Time-Dependent Computed Tomographic Perfusion Thresholds for Patients With Acute Ischemic Stroke
Background and Purpose—Among patients with acute ischemic stroke, we determine computed tomographic perfusion (CTP) thresholds associated with follow-up infarction at different stroke onset-to-CTP and CTP-to-reperfusion times.
Methods—Acute ischemic stroke patients with occlusion on computed tomographic angiography were acutely imaged with CTP. Noncontrast computed tomography and magnectic resonance diffusion–weighted imaging between 24 and 48 hours were used to delineate follow-up infarction. Reperfusion was assessed on conventional angiogram or 4-hour repeat computed tomographic angiography. Tmax, cerebral blood flow, and cerebral blood volume derived from delay-insensitive CTP postprocessing were analyzed using receiver–operator characteristic curves to derive optimal thresholds for combined patient data (pooled analysis) and individual patients (patient-level analysis) based on time from stroke onset-to-CTP and CTP-to-reperfusion. One-way ANOVA and locally weighted scatterplot smoothing regression was used to test whether the derived optimal CTP thresholds were different by time.
Results—One hundred and thirty-two patients were included. Tmax thresholds of >16.2 and >15.8 s and absolute cerebral blood flow thresholds of <8.9 and <7.4 mL·min−1·100 g−1 were associated with infarct if reperfused <90 min from CTP with onset <180 min. The discriminative ability of cerebral blood volume was modest. No statistically significant relationship was noted between stroke onset-to-CTP time and the optimal CTP thresholds for all parameters based on discrete or continuous time analysis (P>0.05). A statistically significant relationship existed between CTP-to-reperfusion time and the optimal thresholds for cerebral blood flow (P<0.001; r=0.59 and 0.77 for gray and white matter, respectively) and Tmax (P<0.001; r=−0.68 and −0.60 for gray and white matter, respectively) parameters.
Conclusions—Optimal CTP thresholds associated with follow-up infarction depend on time from imaging to reperfusion.
Progression to infarction after acute ischemic stroke onset is time-sensitive and has substantial intersubject variability.1,2 Computed tomographic (CT) perfusion (CTP) measurement of brain parenchyma can be used to estimate ischemic core and penumbra and, therefore, provide immediate information for treatment decision-making. Current CTP thresholds that estimate these tissue states are generally derived either by comparison with magnetic resonance (MR) diffusion–weighted imaging (DWI), often done within an hour of CTP, or with follow-up infarction in patients who have reperfused sometime within 24 hours.3–11 Because infarcts grow over time and final tissue fate depends greatly on what happens in the minutes to hours immediately after this imaging snapshot, CTP thresholds predicting infarction are likely to depend on the time from stroke symptom onset to imaging, time from imaging to reperfusion, and the quality of reperfusion.
Five recent trials have showed efficacy of endovascular treatment in patients with intracranial occlusions.12–16 Because of the recent availability of rapid and efficient endovascular treatment techniques, CTP thresholds used to predict irreversibly injured brain tissue may require revision, with attention to reperfusion quality and reperfusion time. Using a group of patients undergoing current endovascular treatment, we sought to determine CTP thresholds associated with follow-up infarction confirmed at 24 to 48 hours in patients achieving early, quality reperfusion (<90 min from CTP), in patients with reperfusion within 90 to 180 min from CTP, and in patients not achieving reperfusion acutely (ie, within 4 hours of CTP).
Data are from PRove-IT—a prospective, multicenter cohort study with CTP imaging17 acute ischemic stroke patients, were included in the study if they presented within 12 hours from last seen normal. Inclusion criteria for the present study were as follows: (1) age >18 years; (2) known symptom onset time; (3) complete anterior circulation occlusion at admission or any complete posterior cerebral artery occlusion; (4) had recanalization assessed on conventional angiography at end of the endovascular treatment (modified thrombolysis in cerebral infarction [mTICI])12 or on follow-up computed tomographic angiography (CTA) within 4 hours of admission imaging (patients who recanalized on follow-up CTA [modified arterial occlusive lesion score 2–3]13 were excluded from this analysis because recanalization time could not be inferred from this cohort); and (5) had 24-hour follow-up imaging on MR DWI or noncontrast CT (NCCT). Demographic and clinical characteristics, medical history, physical examination, relevant workflow, and interval times were collected prospectively. The local ethics boards approved the study.
