Magnetic resonance imaging versus computed tomography for detection of acute vascular lesions in patients presenting with stroke symptoms

Abstract

Background

Magnetic resonance imaging (MRI) is increasingly used for the diagnosis of acute ischaemic stroke but its sensitivity for the early detection of intracerebral haemorrhage has been debated. Computed tomography (CT) is extensively used in the clinical management of acute stroke, especially for the rapid exclusion of intracerebral haemorrhage.

Objectives

To compare the diagnostic accuracy of diffusion‐weighted MRI (DWI) and CT for acute ischaemic stroke, and to estimate the diagnostic accuracy of MRI for acute haemorrhagic stroke.

Search methods

We searched MEDLINE and EMBASE (January 1995 to March 2009) and perused bibliographies of relevant studies for additional references.

Selection criteria

We selected studies that either compared DWI and CT in the same patients for detection of ischaemic stroke or examined the utility of MRI for detection of haemorrhagic stroke, had imaging performed within 12 hours of stroke onset, and presented sufficient data to allow construction of contingency tables.

Data collection and analysis

Three authors independently extracted data on study characteristics and measures of accuracy. We assessed data on ischaemic stroke using random‐effects and fixed‐effect meta‐analyses.

Main results

Eight studies with a total of 308 participants met our inclusion criteria. Seven studies contributed to the assessment of ischaemic stroke and two studies to the assessment of haemorrhagic stroke. The spectrum of patients was relatively narrow in all studies, sample sizes were small, there was substantial incorporation bias, and blinding procedures were often incomplete. Amongst the patients subsequently confirmed to have acute ischaemic stroke (161/226), the summary estimates for DWI were: sensitivity 0.99 (95% CI 0.23 to 1.00), specificity 0.92 (95% CI 0.83 to 0.97). The summary estimates for CT were: sensitivity 0.39 (95% CI 0.16 to 0.69), specificity 1.00 (95% CI 0.94 to 1.00). The two studies on haemorrhagic stroke reported high estimates for diffusion‐weighted and gradient‐echo sequences but had inconsistent reference standards. We did not calculate overall estimates for these two studies. We were not able to assess practicality or cost‐effectiveness issues.

Authors' conclusions

DWI appears to be more sensitive than CT for the early detection of ischaemic stroke in highly selected patients. However, the variability in the quality of included studies and the presence of spectrum and incorporation biases render the reliability and generalisability of observed results questionable. Further well‐designed studies without methodological biases, in more representative patient samples, with practicality and cost estimates are now needed to determine which patients should undergo MRI and which CT in suspected acute stroke.

Plain language summary

[Plain language title]

[Summary text]

Summary of findings

Summary of findings'. 'Results of studies on ischaemic stroke.

Review question: Comparison of diffusion‐weighted magnetic resonance imaging with conventional computer tomography for the early detection of ischaemic brain lesions in patients suspected of stroke Patient population: adults suspected of acute stroke Setting: hospital departments Geographical location: studies were conducted in Europe (3 studies), the USA (3 studies), and in Australia (1 study) Index test: diffusion‐weighted magnetic resonance imaging (DWI) performed within 12 hours of stroke onset Alternative test: computer tomography (CT) performed within 12 hours of stroke onset Reference standard: clinical assessment and imaging follow up Included studies: 7 comparative studies that evaluated DWI and CT in the same patients Total number of patients assessed: 226
Limitations of included studies Limited number of included studies (7 studies); small sample sizes; presence of incorporation bias DWI and CT were evaluated in highly selected patient samples (patients with high probability of stroke), which therefore are not representative of the typical population of patients presenting with 'suspected acute stroke' to an emergency department (poor generalisability of results) The stroke vascular territory was not reported in the majority of included studies although it is likely that they enrolled patients with typical anterior circulation stroke Only a minority of the studied patients had severe strokes (in whom DWI might be contraindicated) The high proportion of mild strokes and reclassification of TIA cases with a positive DWI lesion as strokes might have inflated the DWI sensitivity estimate In most of the studies stroke mimics were not included In all but one study CT was performed before DWI (reducing the sensitivity of CT to detect ischaemia)
CT results TP       73 FP        0 FN       88 TN       65 Total   226	DWI results  TP       147 FP        5 FN       14 TN       60 Total   226	Summary effect (95% CI) DWI sensitivity 0.99 (0.23 to 1.00) DWI specificity 0.92 (0.83 to 0.97) CT sensitivity 0.39 (0.16 to 0.69) CT specificity 1.00 (0.94 to 1.00)
Conclusions and comments In the small cohort of included studies, DWI is more sensitive than CT ‐ but not more specific ‐ for the early detection of ischaemic stroke. The small amount of data and the presence of methodological biases preclude any reliable calculation ‐ from the sensitivity and specificity estimates of CT and DWI ‐ of a positive or negative stroke diagnosis at different rates of stroke prevalence.
Applicability of tests in clinical practice None of the studies addressed practicality. CT is known to be quicker to perform and more readily available in most emergency care settings than magnetic resonance imaging (MRI). MRI is contraindicated in patients with pacemakers and some metal implants. In acutely ill stroke patients it may be difficult to monitor the patient's condition while being MR scanned (and this increases the risk of any respiratory difficulty or cardiovascular compromise that develops during the scan which passes undetected and may have adverse effects for the patient). If the patient is confused or restless as a result of the stroke, the patient may not be able to co‐operate for the longer scan times of MRI.
Costs None of the studies included a cost‐effectiveness evaluation. MRI is known to be more expensive than CT.

