Imaging perfusion deficits, arterial patency and thrombolysis safety and efficacy in acute ischaemic stroke. An observational study of the effect of advanced imaging methods in The Third International Stroke Trial (IST-3) - a randomised controlled trial.

Participants

Patients with suspected acute ischaemic stroke who reached hospital, could be assessed and treated within 6 hours of stroke onset. Patients in whom rt-PA was ‘promising but unproven’ could be randomised in the trial after informed consent was obtained.

Inclusion criteria

(a) Symptoms and signs of clinically definite acute stroke, (b) time of stroke onset definitely < 6 hours previously, (c) CT or MR brain scanning has excluded intracranial haemorrhage and (d) treatment can be started within 6 hours of stroke. Patients with symptoms of large and medium-sized cortical, lacunar and posterior circulation stroke were all included, with no upper age limit. Patients with early visible infarct signs were also included (though not if established infarct signs were present, as these suggest a stroke onset of more than > 6 hours previously).

Exclusion criteria

Age < 18 years, imaging signs that the stroke might be older than 6 hours, and usual contraindications to rt-PA. 60

Interventions

Intravenous rt-PA (total dose 0.9 mg/kg to a maximum of 90 mg, 10% as bolus and the rest infused over 1 hour) compared with ‘open control’ (avoid rt-PA and receive stroke care in exactly the same clinical environment as those allocated ‘immediate rt-PA’).

Baseline assessment

All patients were assessed for stroke severity [National Institutes of Stroke Scale (NIHSS) score], stroke subtype [total anterior circulation infarct (TACI); partial anterior circulation infarct (PACI); lacunar infarct (LACI); or posterior circulation infarct (POCI), clinical syndrome], presence of atrial fibrillation (AF), systolic and diastolic blood pressure and blood glucose.

Objectives

To determine if rt-PA, given to a wider range of patients up to 6 hours after stroke, would improve functional outcome by 6 months net of any hazard.

Outcomes

The primary outcome was alive and independent [Oxford Handicap Score (OHS) 0–2,61 which is very similar to modified Rankin 0–262] at 6 months after stroke. Symptomatic and fatal intracranial haemorrhage, death and recurrent stroke within 7 days and death at 6 months were also assessed.

Brain scanning

All patients had a CT or MR brain scan before randomisation and a follow-up scan at 24–48 hours. A repeat brain scan was required if the patient deteriorated neurologically or intracranial haemorrhage was suspected for any reason. All scans were sent to the trial centre in Edinburgh for blinded central rating of any signs of visible early ischaemia (presence and extent of hypoattenuation, swelling, hyperattenuated artery), haemorrhage, and background brain changes (leukoaraiosis, atrophy, prior stroke lesions, non-stroke lesions) with validated rating tools. 15,63–67 Images were assessed blindly, and assessed via a secure web-based image viewing system by an international panel of expert radiologists.

Sample size

The IST-3 main trial was powered to detect a 4.7% absolute improvement with rt-PA compared with no rt-PA in the number alive and independent at 6 months with power 80% at p = 0.05 with 3100 patients. 57 This effect size was based on the Cochrane Thrombolysis Review in 2000,2 but remained unaltered following the update in 2008. 32

Randomisation

Randomisation was via a secure central telephone or web-based computer system, which recorded all of the baseline data and generated the treatment allocation. A minimisation algorithm was used to achieve optimum balance for key prognostic factors. 57,59

Follow-up

Follow-up of 6-month outcomes was by central office staff blinded to treatment allocation, by postal questionnaire or telephone for non-responders (by an experienced, blinded assessor).

Statistical methods59

The statistical analysis plan was published59 prior to unblinding to the data. To avoid complicating the estimation of the effect of treatment overall and in subgroups,57 the primary analysis was logistic regression for linear effects adjusted for the following covariates: age; NIHSS score; time from onset of stroke symptoms to randomisation; and presence (vs. absence) of ischaemic change on the pre-randomisation brain scan according to the expert read. Unadjusted analyses were also performed. 60 The statistical analysis plan writing committee, while still blinded, adopted the ordinal method, as it is statistically more efficient (effectively reducing the sample size required in stroke trials68). The OHS as a dependent variable had five levels: levels four, five and six were combined into a single level and levels zero, one, two and three were retained as distinct. In this model, the treatment odds ratios between one level and the next are assumed constant, so a single parameter summarises the shift in outcome distribution between treatment and control groups. Analyses were carried out with SAS version 9.2 (SAS Institute Inc., Cary, NC, USA).

Any changes to protocol

Two changes occurred. The first was the change from placebo-controlled to open-label treatment after the first 297 patients due to withdrawal of support for the trial by Boehringer Ingelheim (Bracknell, UK). The second was the revised sample size estimation and introduction of the ordinal analysis described above as a secondary outcome analysis.

Objectives

The basic questions to be addressed are ‘should “perfusion-structural imaging mismatch” or “arterial occlusion” influence whether or not patients receive rt-PA?’ Here, the key question was whether or not rt-PA is more effective in patients with imaging evidence of tissue at risk than in those without apparent tissue at risk. Tissue at risk was defined as:

1. the difference between the extent of core damaged tissue on MR DWI or plain CT and the extent of the MR or CT perfusion lesion (further details of perfusion lesion measurements and comparisons below); or

2. evidence of arterial occlusion on CT or MR angiography.

Imaging acquisition

Where possible, patients were to be examined on the same scanner at baseline and at follow-up, although combinations, for example CT pre randomisation and MR at 24-hour follow-up, were allowed as local clinical practice dictated. Basic minimum acquisition standards were required (see Appendix 1). We provided basic minimum acquisition standards to encourage best practice in perfusion or angiography imaging while allowing for the considerable variation that exists in available scanning technology. Thus, it would have been counterproductive to provide overly narrow acquisition criteria that only a proportion of centres might have been able to meet, as that would have further limited the sample size and generalisability of the data. Full details of the minimum acquisition criteria as sent to participating centres are given in Appendix 1. In addition, before a centre could participate in the Perfusion and Angiography Study, a test perfusion and/or angiogram image data set had to be sent to the IST-3 trial co-ordinating centre to ensure that the imaging met minimum standards and that the data could be processed centrally.

The trial image data were received at the IST-3 trial co-ordinating centre, linked with their demographic data and trial records, anonymised and transferred into the image-processing pipeline. Plain CT and MR images were read according to the IST-3 established structured image analysis protocol by a panel of experts via a web-based image reading system, the Systematic Image Review System (SIRS: see www.neuroimage.co.uk/) as detailed above.