Imaging Protocol and Analysis
CT imaging was conducted on 64–slice CT scanners (Lightspeed; General Electric Healthcare, Waukesha, WI). At admission, all patients had an NCCT scan, head/neck multiphase CTA, and CTP. For the CTP protocol, 45 mL CT contrast agent (Optiray 320; Mallinckrodt Pharmaceuticals, Dublin, Ireland) was power-injected at 4.5 mL/s followed by a saline chase of 40 mL at 6 mL/s. Sections of 8 cm (n=103) and 4 cm (n=29) thickness were acquired at 5 mm slice thickness. Scanning began after a delay of 5 s from contrast injection in ≤2 phases (scanning intervals): first phase every 2.8 s for 60 s (in 40 patients) and an additional second phase every 15 s for 90 s (in 92 patients). Between 24 and 48 hours, an NCCT or whole brain DWI scan was acquired for final infarct delineation in all patients.
An expert (Dr d’Esterre, 9 years of experience) processed each study using commercially available delay-insensitive deconvolution software (CT Perfusion 4D; General Electric Healthcare, Waukesha, WI). For each study, the arterial input function was manually selected from the basilar artery or contralateral ICA using a 2 voxel×2 voxel (in-slice) region of interest (ROI). For all arterial input functions, baseline to peak height Hounsfield unit (HU) differences matched those from respective sagittal sinus. Absolute maps of cerebral blood flow (CBF; mL·min−1·100 g−1), cerebral blood volume (CBV; mL·100 g−1), mean transit time (seconds), start time of the impulse residue function (ie, delay of the tissue time–density curve with respect to the arterial input function; To; s), and Tmax=To+0.5×area under the impulse residue function (s) were calculated by deconvolution of tissue time–density curves and the arterial input function using a delay-insensitive algorithm (CT Perfusion 4D, GE Healthcare). (See Section I in the online-only Data Supplement for details on the Tmax calculation.) Average maps were created by averaging the serial (dynamic) CTP images over the duration of the first pass of contrast—these average maps have excellent anatomic detail and are used for gray/white matter (GM/WM) segmentation and as the source image for coregistration with follow-up imaging. In-plane patient motion was corrected in the x/y-axis using automated software (CT Perfusion 4D), and in cases with extreme motion, time points were manually removed as needed.17
Follow-up imaging (DWI n=77, NCCT n=55) and admission CTP average maps were coregistered using in-house software (Calgary Image Analysis and Processing Center, Calgary, AB) before being analyzed using custom software (IDL, version 6.3; RSI, Boulder, CO) using 2 different target references. For MR follow-up data, the DWI target images were resampled to match the matrix size, slice thickness, and geometry of the admission CTP average maps (Figure I in the online-only Data Supplement). For CT follow-up data, the NCCT and admission CTP average maps were resampled to match the slice geometry of the admission CTA (0.625-mm slice thickness) acquisition. Both MR and CT coregistrations used the mutual information image similarity metric and were performed via a multiresolution pyramid,18 starting with target images that have been blurred to 4 times their original pixel spacing, proceeding to images that have been blurred to twice their original spacing, and finishing with unblurred, full-resolution images.19 In addition, registration parameters for the admission CTP average maps were applied to the admission CTP functional maps to register them with the follow-up MR or CT imaging data.
