Acute stroke: a comparison of different CT perfusion algorithms and validation of ischaemic lesions by follow-up imaging

Eur Radiol (2012) 22:2559 2567 DOI 10.1007/s00330-012-2529-8 NEURO Acute stroke: a comparison of different CT perfusion algorithms and validation of ischaemic lesions by follow-up imaging Benjamin Abels & J. Pablo Villablanca & Bernd F. Tomandl & Michael Uder & Michael M. Lell Received: 26 February 2012 /Revised: 21 April 2012 /Accepted: 6 May 2012 /Published online: 21 June 2012 # European Society of Radiology 2012 Abstract Objectives To compare ischaemic lesions predicted by different CT perfusion (CTP) post-processing techniques and validate CTP lesions compared with final lesion size in stroke patients. Methods Fifty patients underwent CT, CTP and CT angiography. Quantitative values and colour maps were calculated using least mean square deconvolution (LMSD), maximum slope (MS) and conventional singular value decomposition deconvolution (SVDD) algorithms. Quantitative results, core/penumbra lesion sizes and Alberta Stroke Programme Early CT Score (ASPECTS) were compared among the algorithms; lesion sizes and ASPECTS were compared with final lesions on follow-up MRI + MRA or CT + CTA as a reference standard, accounting for recanalisation status. Electronic supplementary material The online version of this article (doi:10.1007/s00330-012-2529-8) contains supplementary material, which is available to authorized users. B. Abels : M. Uder : M. M. Lell Institute of Radiology, University Hospital Erlangen, Erlangen, Germany J. P. Villablanca Department of Neuroradiology, UCLA Medical Center, Los Angeles, CA, USA B. F. Tomandl Department of Neuroradiology, Klinikum Bremen-Mitte, Bremen, Germany B. Abels (*) Diagnostic and Interventional Radiology, University Hospital Heidelberg, Im Neuenheimer Feld 110, 69120, Heidelberg, Germany e-mail: benjamin.abels@med.uni-heidelberg.de Results Differences in quantitative values and lesion sizes were statistically significant, but therapeutic decisions based on ASPECTS and core/penumbra ratios would have been the same in all cases. CTP lesion sizes were highly predictive of final infarct size: Coefficients of determination (R 2 ) for CTP versus follow-up lesion sizes in the recanalisation group were 0.87, 0.82 and 0.61 (P<0.001) for LMSD, MS and SVDD, respectively, and 0.88, 0.87 and 0.76 (P< 0.001), respectively, in the non-recanalisation group. Conclusions Lesions on CT perfusion are highly predictive of final infarct. Different CTP post-processing algorithms usually lead to the same clinical decision, but for assessing lesion size, LMSD and MS appear superior to SVDD. Key Points Following an acute stroke, CT perfusion imaging can help predict lesion evolution. Delay-insensitive deconvolution and maximum slope approach are superior to delay-sensitive deconvolution regarding accuracy. Different CT perfusion post-processing algorithms usually lead to the same clinical decision. CT perfusion offers new insights into the evolution of stroke. Keywords Brain ischemia/diagnosis. Cerebral infarction/ diagnosis. Tomography, X-ray computed/methods. CT perfusion. Multidetector computed tomography Abbreviations ASPECTS Alberta Stroke Programme Early CT Score CBF cerebral blood flow CBV cerebral blood volume CTA CT angiography CTP CT perfusion DC deconvolution

2560 Eur Radiol (2012) 22:2559 2567 DWI FLAIR LMSD MIP MRS MS MTT NIHSS NPV NVT PPV PWI SVDD STARD TAR ToF-MRA TTD TTP Introduction diffusion-weighted imaging fluid attenuation inversion recovery least mean square deconvolution maximum intensity projection modified Rankin scale maximum slope mean transit time National Institutes of Health Stroke Scale negative predictive value non-viable tissue positive predictive value perfusion-weighted imaging singular value decomposition deconvolution Standards for Reporting of Diagnostic Accuracy Studies tissue at risk time-of-flight MR angiography time to drain time to peak Most dedicated stroke centres nowadays rely on diffusion-/ perfusion-weighted MRI (DWI/PWI), or multimodal CT, including unenhanced CT, CT perfusion (CTP) [1] and CT angiography (CTA), for state-of-the-art imaging in the setting of acute ischaemic stroke. Compared with unenhanced CT, these techniques can detect hyperacute ischaemia more reliably [2 4] and may improve risk stratification and selection of candidates for reperfusion therapy by differentiating between penumbra (tissue at risk, TAR) and infarct core (non-viable tissue, NVT). With regard to the results of large thrombolysis trials, the use of new imaging techniques as a surrogate outcome measure in future thrombolytic trials has been recommended [5, 6]. Different approaches have been proposed to distinguish NVT and TAR. On MRI, DWI/PWI mismatch is commonly used. On CTP images, cerebral blood volume/flow (CBV/CBF) mismatch [7] and CBV/ mean transit time (CBV/MTT) [8] mismatch have been suggested. While there is evidence that DWI/PWI mismatch on MRI is suitable for NVT/TAR differentiation [9, 10], the feasibility of CTP for this task has not yet been sufficiently validated [11]. Further, a variety of different CTP postprocessing algorithms are available in commercial software packages, but there is hardly any information about the comparability and accuracy of lesion sizes indicated by individual post-processing tools compared with the final infarct size [11 13]. Studies validating ischaemic lesions predicted by different CTP approaches against final outcome and accounting for therapeutic interventions are lacking. Therefore, the objectives of this study were to: 1. Analyse the comparability of different algorithms in determining quantitative perfusion values and ischaemic lesion sizes using the same CTP data sets and processing settings. 2. Validate TAR and NVT lesions predicted by different CTP algorithms with regard to final lesion size on follow-up imaging as well as clinical outcome. Materials and methods Study design This retrospective study was approved by our institutional review board. Imaging data and clinical information collected in a prospective trial on stroke patients over a 6-month-period, stored in our picture archiving and communication system (PACS) and clinical information system (CIS), were analysed. All patients included in this study presented with symptoms of acute stroke within 6 h of onset (standardised neurological status, including National Institutes of Health Stroke Scale score, NIHSS), underwent unenhanced CT, CTP and CTA on admission (to rule out intracranial haemorrhage, quantify brain perfusion and assess vessel occlusion), as well as multimodal follow-up imaging, consisting of MRI including timeof-flight MRA (ToF-MRA) or unenhanced CT and CTA, and follow-up neurological examination (including modified Rankin scale, MRS). Exclusion criteria were intracranial haemorrhage on admission unenhanced CT, non-diagnostic admission or follow-up imaging study (strong motion artefacts), and incomplete neurological assessment. Intravenous thrombolysis was indicated on the basis of clinical symptoms and imaging results in agreement with current guidelines [14, 15]. CTP imaging and data processing Data were acquired on 16-multi-slice CT (Siemens Healthcare, Erlangen, Germany) with a standard protocol (80 kvp, 250 mas, 4 s delay, 40 s imaging time, 1 s temporal resolution). Two 10-mm sections were imaged, one at the level of the basal ganglia and one contiguous upper section. Perfusion parameters and colour maps were calculated using two software packages: VPCT Neuro (Version 2008-B, Siemens Healthcare, Erlangen, Germany) implementing LMSD and MS, and Vitrea2 CT Brain Perfusion (Version 4.1, Vital Images, Plymouth, MN, USA) implementing conventional SVDD. Details of these post-processing algorithms have been described previously [16 18]. Data processing was performed in a standardised way [16] using default settings as employed in clinical practice. In all cases, perfusion parameters were calculated using LMSD, MS and SVDD, and displayed as colour maps.

