic markings such as ventricles and large sulci were taken into account, although with cortically based infarcts the outline bridged small cortical sulci. A difference in lesion volumes of more than 10% was considered significant and unlikely to be due to measurement error. Analyses used the average of two measurements taken on separate occasions by two neurologists trained in the technique and blinded to clinical data. Intra- and inter-observer agreement was high (variability less than 5%), which compared favorably with other reports.¹⁶

Statistical analysis. Demographic and time-of-scan data are presented as mean values ± SD. Dependent variables are compared using nonparametric techniques except when normality of data could be proved, in which case parametric equivalents are preferred and presented as mean difference with 95% CIs. Corrections are made for paired data and unequal variance when required. Pearson's product moment correlation coefficient was used to measure the strength of association between lesion volumes and clinical measures (CNS, BI, and RS), and final infarct size. Results were considered statistically significant at the 5% level, corrected for multiple comparisons when appropriate.

Results. Eighteen patients (12 men, 6 women; age, 67.4± 13.6 years; range, 37 to 90 years) were recruited prospectively between September 1996 and June 1997. Clinical demographic data are presented in table 1. Fourteen patients had cortically based lesions in the middle cerebral artery (MCA) territory, one had a posterior cerebral artery territory infarct, and three had striatocapsular lesions. One patient was enrolled in the European Cooperative Acute Stroke Study II trial of tissue plasminogen activator,³² and three others were enrolled in trials of putative neuroprotective agents (one of tirilazad and two of Cerestat), although assigned study treatments are not yet known for these patients.

A total of 52 MRI studies were performed. Time from infarct onset to acute study was 12.2 ± 7.8 hours (range, 3.0 to 23.3 hours), with seven studies within 6 hours. Time to subacute study was 4.7 ± 2.6 days(range, 1.9 to 11 days), and time to outcome study was 84 ± 32 days(range, 7 to 120 days). Fifteen patients completed the full protocol. Patient 6 (see table 1) died of an unrelated cardiac event 57 days after a large left MCA territory infarction, Patient 16 refused outcome MRI studies, and acute PI was not obtained in Patient 18 due to technical difficulties. Results from the remainder of the protocol in these three patients are included in the analysis when possible.

Group clinical comparisons. Acute neurologic deficits as measured by the CNS score were predictive of outcome clinical severity (CNS, r = 0.73, p < 0.001; BI, r = 0.67, p < 0.005; RS, r = -0.67, p < 0.005, where n = 17) and final infarct size (n = 16, r = -0.54, p < 0.05).

PI deficit correlations. Significant correlations were found between the acute PI volumes and the acute, subacute, and outcome CNS scores(table 2). There were also significant correlations with outcome BI and RS. Hence, acute PI deficit volumes correlated with the clinical state at all time points. In contrast, subacute PI deficit volumes only correlated significantly with outcome BI. Both acute and subacute PI deficit volumes correlated with final infarct size on T2-weighted imaging.

DWI abnormality correlations. The acute DWI abnormality volumes correlated with the acute CNS score, although not as strongly as the acute PI deficits (r = -0.44 versus r = -0.63; see table 2). There were also correlations with all clinical outcome measurements (CNS, BI, and RS). Subacute DWI lesion volumes correlated with the subacute CNS score and all outcome clinical measurements(CNS, BI, and RS). Both acute and subacute DWI lesion volumes correlated strongly with final infarct size on T2-weighted imaging.

Perfusion and diffusion abnormality patterns. Comparison of acute PI and DWI lesions showed three distinct patterns of abnormality in the 17 patients who had both studies (table 3). In 11 patients (65%), the acute PI lesion was larger than the DWI lesion (PI > DWI group). In the remaining six patients, the PI deficit was either smaller than the DWI deficit (two patients, 12%) or was absent (four patients, 23%). Combined, these patients comprised a perfusion-smaller-than-diffusion-lesion group (PI < DWI group). Six of the 17 patients who had both acute PI and DWI studies were studied within 6 hours of stroke onset. All six of these patients were in the PI > DWI group. Of the 11 patients first studied between 6 and 24 hours, five were in the PI > DWI group and six were in the PI < DWI group. This difference was significant statistically, with earlier time of acute study predicting greater likelihood of the PI lesion being larger than the DWI lesion (Fisher's exact probability test, p < 0.05).