Outcomes

The primary outcome measures were the same clinical measures as for the IST-3 main trial above: functional outcome (OHS 0–2), symptomatic and fatal intracranial haemorrhage, early and late death and massive infarct swelling.

The secondary outcomes were absolute infarct growth, defined qualitatively as a change in the extent of hypoattenuated tissue on CT or of hyperintense tissue on MR FLAIR between baseline and 24–48-hour follow-up, of one point or more on either the IST-3 scale score,64,70 in any arterial territory, or the Alberta Stroke Program Early CT score (ASPECTS)63 if in the middle cerebral artery (MCA) territory; defined quantitatively as the difference in measured lesion volume on plain CT or MR DWI between randomisation and follow-up scans.

Blinding

All image data were collected centrally in Digital Imaging and Communications in Medicine (DICOM) format, matched with the patient record, anonymised and identified only by the study identification number. All image data analyses were performed centrally, blind to treatment allocation, baseline demographic information and follow-up.

Perfusion analysis

We produced perfusion parameter maps for each patient for visual rating and measurement of lesion volume without any threshold applied (Figure 1; Table 1): quantitative (q) perfusion with deconvolution CBF quantitative (CBFq), CBV quantitative (CBVq), MTT quantitative (MTTq), time to maximum flow quantitative (time to peak of the residue function) (Tmaxq) and relative (r) perfusion, that is to say without deconvolution relative CBF (rCBF); relative arrival time fitted; relative time to peak; relative peak time fitted; relative maximum concentration peak; relative full width at half maximum. Full details of the perfusion processing are given in Appendix 2. We also produced a set of parameter maps with thresholds applied (see Table 1). These parameters and thresholds were based on literature values that had been proposed for identifying still-viable but at-risk tissue and core tissue, of which there were many, but none had been validated independently. 71 This was because the IST-3 perfusion substudy was not large enough to generate new thresholds in one half of the data set and validate these in the other half. Therefore, we focused on validating ones which had been reported previously.

FIGURE 1.

Example of CT perfusion parameter maps. (a) CT with plain structural image at randomisation and post randomisation, with infarct outlined, perfusion maps and various thresholds below; (b) MR with acute DW1 and T2, follow-up T2, perfusion maps and various thresholds below. ROI, region of interest.

Maps of the following perfusion thresholds were produced for volumetric and visual measurement (details in Table 1):

• Representing non-salvageable tissue:

  • on CTP: absolute CBV< 2 ml/100 g;20

  • on MRP: rCBF< 31%72 and rCBF< 40%. 73,74

• Representing at-risk tissue:

  • on CTP: relative MTT (rMTT) > 145%;20 rMTT > 125%. 74

  • on MRP: time to maximum flow (Tmax) > 6 seconds75–81 [note that Tmax > 2 seconds was originally identified in the Echoplanar Imaging Thrombolytic Evaluation Trial (EPITHET) but subsequent analyses and other groups have identified Tmax > 6 seconds as a preferred threshold].

These perfusion parameters reflected commonly applied thresholds and image types while keeping the total number of comparisons manageable and reducing the potential for false-positive results. Our systematic review had not identified a specific parameter or threshold that seemed optimum;71 different research groups had not identified an agreed perfusion parameter/threshold since the systematic review. We therefore tested several perfusion parameters/thresholds which covered the most easily available and most promising derived from the most recent research. Many of these thresholds have been defined for one modality only (mostly CTP) but could equally be applied to MR data and, therefore, were tested.

Perfusion lesion extent was quantified visually by one expert neuroradiologist, blind to all other data. We used the ASPECTS,63 subtracting one point from a total of 10 for each MCA ASPECTS region that is in part or wholly affected by the perfusion lesion, even where perfusion image does not cover the whole ASPECTS region. We also recorded if there was (a) no visible perfusion lesion, (b) a visible perfusion lesion that was < 80%, (c) about the same size as or (d) ≥ 20% larger than the structural ischaemic lesion by visually-estimated volume on plain CT or MR DWI/FLAIR. 27,30 ‘Mismatch’ was defined as a perfusion lesion > 20% larger than the structural lesion. We validated these methods in a separate three-centre study (Translational Medicine Research Collaboration Multicentre Acute Stroke Imaging Study47). Visual coding forms are available in Appendix 3. Perfusion lesion volume was measured by manual outlining the lesion by a trained observer blind to all other data on two of the unthresholded parameter maps from above (MTTq and rCBF perfusion lesions) to represent at-risk tissue and non-salvageable tissue respectively. The perfusion lesion volume was also measured on thresholded parameter maps listed in Table 1 using automated thresholding.

Angiography analysis

The randomisation CT angiography images were scored by a blinded neuroradiologist who also measured thrombus density on a workstation in Hounsfield Units (HU). The hyperattenuated artery sign (HAS) was scored on the available imaging, that is to say thin section if available or routine 5 mm section if not, depending on what imaging had been received. Separately, a panel of 11 experts also read all of the CT and MR angiograms using the web-based SIRS (SIRS2), which we modified so as to be able to see the plan scan and angiographic image on the same screen and record both the plain-scan findings and angiographic appearance (SIRS2, sirs2/neuroimage.co.uk/sirs2; Figure 2).

FIGURE 2.

Screen capture of web interface used to visually assess angiographic images.

We assessed the location, extent of vessel affected and degree of obstruction to the lumen of any arterial occlusion, the presence of collateral pathways, the clot burden53 and the attenuation properties of the occluding thrombus. Location and extent of thrombus was coded in the internal carotid artery (ICA), MCA mainstem or sylvian branch, anterior cerebral artery (ACA), posterior cerebral artery (PCA), basilar artery, vertebral artery or combinations thereof. 15,38,40 We debated, at length, the best score to use. Several scores are available to classify the degree of major arterial obstruction (Table 2). These mostly conflate three different concepts – peripheral microvascular tissue perfusion, primary arterial patency, and recanalisation – in a single score, thereby mixing three separate and probably semi-independent entities. 87 This, no doubt, contributes to the poor observer reliability. We previously used the Mori82 and Thrombolysis in Myocardial Infarction (TIMI)83 scores purely to classify arterial patency at the primary point of obstruction on CTA and MRA and, separately, used CTP or MRP to classify tissue-level perfusion and reperfusion which worked well. Other scores (summarised in Table 2) mixed primary occlusion, perfusion and recanalisation. 37,56,84,85 Therefore, in IST-3 we used a score that combines the best elements of the TICI (including 2a and 2b) and arterial occlusive lesion (AOL) scores that only scored angiographic patency at the main point of occlusion and filling of immediate distal vessels, but not tissue perfusion or recanalisation. This score, used in DIAS 3 and 4 and IMS-3,43–45 is described in the protocol paper. 47

Recanalisation was indicated by a change of one point or more on the scale between randomisation and follow-up scans.