Eur Radiol (2012) 22:2559 2567 2561 Quantitative perfusion measurements For the quantitative comparison, two regions of interest were drawn manually in each case: (1) healthy brain tissue (cortical grey matter in the non-affected hemisphere), and (2) ischaemic brain tissue (basal ganglia and cortex with reduced relative CBF<0.60 and TTP delay>2 s). CTP lesion sizes The NVT and TAR lesion sizes on CTP images were obtained from CBV and CBF maps, respectively, by employing predefined thresholds as discussed by Abels et al. [16]: NVT, relative CBV<0.40, corresponding to an absolute CBV threshold of ca. 1.2 ml/100 g; TAR, relative CBF<0.60, corresponding to an absolute CBF LMSD and CBF SVDD threshold of ca. 27 ml/100 g/min, and CBF MS threshold of ca. 34 ml/100 g/min. Lesion sizes were each measured in the section showing the maximum lesion extent. Additionally, an ischaemic lesion scoring system accounting for ten regions in the middle cerebral artery (MCA) territory, adopted from the Alberta Stroke Programme Early CT Score (ASPECTS) [19], as well as a frontal region (anterior cerebral artery territory), thalamus, and occipital lobe (posterior cerebral artery territory) was used to account for three-dimensional lesion extent. Modified NVT and TAR ASPECTS were determined from CBV and CBF maps, respectively. CTP lesion validation Final lesion sizes and final ASPECTS were determined on follow-up MR imaging. An infarct was defined as DWI/ADC restricted tissue that was also hyperintense on T2W and/or fluid attenuation inversion recovery (FLAIR) sequences with or without mass effect. If no follow-up MRI was available or contraindicated, infarct on follow-up unenhanced CT imaging was defined as parenchymal hypodensity with loss of grey white matter differentiation, sulcal effacement and/or mass effect. Recanalisation status of the primary arterial occlusive lesion (AOL) was determined using follow-up TOF-MRA or CTA. Three patient groups were identified: group A, patients with ischaemic lesion on CTP, and partial (AOL II) or full (AOL III) recanalisation on follow-up imaging; group B, patients with ischaemic lesion on CTP, and no recanalisation (AOL 0) or incomplete recanalisation with no distal flow (AOL I) on follow-up; group C, patients with no ischaemic lesion defined as relative CBV<0.40 or relative CBF<0.60 on CTP colour maps. For group A, we hypothesised that NVT on CTP might be infarcted on follow-up imaging, whereas TAR would recover. For group B, we hypothesised that both NVT and TAR might be infarcted on follow-up imaging. For group C, we hypothesised that there would be no infarct on followup imaging. CTP and follow-up studies were read in consensus by one author with more than 3 years and one author with more than 10 years of experience in stroke imaging who were blinded to the clinical information. CTP and follow-up studies were read with an interim time interval of more than 6 months. Comparison of ASPECTS as well as NVT/TAR ratio To investigate whether different algorithms lead to the same therapeutic decision, ASPECTS as well as NVT/TAR ratios were compared among the algorithms. Also, correlation of NVT/TAR ratios, recanalisation rates and clinical outcome (MRS) was analysed. Favourable and unfavourable NVT/ TAR ratios were defined as no greater than 0.50 and greater than 0.50, respectively. Good and bad clinical outcomes were defined as MRS score no greater than 2 and greater than 2, respectively. Statistical analysis Quantitative values obtained with LMSD, MS and SVDD were compared using descriptive statistics. Ischaemic lesion sizes obtained with LMSD, MS, SVDD and follow-up imaging were compared using Pearson s correlation coefficient and linear regression goodness-of-fit analysis. CTP and follow-up ASPECTS were compared using Spearman s rank correlation. The Friedman test was used to test for statistically significant differences of lesion sizes and ASPECTS among algorithms and follow-up. If a statistically significant difference was found, the Wilcoxon signed rank test for paired samples was used in addition. Threshold for significance was set at P<0.05. Group C was not included in the statistical analysis of lesion sizes and ASPECTS because with no lesion present, no lesion could be correlated with follow-up; however, sensitivity/specificity was analysed including all cases from all groups. Correlation of NVT/TAR ratios, recanalisation rates and clinical outcome (MRS) was determined using Spearman s rank correlation. Statistical analysis was performed with SPSS 15.0 (SPSS Inc., Chicago, IL, USA). Results Fifty patients (31 men, 19 women), mean age 70.9±12 years, admitted from October 2007 to April 2008 were included in this study. A flow chart is presented in Fig. 1. Median NIHSS was 10 (interquartile range [IQR] 6 15), median MRS, 3 (IQR 1 4), median time to follow-up, 1 day (IQR 1 2). The average effective radiation dose of unenhanced CT, CTP and CTA was 1.5 msv, 3.2 msv and 3.0 msv, respectively, and 7.7 msv in total. An overview of the patient groups and treatment is given in Table 1.