Two experts (Drs Ahn and Minhas) manually delineated the follow-up infarct region (ROI-1) on DWI or NCCT (when MR was unavailable) acquired between 24 to 48 hours after admission using custom software (IDL, version 6.3). Briefly, a standard window-level set-up was used to segment infarct ROIs on the coregistered follow-up MR or NCCT images. These segmentations were superimposed onto coregistered admission CTP functional maps. HU thresholds of CTP average maps were used to exclude cerebrospinal fluid and skull (<25–30 and >300 HU, respectively) manually and to separate GM from WM (35–45 HU) within the segmented infarct to arrive at ROI-1 for GM and WM separately. Regions of chronic/old infarct and evident leukoaraiosis confirmed on admission NCCT and follow-up FLAIR or NCCT (ROI-2) were removed from primary analysis. ROI-3 was defined as any ipsilateral hemisphere brain tissue outside of ROI-1 and ROI-2. ROI-1 was mirrored onto the contralateral hemisphere at the brain midline (ROI-4), excluding chronic/old infarct and leukoaraiosis. For ROI-1 and ROI-3 individually, histograms (with 100 bins each) were created for each CTP parameter (CBF, CBV, Tmax). Two experts (Drs Ahn and Minhas) used the mTICI score at end of intra-arterial therapy to assess efficient reperfusion (mTICI 2b/3). The same experts assessed recanalization on 4-hour follow-up CTA (modified arterial occlusive lesion score 2–3 versus 0–1).13,17
Clinical data were summarized using standard descriptive statistics. Between group (at the CTP-to-reperfusion level stratification), differences were tested using 1-way analysis of variance or Student’s t test (as appropriate) for parametric data, Kruskal–Wallis test for nonparametric data, and the Fisher’s test for categorical outcomes (Table 1).
Patients were stratified into 2 main groups: (1) stroke symptom onset to CTP <180 min and (2) ≥180 min. Each of these 2 groups were then subdivided into 3 subgroups according to time from CTP to reperfusion, that is, (1) <90 min reperfusion, (2) 90 to 180 min reperfusion, and (3) no acute reperfusion.
Combined patient data (pooled) CTP parameter histograms of all subjects from each group were created for ROI-1 and ROI-3, respectively. These histograms were then analyzed using ROC curve and Youden’s method to determine thresholds most associated with follow-up infarction at different times of reperfusion along with respective sensitivities and specificities for each threshold. The areas under the ROC curves (c-statistics) of all parameters were compared for significance using the Delong method.20 Corresponding relative CTP thresholds (for CBF) were determined by dividing the optimal thresholds by the means of the respective perfusion parameter from ROI-4.
Using individual patient-level data, CTP parameter thresholds most associated with follow-up infarction at different times of reperfusion were again derived. This was done by calculating mean±standard deviation of the ROC-derived parameters from individual patient’s histograms within a group (Table I in the online-only Data Supplement).
To test the hypothesis that optimal CTP-derived thresholds associated with follow-up infarction are dependent on time to reperfusion, we performed (1) 1-way analysis of variance across the different time strata for each optimal CTP threshold derived from patient-level data (discrete-time analysis) and (2) a locally weighted scatterplot smoothing regression to determine the association between optimal CTP thresholds derived from patient-level data and onset to CTP time and CTP to reperfusion time (continuous-time analysis).
A 2-sided P value <0.05 was considered statistically significant. All analyses were performed using R (version 3.2.1), STATA (version13, StataCorp LP, College Station, TX), and MATLAB (R2015a, version 8.5, Mathworks Inc, Natick, MA) statistical packages.
Of a total of 146 patients satisfying study inclusion/exclusion criteria, 132 patients were included for analysis. Patients excluded had inadequate coregistration of admission and follow-up imaging studies (n=11) and CTP acquisition errors (truncated scan, n=3). Clinical demographics for the 132 patients are summarized in Table 1. There were no statistically significant baseline differences between groups by age, sex, stroke severity, or site of occlusion.
Results from pooled-data analysis are described below. Results from patient-level analysis are shown in the Table I and Figure II in the online-only Data Supplement.
Group 1: Stroke Symptom Onset to CTP <180 min
<90 Min CTP-to-mTICI-2b/3 Reperfusion Group (n=28)
Tmax thresholds had the highest accuracies for tissue that will infarct if reperfused early (Table 2). Tmax optimal thresholds were >16.2 s (sensitivity =0.85 and specificity =0.83) and >15.8 (sensitivity =0.83 and specificity =0.81) for GM and WM, respectively. CBF optimal thresholds were <8.9 mL·min−1·100 g−1 (sensitivity =0.80 and specificity =0.83) and <7.4 mL·min−1·100 g−1 (sensitivity =0.78 and specificity =0.81) for GM and WM, respectively (Table 2). In a subset of patients achieving mTICI-3 reperfusion (n=13), Tmax and CBF best predicted infarct if reperfused early; Tmax was significantly better for GM (P<0.05; Table 2). CBV had the least discriminative value among the 3 CTP parameters (Table 2).