We also coded thrombus burden using the clot burden score,53 where one or two points are subtracted from a normal score of 10 for each segment of the main intracranial arteries or their branches that is abnormal on angiography; thus, a score of zero indicates that all major intracranial arteries on one side of the head are thrombosed. We also scored visible HASs on CT and abnormal arteries on CTA whether the abnormality was in a proximal (internal carotid or basilar artery), middle (ACA, MCA or PCA) or distal (sylvian branches of MCA) artery. The resulting proximal-middle-distal (PMD) score value ranges from 1 to 6, where 1 = only distal vessel, 2 = only middle vessel, 3 = middle and distal vessels, 4 = proximal vessel, 5 = proximal and middle vessels and 6 = proximal, middle and distal vessels.

We scored the adequacy of the collateral pathways54 in patients with ICA/MCA main stem occlusion only using the Score for Collateral Status,55 a three-point scale of good, moderate or poor based on the number of opacified arteries visible in the peripheral parts of the affected tissue. Examples are provided for comparison.

The resulting coding forms are available in Appendix 3 and can be seen at www.bric.ed.ac.uk/research/imageanalysis.html#ais.

A neuroradiologist also measured the mean density of any HAS in HU (i.e. standard units used to assess tissue X-ray beam attenuation) using a region of interest cursor placed on the affected artery and also of the unaffected arteries (i.e. basilar, left or right middle cerebral arteries) on a personal computer workstation running Digital Jacket software (an in-house image server application allowing manipulation of DICOM data sets; DesAcc, Bellevue, WA, USA). Ovoid ‘regions of interest’ were applied to the HAS or normal artery and three measurements were taken from each artery at similar locations for each patient; natural anatomical and scan parameter variability meant that measurement location could not always be identically reproduced, though this was attempted as near as possible.

Observer reliability

We tested the interobserver reliability of angiographic image analysis by inviting the expert panel to read the same 10 angiograms using the SIRS2 blind to all other data including their initial analysis. We tested observer reliability of the perfusion imaging by inviting as many raters as possible to rate 20 perfusion images using the modified SIRS2 system that was able to handle colour images and to view two image modalities from the same acquisition time point (e.g. a perfusion and a structural CT image) side by side (SIRS2: sirs2/neuroimage.co.uk/sirs2). These analyses are ongoing at the time of writing.

Sample size

The IST-3 perfusion analysis aimed to examine primarily whether or not rt-PA improves functional outcome more in those with, than without, tissue at risk and secondarily reduced infarct growth. Based on systematic reviews of all available data34 and recent studies,27,30 we estimated that 60% will have mismatch overall;27 70% with mismatch will have infarct growth compared with 30% without mismatch; and rt-PA will reduce infarct growth by 20% in those with, but not those without, mismatch. 34 At 80% power and α = 0.05, a sample of 100 patients would detect a 27% difference in infarct growth, with rt-PA compared without rt-PA, in the presence of mismatch compared with the absence of mismatch; 160 patients would detect a 20% difference in infarct growth; and 400 patients would detect a 15% difference in infarct growth. We acknowledged that, with, at most, 300 patients in the perfusion study, the perfusion study would be underpowered to detect a ‘mismatch × treatment effect’ interaction on the primary clinical outcome. We therefore selected infarct growth as an outcome for the perfusion analysis (as in EPITHET)30 to increase statistical power.

The centres that were using perfusion imaging were among the most active in IST-3. Therefore, we estimated that in 3 years, in up to 15 active centres recruiting between four and eight patients per year each, a total of between 100 and 300 patients with baseline perfusion and or angiography data would be recruited. We estimated that approximately two patients would have CTA/CTP for every one with MRA/MRP. However, that may change as more centres are now acquiring CT perfusion equipment, and so the proportion may end up being nearer to four patients having CTA/CTP for every one with MRA/MRP.

Statistical methods

We first compared imaging variables with each other, then with clinical features and clinical outcomes, and then tested for interactions between imaging variables and rt-PA effects. Thus, we assessed:

• variation in the size of perfusion lesions and proportion with mismatch for each perfusion parameter tested

• associations between clinical and structural imaging variables at baseline, perfusion lesion extent and presence/absence of angiography lesions

• associations between baseline perfusion or angiography imaging variables and subsequent infarct growth, swelling and haemorrhagic transformation on follow-up scanning

• associations between baseline perfusion and angiography lesions and 6-month functional outcome

• test for an interaction between treatment with rt-PA and perfusion lesion extent, presence or absence of mismatch, angiographic arterial occlusion and SICH and 6-month functional outcome.

All analyses were unadjusted and adjusted for key baseline variables using an established prognostic model determined in the IST-3 main trial analysis. 59 We also performed ordinal analysis as this increases the statistical power. 68,88

Secondly, we also compared quantitative perfusion lesion volume with qualitative visual perfusion lesion assessment as coded by the ASPECTS; different perfusion processing algorithms [in this case, the in-house software and MiStar (Apollo Medical Imaging Technology Pty. Ltd, Melbourne, VIC, Australia)]; and test if relative (i.e. to the contralateral hemisphere) parameters are more consistent than quantitative parameters between different software, by comparing (a) the measured volumes of different perfusion parameter lesions, that is mm³, and (b) also by taking account of geometric concordance.

Statistical analyses for the CTA data presented here were performed with Statistical Product and Service Solutions (SPSS) software (v. 20, IBM, New York, NY, USA). Chi-squared testing was used for comparisons between dichotomous data. Simple t-tests were employed to compare normally distributed continuous and dichotomous data, while Mann–Whitney U-testing was employed where continuous or ordinal data were skewed (ASPECT and clot burden scores, HAS length). Similarly, both Pearson and Spearman’s rank-order correlations were applied as appropriate. Significance was taken as p < 0.05.

Any changes to protocol

There were two minor changes to process rather than to fundamental study design.