2562 Eur Radiol (2012) 22:2559 2567 Table 1 Patient groups and therapeutic options Thrombolysis Controls Totals Group A 14 2 16 (32 %) Group B 8 11 19 (38 %) Group C 3 12 15 (30 %) Totals 25 (50 %) 25 (50 %) 50 (100 %) Group A patients with CT perfusion (CTP) lesion + recanalisation, group B patients with CTP lesion + no recanalisation, group C patients without ischaemic lesion on CTP Comparison of lesion sizes Fig. 1 Standards for Reporting of Diagnostic Accuracy Studies (STARD) flow chart Sensitivity of CTP was 83 %; specificity, 100 %; negative predictive value (NPV), 47 %; and positive predictive value (PPV), 100 %, regardless of the algorithm used. Of the false negative cases, five lesions were located outside the imaging volume (not covered by the CTP sections). When correcting for lesions outside of the imaging volume, sensitivity rises to 92 %; specificity, 100 %; NPV, 80 %; PPV, 100 %. Detailed results can be found in Online Table 1. Suspected causes of stroke symptoms were cardiogenic embolism (n022), arterio-arterial embolism (n015), carotid dissection (n02), basilar thrombosis (n01), microangiopathic (n02), TIA (n05), Todd s paresis (n02), migraine with aura (n01). Sites of occlusion were ICA (n04), MCA (n030), PCA (n06), pontine arteries (n01), cerebellar arteries and MCA (n01), MCA and PCA (n01), none (n07). Carotid stenosis greater than 70 % was present in 8 cases and affected time attenuation curve and perfusion quantification in 3 cases. Comparison of quantitative perfusion measurements Perfusion values obtained with the different algorithms revealed statistically significant differences (Fig. 2). However, CBF and CBV values were within a similar range, and average values were highly comparable. The distribution of CBF and CBV values obtained with SVDD was markedly wider when compared with LMSD and MS. In ischaemic tissue, CBF values obtained with MS were higher than those obtained with LMSD and SVDD. Detailed results of the quantitative analysis are presented in the Online Appendix and Online Table 2. Comparing lesion sizes obtained with different algorithms, we found that NVT sizes obtained with SVDD were smaller than those obtained with LMSD and MS (P00.005 and P0 0.002). Differences in TAR sizes among LMSD, MS and SVDD were not statistically significant. Differences in NVT as well as TAR ASPECTS were not statistically significant among the algorithms (Friedman test, P00.422 and P0 0.504, respectively). Correlations of ASPECTS are presented in Online Tables 3 and 4. Comparison of CTP and follow-up lesions Comparing NVT lesion sizes obtained with LMSD, MS and SVDD with follow-up lesion sizes in group A, we found excellent correlations (Table 2). Linear regression goodnessof-fit analyses demonstrated that NVT LMSD was most predictive of final infarct size (R 2 00.87, P<0.001), but NVT MS and NVT SVDD were also well predictive (R 2 00.82 and R 2 00.61; both P<0.001). NVT LMSD and NVT MS lesions tended to be slightly smaller than final lesions (Fig. 3), but differences did not reach statistical significance. NVT SVDD lesions were significantly smaller than NVT LMSD,NVT MS and follow-up lesions (all P<0.01). Statistical analysis of ASPECTS revealed no significant difference for lesions predicted by LMSD, MS, SVDD and final lesions on follow-up (Friedman test, P0 0.254). Correlations between CTP and follow-up ASPECTS in group A were r00.86 (P<0.01), r00.77 (P<0.01) and r0 0.58 (P00.02) for LMSD, MS and SVDD, respectively. Comparing TAR lesion sizes obtained with LMSD, MS and SVDD with final lesion sizes in group B, we found excellent correlations (Table 3). Linear regression goodnessof-fit analyses demonstrated that TAR LMSD was most predictive of final infarct size (R 2 00.88, P<0.001), but TAR MS and TAR SVDD were also highly predictive (R 2 00.87 and R 2 00.76; both P<0.001). TAR LMSD and TAR MS lesions tended to be slightly smaller than follow-up (P00.008 and P00.011), whereas no systematic difference was found for TAR SVDD and follow-up (Fig. 4). Statistical analysis of

