J Neurol (2014) 261:1097–1103 DOI 10.1007/s00415-014-7324-9
ORIGINAL COMMUNICATION
A simple brain atrophy measure improves the prediction of malignant middle cerebral artery infarction by acute DWI lesion volume Christoph Beck • Anna Kruetzelmann • Nils D. Forkert • Eric Juettler • Oliver C. Singer Martin Ko¨hrmann • Jan F. Kersten • Jan Sobesky • Christian Gerloff • Jens Fiehler • Peter D. Schellinger • Joachim Ro¨ther • Go¨tz Thomalla
•
Received: 9 February 2014 / Revised: 20 March 2014 / Accepted: 20 March 2014 / Published online: 1 April 2014 Ó Springer-Verlag Berlin Heidelberg 2014
Abstract In patients with malignant middle cerebral artery infarction (MMI) decompressive surgery within 48 h improves functional outcome. In this respect, early identification of patients at risk of developing MMI is crucial. While the acute diffusion weighted imaging (DWI) lesion volume was found to predict MMI with high predictive values, the potential impact of preexisting brain atrophy on the course of space-occupying middle cerebral artery (MCA) infarction and the development of MMI remains unclear. We tested the hypothesis that the combination of the acute DWI lesion volume with simple measures of
For the Clinical Trial Net of the German Competence Network Stroke. C. Beck (&) A. Kruetzelmann C. Gerloff G. Thomalla Klinik und Poliklinik fu¨r Neurologie, Kopf- und Neurozentrum, Universita¨tsklinikum Hamburg-Eppendorf, Martinistraße 52, 20246 Hamburg, Germany e-mail: [email protected] N. D. Forkert J. Fiehler Klinik und Poliklinik fu¨r Neuroradiologische Diagnostik und Intervention, Diagnostikzentrum, Universita¨tsklinikum Hamburg-Eppendorf, Martinistr. 52, 20246 Hamburg, Germany E. Juettler Neurologische Klinik, RKU-Universita¨ts- und Rehabilitationskliniken Ulm, Oberer Eselsberg 45, 89081 Ulm, Germany O. C. Singer Zentrum der Neurologie und Neurochirurgie, Klinik fu¨r Neurologie, Klinikum der Johann Wolfgang Goethe-Universita¨t, Theodor-Stern-Kai 7, 60590 Frankfurt, Germany
brain atrophy improves the early prediction of MMI. Data from a prospective, multicenter, observational study, which included patients with acute middle cerebral artery main stem occlusion studied by MRI within 6 h of symptom onset, was analyzed retrospectively. The development of MMI was defined according to the European randomized controlled trials of decompressive surgery. Acute DWI lesion volume, as well as brain and cerebrospinal fluid volume (CSF) were delineated. The intercaudate distance (ICD) was assessed as a linear brain atrophy marker by measuring the hemi-ICD of the intact hemisphere to account for local brain swelling. Binary logistic regression analysis was used to identify significant predictors of MMI. Cut-off values were determined by Classification and J. F. Kersten Institut fu¨r Medizinische Biometrie und Epidemiologie, Universita¨tsklinikum Hamburg-Eppendorf, Martinistr. 52, 20246 Hamburg, Germany J. Sobesky Center for Stroke Research Berlin (CSB) Charite´, Universita¨tsmedizin Berlin, Charite´ Campus Mitte, Charite´platz 1, 10117 Berlin, Germany P. D. Schellinger Neurologische Klinik, Johannes Wesling Klinikum Minden, Hans-Nolte-Straße 1, 32429 Minden, Germany J. Ro¨ther Neurologische Abteilung, Asklepios Klinik Altona, Paul-Ehrlich Straße 1, 22763 Hamburg, Germany
M. Ko¨hrmann Neurologische Klinik, Universita¨tsklinikum Erlangen, Maximiliansplatz 2, 91054 Erlangen, Germany
123
1098
Regression Trees analysis. Sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) of the resulting models were calculated. Twentyone (18 %) of 116 patients developed a MMI. Malignant middle cerebral artery infarctions patients had higher National Institutes of Health Stroke Scale scores on admission and presented more often with combined occlusion of the internal carotid artery and MCA. There were no differences in brain and CSF volume between the two groups. Diffusion weighted imaging lesion volume was larger (p \ 0.001), while hemi-ICD was smaller (p = 0.029) in MMI patients. Inclusion of hemi-ICD improved the prediction of MMI. Best cut-off values to predict the development of MMI were DWI lesion volume [ 87 ml and hemi-ICD B 9.4 mm. The addition of hemi-ICD to the decision tree strongly increased PPV (0.93 vs. 0.70) resulting in a reduction of false positive findings from 7/23 (30 %) to 1/15 (7 %), while there were only slight changes in specificity, sensitivity and NPV. The absolute number of correct classifications increased by 4 (3.4 %). The integration of hemi-ICD as a linear marker of brain atrophy, that can easily be assessed in an emergency setting, may improve the prediction of MMI by lesion volume based predictive models. Keywords Stroke Acute Malignant cerebral infarction Magnetic resonance imaging Diffusion-weighted Brain atrophy
Introduction Malignant middle cerebral artery infarctions (MMI) are characterized by consecutive impairment of consciousness due to space occupying cerebral edema with subsequent herniation, mostly leading to death. Mortality rates for conservative treatment strategies range around 80 %, while decompressive surgery within 48 h reduces mortality and improves functional outcome [1–3]. Hence, the early identification of patients with a malignant course of brain edema is essential to guide an improved patient-specific treatment. So far, the extent of ischemic infarction, particularly the lesion volume measured by early diffusion weighted imaging (DWI), was identified as a major determinant for the prediction of MMI with a high positive and negative predictive value [4–8]. However, the space-occupying effect of ischemic edema not only depends on lesion size, but also on the intracranial volume reserve that is mainly determined by preexisting brain atrophy. Moderate to severe brain atrophy might protect stroke patients from the development of MMI [9, 10]. Just recently, a CT-based study demonstrated that cerebrospinal fluid (CSF) volume
123
J Neurol (2014) 261:1097–1103
measurements improve the prediction of a malignant course compared to the prediction using only the cerebral blood volume (CBV) lesion volume [8]. While volumetric measurements of brain and CSF volume usually require a more or less complex post-processing, brain atrophy can also be assessed by determining the intercaudate distance (ICD), which represents a linear brain atrophy marker that can be easily applied even in an emergency setting [11]. We aimed to test the hypothesis, that the integration of a simple brain atrophy measure improves the prediction of MMI based on acute DWI lesion volume as recently proposed [4, 6, 10].