Comparison of the groups showed that PI > DWI patients were studied earlier, and acute PI and DWI lesions were significantly larger than the combined PI < DWI group (see table 3). By the outcome studies, patients in the PI > DWI group had worse outcome, with a larger final infarct volume (difference in mean volumes, 58.2 cm³; CI, 7.2 to 109.8 cm³; p < 0.05; Student's t-test), lower CNS and BI scores, and a worse RS score. However none of the clinical scale score differences reached statistical significance(Mann-Whitney U test, p > 0.05), presumably because of small patient numbers.

Changes in lesion volumes over time. Serial evaluation of PI and DWI showed changes in lesion volume over time. Overall, the acute DWI lesion was only 63 ± 67% of the acute PI deficit volume. By the subacute study, however, the DWI lesion had expanded by 60 ± 83% (mean volume difference, 17.8 cm³; CI, 3.4 to 32.2 cm³;p < 0.05), whereas the PI deficit had contracted 65 ± 32% (mean volume difference, 46.1 cm³; CI, 15.0 to 77.2 cm³;p < 0.05).

When the 11 patients in the PI > DWI group were analyzed separately, the same pattern of evolution was seen. The PI deficit shrunk by 64 ± 34% (mean volume difference, 69.7 cm³; CI, 26.6 to 112.8 cm³;p < 0.05), whereas the DWI lesion expanded by 62 ± 81% between the acute and subacute studies (mean volume difference, 27.9 cm³; CI, 5.4 to 50.5 cm³;p < 0.05), although it remained smaller than the initial PI deficit. This suggests that hypoperfused tissue surrounding a diffusion lesion core is at risk. Figures 1 and 2 show this evolution in a patient with an acute PI > DWI pattern. In contrast, in the other six patients combined (PI < DWI group), the DWI lesion remained stable between the acute and subacute studies (mean volume difference, 0.75 cm³; CI, -0.16 to 1.67 cm³;p > 0.05), and similarly there was no significant change in the PI deficit volume (mean volume difference, -3.9 cm³; CI, -0.4.2 to 12.0 cm³; p > 0.05). These results suggest that in patients with an acute PI deficit that is smaller (or absent) compared with the DWI lesion, no significant expansion of the DWI deficit occurs and therefore there may be no tissue at risk of ischemic expansion.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1. Acute (A-C) and subacute (D-F) MR studies of Patient 11. (A) Acute magnetic resonance angiography (MRA) shows occlusion of the left middle cerebral artery (MCA; arrow). (B) Acute perfusion imaging (PI) deficit (relative mean transit time map) in the left MCA territory shows as a hyperintense region. (C) Acute isotrophic diffusion-weighted imaging (DWI) shows infarct as hyperintensive. Note that the acute PI deficit is a larger than the DWI lesion. Acute image quality is slightly degraded by patient movement. (D) Subacute MRA shows reperfusion of distal left MCA. (E) Subacute PI shows resolution of PI deficit consistent with reperfusion. (F) DWI lesion has expanded between the acute and subacute studies.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2. Outcome T2-weighted image (Patient 11) shows final infarct size approximate subacute diffusion-weighted imaging lesion.