Eur Radiol (2012) 22:2559 2567 2563 Fig. 2 Box plots of quantitative cerebral blood volume (CBV) and cerebral blood flow (CBF) perfusion measurements in healthy brain tissue (top) and ischaemic regions (bottom) obtained with three different CT perfusion (CTP) algorithms (least mean square deconvolution [LMSD], maximum slope [MS], singular value decomposition deconvolution [SVDD]) modified ASPECTS revealed no significant difference in lesions predicted by LMSD, MS, SVDD and final lesions on follow-up imaging (Friedman test, P00.729). Correlations between CTP and follow-up ASPECTS in group B were r00.90, r00.89 and r00.95 (all P<0.01) for LMSD, MS and SVDD, respectively. Comparison of NVT/TAR ratios, recanalisation rate and clinical outcome Correlation between the NVT/TAR ratio on CTP images and follow-up MRS scores was r00.55 (P<0.001), r00.46 (P< Table 2 Pearson correlation of infarct size (NVT) on CTP and followup images in patients with successful recanalisation (n016) Lesion size Follow-up LMSD MS SVDD Follow-Up 1.00 0.94** 0.91** 0.78** LMSD 0.94** 1.00 0.90** 0.70** MS 0.91** 0.90** 1.00 0.61* SVDD 0.78** 0.70** 0.61* 1.00 LMSD least mean square deconvolution, MS maximum slope, NVT non-viable tissue, SVDD singular value decomposition deconvolution **Correlation is significant at the 0.01 level (two-tailed) *Correlation is significant at the 0.05 level (two-tailed) 0.005) and r00.36 (P<0.035) for LMSD, MS and SVDD, respectively. Favourable/unfavourable NVT/TAR ratio correlated well with good/bad clinical outcome (LMSD, r00.52, P< 0.002; MS, r00.53, P<0.01; SVDD, r00.45, P<0.007). Fifty per cent of the patients received thrombolytic therapy (n025) and 50 % received supportive treatment (n025). Clinical outcome (median MRS scores) in patients who had received recombinant tissue plasminogen activator (rtpa) tended to be better than in the control group (median 3 [IQR 1 4] versus 5 [IQR 3 5]), but differences did not reach statistical significance. However, clinical outcome was significantly better in patients who showed a favourable NVT/TAR ratio on CTP compared with those with an unfavourable NVT/TAR ratio. In patients with favourable mismatch, follow-up MRS scores were significantly better than in patients with an NVT/TAR ratio greater than 0.50 (median 2 [IQR 1 3] versus 5 [IQR 4 6], P00.0011). Recanalisation rate was significantly higher in patients with a favourable NVT/TAR ratio less than 0.50 compared with those with a ratio greater than 0.50 (64 % versus 15 %, P<0.0185). Discussion As evidence of the comparability of different CTP algorithms regarding ischaemic lesion accuracy and diagnostic

2564 Eur Radiol (2012) 22:2559 2567 Fig. 3 Recanalisation group (group A). Top panel correlations of infarct size on CTP images and final infarct size on follow-up imaging (NVT predicted by LMSD, MS and SVDD, respectively). Bottom panel 69-year-old man with left middle cerebral artery (MCA) occlusion. Lentiform nucleus as well as portions of the caudate head and the insular cortex are NVT on CTP maps, whereas M2 and parts of M1 and M3 are tissue at risk (TAR), as per nomenclature adopted from ASPECTS [19]. The NVT lesion size corresponds well with the final infarct size after thrombolysis on MRI (FLAIR) value is lacking [11], this study aimed to compare and validate ischaemic lesions as predicted by different CTP approaches. Using three common post-processing algorithms, preset thresholds, and recanalisation status, we Table 3 Pearson correlation of infarct size (NVT + TAR) on CTP and follow-up images in patients without successful recanalisation (n019) Lesion size Follow-up LMSD MS SVDD Follow-up 1.00 0.95** 0.94** 0.89** LMSD 0.95** 1.00 0.98** 0.87** MS 0.94** 0.98** 1.00 0.83** SVDD 0.89** 0.87** 0.83** 1.00 LMSD least mean square deconvolution, MS maximum slope, NVT non-viable tissue, SVDD singular value decomposition deconvolution, TAR tissue at risk **Correlation is significant at the 0.01 level (two-tailed) *Correlation is significant at the 0.05 level (two-tailed) found that different CTP approaches yield significantly different quantitative values and some variations of ischaemic lesion sizes, but overall agreement of ASPECTS and core/ penumbra ratios among algorithms as well as agreement of CTP and final lesions was very good. In previous studies, significant differences in quantitative CTP measurements have been reported. These differences may be attributed to a number of variables, such as contrast bolus distribution and imaging protocol (tube voltage, current, temporal resolution), variations in data processing (arterial input function, output vein selection, tissue segmentation) [13, 20], but also to different algorithms. Kudo et al. found substantial discrepancy in quantitative measurements as well as ischaemic lesions among different algorithms [13]. In our study, we used identical source data and standardised processing settings to minimise confounders. Nevertheless, significant quantitative differences were observed among different algorithms, which confirmed the