Methods Study design, patients, and clinical assessment We analyzed clinical and MRI data of a prospective observational multicenter study of patients with an acute stroke due to middle cerebral artery (MCA) main stem occlusion. The current study provides new and original results, which have not been reported in an original publication [6]. In brief, inclusion criteria in this study comprised an acute ischemic stroke diagnosed by MRI within 6 h of symptom onset as a consequence of MCA main stem occlusion assessed by time of flight-MR angiography. Patients were enrolled prospectively and the diagnosis of MMI was made based on the definitions used in the European randomized controlled trials of decompressive surgery in malignant MCA infarction [3]. Severity of neurological deficit on admission was evaluated using the National Institutes of Health Stroke Scale (NIHSS) [12]. Clinical outcome was assessed by the modified Rankin Scale 90 days after symptom onset [13]. The study was approved by the local ethics committees at each site. MRI protocol and image analysis The acute stroke MRI protocol has been described elsewhere [6]. For volumetric analysis, the infarct lesion was semi-automatically segmented on ADC maps using two manually defined volumes-of-interest (VOIs) comprising a coarse infarct lesion definition and the contralateral healthy hemisphere. The latter VOI was used for an automatic refinement of the infarct lesion. More precisely, voxels in the infarct lesion VOI exceeding 80 % of the mean healthy ADC intensity were rejected from the coarse manual lesion definition resulting in a refined infarct segmentation. A brain segmentation method adapted to ADC image sequences was then performed for exclusion of non-cerebral tissue [14]. This brain segmentation was then further subdivided in CSF and brain tissue by segmenting the CSF
J Neurol (2014) 261:1097–1103
1099
dependent variable, including DWI lesion volume, NIHSS on admission, and vessel occlusion. Vessel occlusion was dichotomized into isolated MCA occlusion (‘‘MCA alone’’) and vessel occlusion involving the extra- or intracranial ICA plus the MCA main stem (‘‘MCA ? ICA’’). In a second model, the hemi-ICD parameter was added to the model. Area under the curve of the models was calculated and the performance of both models was compared using the likelihood-ratio test. To identify the optimal imaging predictors of MMI, we further used Classification and Regression Trees (CART) analysis to generate binary trees and determined optimal thresholds of acute DWI lesion volume and hemi-ICD to predict MMI. We calculated sensitivity, specificity, PPV and NPV for two simple decision trees based on the cut-offs generated by CART analysis.
Results
Fig. 1 Measurement of DWI lesion volume, cerebrospinal fluid (CSF) volume, brain volume (left side) and hemi-intercaudate distance (hemi-ICD, right side) in a MMI and non-MMI patient. Values were: hemi-ICD: 7.1 vs. 9.7 mm; DWI lesion volume: 153 vs. 128 ml; CSF volume: 170 vs. 298 ml; brain volume: 910 vs. 1,041 ml
using a global lower threshold of 1,500 9 10-6 mm2/s. The ICD was measured at the minimum distance between the medial borders of the head of the caudate nuclei at the level of the foramen Monroi in axial b = 0 trace diffusionweighted MR slices using standardized window settings (center 300, width 300) (Fig. 1). To account for spaceoccupying edema, the hemi-ICD, ranging to the midline, was assessed on the side of the non-infarcted hemisphere [10, 11]. The hemi-ICD was measured by one observer (CB) blinded to clinical outcome, but not DWI lesion volume. To enhance precision, the mean value out of three measurements was used. Statistical analysis Demographic, clinical and imaging parameters were compared between the two groups using Fisher’s exact test or Student’s t test as appropriate. In case of asymmetric distribution, values were logarithmized and the geometric instead of the arithmetic mean was used for group comparison. As this is a post hoc analysis, within a population from a previous study, all results are considered exploratory. Thus, we refrained from correction for multiple testing but descriptively report p-values for all analysis. To identify predictors of MMI, we performed a multivariate binary logistic regression analysis with MMI as a
Of 140 patients enrolled in the original study, 24 had to be excluded for technical reasons, as not all diffusion weighted images were available, and, thus, the post-processing streamline could not be completed in these patients. Of the remaining 116 patients, 21 (18 %) developed MMI and 95 (82 %) did not (non-MMI). Both groups were comparable regarding age, sex, side of infarction, and time from symptom onset to MRI (Table 1). Malignant middle cerebral artery infarction patients presented with higher NIHSS scores (20 vs. 16, p = 0.001) and more frequently reduced consciousness (57 % vs. 24 %, p = 0.007) on admission. In addition, large artery atherosclerosis tended to be more frequent in MMI patients, while cardioembolism was less likely to be the origin of stroke. As to imaging findings, acute DWI lesion volume was noticeably larger in MMI patients (97 vs. 27 ml, p \ 0.001), and combined MCA ? ICA occlusion tended to be more frequent in MMI patients (62 % vs. 39 %, p = 0.087). With regards to atrophy measurements, hemiICD was shorter in MMI patients (8.2 vs. 9.6 mm, p = 0.029) while intracranial volume, brain volume, and CSF volume were comparable between groups. Results of logistic regression analysis are given in Table 2. Inclusion of hemi-ICD into the predictive model (model 2) significantly improved the model performance (-2 log likelihood-ratio 63.7 vs. 58.1, p = 0.018). Classification and Regression Trees analysis identified a DWI lesion volume of [87 ml and a hemi-ICD of B9.4 mm as optimal cut-off values (Fig. 2). Using the hemi-ICD together with DWI lesion volume lead to an increased PPV (0.93 vs. 0.70) and specificity (0.99 vs. 0.93), while sensitivity decreased (0.67 vs. 0.76) and NPV remained largely unchanged (0.93 vs. 0.95) (Table 3) compared to
123
1100
J Neurol (2014) 261:1097–1103
Table 1 Group comparison— clinical and imaging characteristics
MMI (n = 21)
Non-MMI (n = 95)
Group comparison (p values)
Baseline Age (years), mean (95 % CI)
62 (57–68)
66 (63–69)
0.249
Female, n (%)
8 (38 %)
49 (52 %)
0.337
Left hemispheric stroke, n (%)
14 (67 %)
59 (62 %)
0.805
NIHSS on admission, mean (95 % CI)
20 (18–22)
16 (15–17)
0.001
LOC on admission [ 0, n (%)
12 (57)
22 (24) [93]
0.007
Large-artery
12 (57)
29 (31)
Cardioembolism
4 (19)
36 (38)
Other
0
3 (3)
5 (24)
27 (28)
162 (136–193)
135 (122–150)
ICA ? MCA occlusion
13 (62 %)
37 (39 %)
Isolated MCA occlusion
8 (38 %)
58 (61 %)
97 (69–136)
27 (23–33)
Etiology, n (%)
0.066
Undetermined Baseline MRI Findings Time to MRI (min), mean (95 % CI) Vessel occlusion dichotomized, n (%)
Values are given as mean (95 % CI) or count (percentage) as appropriate. Numbers of patients with data available are given in brackets for each group MMI malignant middle cerebral artery infarction, NIHSS National Institutes of Health Stroke Scale, LOC level of consciousness (reaching from 0 = alert to 3 = coma), ICA internal carotid artery, MCA middle cerebral artery, MRS modified Rankin scale, CSF cerebrospinal fluid, Hemi-ICD hemi-intercaudate distance
0.087
Imaging Volumetric analysis (ml), mean (95 % CI) Diffusion lesion volume
\0.001
Full intracranial volume
1,177 (1,123–1,231)
1,158 (1,115–1,201)
0.699
Brain volume
902 (851–954)
854 (822–885)
0.179
CSF volume
274 (232–317)
306 (281–331)
0.274
8.2 (7.4–9.1)
9.6 (9.1–10.2)
0.029
Brain volume
0.768 (0.736–0.801)
0.741 (0.725–0.758)
0.167
CSF volume
0.232 (0.199–0.264)
0.259 (0.243–0.276)
0.157
Linear atrophy marker Hemi-ICD (mm), mean (95 % CI) Normalized values
solely using the DWI lesion volume. The absolute number of correct classifications could be increased by 4 (3.5 %). The increased PPV is mirrored by a reduction of the number of false positive classifications from 7 (30.4 %) to only 1 (6.7 %) (Fig. 2).