1. We originally planned to analyse infarct growth as the primary outcome and functional status, with death and SICH as secondary outcomes. However, in view of the clinical importance of functional outcome, and because infarct growth is less clinically relevant to patients, we reordered the primary outcome to be clinical and the secondary outcome to be infarct growth. Additionally, infarct growth can be assessed only in patients with visible infarction – those without a visible infarct do not contribute to this analysis, leading to distorted and potentially misleading results. Hence we focused on the influence of baseline perfusion imaging on clinical outcomes.

2. At the time of original submission, perfusion imaging was thought to be the more important advanced imaging modality to test in stroke and, hence, the focus of planned analysis was on perfusion imaging, which draws heavily on centralised computational analysis. However, in the 4 years since the original submission, angiographic imaging has come into prominence in stroke, and indeed we received far more angiography images than originally expected, almost three times as many as we received of perfusion images. The original planned analysis had been set up for perfusion imaging; angiographic imaging analysis is largely visual and so required a completely different approach. In the event, in order to cope with the number of angiography data efficiently, we had to redesign a visual web-based image viewing and data recording tool (SIRS2) to handle plain scan and angiographic images and then identify several expert neuroradiologists to assist with reading the angiograms. This took extra time, and hence the completion of the angiographic imaging analysis has been delayed. We were, however, fortunate to attract a senior neuroradiology trainee to the project who has been assisting by reading the CT angiograms and measuring thrombus density on a workstation. The results of this latter analysis are included in the report. However, to avoid biasing the analyses, the observers are all still blinded to treatment allocation and the final unblinded analysis has not yet been performed. The unblinded analysis will be presented to the investigators before being presented in public or submitted for publication, as with the perfusion imaging results, and as is proper in clinical trials.

The costs of the programming to redesign the SIRS2 web-based scan-viewing system, the time of the additional neuroradiology expert readers and the neuroradiology trainee’s time to undertake the work on the angiography is all outside the Efficacy and Mechanism Evaluation (EME) funding provided for the original project. There were no other changes.

Patient and public involvement

The IST-3 trial was designed with input from focus group discussions with stroke patients and their carers in the late 1990s. 89 A lay representative was on the IST-3 steering committee and also contributed to the discussions on design of the perfusion and angiography substudy. The lay representative also contributed to the writing of the main trial primary results paper60 and accompanying systematic review. 8 Her input will be sought on all publications arising from the perfusion and angiography substudy.

Baseline clinical and plan scan data

The baseline demographic data of the patients recruited with perfusion or angiography imaging compared with patients recruited without perfusion or angiography imaging are given in Table 3.

The median age of the 141 patients randomised in IST-3 with perfusion imaging was 81.0 years, the NIHSS was 11.0, the median time to randomisation was 3.9 hours and 48% were male. These data were identical for the 2986 patients randomised in IST-3 without perfusion imaging.

Of the patients with angiography imaging, we will focus on those with CTA at randomisation as there were many more with CTA than with MRA. The median age of the 271 patients with CTA at randomisation was 81 years [interquartile range (IQR) 71–85 years], the youngest patient was 18 years and the oldest was 102 years; 41.5% were male. The stroke syndrome was TACI in 34%, PACI in 39%, LACI in 10% and POCI in 17%. The median time to pre-randomisation CT (taken from the CT scan) was 2.8 hours (IQR 1.8–4.2 hours), minimum 0.5 hours and maximum 5.4 hours. AF was noted in 61 out of 234 (26.1%) patients at admission.

We were concerned that patients randomised in IST-3 with perfusion or angiography imaging would be different to those randomised with a plain CT or MR scan. However, the only difference in patients with perfusion imaging (but not those with angiography) was that the randomising clinician thought that more patients had a possible or definite visible ischaemic lesion on structural imaging (63% vs. 40%, respectively; p < 0.0001) than patients without perfusion imaging. However, the blinded central expert panel image readings (which were performed without knowledge of the perfusion imaging) showed no difference in the proportion of patients with visible infarction at randomisation (41% vs. 35% had visible infarction on structural scanning with vs. without perfusion imaging; p = not significant). Among the patients randomised with angiography, there was no difference in the proportion with visible infarction according to either the randomising clinician (46% vs. 40%) or the expert panel. Otherwise, there was no difference in age, NIHSS, proportion with AF, predicted outcome, or in any other variables. The blinded expert scan readers did not have access to the perfusion and angiography imaging. This illustrates the importance of separating the perfusion/angiography images from the structural image interpretation when trying to determine the true additional contribution of the perfusion and angiography.

Numbers analysed

We received perfusion imaging on 151 patients in total (see Figure 3). Of these, 10 had perfusion imaging performed post randomisation only. In 21 it was not possible to process the image data, mainly due to incomplete image acquisition, leaving 123 with perfusion imaging at randomisation and that were rated visually. Of these, in 16 patients we did not receive ‘raw’ perfusion data, only the ‘already processed’ screen capture images created on the scanner where the images were obtained, on which it was not possible to measure lesion volume, although it was possible to perform visual scoring. Hence there were more visual readings than volume measures. Additionally, the ‘already processed’ images tended to have fewer perfusion parameters and, therefore, the full list of perfusion parameters was incomplete for some patients. The majority were CT perfusion imaging. Details of the data available for volume measures are given in Figure 3.

Perfusion parameters variation in perfusion lesion size

We compared CBVq, CBFq, Tmaxq and MTTq lesions at randomisation using ASPECTS and relative to the plain-scan acute ischaemic lesion. Lesion size was smallest for CBVq; CBVq was significantly smaller than CBFq (p < 0.000), which was the same as Tmaxq, which was significantly smaller than MTTq (p < 0.002) (Figure 6). We found similar lesion size variation when the perfusion lesion was expressed in terms of ‘mismatch’ with respect to the plain-scan lesion size. On Tmaxq, 53 out of 116 (46%) patients had mismatch (perfusion lesion 20% larger by visual estimation than the plain-scan lesion).

FIGURE 6.

Perfusion lesion size (a) on ASPECTSs scores by different parameters: 8–10 is no or small lesion, 0–3 is large lesion; (b) relative to the structural imaging lesion by different perfusion parameters.

Perfusion parameters and plain scan findings

The acute ischaemic lesion on the plain scan was not significantly different in size to the CBVq perfusion lesion, but was significantly smaller than the CBFq, Tmaxq and MTTq lesion sizes (all p < 0.0000, t-tests). For example, the CBFq lesion had, on average, an ASPECTS that was 2.1 [standard deviation (SD) 3.4] points larger than the plain scan (p < 0.000); the MTTq lesion ASPECTS was 2.7 (SD 3.6) points larger than the plain-scan lesion (p < 0.000).