Eur Radiol (2012) 22:2559 2567 2565 Fig. 4 Non-recanalisation group (group B). Top panel correlations of infarct size on CTP images and final infarct size on follow-up imaging (NVT + TAR predicted by LMSD, MS and SVDD, respectively). Bottom panel 74-year-old man with right MCA occlusion. On CTP maps, insular ribbon, M2, M5 and M6 are NVT, whereas M1, M3 and findings of previous studies. However, we found that perfusion values were within a comparable range, and our findings indicate that quantitative variations do not necessarily result in significantly different ischaemic lesion extents or ASPECTS. Average perfusion values found in our study were in accordance with those reported in the literature [7, 8, 16, 21]. With regard to lesion size accuracy, Wintermark et al. [22] found that CTP and DWI/PWI MRI were equivalent in a small sample size (n013). Further, good agreement between CTP and MRI lesions has been demonstrated [23, 24]. In another study, Wintermark et al. determined CBV and MTT thresholds for core/penumbra differentiation (2 ml/100 g and 145 %, respectively), employing a deconvolution algorithm developed by Philips Medical Systems [8]. However, Dani et al. [25]demonstrated that adapted thresholds may apply depending on the type of algorithm. Comparing different algorithms, Kudo et al. found that delay-insensitive approaches were superior to M4 portions of the right MCA territory are TAR as per nomenclature adopted from ASPECTS [19]. The basal ganglia are spared. Thrombolysis was not indicated. On the follow-up MRI study, not only NVT but also TAR is infarcted delay-sensitive ones [13], prompting ASIST-Japan to recommend the use of delay-insensitive techniques in clinical routine work. The findings of our study are basically in agreement with those made in previous studies, but they add to the current knowledge by investigating different CTP algorithms in a larger sample size and, in particular, accounting for thrombolysis outcome and recanalisation status. Also, our study adds to the literature, because to our knowledge it is the first to validate different algorithms using preset thresholds for NVT/TAR definition on CTP images and demonstrate excellent correlations of CTP lesions with final lesions on follow-up imaging. Given the superior agreement of CTP lesion sizes obtained with LMSD and MS compared with SVDD, our findings support the suggestion of the Acute Stroke Imaging Research Roadmap [12] and ASIST-Japan that delay-insensitive deconvolution and maximum slope algorithms might be preferable to delay-sensitive

2566 Eur Radiol (2012) 22:2559 2567 deconvolution. However, considering that differences in ASPECTS and NVT/TAR ratios among the algorithms did not reach statistical significance and clinical decisions would have been the same in all cases in this study regardless of the algorithm used, there is no evidence that any of the algorithms results in different clinical decision making and, even less so, clinical outcome. In terms of correlation of CTP findings and clinical outcome, significant correlation of NVT/TAR ratio and MRS scores was demonstrated in our study, using LMSD, MS and SVDD. This confirms the previous findings of Kloska et al. that CTP is a feasible means of predicting clinical outcome [26] and indicates that CBV/CBF mismatch can be used for advanced risk stratification (prediction of clinical outcome) before thrombolytic therapy, regardless of the algorithm used. A relative CBV threshold of 0.40 0.50 (corresponding to an absolute CBV no greater than 1.2 2.0 ml/100 g) may be used for guidance to identify NVT, that is, infarct core when recanalisation is successful, and a preset relative CBF threshold of 0.50 0.60 (corresponding to an absolute CBF no greater than 27 35 ml/100 g/min) may be used for guidance to identify TAR, or lesion extent in case recanalisation is not successful. However, it must be considered that different pathophysiological parameters such as fever and hypotension may affect TAR/NVT development and the results provided in this study. Given the complexities of neurovascular uncoupling and the diversity of mismatch patterns during ischaemia, one must always bear in mind the limitations and pitfalls of quantitative functional imaging when it comes to making therapeutic decisions. Limitations In this context, one major limitation of this study is that the exact time of recanalisation could not be determined. If recanalisation took place at a delayed time, final lesion size may be larger than predicted by NVT on CTP images in group A. This is one possible explanation for NVT lesions being slightly smaller than final infarct sizes in our study. This limitation could not be avoided given the retrospective study design. This study had several other limitations. First, only three algorithms were analysed; however, there are a number of other algorithms available in commercial software packages that have not been compared. Thus, the generalisability of our findings is restricted and investigation of additional algorithms is necessary. The use of preset absolute thresholds in this study can be considered