CI: confidence interval 
 CT: computed tomography 
 DWI: diffusion‐weighted magnetic resonance imaging 
 FN: false negative 
 FP: false positive 
 MR/MRI: magnetic resonance imaging 
 TN: true negative 
 TP: true positive

Background

Target condition being diagnosed

Stroke is the third leading cause of death in Western societies and the leading cause of long‐lasting severe disability (Warlow 2003). The overall incidence of stroke is about 2.4 per 1000 population, with modest geographical variations (Feigin 2003).

The two main categories of stroke are ischaemic and haemorrhagic. The latter occurs when a blood vessel in the brain ruptures causing bleeding either within the brain (intracerebral haemorrhage) or between the brain and the thin membrane that surrounds the brain (subarachnoid haemorrhage). Haemorrhagic strokes account for about 20% of all strokes. An ischaemic stroke occurs when an artery in the brain becomes blocked and blood flow suddenly decreases or stops causing a brain infarction. It is the most common form of stroke and accounts for about 80% of all strokes.

According to the location of the vascular event, ischaemic strokes may be classified as: (1) lacunar syndromes (LACS); (2) posterior circulation syndromes (POCS); (3) total anterior circulation syndromes (TACS); and (4) partial anterior circulation syndromes (PACS) (Bamford 1991).

A transient ischaemic attack (TIA) starts suddenly, like a stroke, but symptoms last only for a short period of time (usually minutes or hours) and then resolve without leaving any noticeable sign or deficit. Symptoms of a TIA disappear completely within 24 hours from onset but they are associated with a high risk of a subsequent stroke, especially within the first few weeks (8% to 11.5% within the first seven days) (Coull 2004).

Stroke is usually diagnosed by a combination of clinical examination and imaging procedures, though it is accepted that some patients with a clinically definite stroke may have normal brain imaging appearances.

An accurate and timely diagnosis is crucial in acute stroke both for decision‐making and for establishing appropriate patient management. The continuing advances in neuroimaging techniques and the advent of thrombolytic therapy and other emergency neurointerventional procedures for ischaemic stroke (for example mechanical clot retrieval and intra‐arterial thrombolysis), whose efficacy is greatest when given within a few hours of stroke onset (NINDS rt‐PA Stroke Study Group 1995; Furlan 1999), have increased the need for a rapid and reliable diagnosis. The main objectives of neuroimaging in acute stroke are to distinguish between stroke and non‐stroke lesions (for example brain tumour, abscess), to distinguish haemorrhagic from ischaemic stroke, and to identify the anatomic localisation of the vascular lesion. In particular, positive signs of acute cerebral ischaemic lesions on imaging may contribute to diagnosis within the first few hours. For example, if the clinical symptoms and radiological localisation of stroke match, or if the scan gives a clue to aetiology (such as the hyperdense artery sign suggesting large artery occlusion), this may give clinicians greater confidence to administer thrombolysis. However, it is not known whether the appearance of the acute ischaemic lesion influences stroke treatment, management, or outcome.