Perfusion parameters and baseline clinical features

Older patients had larger perfusion lesions on all perfusion parameter maps, and more often had perfusion–plain scan lesion mismatch (Figure 7). For example, in patients aged > 80 years, 67% had mismatch on Tmaxq compared with 41% of patients aged ≤ 80 years (p < 0.04). ASPECTS were lower (i.e. larger lesion) for CBFq and Tmaxq in patients aged >80 years compared with ≤ 80 years (p < 0.05).

FIGURE 7.

Perfusion lesion and plain scan mismatch extent by age ≤ 80 years vs. > 80 years. Error bars represent 95% CIs. Older patients had larger perfusion lesions with more perfusion–plain scan mismatch: CBFq, Tmaxq; p = 0.04. They also had lower ASPECTS: CBFq, Tmaxq; p < 0.05.

Patients scanned within 3 hours of stroke had larger perfusion lesions (lower ASPECTS) and more often had perfusion–plain-scan lesion mismatch than did patients scanned between 3 and 6 hours after stroke, which was significant for CBFq (ASPECTS < 3 hours, 5.4; 4.5–6 hours, 7.7; p < 0.05) (Figure 8).

FIGURE 8.

Perfusion lesion: mismatch and ASPECTS by time to randomisation 0–3 hours vs. 3–6 hours. Error bars represent 95% CIs. Larger perfusion imaging lesions at 0–3 hours than at 4.5–6 hours: more mismatch; lower ASPECTSs e.g. for CBFq < 3 hours, 5.4; 4.5–6 hours, 7.7; p < 0.05.

There was a strong correlation between increasing stroke severity as measured by NIHSS score and larger perfusion lesions on all perfusion parameters, according to both the ASPECTS and perfusion–plain scan mismatch (Table 4). The Spearman rank-order correlation coefficients ranged from 0.54 to 0.58 (all p < 0.000) for ASPECTS and 0.38 to 0.48 (all p < 0.000) for mismatch.

Patients with higher NIHSS scores were also more likely to have mismatch (Figure 9); 72% of patients with NIHSS ≥ 15 had mismatch on Tmaxq or MTTq compared with about 45% of those with NIHSS < 15.

FIGURE 9.

Proportion of patients with perfusion–plain scan mismatch by NIHSS score. Error bars represent 95% CIs.

Perfusion parameters and symptomatic intracerebral haemorrhage, death and functional outcome

Larger perfusion lesions were associated with worse functional outcomes (Table 5). Patients with perfusion lesion–plain-scan mismatch on CBFq, Tmaxq or MTTq were less likely to be alive and independent at 6 months, all significant at the p < 0.05 level on unadjusted analyses; after adjusting for baseline prognosis, only CBFq and MTTq mismatch was significantly associated with worse outcome. Similarly, larger perfusion lesion ASPECTSs were associated significantly with poor functional outcome, for CBVq, CBFq, Tmaxq and MTTq on unadjusted analyses (all p < 0.001); on adjusted analyses, all parameters except MTTq retained their significance. The odds of being alive and independent decreased by about 20% for every point worsening of the ASPECTS, significant for CBVq and Tmaxq.

More patients with larger perfusion lesion ASPECTSs on all parameters except CBFq were dead at 6 months (p < 0.05 unadjusted), although not on adjusted analyses. There were more patients with SICH and who died within the first 7 days after stroke with larger perfusion lesions on all parameters but these differences were not statistically significant. Similar associations were seen for perfusion lesion size expressed as the perfusion–plain-scan mismatch.

Treatment interaction

We examined for any evidence that the effect of rt-PA was different in the presence of a perfusion lesion mismatch or with increasing perfusion lesion size by testing for an interaction between the perfusion lesion size and rt-PA on the proportion of patients with SICH within 7 days, who had died, or who were alive and independent at 6 months on ordinal analysis (Table 6). We found no evidence that rt-PA effects differed in patients with or without mismatch, or who had large or small perfusion lesions by ASPECTSs on any perfusion parameter. This was true for OHS 0–1, 0–2, SICH, death within 7 days or death within 6 months.

Ancillary analyses

Lesion volumes were measurable in 56 out of 103 patients with perfusion data that could be processed centrally. Additionally, infarct size was measured on the post-randomisation follow-up scan in 63 patients with a visible lesion. Figure 10 shows the breakdown of patients by lesion presence for analysis. These data are still being analysed.

FIGURE 10.

Perfusion imaging flow chart showing numbers for computational analysis, lesion presence or absence. ROI, region of interest.

Hyperdense artery sign and clinical findings

Among these 271 patients with CTA at randomisation, a recent acute ischaemic lesion was visible in 74 (27%, Table 7), most in the MCA territory (67 out of 74, 91%, Table 8). The median ASPECTS on these initial scans was 10 (IQR 9–10).

On plain CT, 69 out of 271 patients (25.5%) had a HAS indicating arterial thrombus. Hyperdensity involved a MCA main stem in 44 cases (64%) and had a mean length of 17.5 mm (SD  8.9 mm). The next most frequent site was the MCA sylvian branch (27%). The mean density within these hyperdense vessels was 51.0 HU (SD  8 HU). This compares with mean densities of non-hyperdense arteries of 40.1 HU (SD 5.6 HU), 40.3 HU (SD  7.0 HU) and 39.2 HU (SD  7.1 HU) for the basilar, left and right MCAs, respectively (these differences are significant, with p < 0.001 in all cases). The density ratio of normal to abnormal vessel was 1.4 in those with a HAS in one artery compared with 1.0 in those without any HAS (p < 0.001).

Patients with a HAS were more likely to have acute ischaemia on the plain CT scan (χ² = 67; p < 0.001); acute ischaemia was found at the higher-than-background rate of 65.2%.

Patients with a HAS were significantly more likely to be female (χ² = 4.9; p = 0.028) and have a higher NIHSS score at presentation (mean difference 7.2 points; p < 0.001) than those without a HAS. In addition, patients with a HAS were more likely to have a TACI or PACI syndrome (χ² = 33.3; p < 0.001) with nearly two-thirds of all TACI being associated with a HAS. There was no difference in age, presence of AF, hypertension, previous stroke or time to CT between those with or without a HAS. Neither HAS length, PMD score (or a combination of the two) nor HAS density were related to NIHSS, time to CT, or the presence of AF.

