点击显示 收起
1 From the Department of Neurology (H.J.W., A.R., M.S., R.J.S., H.J.F.) and Institute of Diagnostic Radiology (H.J.W., F.W., U.M.), Heinrich-Heine University of Düsseldorf, MNR-Klinik/MRT-2, Moorenstrasse 5, D-40225 Düsseldorf, Germany; Institute for Medicine, Forschungszentrum Jülich, Germany (G.R.F.); and Department of Neurology, University of Aachen, Germany (G.R.F.). Received October 30, 2000; revision requested December 22; final revision received July 30, 2001; accepted August 24. Supported by SFB 194 (TP A9, TP A13) and Kompetenznetz-Schlaganfall (BMBF, TP B5, TP C4).
| ABSTRACT |
|---|
| Top ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS DISCUSSION REFERENCES |
|---|
MATERIALS AND METHODS: Fifty patients who underwent diffusion-weighted and perfusion imaging within 1–24 hours after stroke onset and a follow-up T2-weighted investigation after about 8 days were included. Lesion volumes were evaluated by using a semiautomatic thresholding technique. Volumetric results of acute diffusion-weighted and perfusion imaging were analyzed in comparison with follow-up T2-weighted images and in terms of the time difference between symptom onset and initial MR imaging.
RESULTS: At diffusion-weighted imaging, the acute lesion defined by a signal intensity increase of more than 20%, compared with the contralateral side, showed the best correlation with the infarct size after 1 week. At perfusion imaging, the best predictor relative to the contralateral side was a delay of more than 6 seconds on TTP maps. Temporal analysis of volumetric results, which depended on the time difference between symptom onset and examination, revealed two patient subgroups.
CONCLUSION: Diffusion-weighted imaging helped to predict the size of the lesion on T2-weighted images obtained after about 8 days in patients with a symptom onset of more than 4 hours (r = 0.96), while in patients with a symptom onset of less than 4 hours, perfusion imaging provided important additional information about brain tissue with impaired perfusion.
Index terms: Brain, infarction, 13.78 • Brain, ischemia, 13.781 • Brain, MR, 13.121411, 13.121412, 13.121416, 13.12143, 13.12144 • Magnetic resonance (MR), diffusion study, 13.12144 • Magnetic resonance (MR), perfusion study, 13.12144
| INTRODUCTION |
|---|
| Top ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS DISCUSSION REFERENCES |
|---|
Diffusion-weighted imaging depicts random movements of water protons. It has been shown that in ischemic brain tissue, the movements of proton spins are restricted due to cytotoxic edema. This leads to a decrease in the apparent diffusion coefficient (ADC) of water and an increase in diffusion-weighted signal intensity in areas of acute ischemic injury. The amount of change of the ADC depends on the temporal course of ischemia, the development of the edema, and the resulting increase in cell volume (2–4). Additionally, lesions at diffusion-weighted imaging are inhomogeneous and differ from the normal tissue in various ways because of different temporal rates of lesion evolution toward infarction (5).
By contrast, perfusion imaging findings provide information on the momentary hemodynamic state of brain tissue, as they reveal impaired tissue perfusion caused by blood vessel obstruction (3,6,7). Therefore, perfusion imaging findings may yield information about pathologically hypoperfused regions, even before genuine structural brain tissue damage has taken place.
However, the area with reduced tissue perfusion, as indicated with perfusion imaging, is, in most cases, much larger than the lesion at diffusion-weighted imaging (8–10). This has led to the suggestion that the region of mismatch between acute perfusion imaging and diffusion-weighted imaging may represent pathologically underperfused tissue where structural damage has not yet taken place. But time-to-peak (TTP) maps derived from perfusion imaging also show temporally delayed perfused areas where perfusion is reduced but where blood flow remains well above the 15–20 mL per 100 g of tissue, which is the critical viability limit derived from animal experiments (7,11) and blood flow measurements in patients at positron emission tomography (12). Therefore, TTP maps without further thresholding typically show an overestimation of the final infarct size.
The issue is further complicated by the fact that both diffusion-weighted and perfusion imaging findings are affected, at least in part, by the time course of the ischemic lesion evolution (2,13). It has been shown that the ADC in the acute stage of ischemia continuously decreases to its minimum value of about 50% 24–48 hours after symptom onset. Thereafter, the ADC values increase again and after about 10 days they reach values that are even higher than its normal value (10,14). Perfusion imaging, by contrast, shows a variety of temporal patterns, depending on the degree of spontaneous or treatment-induced reperfusion of the tissue (10).