problematic in so far as quantitative perfusion measurements retrieved from different algorithms may vary significantly and adapted thresholds might apply for each individual algorithm [16, 25]. We accounted for this by using adapted thresholds for LMSD and MS, but no adapted thresholds have been reported for SVDD, and different thresholds may apply for other algorithms. Further, differences found may not be attributed to the post-processing algorithms alone, but may also be due to technical restrictions such as input and output function detection and vascular pixel segmentation by different commercial software packages. Another limitation of this study is the relatively small sample size. Although 50 patients were included in this study, 15 CTP images were negative for cerebral infarction (e.g. TIA, Todd s paresis, migraine with aura), so final lesions could be validated in only 35 patients (recanalisation group, n016; non-recanalisation group, n019). Nevertheless, formal sample size analysis demonstrated that the sample size of this study is sufficient for stratified analysis, with a hypothesised correlation greater than 0.66, alpha set at 0.05 and beta set at 0.20. Nevertheless, future studies with larger sample sizes are desirable. Also, to reduce a possible bias that may be caused by using two different imaging modalities as a reference standard in this study (39 patients [78 %] underwent follow-up MRI + MRA; 11 patients [22 %] underwent follow-up unenhanced CT + CTA), future studies should use only one reference standard. Finally, spatial coverage of 2 10 mm can be considered a limitation. However, the main intention of this study was not to investigate the sensitivity of CTP, but rather to analyse the comparability of different CTP techniques, which does not necessarily require multiple brain sections. Of note, most hospitals are not yet equipped with 128-MSCT but only 16- or 64-MSCT systems, so this study is actually representative of today s standards. In future studies, limited coverage could be addressed by using state-of-the-art CT systems that allow whole-brain perfusion [27]. To identify the infarct core and penumbra, this study relies on CBV/CBF mismatch. However, there are other approaches, such as mismatch of CBV and mean transit time (MTT), time to peak (TTP 0 Tmax) or time to drain (TTD). In particular, CBV/MTT mismatch may allow for better grey white matter differentiation than CBV/CBF mismatch [8], and may be of similar diagnostic value. We decided on CBV/CBF mismatch because it is applicable to all kinds of algorithms, whereas CBV/MTT mismatch can only be used with deconvolution algorithms and is less suitable for an inter-vendor comparison. Moreover, CBF is more specific for stroke than MTT [4]. For clinical practice, we suggest that MTT, TTP or TTD be used initially to detect the presence of ischaemia with high sensitivity, and, as a second step, CBV/CBF or CBV/MTT mismatch might be used for NVT/TAR differentiation. The question which, if any, CTP mismatch approach is really qualified as a biomarker [28] remains controversial. In conclusion, this study indicates that different CTP post-processing techniques are feasible for core/penumbra differentiation and prediction of final tissue outcome. The use of different algorithms does not usually affect clinical

Eur Radiol (2012) 22:2559 2567 2567 decision making, but as delay-insensitive LMSD and MS demonstrate better agreement with final lesion size compared with delay-sensitive SVDD, the former approaches might be preferable. References 1. Miles KA, Hayball M, Dixon AK (1991) Colour perfusion imaging: a new application of computed tomography. Lancet 337:643 645 2. Hopyan J, Ciarallo A, Dowlatshahi D et al (2010) Certainty of stroke diagnosis: incremental benefit with CT perfusion over noncontrast CT and CT angiography. Radiology 255:142 153 3. Kloska SP, Nabavi DG, Gaus C et al (2004) Acute stroke assessment with CT: do we need multimodal evaluation? Radiology 233:79 86 4. Wintermark M, Fischbein NJ, Smith WS, Ko NU, Quist M, Dillon WP (2005) Accuracy of dynamic perfusion CT with deconvolution in detecting acute hemispheric stroke. AJNR Am J Neuroradiol 26:104 112 5. Roberts HC, Dillon WP, Furlan AJ et al (2002) Computed tomographic findings in patients undergoing intra-arterial thrombolysis for acute ischemic stroke due to middle cerebral artery occlusion: results from the PROACT II trial. Stroke 33:1557 1565 6. Higashida RT, Furlan AJ, Roberts H et al (2003) Trial design and reporting standards for intra-arterial cerebral thrombolysis for acute ischemic stroke. Stroke 34:e109 e137 7. König M, Kraus M, Theek C, Klotz E, Gehlen W, Heuser L (2001) Quantitative assessment of the ischemic brain by means of