Computed tomography angiography abnormalities, plain computed tomography and clinical findings

There were 113 out of 253 patients (41.7%) with an abnormality on CTA at randomisation (see Table 8). The MCA main stem was most frequently affected (53%), followed by a MCA main stem plus sylvian branch (30%) or a sylvian branch alone (17%). Among patients with an abnormal artery, the TIMI scores were equally spread across degrees of obstruction from complete to sight. Collateral status was defined as ‘good’ (32.5%), ‘moderate’ (37.3%) or ‘poor’ (30.1%). A ‘poor’ collateral supply was associated with lower initial ASPECT scores (p = 0.014) and an increased likelihood of acute ischaemia on the initial scan (χ² = 4.8; p = 0.028). Similar trends were demonstrated for ischaemic change on follow-up scans but these changes were not significant.

Patients with an abnormal CTA were more likely to have a visible infarct on plain CT at randomisation (47%, χ² = 50.6; p < 0.001). Patients with an abnormal CTA were significantly more likely to be female (χ² = 7.0; p = 0.008), older (median 82 years vs. 78 years without abnormal CTA, mean difference 4.0 years; p < 0.001), have a higher NIHSS score (16 vs. 6, mean difference 10; p < 0.001) and be scanned earlier after stroke (mean difference 0.49 hours; p = 0.005) than those with a normal CTA. Patients with abnormal CTA were more likely to have a TACI or PACI clinical syndrome (χ² = 58.0; p < 0.001), 59% of all TACIs being associated with an abnormal CTA. AF was associated with abnormal CTA: 50% of patients in AF (32 out of 61) had an abnormal CTA, compared with around 36% (62 out of 173) of those in sinus rhythm (χ² = 3.8; p = 0.05).

Clot burden score (where 10 indicates no thrombus and 0 indicates all major intracranial arteries and their branches are thrombosed) showed a strong inverse relationship with NIHSS (–0.62; p < 0.001). Clot burden score was also associated with faster time to CT (0.18; p = 0.008). There was a trend towards a similar association with CTA PMD score but this was not significant. Patients with AF compared with those without had more thrombus (lower clot burden score; mean difference −0.72; p = 0.007). Collateral status was not related to NIHSS score.

Hyperdense artery sign compared with computed tomography angiography

The presence of a HAS was strongly and significantly related to the finding of an abnormal CTA (χ² = 80.3; p < 0.001) (see Table 8). The location of the HAS and CTA abnormality was the same in 63 patients (major vessel involved matched but CTA was more sensitive at detecting thrombus extending distally into small branches); this represents 96.3% of those with a HAS (high specificity) and 69.3% of the CTA abnormalities identified (moderate sensitivity). There were six false-positive HASs (i.e. HAS not associated with any CTA abnormality); despite the appearance of increased attenuation, the measured arterial attenuation in these segments was not significantly different to that within normal vessels (mean 38.3, SD  10.2). A significant inverse relationship of −0.41 (p = 0.001) was demonstrated between the length of the HAS and the clot burden score as calculated from angiography.

Abnormal arteries and follow-up plain scan findings

Follow-up imaging demonstrated a recent infarct in 192 patients (71%), most commonly in the MCA territory (80%), more than double the visible infarction rate on the initial scans. The median ASPECTS on follow-up imaging was 9 (IQR 6–10). Patients with a HAS on the randomisation CT scan were more likely to have ischaemia on follow-up CT (χ² = 31; p < 0.001), as were those with an abnormal CTA at randomisation (χ² = 49.0; p < 0.001). All patients with a HAS at randomisation had a visible infarct on the follow-up CT; in 9% of patients, an abnormal CTA was not related to ischaemia on follow-up. Those with a HAS had significantly different IST-3 and ASPECTs on both pre randomisation and follow-up imaging. The median IST-3 scores were 1 at pre randomisation and 3 on follow-up imaging for those with a HAS compared with 0 and 1 in those patients without a HAS (p < 0.001 in both cases). Similarly, the median ASPECTs for those with a HAS were 9 at pre randomisation and 5 at follow-up, compared with 10 and 9 for those without a HAS (p < 0.001 in both cases). Those with a HAS were more likely to undergo a change in IST-3 ischaemia score between pre randomisation and follow-up CT and for that change to be greater than in those without a HAS. Similar results were obtained if this analysis was performed using change in ASPECTS.

There was complete clearance of HAS in 14 cases (23.0%) at follow-up, while three patients developed a new HAS between randomisation and follow-up imaging. Taking all patients with a HAS at either time point, the relationship between time and density of HAS was highly significant, with a correlation coefficient of −0.36 (p < 0.001) (Figure 11).

FIGURE 11.

Relationship between time of scan after stroke and mean density in HU of the hyperattenuated artery sign (HAS).

Haemorrhagic transformation was present in 46 out of 271 patients (17%), mostly small areas of petechial haemorrhage (64%) unlikely to be associated with neurological deterioration. More pronounced haemorrhagic transformation (larger haematoma in infarct) was less common (12 patients, 26%).

The presence of a HAS was not associated with SICH. HAS length was, however, found to be greater in those patients with SICH (mean difference 10.2 mm; p = 0.048). This relationship was improved with the inclusion of PMD data (mean difference 33.9; p = 0.012). The density of HAS or the hourly change in this density were not related to SICH. Neither an abnormal CTA, the clot burden score or collateral status was associated with SICH.

We tested the effect of plain CT HAS + visible ischaemic change and CTA (individually or in combination: either, or, both) in predicting a visible infarct on follow-up CT in the 271 patients. The randomisation CT was abnormal (acute ischaemia or HAS) in 37% (95% CI 31% to 43%); CTA was abnormal in 40% (95% CI 34% to 47%); either randomisation CT or CTA was abnormal in 50% (95% CI 43% to 56%); both were abnormal in 27% (95% CI 22% to 33%). The sensitivity and specificity for infarct on follow-up CT were (respectively) of abnormal randomisation CT alone, 57% and 96%; of abnormal CTA alone, 55% and 91%; of both pre-randomisation CT and CTA, 44% and 100% (compared with pre-randomisation CT alone χ² = 22; p < 0.001); and of either pre-randomisation CT or CTA, 71% and 87% (compared with pre-randomisation CT alone χ² = 27; p < 0.001) (Figure 12). Thus, combining pre-randomisation CT with CTA in acute stroke significantly increases sensitivity (if either non-contrast CT or CTA are abnormal) or specificity (if both are abnormal) for predicting infarct on 48-hour follow-up CT.

FIGURE 12.

