Fluid-Attenuated Inversion Recovery Evolution Within 12 Hours From Stroke Onset
A Reliable Tissue Clock?
Background and Purpose—It has recently been proposed that fluid-attenuated inversion recovery (FLAIR) imaging may serve as a surrogate marker for time of symptom onset after stroke. We assessed the hypothesis that FLAIR imaging could be used to decide if an MRI was performed within 4.5 hours from symptom onset or later.
Methods—All consecutive patients with presumed stroke who underwent an MRI within 12 hours after known symptom onset were included regardless of stroke subtype and severity between May 2008 and May 2009. Blinded to time of symptom onset, 2 raters judged the visibility of lesions on FLAIR. Apparent diffusion coefficient values, lesion volume on diffusion-weighted imaging, and relative signal intensity of FLAIR lesions were determined.
Results—In 94 consecutive patients with stroke, we found that median time from symptom onset for FLAIR-positive patients (189 minutes; interquartile range, 110 to 369 minutes) was significantly longer compared with FLAIR-negative patients (103 minutes; interquartile range, 75 to 183 minutes; P=0.011). Negative FLAIR had a sensitivity of 46% and a specificity of 79% for allocating patients to a time window of less than 4.5 hours. FLAIR positivity increased with diffusion-weighted imaging lesion volume (P<0.001) but showed no correlation with apparent diffusion coefficient values (P=0.795). There was no significant correlation between relative signal intensity and time from symptom onset (Spearman correlation coefficient −0.152, P=0.128).
Conclusions—Based on our findings, we cannot recommend the use of FLAIR visibility as an estimate of time from symptom onset within the first 4.5 hours.
Stroke is a leading cause of death and disability.1 Approximately one fourth of patients with ischemic stroke become aware of their neurological deficits on awakening.2 Current guidelines for intravenous thrombolytic therapies exclude patients with unknown time of symptom onset and only allow for the inclusion of patients within 4.5 hours.3 However, a recent Korean study suggested that thrombolysis may be safely applied in patients with unknown-onset stroke who fulfilled certain MRI-specific eligibility criteria such as a positive perfusion–diffusion mismatch and absence of well-developed fluid-attenuated inversion recovery (FLAIR) changes of acute diffusion lesions.4 Although diffusion-weighted imaging (DWI) alone allows for the detection of acute ischemic lesions within minutes of stroke onset,5 it does not offer exact information on the age of these lesions within the first 12 hours. Assessment of signal changes on FLAIR images in areas with diffusion restriction may provide an estimate of ischemic lesion age.6 Most previous studies have focused on the evolution of FLAIR imaging in the subacute and chronic time window.7 One recent study showed that a “mismatch” between a positive DWI and a negative FLAIR might allow for the identification of patients who are highly likely to be within the 3-hour time window from stroke symptom onset.8 The primary aim of this study was to determine whether FLAIR imaging could be used to decide if an MRI was performed within 4.5 hours from symptom onset or later.
Patients and Methods
This is a substudy of the “1000+” study, a prospective, single-center observational study with at least 1200 patients to be recruited. Details of the study have been described elsewhere (http://clinicaltrials.gov; NCT00715533).
We included 177 consecutive patients who presented to our hospital between May 2008 and May 2009 with transient ischemic attack or ischemic stroke and underwent an MRI ≤12 hours after known symptom onset. Patients were excluded when the exact time of symptom onset was unknown. Patients were included regardless of stroke subtype and severity. Informed consent was obtained from all patients. Demographic data, side of infarction, and severity of neurological deficit on admission (National Institutes of Health Stroke Scale) were recorded. Up until September 26, 2008, eligible patients were treated with intravenous tissue plasminogen activator within 3 hours of symptom onset according to international guidelines. From that date onward, patients received tissue plasminogen activator up to 4.5 hours after symptom onset due to new scientific evidence.9
MRI studies were performed on a 3-T clinical whole-body scanner (Tim Trio; Siemens AG, Erlangen, Germany) fully dedicated to clinical research. The MRI protocol for patients with acute stroke included the following sequences: T2*-weighted imaging; DWI (TE=93.1 ms, TR=7600 ms, field of view=230 mm, matrix 192×192, 2.5-mm slice thickness with no interslice gap); time-of-flight MR angiography; FLAIR (TE=100 ms, TR=8000 ms, TI=2370.5 ms, field of view=220 mm, matrix 256×232, 5-mm slice thickness with a 0.5-mm interslice gap); and perfusion imaging.