Index test(s)

Non‐contrast computed tomography (CT) is a cost‐effective and widely used neuroimaging method for the initial evaluation of patients presenting with stroke symptoms (Wardlaw 2004). In particular in the acute phase of stroke, when patients are scanned within a few hours of symptoms onset, CT is quick to perform, easy to tolerate, and is known to be very reliable for the detection of intracerebral haemorrhage. Early detection of haemorrhage is essential since the presence of blood in the brain or subarachnoid space is the main contraindication for the administration of aspirin, an anticoagulant, or thrombolytic therapy (Hacke 1998). In contrast, the early infarct signs on CT can be subtle and consequently difficult to detect. It is now widely recognised that patients with an acute ischaemic stroke (especially lacunar stroke and stroke in the brainstem) can have normal CT appearances.

Magnetic resonance imaging (MRI) with diffusion‐weighted sequences (DWI) has been increasingly used in the assessment of patients with stroke and TIA because of its sensitivity in detecting the early changes associated with ischaemia (Lansberg 2000), especially in patients with mild events or with small infarcts (for example lacunar or brainstem infarcts). On the other hand, the detection of acute cerebral haemorrhage on MRI is not as straightforward as on CT. There have been suggestions that MR gradient‐echo sequences (GRE) can be as sensitive as CT for excluding intracerebral haemorrhage before the administration of thrombolysis (Linfante 1999; Lin 2001; Fiebach 2004; Kidwell 2004). However, the utility of MRI as an alternative to CT amongst patients with suspected acute stroke has yet to be fully demonstrated. Moreover, in many countries MRI is not available for stroke patients in many hospitals, it is contraindicated in patients with pacemakers and metal implants, and is difficult to tolerate for severely affected or confused patients.

We, therefore, conducted a systematic review of the literature to estimate the accuracy of DWI compared with CT for the diagnosis of acute ischaemic stroke and to assess the accuracy of MRI (all feasible sequences, for example diffusion‐weighted, gradient‐echo sequences) for the early detection of haemorrhagic stroke.

We restricted the scope of this review to assess the diagnostic accuracy of both CT and MRI in patients suspected of acute stroke. We did not address the issue of the use of imaging methods (such as perfusion‐diffusion mismatch) to identify acute stroke patients who might benefit from thrombolytic therapy outside the conventional therapeutic time window. Nor did we address the consequences of identifying an acute ischaemic lesion on imaging in terms of management or outcome.

Objectives

Primary objectives

• To compare MR diffusion‐weighted images (DWI) and CT scans with respect to the accuracy of the localisation of acute ischaemic lesions. In particular, the review aimed to assess whether DWI could be considered superior to CT for the detection of acute ischaemic lesions within 12 hours (replacement of CT with DWI) or as an additional investigation for patients with negative or inconclusive CT scans.

• To assess the accuracy of MRI for the detection of acute haemorrhagic lesions within 12 hours. This time window was selected because this is the time when antithrombotic or thrombolytic therapy is most likely to be beneficial for patients with ischaemic stroke (and hence the reliable exclusion of haemorrhage is a clinically urgent priority).

Secondary objectives

As imaging results may vary depending upon the technical characteristics of the imaging test and the time when the imaging test is performed, we planned to analyse diagnostic data according to:

• the time of imaging from onset of symptoms (e.g. patients scanned within three, six, and 12 hours of symptoms onset);

• choice of imaging test for detection of haemorrhagic stroke (e.g. diffusion‐weighted sequences, gradient‐echo sequences).

Investigation of sources of heterogeneity

We were not able to investigate methodological and clinical sources of heterogeneity due to the relatively limited number of studies included in this review.

Methods

Criteria for considering studies for this review

Types of studies

We aimed to include studies published in any language. However, we did not include non‐English articles for which a full‐text translation or evaluation could not be obtained.

For the detection of ischaemic lesions, studies were eligible if:

1. both DWI and CT were evaluated in the same patient population (direct comparison) against an acceptable reference standard (as defined later), or if patients were randomised within a study to DWI or CT;

2. clinical and imaging assessments were performed within 12 hours of onset of symptoms;

3. the absolute numbers of observations of true positives, false positives, false negatives, and true negatives were available or derivable from the data reported in the primary studies.

