Correlation of perfusion- and diffusion-weighted MRI with NIHSS score in acute (<6.5 hour) ischemic stroke

Echo planar diffusion-weighted (DWI) and perfusion-weighted (PWI) MRI are novel MRI imaging methods shown to be exquisitely sensitive to cerebral ischemia even minutes after onset.^1-3 Because of their high sensitivity to early cerebral ischemia, both DWI and PWI may be of great value in the assessment of acute stroke patients. Further, because most potential stroke treatments are administered within the first 6 to 8 hours after symptom onset, some researchers speculate that DWI and PWI could be used as adjunctive measures of the effectiveness of stroke interventions, potentially reducing the need for costly clinical trials.^4,5 However, few data are available concerning the relationship between DWI- and PWI-detected abnormalities and severity of neurologic deficit within the <8-hour window commonly used for treatment of acute stroke. Such studies are of critical importance before these imaging techniques can be considered as measures of neurologic outcome.

This study sought to determine the association between MRI lesion size and location as detected by early (<6.5 hour) DWI and PWI, and the degree of neurologic deficit as assessed by an established neurologic measure of stroke severity, the National Institutes of Health Stroke Scale (NIHSS). We hypothesized that lesion size as determined by these early echo planar-generated MRI scans would be highly predictive of early neurologic outcome as assessed by the 24-hour NIHSS score. In addition, the relationship between DWI and PWI findings and subsequent T2W infarct size at 7 days after infarction was assessed.

Methods and patients. Research subjects were selected from patients evaluated in a variety of acute DWI and PWI studies currently underway at Stanford Hospital. Patients were enrolled using DWI and PWI protocols approved by the Stanford Human Subjects Committee. Patients who were admitted to the Stanford Stroke Service with an acute ischemic stroke within 6.5 hours of symptom onset were eligible for enrollment. Time of stroke onset was determined by history to an accuracy of 1 hour or less. Patients who awoke from sleep with their deficit were excluded if the onset time could not be determined within 1 hour. Only patients with presumed supratentorial stroke were eligible for participation. Patients participating in acute stroke treatment trials with investigational agents as well as treatment with open label rt-PA were also eligible. All patients underwent MRI scanning within 6.5 hours of symptom onset. There was no limitation based on the severity of clinical deficit. Patients who could not tolerate the MRI scan due to claustrophobia or who were believed to be too unstable for MRI scanning were excluded from the study. Informed consent was obtained from the patient or an appropriate representative in all cases.

All MRI scanning was performed by echo planar imaging (EPI) using the same 1.5-T General Electric Medical Systems (Waukesha, WI) Signa magnet. Multislice whole brain DWI was performed using 12 slices (TR = 6,000 msec, TE= 110 msec, field of vision = 24 cm, slice thickness = 5 mm with a 2.5-mm skip between slices, image size = 128 × 128, B values = 0, 741). DWI images were acquired in X, Y, and Z planes with and without an inversion pulse (FLAIR sequence). EPI diffusion images were processed to generate average (trace) apparent diffusion coefficient (ADC) maps. B values (0, 741) were measured twice and averaged to improve the accuracy of the ADC determination.

Perfusion-weighted imaging was performed using the same EPI MRI equipment after injection of 20 mL gadolinium (0.2 mmol/kg). PWI acquisition values were TR = 2,000 msec, TE = 40 msec, and 35 time points. Other parameters were the same as for DWI. Total imaging time was approximately 3 minutes. Perfusion images were processed to generate maps of time-to-bolus peak (TTP), which is the time from the start of the scan to the peak bolus effect, and relative cerebral blood flow (rCBV), which is the integral of the area under the bolus transit curve. Approximately 400 individual scans were acquired within 3 minutes. Total scan time for the entire protocol was between 15 and 20 minutes. Postprocessing required an additional 15 to 20 minutes, but raw data images were immediately available for interpretation.

Scores on the NIHSS were determined at the time of presentation and 24 hours later by a neurologist specializing in stroke treatment. All physicians involved in this study had been certified in the performance of NIHSS examinations as part of ongoing research studies.