Discussion In this retrospective analysis of data from a larger prospective observational study, we demonstrated that the prediction of malignant middle artery infarction by DWI lesion volume can be improved by inclusion of the ICD, a simple linear marker of brain atrophy. This complies with the pathophysiological assumption that cerebral atrophy may protect patients from a malignant course in spaceoccupying MCA infarction by given space for ischemic edema. It also reflects the clinical apprehension that a large
123
0.123
cerebral infarction is more likely to result in a malignant cerebral edema in younger patients, which was attributed to age related brain atrophy [7, 9, 15]. Recently, a CT-based study brought up similar findings. Here, the ratio of the ischemic lesion volume to intracranial volume reserve, measured as CSF volume, was a strong predictor of a malignant MCA infarction [9]. While volumetric analysis to determine the grade of cerebral atrophy is time consuming and not clinically applicable, linear brain atrophy markers are easily assessable. We focused on simple measures that can easily be adapted into clinical practice to guide acute treatment decisions without the requirement of complex post-processing procedures. In this respect, the ICD has been previously used as a valid marker with a good longitudinally correlation to age related brain atrophy and high inter-rater reliability measures indicated an intraclass correlation coefficient of 0.89 [11]. In our study, this simple measure
J Neurol (2014) 261:1097–1103
1101
Table 2 Prediction of MMI: multivariate binary regression analysis OR
95 % CI
p values
\0.001
A
Model 1—without hemi-ICD AUC = 0.893, 95 % CI: (0.802–0.983) Correct classifications: 103/116 (88.8 %) DWI
1.033
1.017–1.048
NIHSS
1.132
0.961–1.334
0.138
ICA ? MCA occlusion
2.652
0.725–9.706
0.141
Model 2—with hemi-ICD AUC = 0.902, 95 % CI: (0.809–0.995) Correct classifications: 107/116 (92.2 %) DWI
1.003
1.016–1.049
\0.001
Hemi-ICD NIHSS
0.699 1.167
0.502–0.974 0.975–1.396
0.034 0.092
ICA ? MCA occlusion
3.344
0.828–13.511
0.090
B
Results of a multivariate binary regression analysis with malignant cerebral artery incarction (MMI) as dependent variable OR odds ratio, CI confidence interval, AUC area under curves, ICA internal carotid artery, MCA middle cerebral artery, NIHSS National Institutes of Health Stroke Scale, DWI diffusion weighted imaging, Hemi-ICD hemi-intercaudate distance
improved the prediction of a malignant brain swelling in large MCA stroke significantly. Contrary to a previous study, overall CSF-volume did not turn out to be predictive of a malignant course [9]. In our sample, total brain and CSF volumes did not differ between MMI and non-MMI patients. This finding suggests that at the very early acute stage (i.e., within the first hours of stroke, the window in which our sample was studied) local swelling at the level of the basal ganglia as measured by the hemi-ICD might be more sensitive and strongly related to the development of malignant brain swelling than overall volumes of brain and CSF volume. This could be explained by a compartmentalization of intracranial space where mass effects at different locations carry a different impact. Thus, in this early stage of infarction, the hemiICD may rather reflect local brain edema at the level of the basal ganglia than global cerebral atrophy. As a consequence, we may speculate that a smaller hemi-ICD at this stage identifies patients with a tendency to early swelling at the level of the basal ganglia resulting from both extent and location of their ischemic lesion. Differences in the results of the two studies may also result from the fact that in the study by Minnerup et al., median size of the acute CBV lesion reflecting infarct core was much larger than in our cohort resulting in a more pronounced overall displacement of CSF [9]. Finally, while delineation of the infarct core in acute ischemic stroke by CT is still a controversial issue and not well established as a predictor of MMI, the predictive value of a large acute DWI lesion is beyond doubt [4–6].