For the detection of haemorrhagic lesions, studies were eligible if:

1. MRI sequences were evaluated against a clinical diagnosis of stroke supported by CT findings (reference standard) in cross‐sectional studies;

2. clinical and imaging assessments were performed within 12 hours of symptoms onset;

3. the absolute numbers of observations of true positives, false positives, false negatives, and true negatives were available or derivable from the data reported in the primary studies.

We included both prospective and retrospective studies.

We excluded studies that focused on patients presenting exclusively with a clinical syndrome suggesting either subarachnoid haemorrhage or isolated intraventricular haemorrhage since they are very distinctive clinical syndromes not directly relevant to patients presenting with the focal neurological deficits of acute stroke.

We also excluded studies that: addressed specific anatomical, metabolic, microvascular, or volumetric aspects of stroke; focused on specific technical aspects of CT and MRI; analysed perfusion versus diffusion imaging differences in patients with acute cerebral ischaemia.

Where investigators published several reports based on data from a single study population, we selected the updated or most complete report.

Participants

Adult patients with clinical symptoms suggestive of acute stroke, including patients in whom the subsequent diagnosis proved to be TIA.

Index tests

• DWI and CT performed within 12 hours of onset of symptoms for the detection of ischaemic brain lesions (CT is regarded here as the alternative test for detection of ischaemic lesions).

• MRI (all suitable sequences) performed within 12 hours of onset of symptoms for the detection of haemorrhagic brain lesions.

Target conditions

• Acute ischaemic stroke

• Acute haemorrhagic stroke

Reference standards

A single 'gold standard' for the diagnosis of stroke does not exist. In clinical practice however, expert assessment based on the combination of clinical features, imaging appearances, laboratory tests, and clinical follow up does provide the most comprehensive diagnosis. A diagnosis of stroke based only on the clinical and imaging data available to the clinician within the first few hours is unlikely to be sufficiently accurate.

• For the diagnosis of acute ischaemic stroke we considered an acceptable reference standard to be: a combination of clinical and imaging information supported by clinical or imaging follow up (CT or MRI) or autopsy. Any elaboration of this definition was however deemed suitable for inclusion (e.g. studies that relied exclusively on a clinical diagnosis or exclusively on a CT or MRI follow up).

• For the diagnosis of acute haemorrhagic stroke we considered a valid reference standard to be: a clinical diagnosis supported by CT or autopsy.

Note: in some studies, patients whose symptoms lasted less than 24 hours but who had evidence of an ischaemic lesion on imaging were counted as having had strokes and hence analysed as true positive cases. In other studies, however, patients with symptom duration less than 24 hours and an ischaemic lesion on imaging were analysed as being false positive cases.

Search methods for identification of studies

Electronic searches

We identified eligible studies by searching the following electronic databases:

• Cochrane Stroke Group Trials Register (last searched by the Trials Register Administrator in March 2009);

• Cochrane Central Register of Controlled Trials (CENTRAL) (The Cochrane Library Issue 1, 2009);

• MEDLINE ‐ Ovid (from January 1995 to March 2009);

• EMBASE ‐ Ovid (from January 1995 to March 2009);

• MEDION (last searched in March 2009 using the 'Systematic Reviews and Diagnostic Studies' search filter, the ICPC code = 'Neurological' and the signssymp = 'Medical Imaging').

MEDLINE and EMBASE searches

We searched indexed records which appeared in MEDLINE (January 1995 to March 2009). The choice of this time period was justified by the introduction in the early 1990s of MR diffusion‐weighted and gradient‐echo sequences into clinical practice. The MEDLINE search strategy included both subject headings (MeSH terms) and text words for the target condition (stroke) and the imaging techniques under investigation (MRI and CT). We also included a methodological filter for studies of diagnostic accuracy. Our methodological filter was based on the diagnostic component of the search strategy developed and validated by Astin and colleagues to identify diagnostic accuracy studies on imaging (Astin 2008). We did not apply any language restrictions. We adapted the MEDLINE search to search EMBASE. In particular, we 'translated' the MEDLINE MeSH terms into the corresponding terms available in the EMTREE vocabulary. Full details of both the MEDLINE and EMBASE search strategies together with a brief summary of the MEDLINE search strategy are presented in Appendix 1. We imported all citations identified by the MEDLINE and EMBASE search strategies into the Reference Manager bibliographic database (RefMan 2001).

