Department of Clinical Neurosciences, Hotchkiss Brain Institute University of Calgary, Calgary, Alberta, Canada
Department of Radiology, Hotchkiss Brain Institute University of Calgary, Calgary, Alberta, Canada
Correspondence: Andrew M. Demchuk, Department of Clinical Neurosciences and Radiology, Hotchkiss Brain Institute, University of Calgary, Room 1162, Foothills Medical Centre, 1403 29 St NW, Calgary, AB, Canada T2N 2T9.
There is no consensus on how the reliability and predictive ability of noncontrast computed tomography (NCCT) and computed tomography angiography source image (CTASI) change over time from acute ischemic stroke onset. We hypothesized that the reliability for detecting early ischemic changes (EIC) would be lower in early time periods and that changes identified on CTASI would be more reliable across examiners than changes identified on NCCT. To address this, we compared the relationships between CTASI, NCCT, and final infarct in patients with initial computed tomography (CT) imaging at different time points after stroke onset. Patients with acute ischemic stroke with proximal anterior circulation occlusions (internal carotid artery, middle carotid artery M1, proximal M2) from Calgary CT Angiography (CTA) database were studied. The cohort was categorized in four groups based on time from stroke onset to baseline NCCT/CTA: 0–90 mins (n = 69), 91–180 mins (n = 88), 181–360 mins (n = 46), and >360 mins (n = 58). Median scores of NCCT-Alberta Stroke Program Early CT Score (ASPECTS), CTASI ASPECTS, and follow-up ASPECTS among different time categories were compared. To determine reliability, a subsample of NCCT brain and CTASI were interpreted at separate sessions weeks apart by two neuroradiologists and two stroke neurologists in random order. Median and mean ASPECTS ratings on NCCT and CTASI were higher than final ASPECTS in each time category (P < 0·001 for all comparisons). CTASI ASPECTS was lower than NCCT ASPECTS in each time category, and differences were significant at 0–90 mins and 91–180 mins (P < 0·001). The least agreement among readers was in detection of EIC on NCCT brain in the ultra-early phase (<90 mins) [intraclass correlation coefficient (ICC) = 0·48. By contrast, there was excellent agreement on EIC on CTASI regardless of time period (ICC = 0·87–0·96). Using ASPECTS methodology, CTASI is more reliable than NCCT at predicting final infarct extent particularly in the early time windows.
Computed tomography (CT) is the modality of choice to image patients with acute stroke due to its speed of acquisition and wide availability . As a result, noncontrast CT (NCCT) remains the most commonly performed study as part of acute stroke clinical trials. Noncontrast CT identifies early ischemic changes (EIC) that can predict irreversible ischemic injury. However, interpretation of EIC on NCCT varies substantially between readers, with limited reliability identified in some studies [2, 3]. Alberta Stroke Program Early CT Score (ASPECTS) is a semiquantitative radiological grading of EIC on NCCT, which aids the reader to systematically evaluate the middle cerebral artery (MCA) territory. It grades subtle hypoattenuation of brain tissue in the MCA region on a 10-point anatomical template [4, 5]. The limitation of accurately detecting irreversible injury with ischemic changes on NCCT has sobering implications for acute stroke trials as significant patient enrollment of subjects continues to occur despite large regions of irreversible brain due to an under appreciation of the extent of ischemia. Such enrollment is likely to introduce significant noise into trials as such patients may be less likely to benefit from endovascular interventions aimed at recanalization [6, 7].
Multimodal CT techniques such as CT angiography (CTA) and computed tomography perfusion (CTP) have the potential to improve such reliability for detection of irreversible injury. Use of CTA in particular has increased over the last few years in the management of acute stroke . Apart from identifying intracranial and extracranial vessel disease, CTA source image (CTASI) detects regions of brain parenchyma with hypoattenuation due to severe reductions in cerebral blood volume (CBV) or due to long delays in contrast arriving at the ischemic brain. It has a better correlation with final infarct volumes in comparison to NCCT , better predicts clinical outcomes in comparison to NCCT , and also correlates better to magnetic resonance (MR) based diffusion weighted imaging (DWI) [11, 12]. Computed tomography angiography can help to triage patients to immediate endovascular therapy as a part of acute stroke treatment trials, when they have proximal vessel occlusion and to plan an endovascular approach with knowledge of the aortic arch, the extracranial vessels, and the circle of Willis. Early ischemic changes on NCCT evolve with duration and severity of ischemia. A progressive increase in the water content of brain tissue results in a drop in attenuation over time . On the other hand, hypoattenuation detected on CTASI may be less influenced by time from stroke onset and is distinctly visible very early in the evolution of an infarct . Based on these differences, we hypothesize that differences in the reliability of interpretation and the extent of abnormality detected will vary by time interval after symptom onset when comparing EIC on NCCT with CTASI hypoattenuation. To address this, we compared the relationships between CTASI, NCCT, and final infarct ASPECTS in patients who underwent initial CT imaging at different time points after stroke and detected interclass variability among different readers.