Fig. 2 Model A shows the single step decision tree dividing into two groups based on DWI lesion volume (B or [ 87 ml). The classification is shown by the color of the bars (MMI = grey, nonMMI = white). Model B shows the two step decision tree model classifying patients based on DWI lesion volume (B or [ 87 ml) and hemi-ICD (B or [ 9.4 mm)
Although ICD significantly improved the prediction of MMI by DWI, the net increase of predictive power was small with an increase of the number of correct classification by 3.4 %. In a previous study using ICD measurement to assess brain atrophy, an ICD [ 20 mm significantly reduced the risk of development of MMI with an odds ratio of 0.137 [10]. Again, this discrepancy may be attributed to different cohort characteristics. Particularly, in the study by Lee et al., mean infarct volume (224 cm3) was more than twice as large as the acute DWI lesion volume in the group of MMI patients in our study (97 cm3). Patients were also older than in our cohort (mean age 70 vs. 62 years). However, while the overall improvement of prediction of MMI by the inclusion of ICD was rather small, the number of false positive findings, i.e. patients predicted to develop MMI while they did not, was markedly reduced from 7/23 (30 %) to only 1/15 (7 %), which is also reflected by a more pronounced increase of specificity
123
1102
J Neurol (2014) 261:1097–1103
Table 3 Prediction of MMI: predictive values for both models Sensitivity
Specificity
PPV
NPV
Correct classifications
0.76 (0.53–0.92)
0.93 (0.85–0.97)
0.70 (0.47–0.87)
0.95 (0.90–0.99)
104/116 (89.7 %)
0.67 (0.43–0.85)
0.99 (0.94–1.00)
0.93 (0.68–1.00)
0.93 (0.86–0.97)
108/116 (93.1 %)
Model A DWI [ 87 ml Model B DWI [ 87 ml and hemi-ICD B 9.4 mm Values are given as mean (95 % CI) DWI diffusion weighted imaging, Hemi-ICD hemi-intercaudate distance, PPV positive predictive value, NPV negative predictive value
and PPV. In the simple decision tree model combining DWI lesion volume and ICD, specificity was as high as 0.99 and PPV increased from 0.70 to 0.93. From a pathophysiological perspective this may easily be explained as a larger hemi-ICD reflects a larger intracranial volume reserve which may protect patients with large brain infarction from malignant swelling [9]. Adding hemi-ICD to the predictive model based on DWI lesion volume, thus, excludes the subgroup of patients with large initial DWI volumes that is protected from malignant swelling by a larger intracranial volume reserve. From a clinical point of view, this simple algorithm might help identify patients for early decompressive surgery without waiting for clinical or radiological signs of massive space-occupying effects. Our study has limitations. First, the MR images of 24 patients had to be excluded from the analysis due to technical reasons. This represents a substantial proportion of the entire population and may have introduced a bias. Moreover, while the optimal DWI lesion volume predicting MMI identified in this analysis ([87 ml) was quite close to the cut-off identified in the original paper ([82 ml), the predictive values showed more remarkable differences. This points towards the vulnerability of the findings to random changes resulting from the small sample size. Also, measurement of the hemi-ICD was performed by one observer not blinded to DWI lesion volume. Therefore, a systematic bias can not be ruled out. Furthermore, while the predictive value of acute DWI lesion volume was confirmed in a prospective study, the findings from the current study result from retrospective analysis and need confirmation in future trials. To summarize, the inclusion of a simple linear marker of brain atrophy like the ICD, which can be measured at the scanner within seconds, significantly improves the prediction of MMI by DWI lesion volume. While the overall effect is small, there is a relevant increase of PPV and specificity, which might help in clinical decision making, when early decompressive surgery is considered. Acknowledgments The study was supported by the German Kompetenznetzwerk Schlaganfall sponsored by the Bundesministerium fu¨r Bildung und Forschung (B5; No. 01GI9902/4).
123
Conflicts of interest On behalf of all authors, the corresponding author states that there is no conflict of interest.
References 1. Hacke W, Schwab S, Horn M, Spranger M, De Georgia M, von Kummer R (1996) ‘Malignant’ middle cerebral artery territory infarction: clinical course and prognostic signs. Arch Neurol 53(4):309–315 2. Schwab S, Steiner T, Aschoff A, Schwarz S, Steiner HH, Jansen O, Hacke W (1998) Early hemicraniectomy in patients with complete middle cerebral artery infarction. Stroke 29(9): 1888–1893 3. Vahedi K, Hofmeijer J, Juettler E, Vicaut E, George B, Algra A, Amelink GJ, Schmiedeck P, Schwab S, Rothwell PM, Bousser MG, van der Worp HB, Hacke W (2007) Earlydecompressive surgery in malignant infarction of the middle cerebral artery: a pooled analysis of three randomised controlled trials. Lancet Neurol 6(3):215–222. doi:10.1016/S1474-4422(07)70036-4 4. Oppenheim C, Samson Y, Manai R, Lalam T, Vandamme X, Crozier S, Srour A, Cornu P, Dormont D, Rancurel G, Marsault C (2000) Prediction of malignant middle cerebral artery infarction by diffusion-weighted imaging. Stroke 31(9):2175–2181 5. Thomalla GJ, Kucinski T, Schoder V, Fiehler J, Knab R, Zeumer H, Weiller C, Rother J (2003) Prediction of malignant middle cerebral artery infarction by early perfusion- and diffusionweighted magnetic resonance imaging. Stroke 34(8):1892–1899. doi:10.1161/01.STR.0000081985.44625.B6 6. Thomalla G, Hartmann F, Juettler E, Singer OC, Lehnhardt FG, Kohrmann M, Kersten JF, Krutzelmann A, Humpich MC, Sobesky J, Gerloff C, Villringer A, Fiehler J, Neumann-Haefelin T, Schellinger PD, Rother J (2010) Prediction of malignant middle cerebral artery infarction by magnetic resonance imaging within 6 hours of symptom onset: a prospective multicenter observational study. Ann Neurol 68(4):435–445. doi:10.1002/ ana.22125 7. Hofmeijer J, Algra A, Kappelle LJ, van der Worp HB (2008) Predictors of life-threatening brain edema in middle cerebral artery infarction. Cerebrovasc Dis 25(1–2):176–184. doi:10.1159/ 000113736 8. Dohmen C, Galldiks N, Bosche B, Kracht L, Graf R (2012) The severity of ischemia determines and predicts malignant brain edema in patients with large middle cerebral artery infarction. Cerebrovasc Dis 33(1):1–7. doi:10.1159/000330648 9. Minnerup J, Wersching H, Ringelstein EB, Heindel W, Niederstadt T, Schilling M, Schabitz WR, Kemmling A (2011) Prediction of malignant middle cerebral artery infarction using computed tomography-based intracranial volume reserve measurements. Stroke 42(12):3403–3409. doi:10.1161/STROKEAHA. 111.619734
J Neurol (2014) 261:1097–1103 10. Lee SH, Oh CW, Han JH, Kim CY, Kwon OK, Son YJ, Bae HJ, Han MK, Chung YS (2010) The effect of brain atrophy on outcome after a large cerebral infarction. J Neurol Neurosurg Psychiatry 81(12):1316–1321. doi:10.1136/jnnp.2009.197335 11. Butzkueven H, Kolbe SC, Jolley DJ, Brown JY, Cook MJ, van der Mei IA, Groom PS, Carey J, Eckholdt J, Rubio JP, Taylor BV, Mitchell PJ, Egan GF, Kilpatrick TJ (2008) Validation of linear cerebral atrophy markers in multiple sclerosis. J Clin Neurosci 15(2):130–137. doi:10.1016/j.jocn.2007.02.089 12. Brott T, Adams HP Jr, Olinger CP, Marler JR, Barsan WG, Biller J, Spilker J, Holleran R, Eberle R, Hertzberg V et al (1989)