Diffusion-weighted and PWI lesion volumes were measured off-line. Regions of increased signal intensity from the DWI scans were traced by hand with the aid of an image analysis system (MRVision, Menlo Park, California). A built-in thresholding program was used to delineate the areas of infarction within this hand-traced area. This routine uses a preset region of interest(ROI) measuring approximately 4 × 4 voxels that is placed in the lesion. A "seeding" function is then used to compute the number of voxels that fall within the range of voxel values contained within the ROI. For DWI the cut-off was determined by the maximum and minimum intensities found around the sample area of interest. DWI abnormalities corresponded to areas of decreased signal intensity on ADC maps in all cases. For PWI, the cut-off was calculated in a similar fashion but on a pixel-by-pixel basis rather than for a larger ROI, because of greater heterogeneity within the PWI TTP maps. These areas were subsequently used to calculate DWI and PWI lesion volumes and then compared with 24-hour NIHSS scores. All DWI and PWI volume measurements were performed by a single blinded observer (D.T.) to avoid interobserver variability. To rule out a major difference between observers, interobserver variability was measured in a subset of scans measured by two different observers (M.Y., D.T.). The mean interobserver variability in this subset was 17%.

Acutely and at 7 days, SE T2W MRI scans were performed. These images were measured independently of the DWI and PWI data. T2W lesion volumes were determined by techniques similar to those used for the DWI images. All diffusion, perfusion, and T2W lesion volumes were measured by observers blinded to the patients' clinical status.

Statistical methods. Pearson product moment correlation was used to determine the association between lesion volume and neurologic stroke scale score using a statistical software package (Sigmastat, Jandel Scientific, San Rafael, CA). Similar calculations were made to compare the significance of the relationship among NIHSS, PWI, and T2W MRI. Paired t-test was used to determine whether there were significant differences in lesion volume between the initial DWI and PWI scans and the 7-day T2W scans.

Results. Ten patients were assessed within 6.5 hours of symptom onset. Mean time between stroke onset and initial MRI scan was 4.8 hours. Both DWI and PWI imaging tests were performed in all patients (mean age = 71 years, 6 women and 4 men). Initial and 24-hour NIHSS scores as well as neuroimaging results are summarized in the table. Seven patients (F.T., E.D., P.R., L.N., W.B., E.W., E.E.) participated in various acute stroke neuroprotective studies (Cervene [nalmafene; Baker Norton, Miami, FL], Prosynap [lubeluzole; Janssen Pharmaceutica, Titusville, NJ], and Cerestat [aptiganel hydrochloride; Boehinger Ingleheim, Richfield, CT]). Two patients were given open label rt-PA (J.B., H.R.). One patient(M.H.) received no therapy. Six patients were participants in other ongoing MRI studies.

Perfusion-weighted TTP maps showed an area of contrast delay in all cases suggesting impaired perfusion in the area of subsequent infarction. There was a strong correlation (r = 0.91, p < 0.001) between PWI lesion volume and 24-hour NIHSS score (figure 1). In all patients, the PWI changes were larger than (n = 9) or equal to (n = 1) the DWI changes. On average, the PWI lesions were 50% larger than the DWI lesions, although this did not reach statistical significance (SD = 30%, p = 0.12). In one patient treated with rt-PA (H.R.), the volumes of the DWI and PWI lesions were approximately equal. rCBV maps did not consistently detect any significant abnormality even when DWI was clearly positive and were not subsequently analyzed in this study.

In nine of 10 patients, DWI detected an abnormality (seetable). In eight of nine patients in whom DWI showed an ischemic lesion, the anatomic distribution of the lesion directly corresponded to the clinically apparent neurologic deficit. In these eight DWI-positive patients, there was a strong linear correlation between 24-hour NIHSS and DWI volume (r = 0.96, p < 0.001, seefigure 1). Although the overall correlation between NIHSS and initial DWI volumes including the equivocal and negative DWI studies was more modest, it was still significant (r = 0.67, p = 0.03). Conventional T2W MRI showed obvious ischemic changes in only one patient (P.R.) who underwent a scan at 5.5 hours (seetable).

Diffusion-weighted imaging failed to detect any changes suggesting acute infarction in one patient (L.N.), despite a large area of perfusion delay(figure 2). This patient was scanned earlier than any other patient (2.5 hours). Clinically, her deficit was substantial (NIHSS = 24). On subsequent 24-hour MRI scan, a DWI abnormality developed that closely matched her initial perfusion deficit. There was no change in the severity of her neurologic deficit over this time.

In another case (M.H.), DWI showed a small (0.5 cm³) lesion despite the patient's major initial neurologic deficit (NIHSS = 17). By 24 hours, the patient had moderately improved (NIHSS = 12). Initial PWI showed a moderate (11.9 cm³) volume of perfusion delay. The patient had complete recovery of her neurologic deficits and returned to her previous baseline by 30 days.