We identified patients presenting with acute ischemic stroke secondary to major vessel occlusion presenting within nine-hours of symptom onset (maximum accepted time line for endovascular treatment trials) or the last seen normal times, from the CTA database of the Calgary Stroke Program at our center from August 2003 to December 2009. The Calgary CTA database is a human research and ethics board approved retrospective study database. As the study was based on the ASPECTS template for grading ischemic changes in the MCA territory, patients with either anterior cerebral artery or posterior circulation involvement were excluded. Patients presenting beyond nine-hours were excluded from the study.
At our center, standard NCCT is performed on a multislice helical scanner (GE Medical Systems, Fairfield, CT, or Siemens Medical Solutions, Erlangen, Germany) using 120 kV, 170 mAs with five-millimeter slice thickness. Noncontrast CT is followed immediately by CTA with a helical scan technique. Acquisitions are obtained after a single bolus intravenous contrast injection of 80–100 ml non-ionic contrast media into an antecubital vein at 3–5 ml/s auto-triggered by the appearance of contrast material in the ascending aorta with coverage from arch to vertex. Minimum coverage is from foramen magnum to centrum semiovale with 0·625–1·25 mm slice thickness. Multiplanar volume-reformatted images are immediately created by the CT technologist with 2·5–4·0 mm slice thickness in axial, sagittal, and coronal planes. Follow-up imaging is performed in all patients on the same CT scanner within at least 24 h from the first CT scan to maximum of seven-days after the initial NCCT.
This is a clinimetric study of ASPECTS scoring among multiple raters. Alberta Stroke Program Early CT Score on NCCT and CTASI in our study were scored independently by a neuroradiologist and two stroke neurologists blinded to clinical outcome and follow-up scans. Computed tomography angiography source image ASPECTS was scored on 3 mm reformatted axial CTA images. Alberta Stroke Program Early CT Score is only scored for acute ischemic changes in MCA territory, a score of 10 indicates a normal scan and a score of 0 indicates complete MCA infarction. Computed tomography angiography source image was evaluated for extent of hypoattenuation. All images were seen on a large high-resolution monitor with optimal window and level settings to maximize the contrast produced by small attenuation differences between normal and ischemic tissue. Time from stroke symptom onset to baseline NCCT and CTA was categorized into four groups: Group 1 (0–90 mins), Group 2 (91–180 mins), Group 3 (181–360 mins), and Group 4 (360–540 mins).
To determine the interobserver reliability in scan reading in different time frames from stroke onset a subsample was randomly chosen, stratified by time category from symptom onset to CT: 0–90 mins (n = 16), 91–180 mins (n = 16), 181–360 mins (n = 17), and >360 mins (n = 15). Noncontrast CT brain and CTASI were interpreted at separate sessions 2–3 weeks apart by two neuroradiologists and two stroke neurologists in a random order. Alberta Stroke Program Early CT Score was scored to determine extent of EIC.
Continuous nonnormal data are described as median [interquartile range (IQR)]. To give a more finely grained measure of the central tendency of the ASPECTS scores, we also display the means and standard deviations. This was justified because the ASPECTS data were approximately normally distributed in our cohort, even though they are noncontinuous ordinal data. Within-patient differences in ASPECTS scores were compared using the nonparametric paired Wilcoxon signed-rank test. The intraclass correlation coefficient was chosen as the measure of interobserver reliability, as in other studies . P-values are not corrected for multiple comparisons. For statistical analyses, sas version 9·1.2 (SAS Institute Inc., Cary, NC) was used. A conventional P-value < 0·05 was considered significant.
Among 1341 patients within the CTA database, 421 patients had documented occlusions. Among these, we identified 261 patients with proximal vessel occlusion in anterior circulation. There were 51% males with a mean age of 66·8 years (±15·1 years). There were 34 (13%) patients with distal internal carotid artery occlusions, 143 (54·8%) with MCA M1 occlusions, and 84 (32·2%) patients had proximal MCA M2 occlusion. Among 261 patients, 151 (60·1%) patients had hypertension, 69 (26·4%) patients had atrial fibrillation, 46 (17%) patients had diabetes mellitus, and 76 (29%) patients were smokers. The median baseline National Institutes of Health Stroke Scale score was 14 (IQR 9–19). Median baseline NCCT ASPECTS and CTASI ASPECTS were 8 (IQR 6–9) and 6 (IQR 5–8), respectively (Supporting Information Table S1).