Searching other resources

We handsearched all proceedings of the International Stroke Conference and the European Stroke Conference (1995 to 2004). These proceedings were published as abstracts in special issues of two peer‐reviewed journals: Stroke and Cerebrovascular Diseases.

We searched the following websites using terms for the target condition (stroke, acute stroke) and for the two imaging techniques under investigation (magnetic resonance imaging, computer tomography):

• American College of Cardiology ‐ ACC (www.acc.org) (last searched March 2009);

• American College of Radiology ‐ ACR (www.acr.uk) (last searched March 2009);

• American Heart Association ‐ AHA (www.americanheart.org) (last searched March 2009);

• American Stroke Association ‐ ASA (www.strokeassociation.org) (last searched March 2009);

• Department of Radiology & Diagnostic Imaging ‐ University of Alberta, Canada (www.radiology.med.ualberta.ca) (last searched March 2009);

• National Institute of Neurological Diseases and Stroke ‐ NINDS (www.ninds.nih.gov) (last searched March 2009);

• National Stroke Association ‐ NSA (www.stroke.org) (last searched March 2009);

• Royal College of Physicians (www.rcplondon.ac.uk) (last searched March 2009);

• Royal College of Radiologists ‐ RCR (www.rcr.ac.uk) (last searched March 2009);

• Scottish Intercollegiate Guidelines Network ‐ SIGN (www.sign.ac.uk) (last searched March 2009).

We perused the reference lists of all relevant articles to identify further published studies for possible inclusion in the review. We also contacted experts in the field to enquire about ongoing or completed but yet not published diagnostic studies.

Data collection and analysis

Selection of studies

One author (MB) initially screened the titles and abstracts of the search results and retrieved all potentially relevant reports in full. Three review authors (MB, MGC, ER) independently reviewed all relevant reports according to the pre‐defined inclusion criteria. We resolved any disagreements by consensus or arbitration. The same three authors extracted data from the selected reports.

Data extraction and management

We designed a data abstraction form specifically to collect details from selected studies. We recorded the following information for each individual study (without concealing the study authorship or other publication details): journal name, year of publication, study design and method of recruitment (systematic review, randomised controlled trial, cross‐sectional survey; prospective study, retrospective study), setting, number and characteristics of participants (age, sex, ethnicity, previous history of stroke, concomitant diseases), classification of stroke, definition of abnormal CT and MR images, time of imaging, the reference standard by which the final diagnosis was established, time interval from index test and comparator tests, time interval from index test(s) and established diagnosis of stroke, technical characteristics of MRI and CT, information related to the clinicians who read and interpreted imaging results (background speciality, level of expertise) and to the clinicians who established a clinical diagnosis of stroke. We resolved any disagreements by consensus or arbitration.

Assessment of methodological quality

Four authors (MB, MGC, ER, NA) independently assessed the methodological quality of each included study using the QUality Assessment of Diagnostic Accuracy Studies (QUADAS) tool developed by the NHS Centre for Reviews and Dissemination at the University of York, UK (Whiting 2003). The QUADAS tool is structured in a series of questions which should be answered 'yes', 'no', or 'unclear', and aims to evaluate the presence of spectrum bias, bias associated with the choice of reference standard, disease progression bias, verification bias, review bias, clinical review bias, incorporation bias, and bias associated with study withdrawals and indeterminate results. In particular, we considered a representative spectrum of patients to be: female and male patients of all ages presenting with mild, moderate, or severe stroke symptoms; with or without previous history of stroke; scanned within a few hours of onset of symptoms. We defined an appropriate reference standard likely to correctly classify the target condition as: an expert clinical assessment coupled with clinical and imaging follow up. We also considered up to seven days an acceptable time period between MRI and CT for the detection of haemorrhagic stroke. For ischaemic stroke we accepted any time period reported by the studies’ investigators between CT and MRI, and the follow up reference test. We also decided to add to the recommended QUADAS questions the following items pertinent to the purpose of this systematic review: expertise of the person interpreting the imaging results; whether the scans were read blind to clinical information; and whether the sequence of imaging tests was determined at random. As regards the expertise of the person interpreting imaging results, we distinguished radiologists and neuroradiologists, who by definition are experts in reading imaging test results, from neurologists, geriatricians, and general medicine doctors. We resolved any disagreements by consensus or arbitration. For each individual study we tabulated the agreed results of the quality assessment.