In an acute stroke setting, TTP maps offer two advantages over other images derived from perfusion imaging. First, they involve the least amount of postprocessing, and thus allows the clinical information to be available within 20 minutes. Second, TTP maps provide a relatively clear picture of the location and extent of the lesion. However, at present, diffusion-weighted imaging still serves most clinicians in the characterization of the ischemic lesion.
The purpose of our study was to investigate the predictive value of the ischemic lesion size, as depicted in the acute stroke phase on diffusion-weighted images and TTP maps derived from perfusion imaging, for the infarct size, as derived from T2-weighted imaging in the postacute phase.
| MATERIALS AND METHODS |
|---|
| Top ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS DISCUSSION REFERENCES |
|---|
Imaging
The MR imaging protocol was part of our routine clinical protocol and comprised T2- and diffusion-weighted sequences, an MR angiographic investigation, and serial T2*-weighted measurements, with 40 images obtained for the measurement of tissue perfusion. The total imaging time with the 1.5-T whole-body MR imager (Magnetom Vision; Siemens, Erlangen, Germany) was approximately 20 minutes. For T2-weighted imaging, we used a turbo gradient-echo spin-echo sequence with a repetiton time (TR) of 7,040 msec, echo time (TE) of 115 msec, echo train length of 69, 20 transverse sections acquired parallel to the base of the skull, 5-mm section thickness, 1.5-mm intersection gap, 160° flip angle, 230-mm field of view (FOV), 345 x 512 matrix, and imaging time of 1 minute 17 seconds.
Diffusion-weighted imaging was performed with a single-shot echo-planar sequence with two diffusion-weighted b values—of 0 and 1,000 sec/mm2—in three dimensions in space—readout, phase-encoding, and section-selection directions—to result in four images per section. The imaging parameters were a TE of 100 msec, 20 transverse sections acquired, a 5-mm section thickness, a 1.5-mm intersection gap, a 96 x 128 matrix, a 240-mm FOV, and imaging time of 26 seconds.
For MR angiographic measurement, we used a three-dimensional time-of-flight sequence with a TR of 35 msec, TE of 7.2 msec, 20° flip angle, 1.5-mm section thickness, 200 x 512 matrix, 108 sections acquired, 200-mm FOV, and imaging time of 6 minutes 44 seconds. The measurement of perfusion imaging was performed with serial T2*-weighted single-shot echo-planar sequences, with imaging of 12 sections 40 times every 2 seconds. The parameters were a TE of 54 msec, 12 sections with 5-mm thickness acquired, a 1.5-mm intersection gap, a 128 x 128 matrix, and a 240-mm FOV. At the time of the fourth imaging, a 15-mL bolus of gadolinium-enhanced contrast agent (gadopentetate dimeglumine ; Schering, Berlin, Germany) was injected at a rate of 5 mL/sec and was immediately followed by 15 mL of sodium chloride that was injected at the same rate.
Postprocessing and Image Analysis
All data resulting from diffusion-weighted and perfusion imaging were transferred to a workstation (Sun Ultra Sparc 1; Sun Microsystems, Palo Alto, Calif). By averaging the images obtained with the three diffusion-weighting directions (b value of 1,000 sec/mm2), trace diffusion-weighted images were calculated for each section. Additionally, ADC maps were produced from diffusion-weighted imaging. From perfusion imaging data, TTP parameter maps and relative regional cerebral blood volume, or rrCBV, maps were calculated. TTP here refers to the time between the first imaging and the time at which the signal intensity reaches its minimum, which occurs owing to the administration of the bolus of contrast agent at T2*-weighted imaging. rrCBV was calculated by means of a numerical integration of the concentration time curve after converting the signal intensities to concentration values by using the logarithmic relation c(t) -ln[S(t)/S0]/TE, where c(t) is concentration of contrast agent at time t, S(t) is signal intensity at time t, and S0 is precontrast signal intensity (15). All of the image postprocessing was performed with computer programs that we have developed.
Volumetric analysis was performed only on diffusion-weighted trace images and TTP maps, as described next. For visualization of the resulting images, we used the program MPItool (Advanced Tomo Vision, Erftstadt, Germany). The entire procedure of transferring data, calculating images, up to visualization lasted less than 8 minutes. For the evaluation of volumetric results at diffusion-weighted and perfusion imaging, we used a semiautomatic segmentation technique consisting of thresholds that are relative to the unaffected contralateral hemisphere (Fig 1). The segmentation was performed (A.R.) on manually defined regions that excluded artifacts and ventricles. For TTP maps, this procedure has been described previously (16). For perfusion imaging, the thresholds were 2, 4, 6, 8, 10, 12, 14, 16, 18, and 20 seconds of the delay after bolus arrival in a well-perfused tissue. The method of referring to the contralateral well-perfused tissue is designed to minimize the effect of individual hemodynamic circulation time in the patients. Differences in bolus delay within the brain due to individual cardiac output are treated as statistical fluctuations.