90 to 180 Min CTP-to-TICI-2b/3 Reperfusion Group (n=29)
Tmax and CBF had similar accuracies for tissue that will infarct if reperfused late (Table 2). Optimal Tmax thresholds were 12.4 s (sensitivity =0.89 and specificity 0.90) and 11.2 s (sensitivity =0.81 and specificity 0.82) for GM and WM, respectively. Optimal CBF thresholds were 11.1 mL·min−1·100 g−1 (sensitivity =0.87 and specificity =0.86) and 11.2 mL·min−1·100 g−1 (sensitivity =0.76 and specificity =0.73) for GM and WM, respectively.
Tmax and CBF had similar accuracies for infarct if not acutely reperfused (Table 2). Optimal Tmax thresholds were 10.1 s (sensitivity =0.91 and specificity =0.87) and 9.8 s (sensitivity =0.86 and specificity =0.93) for GM and WM, respectively. Optimal CBF thresholds were 15.1 mL·min−1·100 g−1 (sensitivity =0.80 and specificity =0.86) and 14.9 mL·min−1·100 g−1 (sensitivity =0.87 and specificity =0.84) for GM and WM, respectively. Other details are in Table 2.
Group 2: Stroke Onset to CTP ≥180 min
<90 Min CT-to-TICI-2b/3 Reperfusion Group (n=14)
Tmax and CBF thresholds had similar accuracies for tissue that will infarct if reperfused early (Table 2). Optimal Tmax thresholds were 14.6 s (sensitivity =0.83 and specificity =0.85) and 14.6 s (sensitivity =0.80 and specificity =0.81) for GM and WM, respectively. Optimal CBF thresholds were 9.5 mL·min−1·100 g−1 (sensitivity =0.84 and specificity =0.86) and 8.3 mL·min−1·100 g−1 (sensitivity =0.80 and specificity =0.79) for GM and WM, respectively. CBV had the least discriminative power among the 3 CTP parameters (Table 2).
90 to 180 Min CT-to-TICI-2b/3 Reperfusion Group (n=13)
Tmax and CBF had similar accuracies for tissue that will infarct if reperfused late (Table 2). Optimal Tmax thresholds were 10.3 s (sensitivity =0.74 and specificity =0.76) and 12.2 s (sensitivity =0.69 and specificity =0.71) for GM and WM, respectively. Optimal CBF thresholds were 14.3 mL·min−1·100 g−1 (sensitivity =0.70 and specificity =0.71) and 12.6 mL·min−1·100 g−1 (sensitivity =0.79 and specificity =0.73) for GM and WM, respectively. CBV had the least discriminative power among the 3 CTP parameters (Table 2).
Tmax and CBF had similar accuracies for infarct if not reperfused acutely (Table 2). Optimal Tmax thresholds were 9.5 s (sensitivity =0.81 and specificity =0.82) and 10.2 s (sensitivity =0.71 and specificity =0.70) for GM and WM, respectively. Optimal CBF thresholds were 14.4 mL·min−1·100 g−1 (sensitivity =0.82 and specificity =0.80) and 15.1 mL·min−1·100 g−1 (sensitivity =0.80 and specificity =0.82) for GM and WM, respectively. Other details are in Table 2.
Optimal GM and WM Tmax, CBF, and CBV thresholds associated with follow-up infarction from patient-level data analysis were similar to those from the pooled data analysis (Table 2 and Table I in the online-only Data Supplement). There was no relationship between stroke onset-to-CTP time and optimal CTP thresholds derived from patient-level data for all parameters based on discrete and continuous time analysis (P>0.05; Figure 1 and Figure II in the online-only Data Supplement). A statistically significant relationship was seen between CTP-to-reperfusion time and optimal thresholds for the CBF parameter (P<0.001 for both discrete and continuous time analysis; r=0.59 and 0.77 for GM and WM, respectively) and Tmax parameter (P<0.001 for both discrete and continuous time analysis; r=−0.68 and −0.60 for GM and WM, respectively; Figure 2 and Figure II in the online-only Data Supplement). The CBV parameter not did show any significant differences in optimal thresholds for any time stratification (P>0.1).