Statistical analysis and data synthesis

We extracted or derived indices of diagnostic performance from data presented in each primary study for each imaging test. We constructed 2 X 2 contingency tables of true positive cases, false positive cases, false negative cases, and true negative cases. We considered patients with ischaemic stroke as false positives or true negatives when analysing the performance for detecting haemorrhagic stroke, and we counted patients with haemorrhagic stroke as false positives or true negatives when analysing the performance for detecting ischaemic stroke. We calculated sensitivity and specificity with 95% confidence intervals (CI) for each imaging test in each study. We tabulated results for studies on ischaemic stroke separately from those for studies on haemorrhagic stroke.

We drew forest plots to show the variation of sensitivity and specificity estimates together with their 95% CI. For studies on ischaemic stroke where DWI was compared with CT versus a reference standard of clinical diagnosis and imaging follow up we also plotted the imaging test results on a receiver operating characteristic (ROC) plot of true positive rate (sensitivity) against false positive rate (1 ‐ specificity).

We explored the heterogeneity of the sensitivity and specificity estimates amongst studies on ischaemic stroke by examining both the forest plots and the ROC plot. As almost all estimates of specificity were at 'ceiling level' (specificity of 1) there was no evidence of heterogeneity and it was not possible to use statistical methods that rely on estimating correlations between sensitivity and specificity to enable estimation of a summary ROC curve (Macaskill 2004; Harbord 2006). Rather, we separately pooled estimates of sensitivities and specificities across the studies. For CT sensitivity and DWI sensitivity and specificity we undertook meta‐analyses using maximum likelihood estimation of a random‐effects model to pool logit transformed proportions and allow for within‐study binomial variation. We computed confidence intervals using MCMC sampling. We used a fixed‐effect analysis to estimate the pooled specificity of CT as a specificity of 1 was observed in every study (score method was used to compute confidence intervals). We used the DiagMeta package within the R software (The R Foundation for Statistical Computing Version 2.7.1) to carry out the analyses. We were not able to perform a formal statistical comparison between tests due to the zero cell issues and small sample sizes. An informal comparison between tests was made by meta‐analysing each test separately and examining the results. As all studies in the analysis evaluated both tests in all patients this comparison should not be biased by differences between the studies. 

We did not include study‐level covariates in the analyses to assess factors that might have contributed to heterogeneity (such as time of imaging) as in small meta‐analyses this is likely to produce unreliable estimates.

We did not calculate overall estimates for studies on haemorrhagic stroke as we only identified two studies of different methodological quality from the literature.

Results

Results of the search

The MEDLINE and EMBASE searches identified 9961 citations. Of these, we considered 112 relevant to the purpose of our review and we retrieved the full‐text articles (Figure). We subsequently excluded 103 articles (see the Characteristics of excluded studies table). The most common reason for exclusion was that the study was either not a primary diagnostic study of test accuracy or it did not involve appropriate test comparisons. Eight studies, published in nine reports, with a total of 306 participants fulfilled our inclusion criteria. Six studies focused on the comparison between DWI and CT for the detection of ischaemic lesions (Sorensen 1996; Barber 1999; Bozzao 1999; Gonzalez 1999; Urbach 2000; Saur 2003), one study estimated the accuracy of MRI for detection of haemorrhagic lesions (Oppenheim 2005), and one study assessed the use of MRI compared with CT for detection of both ischaemic and haemorrhagic lesions (Chalela 2007). Thus, seven studies contributed to the assessment of acute ischaemic stroke and two studies contributed to the assessment of haemorrhagic stroke. The details of all included studies are reported in the Characteristics of included studies table.

1.