In 8 of 10 patients, T2W MRI scans were obtained at 7 ± 2 days. There was a high correlation between initial DWI and PWI findings and 7-day T2W MRI-scan abnormalities in these patients (r = 0.99, p < 0.00001 for both). In these eight patients the overall T2W lesion size was 32% (p = 0.48) larger than the initial DWI volume(figure 3). In contrast, the T2W lesion averaged 23% smaller than the initial PWI abnormality (p = 0.80). However, there was significant variability among patients (seefigure 3). On serial scans, there was a progressive enlargement in the area of DWI damage seen in seven of eight patients(figure 3 and 4). No neurologic alterations were associated with these MRI changes.

Discussion. We found a significant correlation between lesion size as measured by DWI or PWI performed within 6.5 hours of symptom onset and a commonly used and validated neurologic assessment measure, the NIHSS. These findings suggest that DWI and PWI may serve as appropriate adjunctive measures of the severity of cerebral ischemic injury. In addition, they appear to be accurate predictors of persistent neurologic deficit at 24 hours. Moreover, initial DWI and PWI volumes were highly correlated(r = 0.99, p < 0.00001) with 7-day T2W MRI in all patients, confirming the accuracy of this technique in the detection of early hemispheric ischemia.

The severity of the neurologic deficit was quantified using a well-established and validated neurologic assessment measure,^6-9 the NIHSS. Use of such a neurologic scale permits quantitative rather than simply qualitative measurements regarding the relationship between clinical stroke severity and MRI-determined lesion size. Moreover, because this scale has been shown to be predictive of subsequent neurologic outcome,⁶ our findings support the hypothesis that early DWI and PWI findings strongly reflect the severity of clinically observed neurologic deficits.

Few other studies have attempted to relate neuro-imaging abnormalities to clinical stroke severity. Brott et al.¹⁰ studied the relationship between initial CT findings and a variant of the NIHSS in 65 patients within 48 hours of acute stroke. However, 60% of the patients had negative CT results on initial examination. For the 26 patients with an initial CT abnormality, the Spearman rank correlation coefficient was 0.74. Saunders found a significant correlation between infarct volume and clinical outcome in patients with MRI-evaluated middle cerebral artery (MCA) strokes. However, the mean imaging time was 27.5 hours, and only a rough measure of outcome (independent, dependent, or dead) was used. Large-volume lesions were significantly associated with death, but no such prediction could be made for independent versus dependent groups, likely a reflection of the crude neurologic measuring criteria.

Only a limited number of diffusion or perfusion studies have evaluated the relationship between neurologic deficit and DWI and PWI lesion volumes in humans at early (<6.5 hours) times. Warach et al.¹¹ first compared neurologic outcome with the presence or absence of DWI abnormalities in 12 patients with acute ischemia of less than 6 hours duration. DWI and PWI were found to be highly sensitive and specific in predicting persistent neurologic deficits and accurate determined outcome in 11 (91%) of 12 patients. However, no specialized stroke scales were used to quantify the severity of neurologic deficit, and stroke volumes could not be assessed because only single-slice DWI and PWI images were obtained in most patients. Thus, only general observations regarding neurologic outcome could be made.

Sorensen et al.⁴ used a multislice EPI technique similar to ours to assess 11 patients within 6 hours of symptom onset, and also found that DWI and PWI were highly correlated with subsequent T2W MRI findings. In their series, two patients with severe neurologic deficits did not display DWI or PWI changes, had full recovery, and were subsequently diagnosed with complex migraine rather than ischemic stroke, suggesting that DWI and PWI can accurately differentiate cerebral ischemia from other potential stroke mimics. This study was also limited by the absence of validated neurologic outcome but demonstrated the substantial ability of DWI and PWI to detect early brain ischemia.

Only one other study has correlated DWI with neurologic outcome in patients with early ischemia. Lovblad et al.¹² evaluated 26 patients scanned within 48 hours of the start of symptoms, 13 within 6 hours of onset. This study found a significant correlation between DWI lesion size and both initial (r = 0.53, p = 0.0003) and chronic NIHSS score (r = 0.68, p < 0.0001). The<6-hour patients were not separately analyzed in this study. In addition, there was no comparison of the initial MRI with subsequent neurologic deficit. Our findings complement these results by showing that the initial DWI lesion volume is also predictive of subsequent early (24-hour) neurologic outcome.

We found that PWI correlated better with 24-hour NIHSS than DWI. This difference was primarily attributable to one patient (L.N.) in whom the PWI detected a substantial perfusion delay, whereas DWI showed no abnormality. This indicates that the PWI changes can occur before diffusion changes. To our knowledge, this has not been previously reported. This finding suggests that in certain patients, perhaps only in those evaluated at early times, cerebral perfusion can be delayed before any diffusion abnormalities appear. This would be quite consistent with current theories on the pathogenesis of acute cerebral ischemia that suggest that there is an initial perfusion decrease followed by cellular dysfunction. With the exception of this patient, both DWI and PWI were similarly accurate at predicting the severity of neurologic deficit at 24 hours (r = 0.90, p < 0.001).