We have shown that CTP thresholds associated with final infarct are primarily dependent on the CTP-to-reperfusion time. These thresholds are not dependent on stroke onset to CTP time. In patients undergoing fast and quality reperfusion with modern endovascular treatment, we show that a CTP-derived Tmax threshold of around >16 s on average, in both GM and WM, has the highest sensitivity/specificity for brain tissue that is infarcted even when reperfused early (within 90 min from CTP imaging). Further, progressively lower Tmax thresholds of ≈>12.5 and >9.5 s are associated with GM and WM infarction if reperfusion is achieved between 90 to 180 min from CTP and in the acute nonreperfusers, respectively. The CBF parameter also has a high discriminative ability in all groups, whereas CBV does not (Table 2).
With recent clinical trials showing benefit of fast and effective endovascular treatment in patients with acute ischemic stroke, this therapy will become the standard of care; however, patients with large ischemic core are unlikely to benefit from this therapy.21–24 In this context, it is important for physicians to know the extent of brain that will infarct even if early reperfusion is achieved. Moreover, for physicians dealing with patients in primary stroke centers some distance away from an endovascular capable tertiary hospital, information on brain tissue that is likely to infarct in the time it takes for the patient to be transported to the tertiary center is vital.25,26 Our analysis demonstrates that optimal CTP parameter thresholds exist for identifying brain tissue that will likely infarct at different times from imaging if efficient reperfusion is not achieved (Table 2 and Figure 3; Table I in the online-only Data Supplement). These time-based CTP thresholds have the potential to help physicians in primary stroke centers make appropriate triaging decisions when directing patients for endovascular treatment to tertiary hospitals that may be 90 to 180 min away.26 These thresholds can also help neurointerventionists decide when to stop endovascular treatment if attempts at reperfusion are taking longer than expected. Figures 3 and 4 demonstrate the hypothetical time-based model and patient examples, respectively, of the utility of these time-based CTP thresholds. It is, however, possible that in patients achieving even earlier reperfusion, possibly within 60 min from imaging, hat the CTP thresholds for infarct are more stringent.
In the acute nonreperfuser cohort, reperfusion status was assessed during the acute treatment process and not at 24 hours; this assessment gives us an estimate of tissue that is likely to infarct if not reperfused acutely. This threshold, along with the time-based CTP threshold we have derived for at-risk brain tissue, can give clinicians an estimate of infarct growth over time (Figures 3 and 4). We have not derived CTP parameter thresholds for penumbra (brain tissue that is likely to infarct if never reperfused). As a practical construct, this may be impossible to determine.
The absolute and relative CBF thresholds for GM and WM in early reperfusers are lower than CBF thresholds previously reported for irreversibly infarcted brain.3–7 Factors affecting CTP processing algorithms that can lead to underestimation of CBF include delay (T0) between arterial input and ischemic tissue time density curves and use of venous curve for partial volume correction27; however, the CTP algorithm used in this study is delay-insensitive, and the arterial input curve was chosen from large intracranial arteries at the base of the skull; thus, obviating the need to use a venous curve for partial volume correction.27 Therefore, the lower absolute and relative CBF thresholds we report are likely because of our unique data set, where many patients have undergone ultraearly and efficient reperfusion.
The use of a prolonged CTP acquisition in our study minimizes the underestimation of CBV and mean transit time from the truncation of time-density curves as compared with a shorter acquisition protocols.28 In spite of this, our analysis reveals that CBV is the least discriminative of the CTP parameters. Although low CBV remains strongly predictive of progression to irreversible infarction, normal CBV can be seen in both normal and in severely ischemic brain tissue, and normal and elevated CBV can be seen in reperfused ischemic brain tissue.29,30
Although we chose consecutive patients fulfilling appropriate selection criteria from a prospective study and have a small sample, given the pooling of patient imaging data for ROC analyses, our study is adequately powered to detect optimal thresholds. We also demonstrate consistency between the primary analyses (pooling of patient data) and sensitivity analyses (patient-level data). Nonetheless, validation studies need to be performed to show that these thresholds reliably predict infarction and final clinical outcome. In a minority of patients, we defined final infarct on follow-up NCCT. To address the limitation that NCCT is less sensitive than MR in detecting infarcted brain on follow-up, 2 experts (Drs Ahn and Minhas) with ≥10 years of experience delineated final infarct on NCCT by consensus.