The presence of ischemic lesions on FLAIR images of 177 patients was rated in 2 steps by 2 experienced raters (M. Ebinger and M.R.) blinded to clinical information. Raters underwent a training session using independent data sets (provided by J.B.F.) to become accustomed to the rating procedure.
Based on the recommendation by a third reader (I.G.) unblinded to DWI, the raters were instructed to look for the ischemic lesions on the affected side of the brain. If there were no lesions on DWI or if there were lesions on both sides of the brain, there was no instruction with regard to side. Raters were provided with FLAIR images only and were blinded to time of symptom onset and to DWI. They delineated lesions and the third rater (I.G.) compared them with the lesions visible on DWI to determine whether the blinded raters noticed an ischemic lesion on FLAIR images corresponding to the area of the DWI restriction. Ischemic lesions were judged to be visible if identified by both raters. In addition, both raters judged the extent of white matter lesions for each patient using 3 categories: no, moderate, or severe white matter disease based on the Wahlund score (0 to 4, 5 to 10, >10, respectively).10
In the second step, the raters first viewed the DWI and then the FLAIR images for each patient. They delineated both the DWI restriction and the corresponding lesion on FLAIR images when visible. Only lesions delineated in this reading were used for further postprocessing. Raters judged lesions either as “subtle” or as “obvious” based on their subjective estimate of how easily visible they were on FLAIR.
For both steps, in cases in which a lesion was judged visible by only one of the 2 raters, an adjudication rating was performed by an additional rater (J.B.F.).
A commercially available software package (MRIcro, Version 1.40) was used for delineation of regions of interest and image analysis. For patients with disseminated lesions, more than one region of interest was drawn. Image postprocessing included calculation of volume and apparent diffusion coefficient values from all DWI lesions as well as absolute signal intensity and signal intensity relative to the contralateral area (rSI) for all FLAIR lesions. FLAIR lesions were further segmented into areas of white and gray matter through an automatized procedure using SPM8 (Statistical Parametric Mapping, Version 8) and volumes as well as signal intensity and rSI were separately calculated for both types of brain matter. Masks of gray and white matter as well as cerebrospinal fluid were generated using FLAIR images. SPM8 minimizes the influence of hyperintense lesions by a built-in FLAIR-derived segmentation template. Using a cutoff value of <5% of voxels within white or gray matter, we allocated lesions to cortical and subcortical, respectively. Lesions that had >5% of voxels within gray and white matter were deemed mixed lesions.
To determine sensitivity, specificity, positive predictive value, and negative predictive value for the identification of patients within 3 and 4.5 hours of symptom onset by negative FLAIR, we included only patients with visible DWI lesions, available FLAIR images of sufficient quality for rating, and known time of symptom onset. All results were tested for normality with the one-sample Kolmogorov-Smirnov goodness-of-fit test. Thereafter, means and SDs were used to describe normally distributed data, whereas medians and interquartile ranges were used for nonnormally distributed results. Differences between patient groups were assessed with the Mann–Whitney U test and the Kruskal-Wallis H test. A linear regression model was used to assess the correlation between rSI and time from stroke onset. All statistical analyses were performed using commercial statistical software (SPSS 16.0) and Clinical Calculator 1 (VassarStats, 2001 to 2009).
We have analyzed data of 177 consecutive patients with known symptom onset of transient ischemic attack or ischemic stroke and MRI within 12 hours of symptom onset. Of those, 81 patients (46%) did not have an acute DWI lesion and were subsequently not diagnosed with ischemic stroke but rather transient ischemic attack. One patient was excluded due to missing FLAIR images and another due to poor image quality. Thus, data sets of 94 patients (53%) were included into the final analysis (for baseline characteristics, see Table 1).