Graph comparing tests with differing sensitivities and specificities. The value of an abnormal non-contrast (plain) CT in predicting infarct on follow-up CT is plotted (dark green marker). Solid lines connect this marker to points 1,1 and 0,0, thereby creating likelihood ratios (represented by the slope of the lines) for the sensitivity and specificity of plain CT. The addition of CT angiography to plain CT in the acute setting improves either sensitivity or specificity but not both. NCCT, non-contrast (i.e. plain) CT.

Hyperdense artery, computed tomography angiography and clinical outcomes

At the end of 6 months, 55 out of 271 patients had died (20%), 93 were dependent (34.3%) and 123 (45.5%) were alive and independent (OHS 0–2). Patients with visible ischaemic change on the randomisation CT were less likely to be alive and independent (mean difference in OHS 1.3; p < 0.001). Similarly, significant inverse correlations were identified between 6-month OHS and ASPECTS (r = −0.29; p < 0.001).

The presence of a HAS was significantly associated with death at 6 months (χ² = 15.96; p < 0.001). There was no significant association between length of HAS or density of HAS and death. An abnormal CT angiogram was significantly associated with death at 6 months (χ² = 28.79; p < 0.001). Clot burden scores were significantly lower in those patients who died (mean difference 1.9; p < 0.001). Collateral supply score was not associated with death.

A hyperdense artery was associated with worsening OHS (mean difference 1.7; p < 0.001). There was no significant association between length of HAS or density of HAS and OHS. An abnormal CTA was associated with an increase in OHS (mean difference 2.1; p < 0.001). The clot burden score was significantly inversely correlated with OHS (r = −0.53; p < 0.001). Collateral status was not related to OHS.

Multiple linear regression modelling to assess the predictive value of acute CT signs including visible ischaemic change, the HAS and an abnormal CTA on death and dependency showed that age and either a HAS or an abnormal CTA were independently predictive of outcome (Table 9) but not visible acute ischaemic change on CT. There was no evidence of collinearity. Abnormal pre-randomisation CT in isolation related to worse NIHSS score (seven points higher; p < 0.001) and a greater likelihood of OHS 3–6 (χ² = 20; p < 0.001). The combined effect of abnormal pre-randomisation CT ± CTA provided very similar results: NIHSS score was eight points higher (p < 0.001) with an increased rate of OHS 3–6 (χ² = 29; p < 0.001). Neither NIHSS score nor rate of OHS 3–6 was significantly different between the pre-randomisation CT and pre-randomisation CT ± CTA groups (p = 1.000 in both cases). Thus, including CTA in the imaging assessment of acute stroke improves diagnosis by identifying more patients with changes of the acute ischaemic stroke on imaging but does not translate into better prediction of prognosis over simple clinical variables (age, NIHSS score) and plain CT findings alone.

Interaction between computed tomography angiography findings and recombinant tissue plasminogen activator treatment effect

Of the 271 patients with pre-randomisation CTA, 142 patients were randomly assigned rt-PA and 129 to control. The odds ratios (ORs) for the effect of rt-PA on early and late outcomes, in the presence compared with absence of occlusion on CTA, were symptomatic haemorrhage 1.41 (95% CI 0.86 to 2.31), death 1.23 (95% CI 0.70 to 2.17), and independent survival (OHS 0–2) 1.39 (95% CI 0.59 to 3.27). This indicates no significant interaction between rt-PA and the presence or absence of occlusion on CTA.

Additional analysis of hyperdense artery sign and recombinant tissue plasminogen activator effect

We assessed the effect of rt-PA in patients with or without the HAS in all patients in IST-3 who had a plain CT scan at randomisation and for follow-up (n = 2730). Some patients who had MR at one or other time instead of CT were excluded from this analysis and randomisation CT scans for 19 out of 3035 patients were not received in the central trials office. There were 674 patients randomised in IST-3 with a HAS (24.7% of 2730) and 2056 without a HAS (75.3% of 2730). The clinical and imaging features and associations with SICH and 6-month outcomes were the same for the whole trial as for the subset with angiography imaging reported above.

We will, therefore, not repeat those findings here. Patients allocated to rt-PA were more likely to have the HAS shrink or disappear between randomisation and follow-up scans (OR 1.53; p = 0.011 and OR 1.49; p = 0.010 respectively) and a trend towards lower likelihood of new HAS formation (OR 1.25; p = 0.141). An analysis of the interaction between rt-PA and presence or absence of a HAS on 6-month functional outcome in all IST-3 patients using adjusted ordinal regression analysis is ongoing.

Ongoing analyses

Further analyses pending are as follows.

Third International Stroke Trial Perfusion and Angiography Imaging Study

The EME programme is funded by the Medical Research Council (MRC) and NIHR, with contributions from the Chief Scientist Office (CSO) in Scotland, National Institute for Social Care and Health Research (NISCHR) in Wales and the Health and Social Care Research and Development (HSC R&D), Public Health Agency in Northern Ireland, and is managed by the NIHR. The views and opinions expressed therein are those of the authors and do not necessarily reflect those of the funding agencies or UK Department of Health.

Third International Stroke Trial: main trial

The start-up phase was supported by a grant from the Stroke Association, UK. The expansion phase was funded by the Health Foundation UK. The main phase of the trial is funded by UK MRC (grant numbers G0400069 and EME 09–800–15) and managed by the NIHR on behalf of the MRC–NIHR partnership; the Research Council of Norway; Arbetsmarknadens Partners Forsakringsbolag (AFA) Insurances Sweden; the Swedish Heart Lung Fund; The Foundation of Marianne and Marcus Wallenberg, Stockholm County Council; Karolinska Institute Joint ALF-project grants Sweden, the Polish Ministry of Science and Education (grant number 2PO5B10928); the Australian Heart Foundation; Australian National Health and Medical Research Council; the Swiss National Research Foundation; the Swiss Heart Foundation; the Foundation for Health and Cardio-/Neurovascular Research, Basel, Switzerland; the Assessorato alla Sanita, Regione dell’Umbria, Italy; and Danube University, Krems, Austria. Boehringer-Ingelheim GmbH donated drug and placebo for the 300 patients in the double-blind phase, but thereafter had no role whatsoever in the trial. The UK Stroke Research Network (SRN study ID 2135) adopted the trial on 1 May 2006, supported the initiation of new UK sites, and in some centres, and, after that date, data collection was undertaken by staff funded by the network or working for associated NHS organisations. IST-3 gratefully acknowledges the extensive support of the NIHR Stroke Research Network, NHS Research Scotland, through the Scottish SRN, and the National Institute for Social Care and Health Research Clinical Research Centre. The central imaging work was undertaken at the Brain Imaging Research Centre (www.bric.ed.ac.uk), a member of the Scottish Imaging Network – A Platform for Scientific Excellence (SINAPSE) collaboration (www.sinapse.ac.uk), at the Division of Clinical Neurosciences, University of Edinburgh. SINAPSE is funded by the Scottish Funding Council and the CSO of the Scottish Executive. Additional support was received from Chest Heart and Stroke Scotland, DesAcc, University of Edinburgh, Danderyd Hospital Research and Development Department, Karolinska Institutet, Oslo University Hospital, and the Dalhousie University Internal Medicine Research Fund.