Flow of studies through the selection process

Methodological quality of included studies

Studies on ischaemic stroke (seven studies)

Seven studies compared CT with DWI in the same patients for the detection of acute cerebral ischaemia. The total number of assessed patients was 226. Sample size ranged from 11 to 90 patients (mean 32 patients). The reported mean age was 65.1 years (range 21 to 100 years). The proportion of men ranged from 40% to 73%, with no information on gender distribution in one study. Only three studies clearly reported stroke severity and only three reported the number of patients who were excluded because they could not tolerate MRI. CT and DWI were performed within three hours of symptoms onset in one study, within six hours in four studies, and within 12 hours in the two remaining studies. In all but one study CT was performed before DWI. In five studies the average delay between CT and DWI was 55.3 minutes (SD 24.4 minutes). One study reported that the median interval between the two imaging techniques was 34 minutes, and the remaining study did not provide this information but stated that the interval between DWI and stroke onset was 4.2 hours. In four studies selection of patients was restricted to middle cerebral artery stroke; in two studies the stroke vascular territory was not given even though it is likely that they predominantly enrolled patients with anterior circulation stroke; and in one study patients were not selected according to the type of stroke.

The quality of the seven included studies varied (Figure and Figure). Four studies collected patients’ data prospectively (132 patients in total) and three studies retrospectively (94 patients in total). In all three retrospective studies the original MR and CT images of acute stroke, obtained from the patients’ hospital records, were reviewed de novo by the study investigators. In these three studies, even though brain images were reviewed de novo, there was still a risk of bias due to the retrospective selection of patients' records. Four studies clearly described their inclusion criteria but only one study appeared to include a representative spectrum of stroke patients (a consecutive series of patients referred to hospital because of a clinical suspicion of stroke and irrespective of gender, age, previous medical history, co‐morbidity, symptom severity, or final diagnosis) (90 patients) (Chalela 2007). However, the extremely mild strokes and the absence of any stroke mimics in this study indicated that some clinical exclusion criteria must have been applied after hospital admission and before study inclusion and scanning. Furthermore, in this study patients presenting with TIA but in whom DWI showed a new ischaemic lesion had their diagnosis changed to stroke (incorporation bias). In all the included studies the reference standard for diagnosis of stroke was a clinical diagnosis supported by imaging follow up. The reference standard was independent of the index text in six studies (the acute images were not used in the final diagnosis). Readers of DWI and CT acute images were reported to be blind to patients' clinical details and final diagnosis in only three studies. Information on blinding of the reference standard results was not clearly reported in five studies and in two studies interpretation of the follow up images was not blind to the findings of the acute images. Information on the expertise of clinicians reading imaging results was available in all but one study. None of the studies use formal randomisation methods to determine the sequence of the imaging tests.

2.

Methodological quality of the seven included studies on ischaemic stroke

3.

Methodological quality summary: review authors' judgment on each individual QUADAS item for the seven included studies on ischaemic stroke.

Studies on haemorrhagic stroke (two studies)

The characteristics of the two studies assessing the use of acute MRI for detection of haemorrhagic lesions are summarised in Table.

1. Characterisitcs and diagnostic results of the two included studies on haemorrhagic stroke.

Study	Participants (% men)	Participants assessed	Age (range)	Stroke severity	Time of imaging	MRI results (95% CI)
Chalela 2007 *	450 (unknown)	90	Median 76 years (21 t0 100 years)	Median score at NIHSS = 3 (range 0 to 37)	Within 3 hours of stroke onset	GRE and DWI sensitivity 0.83 (0.52 to 0.98) GRE and DWI specificity 1.00 (0.95 to 1.00)
Oppenheim 2005 **	86 (64)	82	Mean 68.8 years	Mean score at NIHSS = 11.25	Within 6 hours of stroke onset (mean time 2.6 hours)	DWI sensitivity 1.00 (0.91 to 1.00) DWI specificity 1.00 (0.91 to 1.00) GRE sensitivity 1.00 (0.91 to 1.00) GRE specificity 0.98 (0.87 to 1.00)

*: prospective 
 **: retrospective 
 95% CI: 95% confidence intervals 
 DWI: diffusion‐weighted imaging 
 GRE: gradient‐echo 
 MRI: magnetic resonance imaging 
 NIHSS: National Institute of Health Stroke Scale

The prospective study by Chalela and colleagues (Chalela 2007) compared non‐contrast CT with MRI (diffusion‐weighted and susceptibility‐weighted images) in 450 patients referred for emergency assessment of suspected stroke, 90 of whom were scanned within three hours from the onset of symptoms. The patients’ median severity score, assessed by the National Institute of Health Stroke Scale (NIHSS), was 3 (range 0 to 37) indicating the presence of predominantly mild stroke deficits. The proportion of patients with primary cerebral haemorrhage was 13% (12/90). Overall the proportion of patients who could not tolerate MRI, amongst a predominantly mild stroke population, was 11% (49/450).