Finally, comparison of PI and DWI lesions with final infarct size(T2-weighted image) showed the mean final infarct size was 46 ± 44% of the mean acute PI deficit volume (mean volume difference, 34.5 cm³; CI, 6.5 to 62.5 cm³; p < 0.05), whereas the acute DWI lesion was 41 ± 114% the volume of the final infarct size (mean volume difference, 12.6 cm³; CI, -31.9 to 6.7 cm³; p = 0.18). These relations changed over time with no difference between the final infarct size and either the subacute PI deficit (mean volume difference, 15.6 cm³; CI, -34.5 to 3.3 cm³; p > 0.05) or the subacute DWI lesion (mean volume difference, 5.8 cm³; CI, -0.5 to 12.2 cm³; p > 0.05).

Subanalysis of the PI > DWI group showed that the acute PI lesion volume was significantly larger than the final infarct size (mean volume difference, 52.0 cm³; CI, 0.01 to 13.4 cm³;p < 0.05). Although the acute DWI lesion volume was smaller than the final infarct size, this difference did not reach significance (mean volume difference, -22.6 cm³; CI, -53.5 to 8.4 cm³; p= 0.13). By the subacute scan there was no significant difference in volumes between the PI lesion volume (mean volume difference, -22.6 cm³; CI,-53.7 to 8.44 cm³;p > 0.05) or DWI lesion volume (mean volume difference, 5.7 cm³; CI, -4.6 to 15.9 cm³; p > 0.05) and the final infarct size.

In the PI < DWI group, there was no significant difference between the final infarct size and acute PI (mean volume difference, -0.9 cm³; CI,-5.3 to 3.5 cm³; p > 0.05), acute DWI (mean volume difference, 4.6 cm³; CI, -4.5 to 13.8 cm³; p > 0.05), subacute PI (mean volume difference, -5.6 cm³; CI, -12.1 to 0.9 cm³;p > 0.05), and subacute DWI (mean volume difference, 5.5 cm³; CI, -4.6 to 15.6 cm³; p > 0.05).

Discussion. The major findings of this study concern the utility of PI and DWI to predict infarct evolution and eventual outcome from single acute studies. Acute PI lesions correlated well with acute clinical state, as measured by the CNS, so that larger acute PI lesion volumes are associated with more severe clinical deficits at the time of scanning. This finding supports our hypothesis that PI measures of acute hypoperfusion predict the degree of acute clinical impairment, and is in accord with experimental models in which compromise of cerebral function occurs as perfusion falls below a critical threshold.^3,33,34 Between the acute and subacute studies, the PI lesion contracted by 65 ± 32%, consistent with at least partial reperfusion, with almost all of this contraction occurring in the PI > DWI group. In no case was an increase in PI lesion volume seen. Therefore, the initial PI lesion represented the maximum possible infarct size and, in the absence of further vessel occlusions or closure of collateral supply, the worst potential clinical outcome. Acute PI lesion volumes also predicted clinical outcome and final infarct size. To our knowledge these are novel findings using EPI.

Subacute PI lesions correlated poorly with outcome clinical state. That the PI lesion volume predicted clinical state and outcome when measured acutely but not subacutely, with reperfusion between the acute and subacute studies leading to PI lesions smaller than the DWI lesion volume, suggests at least some of the reperfusion was nonnutritional, which is defined as perfusion of irreversibly damaged tissue that does not persist at the chronic phase.^35-37

Because acute PI lesions correlate with acute clinical state, it would be reasonable to expect some clinical improvement with reperfusion occurring before the onset of irreversible tissue damage, as seen dramatically with the"spectacular shrinking deficit."³⁸ When we analyzed individual patients, improvement in acute neurologic state was associated with reduction or resolution of the PI lesion volume by the subacute study in all but three patients (see table 1). Of these exceptions, Patient 4 had a very large acute DWI lesion, suggesting significant damage had occurred before reperfusion. In the other two patients(Patients 3 and 13), the subacute PI deficit was smaller than the initial DWI lesion, suggesting reperfusion had been nonnutritional. It is also possible that the infarctions in these patients were in eloquent regions of the brain such that infarct location determined final clinical score. Patient 6 had a CNS score that remained constant, however there had been no reperfusion between the acute and subacute studies.