(i) Magnetic resonance perfusion

(ii) Computed tomography perfusion

(i) Magnetic resonance angiography

(ii) Computed tomography angiography

Digital Imaging and Communications in Medicine storage and conversion

Subjects in the main arm of IST-3 who have been identified as having perfusion images are transferred to the perfusion substudy file system by DICOM sent to a receiver running on a server attached to a high-performance computing facility. The receiver is implemented using DICOM confidential104 and stores the received images in standard DICOM format as well as cataloguing the images in a database. The cataloguing process records the unique identifiers which reference the image data, the dates and times of the imaging studies as well as information specific to the series modality such as echo time or tube voltage and current. The catalogue makes this information available to other processing steps; it is also used to identify structural and perfusion-imaging series in the studies labelled as R (randomisation) and P (post) time points.

The next step in the initial conversion of the DICOM data is reconstruction. After relevant structural and perfusion series have been identified, each image in the series is composited into a 3D volume; in the case of perfusion series the individual 3D volumes are joined in acquisition order to create a 4D ‘volume’. This composition is carried out by the widely used utility DCM2NII and the final data are stored in NIFTI format for subsequent processing and analysis. In all of the data reconstructed, the actual acquisition order of the data is only represented at the volume level and not the individual slice level, often referred to as slice timing. The slice timing has been shown to be important in the processing of MR perfusion data, especially in the case of the parameter Tmax;75 however, no account could be taken of this in such a chronologically and geographically diverse data set.

Contrast concentration estimation

The first step in the processing is to convert the signal time series within each voxel into a time series proportional to contrast concentration using the relevant relationship between signal and contrast concentration for either MR or CT. Additionally, in MR, the relationship between signal and concentration is non-linear108,109 and a modulation transfer function is applied, as described in Straka et al. ,80 after applying the usual formula. In both cases, this conversion relies upon estimating the mean signal in each voxel prior to the arrival of contrast, and in MR data the first three time points are disposed of to ensure the signal has reached a steady state.

Discretisation

Discretisation is the process used to convert the equations describing the distribution of contrast (perfusion) into a form suitable for solving using the quantities observed in the contrast concentration time series. The first step is defining a quantity referred to as the arterial input function (AIF). Owing to the diverse sources of data, the different perfusion series had no standard anatomical coverage, and therefore the AIF could not be placed in a consistent location. Consequently, the AIF was placed manually in a location adjacent to a vessel with early enhancement where possible in the case of MR, and in CT series which included it, this was in the contralateral middle cerebral artery; in CT series with limited coverage, the anterior cerebral artery was chosen. Locations for two additional time series were also defined, one for the venous out flow (VOF) which was placed in the superior sagittal sinus and another in a region of what was assumed to be normal white matter in the contralateral hemisphere often close to the anterior horns of the ventricles. To mitigate for possible partial volume effects in estimating the AIF, the area under the AIF as adjusted to match that of the VOF by multiplication with a scalar. The normal white matter time series was used as a means to compare data from different sources.

As previously stated, the method used to convert the AIF and voxel time series values into a form suitable for numerical solution was first applied to perfusion imaging by Wu et al. 105 The method uses a set of simultaneous equations expressed as a matrix equation to represent the convolution integral, deconvolution is achieved by inverting the matrix and solving for the vector of unknown quantities. The discretisation is different from other alternative schemes in that the signals are treated as being periodic; this is achieved by using a matrix with a special structure referred to as a circulant Toeplitz matrix.

The individual values of the AIF are used to provide the elements of this matrix, where from left to right each column is a shifted copy of the previous column. Being circulant means that as elements drop off the last row of the matrix they reappear in the first row of the next column. When populated in this way, the first column of the circulant matrix contains the unshifted vector of AIF values in the correct temporal order, whereas in the last row the temporal order is reversed. As copies of a deterministic signal corrupted by noise, the individual elements of a matrix populated in this way are very far from independent and their true values are unknown. Consequently, it can be expect that obtaining the inverse of such a matrix will be problematic. It is for this reason that regularisation is applied to the deconvolution used to obtain the parameter estimates. The advantage of the block circulant method is that the estimated quantities are less sensitive to delay between the AIF and the voxel concentration time series.

Parameter estimation

The parameters of interest were recovered, as follows, from the observed quantities and the residue function obtained from deconvolution of the arterial and voxel time series on a per-voxel basis. The CBV was defined as the ratio of the areas under the voxel and arterial concentration time courses. The CBF and Tmax were defined as the peak value of the residue function and the time at which it occurred. The MTT was defined as the ratio of CBV to CBF.

The bolus arrival time (AT), peak time (PT) and the difference of the two, time to peak (TTP) from bolus arrival as well as the maximum value of contrast concentration (Cmax), etc., were obtained for each voxel as follows. A reconstructed contrast concentration time series was formed by convolving the AIF with the estimated residue function on a per-voxel basis. Owing to the regularisation applied in the deconvolution, the time series formed in this way is much smoother than the original data, and, therefore, initial estimates of parameters such as AT, Cmax, etc., obtained directly from it, are less affected by noise than would be the case if they were taken from the original time series. These initial estimates are then used to obtain the starting parameters for fitting a heuristic model to the reconstructed time series. 110 The parameters of the fitted model are then used to provide the estimates of AT, PT, TTP, etc., used to create the parametric maps.

Parametric map storage

The parametric maps were stored in NIFTI format as single precision floating point values, with each voxel value equal to the parameter value for that voxel, that is to say with out any scaling. The CBV was stored in units of ml/100 g by assuming a fixed value for the density of brain tissue; the CBF was stored in units of ml/100 g/minute and the MTT in seconds. The other parameters either have units of seconds or arbitrary units (e.g. Cmax and rCBF).

Computed tomography image interpretation form

(PDF download)

Magnetic resonance image interpretation form

(PDF download)