1103 Measurements of acute cerebral infarction: a clinical examination scale. Stroke 20(7):864–870 13. van Swieten JC, Koudstaal PJ, Visser MC, Schouten HJ, van Gijn J (1988) Interobserver agreement for the assessment of handicap in stroke patients. Stroke 19(5):604–607 14. Forkert ND, Saring D, Fiehler J, Illies T, Moller D, Handels H (2009) Automatic brain segmentation in Time-of-Flight MRA images. Methods Inf Med 48(5):399–407. doi:10.3414/ME9237 15. Krieger DW, Demchuk AM, Kasner SE, Jauss M, Hantson L (1999) Early clinical and radiological predictors of fatal brain swelling in ischemic stroke. Stroke 30(2):287–292
123
ORIGINAL COMMUNICATION
A simple brain atrophy measure improves the prediction of malignant middle cerebral artery infarction by acute DWI lesion volume Christoph Beck • Anna Kruetzelmann • Nils D. Forkert • Eric Juettler • Oliver C. Singer Martin Ko¨hrmann • Jan F. Kersten • Jan Sobesky • Christian Gerloff • Jens Fiehler • Peter D. Schellinger • Joachim Ro¨ther • Go¨tz Thomalla
•
Received: 9 February 2014 / Revised: 20 March 2014 / Accepted: 20 March 2014 / Published online: 1 April 2014 Ó Springer-Verlag Berlin Heidelberg 2014
Abstract In patients with malignant middle cerebral artery infarction (MMI) decompressive surgery within 48 h improves functional outcome. In this respect, early identification of patients at risk of developing MMI is crucial. While the acute diffusion weighted imaging (DWI) lesion volume was found to predict MMI with high predictive values, the potential impact of preexisting brain atrophy on the course of space-occupying middle cerebral artery (MCA) infarction and the development of MMI remains unclear. We tested the hypothesis that the combination of the acute DWI lesion volume with simple measures of
For the Clinical Trial Net of the German Competence Network Stroke. C. Beck (&) A. Kruetzelmann C. Gerloff G. Thomalla Klinik und Poliklinik fu¨r Neurologie, Kopf- und Neurozentrum, Universita¨tsklinikum Hamburg-Eppendorf, Martinistraße 52, 20246 Hamburg, Germany e-mail: [email protected] N. D. Forkert J. Fiehler Klinik und Poliklinik fu¨r Neuroradiologische Diagnostik und Intervention, Diagnostikzentrum, Universita¨tsklinikum Hamburg-Eppendorf, Martinistr. 52, 20246 Hamburg, Germany E. Juettler Neurologische Klinik, RKU-Universita¨ts- und Rehabilitationskliniken Ulm, Oberer Eselsberg 45, 89081 Ulm, Germany O. C. Singer Zentrum der Neurologie und Neurochirurgie, Klinik fu¨r Neurologie, Klinikum der Johann Wolfgang Goethe-Universita¨t, Theodor-Stern-Kai 7, 60590 Frankfurt, Germany
brain atrophy improves the early prediction of MMI. Data from a prospective, multicenter, observational study, which included patients with acute middle cerebral artery main stem occlusion studied by MRI within 6 h of symptom onset, was analyzed retrospectively. The development of MMI was defined according to the European randomized controlled trials of decompressive surgery. Acute DWI lesion volume, as well as brain and cerebrospinal fluid volume (CSF) were delineated. The intercaudate distance (ICD) was assessed as a linear brain atrophy marker by measuring the hemi-ICD of the intact hemisphere to account for local brain swelling. Binary logistic regression analysis was used to identify significant predictors of MMI. Cut-off values were determined by Classification and J. F. Kersten Institut fu¨r Medizinische Biometrie und Epidemiologie, Universita¨tsklinikum Hamburg-Eppendorf, Martinistr. 52, 20246 Hamburg, Germany J. Sobesky Center for Stroke Research Berlin (CSB) Charite´, Universita¨tsmedizin Berlin, Charite´ Campus Mitte, Charite´platz 1, 10117 Berlin, Germany P. D. Schellinger Neurologische Klinik, Johannes Wesling Klinikum Minden, Hans-Nolte-Straße 1, 32429 Minden, Germany J. Ro¨ther Neurologische Abteilung, Asklepios Klinik Altona, Paul-Ehrlich Straße 1, 22763 Hamburg, Germany
M. Ko¨hrmann Neurologische Klinik, Universita¨tsklinikum Erlangen, Maximiliansplatz 2, 91054 Erlangen, Germany
123
1098
Regression Trees analysis. Sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) of the resulting models were calculated. Twentyone (18 %) of 116 patients developed a MMI. Malignant middle cerebral artery infarctions patients had higher National Institutes of Health Stroke Scale scores on admission and presented more often with combined occlusion of the internal carotid artery and MCA. There were no differences in brain and CSF volume between the two groups. Diffusion weighted imaging lesion volume was larger (p \ 0.001), while hemi-ICD was smaller (p = 0.029) in MMI patients. Inclusion of hemi-ICD improved the prediction of MMI. Best cut-off values to predict the development of MMI were DWI lesion volume [ 87 ml and hemi-ICD B 9.4 mm. The addition of hemi-ICD to the decision tree strongly increased PPV (0.93 vs. 0.70) resulting in a reduction of false positive findings from 7/23 (30 %) to 1/15 (7 %), while there were only slight changes in specificity, sensitivity and NPV. The absolute number of correct classifications increased by 4 (3.4 %). The integration of hemi-ICD as a linear marker of brain atrophy, that can easily be assessed in an emergency setting, may improve the prediction of MMI by lesion volume based predictive models. Keywords Stroke Acute Malignant cerebral infarction Magnetic resonance imaging Diffusion-weighted Brain atrophy
Introduction Malignant middle cerebral artery infarctions (MMI) are characterized by consecutive impairment of consciousness due to space occupying cerebral edema with subsequent herniation, mostly leading to death. Mortality rates for conservative treatment strategies range around 80 %, while decompressive surgery within 48 h reduces mortality and improves functional outcome [1–3]. Hence, the early identification of patients with a malignant course of brain edema is essential to guide an improved patient-specific treatment. So far, the extent of ischemic infarction, particularly the lesion volume measured by early diffusion weighted imaging (DWI), was identified as a major determinant for the prediction of MMI with a high positive and negative predictive value [4–8]. However, the space-occupying effect of ischemic edema not only depends on lesion size, but also on the intracranial volume reserve that is mainly determined by preexisting brain atrophy. Moderate to severe brain atrophy might protect stroke patients from the development of MMI [9, 10]. Just recently, a CT-based study demonstrated that cerebrospinal fluid (CSF) volume
123
J Neurol (2014) 261:1097–1103
measurements improve the prediction of a malignant course compared to the prediction using only the cerebral blood volume (CBV) lesion volume [8]. While volumetric measurements of brain and CSF volume usually require a more or less complex post-processing, brain atrophy can also be assessed by determining the intercaudate distance (ICD), which represents a linear brain atrophy marker that can be easily applied even in an emergency setting [11]. We aimed to test the hypothesis, that the integration of a simple brain atrophy measure improves the prediction of MMI based on acute DWI lesion volume as recently proposed [4, 6, 10].