Twenty-six patients received tissue plasminogen activator. A total of 203 ischemic lesions were identified in the 94 patients with stroke. The median DWI lesion volume for all patients was 1.47 mL (interquartile range [IQR], 0.49 to 4.33 mL). The mean apparent diffusion coefficient value for all lesions was 0.73 (±0.28) 10−3mm2/s. Forty patients had no, 28 patients had moderate, and 26 patients severe white matter disease. There was no significant difference in terms of white matter disease between FLAIR-negative and FLAIR-positive patients (P=0.875). Of all FLAIR-negative patients, 41% had no, 32% moderate, and 27% severe white matter disease, whereas these percentages were 44%, 28%, and 28% for FLAIR-positive patients, respectively.
In the first, blinded read, 23 of 94 (25%) patients were FLAIR-positive. FLAIR was rated negative in 46 of 55 (84%) patients ≤3 hours and in 59 of 70 (84%) patients ≤4.5 hours. Thus, a “mismatch” between a positive DWI and negative corresponding FLAIR had a sensitivity of 84%, a specificity of 36%, a positive likelihood ratio of 1.3, and a negative likelihood ratio of 0.46 for symptom onset within the first 3 hours. Within the first 4.5 hours, a negative FLAIR correctly allocated patients with 84% sensitivity, 50% specificity, 1.69 positive likelihood ratio, and 0.31 negative likelihood ratio. Median time from symptom onset to imaging for FLAIR-positive patients was 292 minutes (IQR, 113 to 599 minutes), whereas median time for FLAIR-negative patients was 123 minutes (IQR. 81 to 230 minutes, P=0.014). Median DWI volume was 3.45 mL (IQR, 1.72 to 25.30 mL) for FLAIR-positive patients and 0.77 mL (IQR, 0.27 to 3.69 mL) for FLAIR-negative patients (P<0.001).
In the unblinded read, 57 of 94 (61%) patients had a visible lesion on FLAIR, corresponding to the acute ischemic lesion seen on DWI. (Examples of FLAIR-positive and -negative patients are provided in Figure 1.) Median time from symptom onset for FLAIR-positive patients was 189 minutes (IQR, 110 to 369 minutes), whereas median time for FLAIR-negative patients was 103 minutes (IQR, 75 to 183 minutes, P=0.011). Figure 2 shows the distribution of FLAIR-positive patients with regard to time elapsed since symptom onset. Note, that there were 2 FLAIR-positive patients imaged within 60 minutes. The first patient had 5 separate lesions, ranging in size from 0.07 mL to 0.4 mL on DWI, and all 5 of these lesions were visible on FLAIR. The second patient presented with only one lesion, 52 mL in volume on DWI. A total of 203 lesions were identified on DWI images with a median volume of 0.45 mL (IQR, 0.16 to 1.60 mL) for FLAIR-positive lesions and 0.19 mL (IQR, 0.06 to 0.82 mL) for FLAIR-negative lesions (P<0.001). Only 13% (3 of 23) of lesions <0.05 mL were found to be FLAIR-positive. Of lesions with a volume between 0.05 mL and 0.10 mL, 31% (11 of 36) were detectable on FLAIR. Between 0.10 and 0.25 mL, the ratio of FLAIR-positive lesions inverted to 69% (25 of 36). Lesions >0.25 mL and <50 mL were found to be FLAIR-positive in 57% of cases (60 of 105). All lesions >50 mL were deemed FLAIR-positive (3 of 3).
Apparent diffusion coefficient values did not differ significantly between patients with a positive and those with a negative FLAIR (0.72±0.12 versus 0.74±0.13×10−3mm2/s, P=0.795; Table 1).
FLAIR was rated negative in 27 of 55 (49%) patients ≤3 hours and in 32 of 70 (46%) patients ≤4.5 hours. FLAIR was rated positive in 29 of 39 (74%) patients >3 hours and in 19 of 24 (79%) patients >4.5 hours.
After exclusion of patients with a DWI lesion <0.5 mL (n=23), 44 FLAIR-positive patients and 27 FLAIR-negative patients remained. For these 71 patients, FLAIR was rated negative in 21 of 41 (51%) patients ≤3 hours and 25 of 51 (49%) patients ≤4.5 hours. Table 2 shows sensitivity, specificity, positive predictive value, and negative predictive value for allocating patients into different time windows based on FLAIR imaging.