The median time from stroke symptom onset to CT was 148 mins (IQR 90–311 mins) and median time from NCCT to CTA was three-minutes (IQR 2–5 mins). Among 261 patients, 69 (27%) patients were scanned between 0 and 90 mins, 88 (34%) were scanned between 91 and 180 mins, 46 (18%) were scanned between 181 and 360 mins, and 58 (22%) were scanned >360 mins but less than nine-hours.
Reliability of NCCT ASPECTS and CTASI ASPECTS with final ASPECTS
Comparing reliability to assess EIC in NCCT and CTASI among four readers, there was a least agreement among readers in detection of EIC on NCCT brain in ultra-early phase (<90 mins) (Table 1). By contrast, the interrater reliability of determination of CTASI abnormality was similarly high irrespective of the time period (Table 1).
Table 1. Interobserver reliability of ASPECTS scoring on NCCT and CTASI in different time categories from stroke onset
95% CI (lower, upper)
95% CI (lower, upper)
aComparison of the intraclass correlation at 0–90 mins vs. 90 mins or greater, P = 0·0001 for NCCT and P = 0·178 for CTASI.
ASPECTS, Alberta Stroke Program Early CT Score; NCCT, noncontrast computed tomography; CTASI, CT angiography source image; CI, confidence interval.
The median and mean differences between baseline ASPECTS (measured on either CTASI or NCCT) and final ASPECTS, according to the prespecified time categories, are shown in Table 2. Both CTASI ASPECTS and NCCT ASPECTS were significantly higher than final ASPECTS in each time category. However, the median and mean differences between CTASI and final ASPECTS were always less than the median and mean difference between baseline NCCT and final ASPECTS, in each time category. The greatest difference was observed in the 0–90 mins time interval (Fig. 1).
Table 2. Median values of NCCT, CTASI, and final ASPECTS across different time categories
Time stroke onset to CT (mins)
Both CTASI ASPECTS and NCCT ASPECTS are significantly higher than final ASPECTS in each time category (P ≤ 0·02). CTASI ASPECTS was significantly lower than the initial NCCT ASPECTS, and thus closer to final NCCT ASPECTS at 0–90 mins (P < 0·001 for comparison of CTASI ASPECTS and NCCT ASPECTS) and 91–180 mins (P = 0·006) but not at 181–360 mins (P = 0·12) or >360 mins (P = 0·83).
ASPECTS, Alberta Stroke Program Early CT Score; NCCT, noncontrast computed tomography; CTASI, CT angiography source image; IQR, interquartile range.
Our study showed a low agreement among readers to delineate EIC on NCCT brain in the ultra-early phase of stroke (<90 mins). In comparison, CTASI had excellent agreement for hypoattenuation irrespective of the time from stroke onset. The present study also showed significant difference between NCCT ASPECTS and CTASI ASPECTS in the ultra-early phase of acute ischemic strokes (0–90 mins from stroke symptom onset). This difference narrows as time progresses with changes on NCCT becoming more evident highlighting the importance of CTASI in delineating abnormality in very early phase of stroke in comparison to NCCT. We also show that CTASI ASPECTS more closely approximates to final infarct extent in all time intervals from stroke onset.
In previous studies, it has been shown that subtlety of EIC in acute ischemic stroke results in poor interobserver reliability especially within the first hours following stroke onset [15-17]. Using ASPECTS score methodology, we have shown moderate to excellent interobserver reliability . However, none of these studies had compared reliability of NCCT ASPECTS with CTASI ASPECTS in different time frames from stroke onset. In the present study, the overall reliability to detect ischemic changes on NCCT and CTASI is similar to that shown earlier,  but the ability of NCCT to guide clinicians for decision making in ultra-early phase after stroke is questioned. In comparison, consistently high reliability of CTASI irrespective of time period from stroke onset makes it a more valuable tool in this phase of stroke. Better approximation of CTASI ASPECTS with final ASPECTS in our study also support findings from earlier studies and may suggest that these changes may be irreversible, but from the present study, it is difficult to make this conclusion without analyzing factors like bolus scan delay, time to recanalizatio,n and leptomeningeal collateral status.