|
The resulting 20 lesion volumes, 10 for diffusion-weighted imaging and 10 for perfusion imaging, were then compared with the volume of the infarction at day 8 (range, 6–11 days). This latter volume was determined manually by two independent observers (A.R., G.R.F.) by tracing the edge of the hyperintense area on T2-weighted MR images. For each threshold value, the volumes on diffusion-weighted images and TTP maps were compared with the defined outcome volume to find the best predictor for the volume on T2-weighted images.
Statistical analyses were performed by using mean and SDs for demographic data. Interobserver errors were calculated from root-mean-square deviations and are given as percentages. Those percentages were averages over the individual root-mean-square deviations relative to the corresponding mean values. Interobserver reliability was calculated from the Pearson product moment correlation coefficient.
Correlations between acute and follow-up volume data were also analyzed with the Pearson product moment correlation coefficient. Additionally, quantitative agreement of the lesion size was tested by taking into account the slope and the intercept of linear regressions.
| RESULTS |
|---|
| Top ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS DISCUSSION REFERENCES |
|---|
Hyperintense areas on T2-weighted images obtained after about 8 days were identified in 49 cases. The mean lesion volume after 1 week was 40 mL ± 49. In the remaining case, the putative lesion on T2-weighted images could not be distinguished from a nearby sulcus. However, diffusion-weighted imaging performed at this same time depicted the lesion (volume was approximately 6 mL).
We then compared the lesion volume sizes on diffusion-weighted and perfusion images obtained at the acute stage with the follow-up T2 measurements by using the thresholds described earlier to get the best predictors for infarction volume.
The best correlation (r = 0.83, slope = 0.8) of the lesion at diffusion-weighted imaging and that at T2-weighted imaging resulted from thresholding the lesion at diffusion-weighted imaging at a pixel intensity 20% higher than the unaffected contralateral tissue (Fig 2, Table). The ratio of the lesion volume at diffusion-weighted imaging (at threshold 20%) and that at follow-up T2-weighted imaging should be about 1. Figure 3 illustrates the two subgroups of patients: group A, in whom investigations were performed within the first 4 hours after symptom onset, and group B, in whom MR imaging was performed within 4–24 hours after symptom onset (Fig 4). When only investigations from group B were taken into account, the correlation coefficient r of a lesion at diffusion-weighted imaging and that at T2-weighted imaging reached a value of 0.96 (Table). This increase in correlation coefficient becomes meaningful when compared with the other subset of patients, which has about the same size (19 patients), with delays to symptom onset of less than 4 hours. In this subgroup, the correlation coefficient was the same, 0.83, as that for the whole set. The data point for the highest volume at follow-up T2-weighted imaging was influential for regression, as seen in Figure 2. However, even omitting this outlier, the 20% threshold yielded the highest correlation.
|
|
|
|
|
By using the 20% threshold for diffusion-weighted imaging and a 6-second delay for perfusion imaging, in 26 subjects there was a perfusion-weighted imaging–diffusion-weighted imaging mismatch in the acute stage. In 12 of these subjects, an average lesion enlargement of 26 mL was observed on diffusion-weighted images obtained in the acute stage compared with T2-weighted images obtained after about 1 week, corresponding to an average relative enlargement of 48%. Seven patients showed a slight lesion enlargement of 5 mL (25%), although there was no perfusion-weighted imaging–diffusion-weighted imaging mismatch.
The methods of volumetric analysis (semiautomatic for diffusion-weighted imaging and TTP maps, manual for T2-weighted MR imaging) also caused errors of measurement. The manual volumetric measurements on the T2-weighted images resulted in an interobserver error of 6%. The interobserver reliabilities for the manually defined reference regions at acute diffusion-weighted imaging and on TTP maps were 0.84 and 0.99, respectively. They resulted in interobserver errors of 4% (diffusion-weighted imaging) and 3% (TTP maps) of the volumes.
| DISCUSSION |
|---|
| Top ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS DISCUSSION REFERENCES |
|---|
The best correlation between acute TTP lesion volume and outcome at T2-weighted MR imaging performed after about 8 days was provided by the 6-second threshold (16). The correlation (r = 0.73) was not as good as that with diffusion-weighted imaging and therefore was not as good a predictor of the infarct size at the approximated day 8, as indicated by the larger CIs for the correlation of perfusion imaging (Table). Nevertheless, the lesion volume on the acute TTP maps, as assessed by using the 6-second threshold, was important for the estimation of the tissue with impaired perfusion and therefore for a possible increase in ischemic lesion size in patients with a symptom onset of less than 4 hours.