Of a total of 203 DWI lesions, 102 were found to be FLAIR-positive with a mean rSI 1.28 (±0.15) and a median volume of 0.25 mL (IQR, 0.13 to 0.73 mL). Lesions deemed by the raters as obvious were larger in volume than subtle lesions (5.73 versus 0.80 mL), but the difference was not statistically significant (P=0.292). Obvious lesions had a higher rSI than subtle lesions (1.39 versus 1.21) and this difference was statistically significant (P<0.001). There was no statistically significant difference in time elapsed since symptom onset between obvious and subtle lesions (172 versus 201 minutes, P=0.149). In a linear regression model, rSI failed to predict time from stroke onset (R2=0.007). Spearman correlation coefficient for rSI and time from symptom onset was −0.152 (P=0.128; Figure 3).
Of the 102 lesions that were positive on FLAIR, 18 were found to be subcortical, 17 were cortical, and 67 were mixed lesions. Median volume for subcortical lesions was 0.22 mL (IQR, 0.15 to 0.43 mL), for cortical lesions 0.13 mL (IQR, 0.08 to 0.42 mL), and for mixed lesions 0.39 mL (IQR, 0.15 to 1.13 mL; P=0.015). Mean rSI for subcortical lesions was 1.29 (±0.22), for cortical lesions 1.32 (±0.20), and for mixed lesions 1.31 (±0.14, P=0.58). Median time from symptom onset was 229 minutes (IQR, 85 to 476 minutes) for subcortical lesions, 220 minutes (IQR, 130 to 517 minutes) for cortical lesions, and 176 minutes (IQR, 95 to 438 minutes; P=0.51) for mixed lesions.
The main finding of our study was that within the current time window for tissue plasminogen activator, FLAIR is not a reliable “chronometer” in stroke. It is not before the seventh hour that FLAIR positivity approaches 100% (Figure 2). Using FLAIR negativity as an indicator of stroke within 4.5 hours, we misclassified 45% of our patients. Mainly, FLAIR-positive patients were falsely regarded as beyond 4.5 hours of symptom onset (40% of all patients). Therefore, we cannot recommend using FLAIR to estimate time of stroke onset with regard to the first 4.5 hours. Similar to our results (Table 2), Thomalla et al using a 1.5-T scanner found that negative FLAIR imaging allocated ischemic lesions to a time window of ≤4.5 hours with better specificity and positive predictive value than sensitivity and negative predictive value.8 Further similarities included an increased likelihood of lesion detection with knowledge of DWI and with greater lesion size; white matter disease had no significant impact on lesion detectability (a counterintuitive finding we had no explanation for); and DWI was more sensitive than FLAIR to the detection of ischemic lesions.
As expected, we found that obvious lesions had a higher rSI than subtle lesions, but time from symptom onset did not differ significantly between obvious and subtle lesions. Overall, rSI on FLAIR did not increase with time elapsed since symptom onset and this random distribution of rSI with regard to time sounds a note of caution on trying to determine time of onset based on FLAIR imaging (Figure 3). Correspondingly, although it is known that apparent diffusion coefficient values decrease within the first days after symptom onset,11 they did not appear to be a useful “tissue clock” in the first 12 hours. Other than lesion volume, we found no statistically significant difference between cortical and subcortical FLAIR-positive infarction in terms of rSI or time from symptom onset. Stratifying our analyses of time dependency into cortical, subcortical, and mixed lesions, we found randomly scattered rSI over time for all 3 groups.
This study has a number of limitations. On average, neurological deficits were mild (median National Institutes of Health Stroke Scale 4), number of patients in different time windows was limited (eg, only 16 patients >6 hours), and lesions on DWI had a small median volume (1.5 mL). Thus, this cohort was not representative of patients with acute stroke eligible for thrombolysis and this may explain the lower predictive values of FLAIR as compared with the findings by Thomalla et al.8 Comparability to studies including more severely affected patients was limited. Furthermore, using relative values (rSI) only, we did not correlate a quantitative measure of FLAIR with time; there were no repeated measurements in individual patients; and, finally, given the heterogeneity of stroke types, this study was not powered to identify subgroups with a better correlation between FLAIR evolution and time elapsed since symptom onset.