One of the critiques of ASPECTS interpretation relates to the challenges of detecting subtle EIC. Ischemic changes on NCCT are particularly subtle in the ultra-early phase after stroke and become more obvious with time from stroke onset. This time dependence of EIC is not surprising given animal model studies of CT-based hypoattenuation with cerebral ischemia. Hypodensity due to EIC on NCCT brain is due to shift in brain tissue water content secondary to significant ischemia (brain parenchyma with blood supply <12 ml/100 gm/min), resulting in a diminished X-ray attenuation secondary to decreased specific gravity of brain tissue. This change is commonly seen within six-hours after the symptom onset in patients with acute ischemic stroke [19, 20]. One percent increase in brain tissue water content results in a drop of 1·8–2·6 Hounsfield units (HU) in attenuation. For hypoattenuation to be visible to the human eye, there has to be a large shift in water uptake in ischemic tissue [3, 21, 22]. Early ischemic changes on CT scans as early as 1·3 h from the onset of stroke causes change of approximately 1·2 ± 0·6 HU , with a decrease in 1–2 HU per hour thereafter (shown in the animal experiments) . Computed tomography angiography source image does not have a dependence on time to reveal hypoattenuation (Figure 2). This is because it predominantly reflects a reduction in CBV that is evident immediately after cerebral ischemia occurs. Computed tomography angiography source image hypoattenuation correlates with decreased CBV on perfusion imaging and DWI lesion on MR imaging consistent with irreversible ischemia [24, 25]. However, some recent studies have shown that CTA hypoattenuation correlates better with reduced cerebral blood flow (CBF) than CBV, thus suggesting that CTASI may in fact also represent regions of ischemic penumbra and may therefore be partially reversible with early recanalization [26, 27].
The present study has implications for clinical practice and trial design. For a treating clinician, the most important imaging question to answer is whether the extent of irreversible ischemia is too large for recanalization (malignant profile) to benefit the patient even within acceptable time windows for systemic thrombolysis (<4·5 h) or intra-arterial treatment (<9 h) . Reliance on NCCT alone risks underestimating the number of cases where a malignant profile of MCA infarction is present. The use of CTASI information, in centers where hyperacute CTA is routinely performed, may help identify patients at risk for a poor prognosis even with arterial recanalization. Equally important is the implications for clinical trial design. Trials of recanalization treatment performed early after symptom onset should incorporate advanced CT imaging at least in the form of CT/CTA to limit trial enrollment of malignant profile subjects who only add ‘noise’ to such studies as benefit from thrombolysis is highly unlikely.
Our study does have limitations. Use of different CT scanners with variable protocols may give venous or arterial phase CTASI images depending upon the time from bolus scan delays. Recent studies have illustrated faster acquisition of CTA data, shifting the images to more arterial weighting which would represent delay of contrast and CBF weighted images rather than the venous weighted images initially reported in the original reports on CTASI correlating with CBV and DWI [26, 27]. With time, CTASI may also underestimate the extent of irreversible injury as a result. Secondly, the present study is a retrospective analysis and does not have documentation of recanalization status which may be an important factor affecting the extent of final infarct or progressive growth of infarct. We investigated the reliability of only one method for scoring CT changes, the ASPECTS system; therefore, our findings may not generalize to other methods of interpreting or scoring EIC. Computed tomography perfusion (CTP) was not performed in the present study. To ascertain the importance of CTASI in hyperacute phase, we call for further studies to compare CTASI with CTP.
Using the ASPECTS methodology, CTASI is more reliable particularly in early time windows in comparison to NCCT. The true extent of ischemic change is more likely to be underestimated with NCCT in very early scanned patients and should not be relied upon when CTASI is available. In centers that perform hyperacute CTA, CTASI should be interpreted as part of standard assessment for stroke diagnosis and prognosis. Patients with more extensive CTASI changes are more likely to have larger final strokes. However, we do not know whether CTASI changes are associated with the response to recanalization therapies; this should be the subject of future research studies. For acute stroke trials, CTA could be considered as part of baseline imaging protocol to identify patients with extensive early changes that may not be desirable clinical trial candidates.
We would like to acknowledge other members of the Calgary CTA Study Group who contributed to this project including: Jayanta Roy, Christine O'Reilly, Kamal Sahi, Christine Stables, Sarah Tymchuk, Sherif Idris, Phil Barber, Dar Dowlatshahi, Tim Watson, Suresh Subramanian, Jean-Martin Boulanger, and Shelagh Coutts.