Signal intensity at diffusion-weighted imaging is influenced by the diffusion of water protons, spin density, and T2 effects (17). Therefore, at diffusion-weighted imaging, possible contributions of the T2 effect, in particular in patients with a symptom onset of more than 4 hours, will only increase the correlation to the postacute T2-weighted image. The standard diffusion-weighted MR sequence (b = 1,000 sec/mm2, TE = 100 msec) used here to determine the lesion size should, nevertheless, be comparable to other MR imagers in which the same measurement parameters are used. A further analysis of the pathophysiologic changes or a differentiation within the ischemic lesion would necessitate a quantitative ADC value analysis.
In our study, we preferred diffusion-weighted imaging to ADC maps for volumetric analysis, because the change in ADC between an ischemic lesion and the surrounding tissue is very small in comparison to the change in signal intensity at diffusion-weighted imaging. Furthermore, the diffusion coefficient of cerebrospinal fluid is much larger than that of brain tissue, and, therefore, the partial volume effect on ADC maps will disturb volumetric measurements near the sulci and ventricles.
TTP maps of perfusion imaging show the arrival time of the contrast agent, which is the same for gray and white matter in the brain, within the temporal resolution of the measurement, and, therefore, these maps are easy to interpret in the acute stage. The numerical calculation of TTP maps from bolus tracking MR perfusion imaging data sets constitutes a fast way of postprocessing in comparison to the quantitative maps of mean transit time or cerebral blood flow. Furthermore, most manufacturers of MR imagers have implemented the calculation of TTP maps in their software. The information derived from quantitative perfusion maps such as mean transit time and cerebral blood flow will nevertheless provide important insights into the process of stroke evolution and should be taken into account in further analyses of ischemic lesions in nonacute stages. At perfusion imaging without the use of thresholds on TTP maps, often there will be an overestimation of the lesion size at the chronic stage of the stroke. This may be due to delayed perfusion that results from hemodynamically relevant occlusive artery disease (18).
The semiautomatic evaluation techniques used in our study at both diffusion-weighted and perfusion imaging improved the stability of volumetric results in comparison to manual tracing of the edge of lesions. The use of thresholds for the further assessment of the lesion increased the objectivity of the volume measurements. Errors that occur when applying different levels of contrast and brightness on the images and when manually tracing the often smooth edges of the ischemic lesions could thus be avoided.
The most important pathophysiologic changes in stroke occur within the first few days after the symptom onset. For this reason, and also for the feasibility of patient handling, we used day 8 for the second investigation, as it was used by other investigators before (9,10,16). In addition, the T2 effect increases within the first 7 days after stroke onset to become the maximum contribution to the signal intensity, compared with diffusion and spin-density effects, so that the T2-weighted image will show a stable representation of the stroke lesion (16).
Nevertheless, there are problems with T2-weighted imaging. For example, the boundaries between hyperintense lesions and cerebrospinal fluid are difficult to determine. Therefore, further studies have to show if, for example, fluid-attenuated inversion recovery imaging can improve the volumetric analysis compared with T2-weighted MR imaging. To minimize such uncertainties about the extent of the infarction in our study, two experienced independent observers analyzed the images manually. Altogether, the errors of volumetric analyses that resulted from technical problems like partial volume effects (particularly in small lesions) and limited anatomic resolution were small compared with the interobserver discrepancy in the assessment of T2-weighted images.
In conclusion, our results show the following: (a) By using a standard diffusion-weighted sequence (b = 1,000 sec/mm2, TE = 100 msec), the best correlation of the lesion at diffusion-weighted imaging with the lesion volume at T2-weighted imaging was achieved by thresholding to a signal intensity increase of greater than or equal to 20%, compared with that for the unaffected contralateral side. (b) For a delay of more than 4 hours from the symptom onset to imaging, the correlation coefficient of lesion volume at diffusion-weighted imaging versus that at T2 imaging was 0.96. (c) At perfusion imaging, a threshold of 6 seconds on TTP maps showed the best correlation with the lesion volume at T2-weighted imaging (r = 0.73). (d) Perfusion imaging findings showed that there were regions with pathologically delayed perfusion that were inconspicuous at diffusion-weighted imaging, especially in cases where the delay between symptom onset and imaging was less than 4 hours.
| ACKNOWLEDGMENTS |
|---|
| REFERENCES |
|---|
| Top ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS DISCUSSION REFERENCES |
|---|