In conclusion, CTP thresholds associated with follow-up infarction vary based on time from imaging to quality reperfusion. These time-based CTP thresholds could potentially have important implications for clinical decision-making and triage of patients with acute ischemic stroke and proximal occlusions for endovascular treatment.
We thank the staff and fellows from the Calgary Stroke Program, Calgary, Alberta, Canada, and University Hospital, Modena, Italy, as well as Andrea Bernardoni, Carmine Tamborino, and Marina Padroni from the University Hospital in Ferrara, Italy.
Sources of Funding
This study is funded through an operating grant (PRoveIT) from the Canadian Institute of Health Research (CIHR). Dr Menon also holds the Heart and Stroke/University of Calgary Professorship in Stroke Imaging. Dr d’Esterre is a Heart and Stroke Foundation of Canada Fellow. Dr Frayne is the Hopewell Professor of Brain Imaging. Dr Lee receives funding from CIHR and Heart and Stroke Foundation of Canada.
Dr Goyal has a patent pending on systems of stroke diagnosis using multiphase computed tomographic angiography and a licensing agreement with GE healthcare for the same. Dr Goyal also provided consulting services to Covidien for design and conduct of SWIFT PRIME trial and for teaching engagements. Dr Hill and Dr Goyal have a research grant with Covidien to conduct clinical trials. Dr Lee receives royalties from licensing of the CT Perfusion software to GE Healthcare. The ESCAPE trial was partially funded through an unrestricted research grant by Covidien to the University of Calgary. The other authors report no conflicts.
Guest Editor for this article was Hans-Christoph Diener, MD, PhD.
The online-only Data Supplement is available with this article at http://stroke.ahajournals.org/lookup/suppl/doi:10.1161/STROKEAHA.115.009250/-/DC1.
- Received May 11, 2015.
- Revision received September 24, 2015.
- Accepted September 30, 2015.
- © 2015 American Heart Association, Inc.
- Astrup J,
- Siesjö BK,
- Symon L.
- Jovin TG,
- Yonas H,
- Gebel JM,
- Kanal E,
- Chang YF,
- Grahovac SZ,
- et al
- Wintermark M,
- Flanders AE,
- Velthuis B,
- Meuli R,
- van Leeuwen M,
- Goldsher D,
- et al
- Murphy BD,
- Fox AJ,
- Lee DH,
- Sahlas DJ,
- Black SE,
- Hogan MJ,
- et al
- Bivard A,
- Spratt N,
- Levi C,
- Parsons M.
- Campbell BC,
- Christensen S,
- Levi CR,
- Desmond PM,
- Donnan GA,
- Davis SM,
- et al
- Schaefer PW,
- Souza L,
- Kamalian S,
- Hirsch JA,
- Yoo AJ,
- Kamalian S,
- et al
- Yoo AJ,
- Zaidat OO,
- Chaudhry ZA,
- Berkhemer OA,
- González RG,
- Goyal M,
- et al
- Souza LC,
- Yoo AJ,
- Chaudhry ZA,
- Payabvash S,
- Kemmling A,
- Schaefer PW,
- et al
- Yoo AJ,
- Verduzco LA,
- Schaefer PW,
- Hirsch JA,
- Rabinov JD,
- González RG.
- Menon BK,
- Saver JL,
- Goyal M,
- Nogueira R,
- Prabhakaran S,
- Liang L,
- et al
- Lee TY,
- Menon BK,
- Goyal M,
- Demchuk AM,
- Frayne R.
- d’Esterre CD,
- Aviv RI,
- Lee TY.
- Hatazawa J,
- Shimosegawa E,
- Toyoshima H,
- Ardekani BA,
- Suzuki A,
- Okudera T,
- et al