Use of quantified perfusion imaging adds a new dimension to the assessment of cerebral hemodynamics during acute stroke. To our knowledge, this study is the first attempt to quantify the size of the perfusion abnormality rather than to describe it in qualitative terms. We found that perfusion abnormalities were generally larger than diffusion changes acutely. This perfusion "mismatch" varied from 1% to 100% of the initial diffusion volume and averaged 50% larger. Even by 7 days T2W abnormalities still averaged 23% smaller than the initial (<6.5 hour) perfusion deficits, although there was substantial variation between patients that may have reflected differences in individual perfusion characteristics.

Warach et al.¹¹ first reported that perfusion qualitatively helped in clinical assessment of acute stroke patients. However, the PWI examinations were frequently limited to one slice and were not reported separately from the diffusion data. More recently, Baird et al.¹³ reporting on 28 patients with acute stroke found that patients studied <6 hours after symptom onset had a 97% increase in DWI lesion size, whereas patients studied at >6 hours had only a 47% increase (p = 0.48). Moreover, in patients with a perfusion-diffusion mismatch, there was a substantial increase in lesion volume compared with those patients without such a mismatch (230% versus-46%, p = 0.04). However, no quantitative measure of perfusion volume was used in these studies.

We found an average 50% increase in DWI lesion size over 7 days. In all patients the initial PWI lesion volume was larger than or equal to the DWI lesion volume, but patients with the largest initial PWI-to-DWI difference did not necessarily have the largest increase in lesion size. This suggests that final infarct size is influenced not only by PWI- and DWI-detected abnormalities but probably by a variety of other factors such as severity of ischemia or perhaps the extent of cerebral edema. Nevertheless, it is clear that PWI adds significant additional information to the MRI assessment of acute ischemia.

It has been hypothesized that PWI detects the maximum extent of the penumbra, whereas DWI identifies irreversibly damaged tissue.¹³ Further research is necessary to clarify the predictive value of DWI and PWI in acute stroke. DWI and PWI may serve as ideal tools for studying stroke during this critical interval because of their ability to generate both functional and anatomic information in a single, relatively short study.

There are several limitations to this study. First, because most (9/10) patients participated in neuroprotective trials or were given open label rt-PA, the natural history of acute ischemia may have been altered by treatment. However, aside from the single untreated patient in whom no persistent lesion was found (M.H.), the correlation of DWI and PWI lesion volumes with neurologic deficit was high. Therefore, it seems unlikely that treatment had any substantial confounding effect, particularly in that patients were assessed relatively early (24 hours) after treatment.

However, the lack of a change in lesion volume associated with treatment does not necessarily suggest that the treatments employed are ineffective. The small size of the study group and the possibility of placebo therapy in many of the drug study patients could easily have masked even a substantial treatment effect. Moreover, the early time to NIHSS assessment may not have afforded sufficient time for the maximum degree of neurologic recovery to be detected. Only larger studies with a sufficient number of controls and longer follow-up can answer this question adequately. In addition, the findings of this study can only be applied to large anterior circulation cortically based strokes, because by chance only patients with this type of stroke were evaluated. It is well known that smaller subcortical and brainstem lesions can cause deficits out of proportion to the volume of ischemia evident by pathology or neuroimaging. Therefore, the correlation of infarct volume with degree of neurologic deficit will likely be significantly lower in lacunar strokes. This effect was reported by Lovblad et al. in their series.¹²

Despite these limitations, it is clear from these data that both DWI and PWI are highly associated with early ischemic injury. Further research is needed to elucidate whether specific patterns of DWI and PWI changes can be identified that can detect patients who might benefit from various treatment modalities. In addition, recent data indicated that ischemic lesions may expand over time,^13,14 causing further neurologic injury even days after the acute insult. This suggests that the window of opportunity for treatment, and perhaps the duration of treatment necessary, could be longer than previously suspected.^14,15 Data from animal studies have shown that DWI lesions can be reversed or reduced in size after treatment with neuroprotective agents.^16-21 In addition, some recent studies of human stroke have also suggested that DWI lesions can be reversed by the use of neuroprotective agents.²² Such data are encouraging, and portend a time when stroke may be treated more effectively and possibly with individually tailored therapy.

Diffusion-weighted and PWI appear to provide us with valuable information about the status of the brain during acute ischemia. In the future, such information will almost certainly be of substantial value to both our understanding and the management of acute ischemic stroke.