Methods Study design, patients, and clinical assessment We analyzed clinical and MRI data of a prospective observational multicenter study of patients with an acute stroke due to middle cerebral artery (MCA) main stem occlusion. The current study provides new and original results, which have not been reported in an original publication [6]. In brief, inclusion criteria in this study comprised an acute ischemic stroke diagnosed by MRI within 6 h of symptom onset as a consequence of MCA main stem occlusion assessed by time of flight-MR angiography. Patients were enrolled prospectively and the diagnosis of MMI was made based on the definitions used in the European randomized controlled trials of decompressive surgery in malignant MCA infarction [3]. Severity of neurological deficit on admission was evaluated using the National Institutes of Health Stroke Scale (NIHSS) [12]. Clinical outcome was assessed by the modified Rankin Scale 90 days after symptom onset [13]. The study was approved by the local ethics committees at each site. MRI protocol and image analysis The acute stroke MRI protocol has been described elsewhere [6]. For volumetric analysis, the infarct lesion was semi-automatically segmented on ADC maps using two manually defined volumes-of-interest (VOIs) comprising a coarse infarct lesion definition and the contralateral healthy hemisphere. The latter VOI was used for an automatic refinement of the infarct lesion. More precisely, voxels in the infarct lesion VOI exceeding 80 % of the mean healthy ADC intensity were rejected from the coarse manual lesion definition resulting in a refined infarct segmentation. A brain segmentation method adapted to ADC image sequences was then performed for exclusion of non-cerebral tissue [14]. This brain segmentation was then further subdivided in CSF and brain tissue by segmenting the CSF
J Neurol (2014) 261:1097–1103
1099
dependent variable, including DWI lesion volume, NIHSS on admission, and vessel occlusion. Vessel occlusion was dichotomized into isolated MCA occlusion (‘‘MCA alone’’) and vessel occlusion involving the extra- or intracranial ICA plus the MCA main stem (‘‘MCA ? ICA’’). In a second model, the hemi-ICD parameter was added to the model. Area under the curve of the models was calculated and the performance of both models was compared using the likelihood-ratio test. To identify the optimal imaging predictors of MMI, we further used Classification and Regression Trees (CART) analysis to generate binary trees and determined optimal thresholds of acute DWI lesion volume and hemi-ICD to predict MMI. We calculated sensitivity, specificity, PPV and NPV for two simple decision trees based on the cut-offs generated by CART analysis.
Results
Fig. 1 Measurement of DWI lesion volume, cerebrospinal fluid (CSF) volume, brain volume (left side) and hemi-intercaudate distance (hemi-ICD, right side) in a MMI and non-MMI patient. Values were: hemi-ICD: 7.1 vs. 9.7 mm; DWI lesion volume: 153 vs. 128 ml; CSF volume: 170 vs. 298 ml; brain volume: 910 vs. 1,041 ml
using a global lower threshold of 1,500 9 10-6 mm2/s. The ICD was measured at the minimum distance between the medial borders of the head of the caudate nuclei at the level of the foramen Monroi in axial b = 0 trace diffusionweighted MR slices using standardized window settings (center 300, width 300) (Fig. 1). To account for spaceoccupying edema, the hemi-ICD, ranging to the midline, was assessed on the side of the non-infarcted hemisphere [10, 11]. The hemi-ICD was measured by one observer (CB) blinded to clinical outcome, but not DWI lesion volume. To enhance precision, the mean value out of three measurements was used. Statistical analysis Demographic, clinical and imaging parameters were compared between the two groups using Fisher’s exact test or Student’s t test as appropriate. In case of asymmetric distribution, values were logarithmized and the geometric instead of the arithmetic mean was used for group comparison. As this is a post hoc analysis, within a population from a previous study, all results are considered exploratory. Thus, we refrained from correction for multiple testing but descriptively report p-values for all analysis. To identify predictors of MMI, we performed a multivariate binary logistic regression analysis with MMI as a
Of 140 patients enrolled in the original study, 24 had to be excluded for technical reasons, as not all diffusion weighted images were available, and, thus, the post-processing streamline could not be completed in these patients. Of the remaining 116 patients, 21 (18 %) developed MMI and 95 (82 %) did not (non-MMI). Both groups were comparable regarding age, sex, side of infarction, and time from symptom onset to MRI (Table 1). Malignant middle cerebral artery infarction patients presented with higher NIHSS scores (20 vs. 16, p = 0.001) and more frequently reduced consciousness (57 % vs. 24 %, p = 0.007) on admission. In addition, large artery atherosclerosis tended to be more frequent in MMI patients, while cardioembolism was less likely to be the origin of stroke. As to imaging findings, acute DWI lesion volume was noticeably larger in MMI patients (97 vs. 27 ml, p \ 0.001), and combined MCA ? ICA occlusion tended to be more frequent in MMI patients (62 % vs. 39 %, p = 0.087). With regards to atrophy measurements, hemiICD was shorter in MMI patients (8.2 vs. 9.6 mm, p = 0.029) while intracranial volume, brain volume, and CSF volume were comparable between groups. Results of logistic regression analysis are given in Table 2. Inclusion of hemi-ICD into the predictive model (model 2) significantly improved the model performance (-2 log likelihood-ratio 63.7 vs. 58.1, p = 0.018). Classification and Regression Trees analysis identified a DWI lesion volume of [87 ml and a hemi-ICD of B9.4 mm as optimal cut-off values (Fig. 2). Using the hemi-ICD together with DWI lesion volume lead to an increased PPV (0.93 vs. 0.70) and specificity (0.99 vs. 0.93), while sensitivity decreased (0.67 vs. 0.76) and NPV remained largely unchanged (0.93 vs. 0.95) (Table 3) compared to
123
1100
J Neurol (2014) 261:1097–1103
Table 1 Group comparison— clinical and imaging characteristics
MMI (n = 21)
Non-MMI (n = 95)
Group comparison (p values)
Baseline Age (years), mean (95 % CI)
62 (57–68)
66 (63–69)
0.249
Female, n (%)
8 (38 %)
49 (52 %)
0.337
Left hemispheric stroke, n (%)
14 (67 %)
59 (62 %)
0.805
NIHSS on admission, mean (95 % CI)
20 (18–22)
16 (15–17)
0.001
LOC on admission [ 0, n (%)
12 (57)
22 (24) [93]
0.007
Large-artery
12 (57)
29 (31)
Cardioembolism
4 (19)
36 (38)
Other
0
3 (3)
5 (24)
27 (28)
162 (136–193)
135 (122–150)
ICA ? MCA occlusion
13 (62 %)
37 (39 %)
Isolated MCA occlusion
8 (38 %)
58 (61 %)
97 (69–136)
27 (23–33)
Etiology, n (%)
0.066
Undetermined Baseline MRI Findings Time to MRI (min), mean (95 % CI) Vessel occlusion dichotomized, n (%)
Values are given as mean (95 % CI) or count (percentage) as appropriate. Numbers of patients with data available are given in brackets for each group MMI malignant middle cerebral artery infarction, NIHSS National Institutes of Health Stroke Scale, LOC level of consciousness (reaching from 0 = alert to 3 = coma), ICA internal carotid artery, MCA middle cerebral artery, MRS modified Rankin scale, CSF cerebrospinal fluid, Hemi-ICD hemi-intercaudate distance
0.087
Imaging Volumetric analysis (ml), mean (95 % CI) Diffusion lesion volume
\0.001
Full intracranial volume
1,177 (1,123–1,231)
1,158 (1,115–1,201)
0.699
Brain volume
902 (851–954)
854 (822–885)
0.179
CSF volume
274 (232–317)
306 (281–331)
0.274
8.2 (7.4–9.1)
9.6 (9.1–10.2)
0.029
Brain volume
0.768 (0.736–0.801)
0.741 (0.725–0.758)
0.167
CSF volume
0.232 (0.199–0.264)
0.259 (0.243–0.276)
0.157
Linear atrophy marker Hemi-ICD (mm), mean (95 % CI) Normalized values
solely using the DWI lesion volume. The absolute number of correct classifications could be increased by 4 (3.5 %). The increased PPV is mirrored by a reduction of the number of false positive classifications from 7 (30.4 %) to only 1 (6.7 %) (Fig. 2).