perfusion-related parameters derived from perfusion CT. Stroke 32:431 437 8. Wintermark M, Flanders AE, Velthuis B et al (2006) Perfusion-CT assessment of infarct core and penumbra: receiver operating characteristic curve analysis in 130 patients suspected of acute hemispheric stroke. Stroke 37:979 985 9. Shih LC, Saver JL, Alger JR et al (2003) Perfusion-weighted magnetic resonance imaging thresholds identifying core, irreversibly infarcted tissue. Stroke 34:1425 1430 10. Kluytmans M, van Everdingen KJ, Kappelle LJ, Ramos LM, Viergever MA, van der Grond J (2000) Prognostic value of perfusion- and diffusion-weighted MR imaging in first 3 days of stroke. Eur Radiol 10:1434 1441 11. Wintermark M, Rowley HA, Lev MH (2009) Acute stroke triage to intravenous thrombolysis and other therapies with advanced CT or MR imaging: pro CT. Radiology 251:619 626 12. Wintermark M, Albers GW, Alexandrov AVet al (2008) Acute stroke imaging research roadmap. AJNR Am J Neuroradiol 29:e23 e30 13. Kudo K, Sasaki M, Yamada K et al (2010) Differences in CT perfusion maps generated by different commercial software: quantitative analysis by using identical source data of acute stroke patients. Radiology 254:200 209 14. Adams HP Jr, del Zoppo G, Alberts MJ et al (2007) Guidelines for the early management of adults with ischemic stroke: a guideline from the American Heart Association/American Stroke Association Stroke Council, Clinical Cardiology Council, Cardiovascular Radiology and Intervention Council, and the Atherosclerotic Peripheral Vascular Disease and Quality of Care Outcomes in Research Interdisciplinary Working Groups: The American Academy of Neurology affirms the value of this guideline as an educational tool for neurologists. Circulation 115:e478 e534 15. Del Zoppo GJ, Saver JL, Jauch EC, Adams HP Jr (2009) Expansion of the time window for treatment of acute ischemic stroke with intravenous tissue plasminogen activator: a science advisory from the American Heart Association/American Stroke Association. Stroke 40:2945 2948 16. Abels B, Klotz E, Tomandl BF, Kloska SP, Lell MM (2010) Perfusion CT in acute ischemic stroke: a qualitative and quantitative comparison of deconvolution and maximum slope approach. Am J Neuroradiol 31:1690 1698 17. Klotz E, Konig M (1999) Perfusion measurements of the brain: using dynamic CT for the quantitative assessment of cerebral ischemia in acute stroke. Eur J Radiol 30:170 184 18. Østergaard L, Weisskoff RM, Chesler DA, Gyldensted C, Rosen BR (1996) High resolution measurement of cerebral blood flow using intravascular tracer bolus passages. Part I: Mathematical approach and statistical analysis. Magn Reson Med 36:715 725 19. Pexman JH, Barber PA, Hill MD et al (2001) Use of the Alberta Stroke Program Early CT Score (ASPECTS) for assessing CT scans in patients with acute stroke. AJNR Am J Neuroradiol 22:1534 1542 20. Ferreira RM, Lev MH, Goldmakher GV et al (2010) Arterial input function placement for accurate CT perfusion map construction in acute stroke. Am J Roentgenol 194:1330 1336 21. Eastwood JD, Lev MH, Azhari T et al (2002) CT perfusion scanning with deconvolution analysis: pilot study in patients with acute middle cerebral artery stroke. Radiology 222:227 236 22. Wintermark M, Reichhart M, Thiran JP et al (2002) Prognostic accuracy of cerebral blood flow measurement by perfusion computed tomography, at the time of emergency room admission, in acute stroke patients. Ann Neurol 51:417 432 23. Schramm P, Schellinger PD, Klotz E et al (2004) Comparison of perfusion computed tomography and computed tomography angiography source images with perfusion-weighted imaging and diffusion-weighted imaging in patients with acute stroke of less than 6 hours duration. Stroke 35:1652 1658 24. Schaefer PW, Barak ER, Kamalian S et al (2008) Quantitative assessment of core/penumbra mismatch in acute stroke: CT and MR perfusion imaging are strongly correlated when sufficient brain volume is imaged. Stroke 39:2986 2992 25. Dani KA, Thomas RG, Chappell FM et al (2011) Computed tomography and magnetic resonance perfusion imaging in ischemic stroke: definitions and thresholds. Ann Neurol 70:384 401 26. Kloska SP, Dittrich R, Fischer T et al (2007) Perfusion CT in acute stroke: prediction of vessel recanalization and clinical outcome in intravenous thrombolytic therapy. Eur Radiol 17:2491 2498 27. Dorn F, Muenzel D, Meier R, Poppert H, Rummeny EJ, Huber A (2011) Brain perfusion CT for acute stroke using a 256-slice CT: improvement of diagnostic information by large volume coverage. Eur Radiol 21:1803 1810 28. Buckler AJ, Bresolin L, Dunnick NR et al (2011) Quantitative imaging test approval and biomarker qualification: interrelated but distinct activities. Radiology 259:875 884