Acute DWI lesion volumes did not correlate as closely with acute clinical deficits as did acute PI lesions. This finding implies acute DWI lesions do not reflect closely the extent of functionally compromised tissue. In a previous report,²⁵ a stronger correlation was found between the acute DWI lesion volume and acute clinical state, although perfusion deficits were not reported in that study. The use of different clinical scales (CNS in our study and National Institutes of Health Stroke Scale²⁵) may account in part for this difference, but our findings suggest that DWI lesions may be unreliable markers of acute clinical state. Overall, acute DWI lesions were smaller than the final infarct size, but increased in volume by a mean of 60 ± 83% (in 56% of patients) between the acute and subacute studies, so that there was no significant difference between the subacute DWI lesion volume and final infarct size. This DWI expansion was almost entirely due to the increase seen in the PI > DWI group of patients, with no enlargement beyond the initial DWI lesion in the other groups. This confirms a previous report in which seven of 10 patients with an acute PI > DWI pattern, showed subsequent expansion of the DWI lesion, with no change in DWI lesion in all three patients with a PI < DWI pattern.¹⁶ This suggests that a poor correlation between the acute DWI lesion volume with acute clinical state is likely if the proportion of PI > DWI patients is large.

In addition, acute DWI lesion volumes correlated with clinical outcome and final infarct size, with correlations of a similar magnitude to previous reports,^16,25 with the exception of a stronger correlation with outcome functional state in the present study. The presence or absence of DWI lesions has been used as a surrogate marker of response to treatment in animal studies,^39-41 and it has been proposed that DWI be used to screen patients before their enrollment in acute stroke trials.^13,25 This study supports such a role for DWI in human studies, but also indicates that PI provides complementary information and is equally important.

Subacute DWI lesion volumes correlated with subacute clinical state and predicted clinical outcome and final infarct volume. In no patient did we see more than a 10% decrease in DWI volume between the acute and subacute studies, and all DWI lesions appeared to result in some permanent residual infarction. By the subacute study, the DWI lesion volume did not differ significantly from the final infarct size, suggesting the final extent of the infarction had been largely determined and was coextensive with the subacute DWI lesions. Therefore, although reperfusion led to subacute PI lesions smaller than the final infarct volume, with poor correlation with clinical outcome, subacute DWI lesions were similar in size to final infarct volume and did correlate with clinical outcome.

Caution is required in equating DWI lesions directly with irreversible tissue damage, because very early reperfusion in animal models may lead to reversal of diffusion abnormalities.^20,21 Although overall there may have been no significant difference between the mean subacute DWI lesion volume and mean final infarct size, in eight patients the subacute DWI volume was more than 10% larger than the final infarct size. This phenomenon has been noted previously and explanations have included the presence of vasogenic edema at the time of scanning and late atrophy.^14,16 An additional possibility is that the DWI lesion may in part be reversible,¹⁴ with a decrease in the diffusional movement of water at the periphery of the lesion, enough to show as hyperintensity on the DWI, but not below a threshold at which irreversible damage occurs. These considerations aside, reversal of DWI lesions has not been shown convincingly in human stroke, and diffusion abnormalities typically evolve into infarction,^10,13,14 which taken in conjunction with these findings suggest DWI may be used as a marker of irreversible infarction.

There are six possible patterns of combined PI and DWI lesion.^11,14,16 The most frequent is 1) PI > DWI lesion, found in 65% of patients and which compares well with 55%¹⁴ and 77%¹⁶ in two previous reports; 2) PI and DWI deficits of similar size, and although this pattern was not seen in the present study, it has been described;^14,16 3) PI < DWI deficit, found in 12% of our patients; 4) DWI deficit but no PI deficit, found in 24% of our patients; 5) PI deficit without DWI deficit, which has not to our knowledge been reported but, anecdotally, has been seen in one of our recent patients in which it was associated with a transient neurologic deficit and resolution of the PI deficit by the subacute study; and 6) neither PI nor DWI lesion despite clinical deficit. Pattern 6 was not seen in the current study; however, it has been described in two patients with transient clinical deficits and a history of complicated migraine.¹⁴ As well as being associated with pattern 5, TIAs might also be expected to be associated with either pattern 6 or pattern 4 (a small subclinical infarct), as suggested by CT studies in which small permanent lesions have been documented despite clinical recovery.⁴²