Discussion In this retrospective analysis of data from a larger prospective observational study, we demonstrated that the prediction of malignant middle artery infarction by DWI lesion volume can be improved by inclusion of the ICD, a simple linear marker of brain atrophy. This complies with the pathophysiological assumption that cerebral atrophy may protect patients from a malignant course in spaceoccupying MCA infarction by given space for ischemic edema. It also reflects the clinical apprehension that a large
123
0.123
cerebral infarction is more likely to result in a malignant cerebral edema in younger patients, which was attributed to age related brain atrophy [7, 9, 15]. Recently, a CT-based study brought up similar findings. Here, the ratio of the ischemic lesion volume to intracranial volume reserve, measured as CSF volume, was a strong predictor of a malignant MCA infarction [9]. While volumetric analysis to determine the grade of cerebral atrophy is time consuming and not clinically applicable, linear brain atrophy markers are easily assessable. We focused on simple measures that can easily be adapted into clinical practice to guide acute treatment decisions without the requirement of complex post-processing procedures. In this respect, the ICD has been previously used as a valid marker with a good longitudinally correlation to age related brain atrophy and high inter-rater reliability measures indicated an intraclass correlation coefficient of 0.89 [11]. In our study, this simple measure
J Neurol (2014) 261:1097–1103
1101
Table 2 Prediction of MMI: multivariate binary regression analysis OR
95 % CI
p values
\0.001
A
Model 1—without hemi-ICD AUC = 0.893, 95 % CI: (0.802–0.983) Correct classifications: 103/116 (88.8 %) DWI
1.033
1.017–1.048
NIHSS
1.132
0.961–1.334
0.138
ICA ? MCA occlusion
2.652
0.725–9.706
0.141
Model 2—with hemi-ICD AUC = 0.902, 95 % CI: (0.809–0.995) Correct classifications: 107/116 (92.2 %) DWI
1.003
1.016–1.049
\0.001
Hemi-ICD NIHSS
0.699 1.167
0.502–0.974 0.975–1.396
0.034 0.092
ICA ? MCA occlusion
3.344
0.828–13.511
0.090
B
Results of a multivariate binary regression analysis with malignant cerebral artery incarction (MMI) as dependent variable OR odds ratio, CI confidence interval, AUC area under curves, ICA internal carotid artery, MCA middle cerebral artery, NIHSS National Institutes of Health Stroke Scale, DWI diffusion weighted imaging, Hemi-ICD hemi-intercaudate distance
improved the prediction of a malignant brain swelling in large MCA stroke significantly. Contrary to a previous study, overall CSF-volume did not turn out to be predictive of a malignant course [9]. In our sample, total brain and CSF volumes did not differ between MMI and non-MMI patients. This finding suggests that at the very early acute stage (i.e., within the first hours of stroke, the window in which our sample was studied) local swelling at the level of the basal ganglia as measured by the hemi-ICD might be more sensitive and strongly related to the development of malignant brain swelling than overall volumes of brain and CSF volume. This could be explained by a compartmentalization of intracranial space where mass effects at different locations carry a different impact. Thus, in this early stage of infarction, the hemiICD may rather reflect local brain edema at the level of the basal ganglia than global cerebral atrophy. As a consequence, we may speculate that a smaller hemi-ICD at this stage identifies patients with a tendency to early swelling at the level of the basal ganglia resulting from both extent and location of their ischemic lesion. Differences in the results of the two studies may also result from the fact that in the study by Minnerup et al., median size of the acute CBV lesion reflecting infarct core was much larger than in our cohort resulting in a more pronounced overall displacement of CSF [9]. Finally, while delineation of the infarct core in acute ischemic stroke by CT is still a controversial issue and not well established as a predictor of MMI, the predictive value of a large acute DWI lesion is beyond doubt [4–6].