Sensitivity and specificity of a negative FLAIR improved after exclusion of patients with very small lesions (<0.5 mL). This secondary analysis was justified by both methodological issues and clinical reasoning. First, small lesions may have escaped notice simply due to different slice thickness of FLAIR and DWI in our standard stroke protocol. Concurring with previous results,8 DWI lesion volume had a significant effect on detectability of areas with restricted diffusion on FLAIR images. Exclusion of small lesions also improved comparability with the work by Thomalla et al, because their sample included no lacunar strokes.8 Second, very small lesions, especially in the absence of vessel occlusion or a larger perfusion deficit, may not be the target of choice for thrombolysis.
In the setting of acute stroke, time is a surrogate marker for processes that evolve during ischemia. Duration of symptoms is therefore generally used to determine a patient’s eligibility for thrombolysis. FLAIR imaging rather than time may, however, be a better indicator of a stage in which treatment is no longer effective or has even become harmful. A recent study showed that a focal FLAIR hyperintensity within acute DWI lesions might serve as a predictor of symptomatic intracerebral hemorrhage after thrombolysis.12 Our results show that FLAIR visibility does not correlate well with time from symptom onset. Therefore, prospective trials are needed to investigate if a pathophysiological approach to treatment decisions using stroke MRI is more appropriate than a mainly time-driven algorithm.
We thank Peter Schlattman from the Institut für Medizinische Informatik, Biometrie und Epidemiologie, Universitymedicine Berlin, for statistical support.
Sources of Funding
The research leading to these results has received funding from the Federal Ministry of Education and Research through the Grant Center for Stroke Research Berlin (01 EO 0801), the Volkswagen Foundation (Lichtenberg program to M. Endres), DFG (NeuroCure), and EU (European Stroke Network).
M. Ebinger and I.G. contributed equally.
- Received September 19, 2009.
- Revision received October 5, 2009.
- Accepted October 7, 2009.
Chaturvedi S, Adams HP Jr, Woolson RF. Circadian variation in ischemic stroke subtypes. Stroke. 1999; 30: 1792–1795.
Moseley ME, Kucharczyk J, Mintorovitch J, Cohen Y, Kurhanewicz J, Derugin N, Asgari H, Norman D. Diffusion-weighted MR imaging of acute stroke: correlation with T2-weighted and magnetic susceptibility-enhanced MR imaging in cats. AJNR Am J Neuroradiol. 1990; 11: 423–429.
Welch KM, Windham J, Knight RA, Nagesh V, Hugg JW, Jacobs M, Peck D, Booker P, Dereski MO, Levine SR. A model to predict the histopathology of human stroke using diffusion and T2-weighted magnetic resonance imaging. Stroke. 1995; 26: 1983–1989.
Lansberg MG, Thijs VN, O'Brien MW, Ali JO, de Crespigny AJ, Tong DC, Moseley ME, Albers GW. Evolution of apparent diffusion coefficient, diffusion-weighted, and T2-weighted signal intensity of acute stroke. AJNR Am J Neuroradiol. 2001; 22: 637–644.
Wahlund LO, Barkhof F, Fazekas F, Bronge L, Augustin M, Sjogren M, Wallin A, Ader H, Leys D, Pantoni L, Pasquier F, Erkinjuntti T, Scheltens P. A new rating scale for age-related white matter changes applicable to MRI and CT. Stroke. 2001; 32: 1318–1322.
Schlaug G, Siewert B, Benfield A, Edelman RR, Warach S. Time course of the apparent diffusion coefficient (ADC) abnormality in human stroke. Neurology. 1997; 49: 113–119.
Cho AH, Kim JS, Kim SJ, Yun SC, Choi CG, Kim HR, Kwon SU, Lee DH, Kim EK, Suh DC, Kang DW. Focal fluid-attenuated inversion recovery hyperintensity within acute diffusion-weighted imaging lesions is associated with symptomatic intracerebral hemorrhage after thrombolysis. Stroke. 2008; 39: 3424–3426.