The most frequent pattern was an acute PI lesion volume greater than the DWI lesion volume (PI > DWI group), present in all patients studied within 6 hours of stroke onset, and less common with later scanning. Acute lesion volumes in this group were not static but approached each other. Mean PI lesions contracted and DWI lesions expanded, so that the subacute DWI lesions approximated final infarct size, but were still smaller than the acute PI deficits. This evolution demonstrates that not all hypoperfused tissue is destined for irreversible infarction. We suggest there is a rim of hypoperfused tissue surrounding the DWI lesion core, in which the rate of transition to viable perfusion is important in determining whether infarction, as represented by the DWI lesion, will spread.

No such evolution was seen in the combined PI < DWI group. We postulate that reperfusion had occurred (partial in the PI < DWI subgroup, and full in the DWI but no PI subgroup) before scanning but after the onset of irreversible tissue damage. Other possible explanations for this pattern might relate to the establishment of collateral circulation despite a persistently occluded feeding artery, or expansion of the ischemic zone beyond the initial perfusion deficit due to excitotoxic damage with spreading depression and/or recurrent depolarizations.^33,43,44 The lack of evolution of the DWI lesion and concordance of final infarct volume with the acute DWI lesion in these PI < DWI groups strongly suggests that there was no potentially salvageable or "at-risk" tissue within the infarct at the acute scan, and hence no enlargement of the DWI lesion was likely. This observation supports the hypothesis that the DWI lesion largely represents irreversibly infarcted tissue.

If DWI lesions are indeed markers of irreversible infarction, then the evolution seen in the PI > DWI patients would be consistent with expansion of the ischemic core into at-risk hypoperfused tissue, mirroring changes in the ischemic penumbra. It is in these patients that treatment strategies aimed at restoring blood flow would be expected to be of most benefit(table 4). Whether patients with a PI > DWI pattern, presenting beyond the current 3-hour thrombolysis time window recommendations,⁴⁵ would benefit from reperfusion therapy remains to be determined. Factors that need to be studied include the current 3-hour time window, which may be able to be extended in the presence of a significant PI > DWI mismatch, and whether the size of the initial DWI lesion will influence decisions on acute stroke therapy. These issues are complex and require further investigation. When no further DWI lesion evolution is likely (for example, in the acute PI < DWI groups or subacutely in all patients), the extent of the DWI lesion is a valid marker of final infarct volume. These patients would not be expected to benefit from thrombolytic or revascularization therapies, with the added risks of such treatment, but may be candidates for neuroprotective strategies. This hypothesis also requires further investigation.

This study has demonstrated the independent prognostic value of EPI, and the importance of combining both PI and DWI in the evaluation of acute ischemic stroke patients so that predictions of infarct evolution and outcome may be made. The early finding of a perfusion deficit that is larger than a diffusion deficit implies the presence of at-risk tissue in the putative ischemic penumbra, which may be salvageable with therapeutic restoration of blood flow. When perfusion deficits are absent or smaller than the diffusion deficit, neuroprotective strategies would be more appropriate. In addition, this study suggests that the infarct core has stabilized by 5 days after stroke onset, and this seems an appropriate time point for reevaluation of the effect of stroke therapy. We believe the addition of PI and DWI sequences to current stroke imaging protocols, with imaging times of less than 2 minutes each, does not slow significantly the acute investigation of stroke patients, nor delay unacceptably delivery of acute stroke therapy. Techniques such as those reported here may, in the future, become very useful in deciding acute stroke therapy in individual patients.