Fig. 2 Model A shows the single step decision tree dividing into two groups based on DWI lesion volume (B or [ 87 ml). The classification is shown by the color of the bars (MMI = grey, nonMMI = white). Model B shows the two step decision tree model classifying patients based on DWI lesion volume (B or [ 87 ml) and hemi-ICD (B or [ 9.4 mm)
Although ICD significantly improved the prediction of MMI by DWI, the net increase of predictive power was small with an increase of the number of correct classification by 3.4 %. In a previous study using ICD measurement to assess brain atrophy, an ICD [ 20 mm significantly reduced the risk of development of MMI with an odds ratio of 0.137 [10]. Again, this discrepancy may be attributed to different cohort characteristics. Particularly, in the study by Lee et al., mean infarct volume (224 cm3) was more than twice as large as the acute DWI lesion volume in the group of MMI patients in our study (97 cm3). Patients were also older than in our cohort (mean age 70 vs. 62 years). However, while the overall improvement of prediction of MMI by the inclusion of ICD was rather small, the number of false positive findings, i.e. patients predicted to develop MMI while they did not, was markedly reduced from 7/23 (30 %) to only 1/15 (7 %), which is also reflected by a more pronounced increase of specificity
123
1102
J Neurol (2014) 261:1097–1103
Table 3 Prediction of MMI: predictive values for both models Sensitivity
Specificity
PPV
NPV
Correct classifications
0.76 (0.53–0.92)
0.93 (0.85–0.97)
0.70 (0.47–0.87)
0.95 (0.90–0.99)
104/116 (89.7 %)
0.67 (0.43–0.85)
0.99 (0.94–1.00)
0.93 (0.68–1.00)
0.93 (0.86–0.97)
108/116 (93.1 %)
Model A DWI [ 87 ml Model B DWI [ 87 ml and hemi-ICD B 9.4 mm Values are given as mean (95 % CI) DWI diffusion weighted imaging, Hemi-ICD hemi-intercaudate distance, PPV positive predictive value, NPV negative predictive value
and PPV. In the simple decision tree model combining DWI lesion volume and ICD, specificity was as high as 0.99 and PPV increased from 0.70 to 0.93. From a pathophysiological perspective this may easily be explained as a larger hemi-ICD reflects a larger intracranial volume reserve which may protect patients with large brain infarction from malignant swelling [9]. Adding hemi-ICD to the predictive model based on DWI lesion volume, thus, excludes the subgroup of patients with large initial DWI volumes that is protected from malignant swelling by a larger intracranial volume reserve. From a clinical point of view, this simple algorithm might help identify patients for early decompressive surgery without waiting for clinical or radiological signs of massive space-occupying effects. Our study has limitations. First, the MR images of 24 patients had to be excluded from the analysis due to technical reasons. This represents a substantial proportion of the entire population and may have introduced a bias. Moreover, while the optimal DWI lesion volume predicting MMI identified in this analysis ([87 ml) was quite close to the cut-off identified in the original paper ([82 ml), the predictive values showed more remarkable differences. This points towards the vulnerability of the findings to random changes resulting from the small sample size. Also, measurement of the hemi-ICD was performed by one observer not blinded to DWI lesion volume. Therefore, a systematic bias can not be ruled out. Furthermore, while the predictive value of acute DWI lesion volume was confirmed in a prospective study, the findings from the current study result from retrospective analysis and need confirmation in future trials. To summarize, the inclusion of a simple linear marker of brain atrophy like the ICD, which can be measured at the scanner within seconds, significantly improves the prediction of MMI by DWI lesion volume. While the overall effect is small, there is a relevant increase of PPV and specificity, which might help in clinical decision making, when early decompressive surgery is considered. Acknowledgments The study was supported by the German Kompetenznetzwerk Schlaganfall sponsored by the Bundesministerium fu¨r Bildung und Forschung (B5; No. 01GI9902/4).
123
Conflicts of interest On behalf of all authors, the corresponding author states that there is no conflict of interest.
References 1. Hacke W, Schwab S, Horn M, Spranger M, De Georgia M, von Kummer R (1996) ‘Malignant’ middle cerebral artery territory infarction: clinical course and prognostic signs. Arch Neurol 53(4):309–315 2. Schwab S, Steiner T, Aschoff A, Schwarz S, Steiner HH, Jansen O, Hacke W (1998) Early hemicraniectomy in patients with complete middle cerebral artery infarction. Stroke 29(9): 1888–1893 3. Vahedi K, Hofmeijer J, Juettler E, Vicaut E, George B, Algra A, Amelink GJ, Schmiedeck P, Schwab S, Rothwell PM, Bousser MG, van der Worp HB, Hacke W (2007) Earlydecompressive surgery in malignant infarction of the middle cerebral artery: a pooled analysis of three randomised controlled trials. Lancet Neurol 6(3):215–222. doi:10.1016/S1474-4422(07)70036-4 4. Oppenheim C, Samson Y, Manai R, Lalam T, Vandamme X, Crozier S, Srour A, Cornu P, Dormont D, Rancurel G, Marsault C (2000) Prediction of malignant middle cerebral artery infarction by diffusion-weighted imaging. Stroke 31(9):2175–2181 5. Thomalla GJ, Kucinski T, Schoder V, Fiehler J, Knab R, Zeumer H, Weiller C, Rother J (2003) Prediction of malignant middle cerebral artery infarction by early perfusion- and diffusionweighted magnetic resonance imaging. Stroke 34(8):1892–1899. doi:10.1161/01.STR.0000081985.44625.B6 6. Thomalla G, Hartmann F, Juettler E, Singer OC, Lehnhardt FG, Kohrmann M, Kersten JF, Krutzelmann A, Humpich MC, Sobesky J, Gerloff C, Villringer A, Fiehler J, Neumann-Haefelin T, Schellinger PD, Rother J (2010) Prediction of malignant middle cerebral artery infarction by magnetic resonance imaging within 6 hours of symptom onset: a prospective multicenter observational study. Ann Neurol 68(4):435–445. doi:10.1002/ ana.22125 7. Hofmeijer J, Algra A, Kappelle LJ, van der Worp HB (2008) Predictors of life-threatening brain edema in middle cerebral artery infarction. Cerebrovasc Dis 25(1–2):176–184. doi:10.1159/ 000113736 8. Dohmen C, Galldiks N, Bosche B, Kracht L, Graf R (2012) The severity of ischemia determines and predicts malignant brain edema in patients with large middle cerebral artery infarction. Cerebrovasc Dis 33(1):1–7. doi:10.1159/000330648 9. Minnerup J, Wersching H, Ringelstein EB, Heindel W, Niederstadt T, Schilling M, Schabitz WR, Kemmling A (2011) Prediction of malignant middle cerebral artery infarction using computed tomography-based intracranial volume reserve measurements. Stroke 42(12):3403–3409. doi:10.1161/STROKEAHA. 111.619734
J Neurol (2014) 261:1097–1103 10. Lee SH, Oh CW, Han JH, Kim CY, Kwon OK, Son YJ, Bae HJ, Han MK, Chung YS (2010) The effect of brain atrophy on outcome after a large cerebral infarction. J Neurol Neurosurg Psychiatry 81(12):1316–1321. doi:10.1136/jnnp.2009.197335 11. Butzkueven H, Kolbe SC, Jolley DJ, Brown JY, Cook MJ, van der Mei IA, Groom PS, Carey J, Eckholdt J, Rubio JP, Taylor BV, Mitchell PJ, Egan GF, Kilpatrick TJ (2008) Validation of linear cerebral atrophy markers in multiple sclerosis. J Clin Neurosci 15(2):130–137. doi:10.1016/j.jocn.2007.02.089 12. Brott T, Adams HP Jr, Olinger CP, Marler JR, Barsan WG, Biller J, Spilker J, Holleran R, Eberle R, Hertzberg V et al (1989)
1103 Measurements of acute cerebral infarction: a clinical examination scale. Stroke 20(7):864–870 13. van Swieten JC, Koudstaal PJ, Visser MC, Schouten HJ, van Gijn J (1988) Interobserver agreement for the assessment of handicap in stroke patients. Stroke 19(5):604–607 14. Forkert ND, Saring D, Fiehler J, Illies T, Moller D, Handels H (2009) Automatic brain segmentation in Time-of-Flight MRA images. Methods Inf Med 48(5):399–407. doi:10.3414/ME9237 15. Krieger DW, Demchuk AM, Kasner SE, Jauss M, Hantson L (1999) Early clinical and radiological predictors of fatal brain swelling in ischemic stroke. Stroke 30(2):287–292
123