The retrospective study by Oppenheim and colleagues (Oppenheim 2005) used data extracted from the acute databases of two university hospitals to evaluate the accuracy of five MR sequences (T₁, GRE, FLAIR, T₂‐EPI, and DWI) to identify within 86 stroke patients those with (43 patients) and without (43 patients) intracerebral haemorrhage. Patients were included if they presented with a stroke severity score of ≥ 3 points on the NIHSS and if they underwent imaging within six hours of stroke onset. The patients’ final diagnosis incorporated all clinical, pathological, and imaging investigations. However, as only a small number of patients underwent CT, the reference standard was highly inconsistent and the final diagnosis of intracerebral haemorrhage was primarily based on multisequence MRI (incorporation bias). Not all patients completed all five MR sequences and we assessed results from 82 patients who completed both the gradient‐echo and diffusion‐weighted sequences.

Figure summarises the results of the quality assessment of the two studies on haemorrhagic stroke. Only five of the 15 QUADAS items were met by both studies. Information on the spectrum of patients was not clearly reported in the Oppenheim study (Oppenheim 2005). In both studies the reference standard was a final clinical diagnosis of haemorrhagic stroke supported by all available imaging investigations including the acute images. However, in both studies it was unclear whether all patients were verified by the same reference standard and there was clear evidence of incorporation bias as MRI findings contributed to the final diagnosis. In both studies the follow up images were not read blind to the findings of acute images. The reading order of MRI examinations was determined at random only in the Oppenheim study (Oppenheim 2005).

4.

Methodological quality summary: review authors' judgment on each individual QUADAS item for the two included studies on haemorrhagic stroke.

Findings

Studies on ischaemic stroke (seven studies)

Figure shows the forest plots of the sensitivity and specificity estimates for DWI and CT for the seven studies that assessed patients with ischaemic stroke. Sensitivity estimates for DWI ranged from 0.73 to 1.00 (median 1.00) and the sensitivity estimates for CT ranged from 0.11 to 0.75 (median 0.45). Specificity estimates for DWI ranged from 0.86 to 1.00 (median 1.00) whilst specificity estimates for CT were all at 'ceiling level' (1.00 specificity).

5.

Forest plots of DWI and CT results for ischaemic stroke

The pairs of observed values of sensitivity and specificity for DWI and CT are presented in a ROC space in Figure. The pooled estimates for DWI sensitivity and specificity were 0.99 (95% CI 0.23 to 1.00) and 0.92 (95% CI 0.83 to 0.97) respectively, whilst the pooled estimates for CT sensitivity and specificity were 0.39 (95% CI 0.16 to 0.69) and 1.00 (95% CI 0.94 to 1.00) respectively.  

6.

ROC plot for the seven studies that compared DWI with CT for the early detection of ischaemic stroke

Studies on haemorrhagic stroke (two studies)

As data for the assessment of haemorrhagic stroke were derived from only two studies with clear evidence of incorporation bias, we did not perform a meta‐analysis of measures of test accuracy. The findings of the two studies suggested that MRI sequences may distinguish between patients with and without acute intracerebral haemorrhage with reasonably high sensitivity and specificity (see Figure). The Chalela study (Chalela 2007) showed a sensitivity of 0.83 (95% CI 0.52 to 0.98) and a specificity of 1.00 (95% CI 0.95 to 1.00) for gradient‐echo and diffusion‐weighted MRI in patients assessed within three hours of stroke, versus CT and clinical assessment. The sensitivity estimate was, however, based on only 12 patients (13% of the 90 patients investigated) who were found to have acute cerebral haemorrhage. Similarly, the Oppenheim study (Oppenheim 2005) reported 1.00 (95% CI 0.91 to 1.00) for both the sensitivity and specificity of diffusion‐weighted MRI and 1.00 (95% CI 0.91 to 1.00) sensitivity and 0.98 (95% CI 0.87 to 1.00) specificity for gradient‐echo MRI performed within six hours of symptoms onset, but with no CT comparator in most of the patients. However, the high proportion (50%) of patients with haemorrhage in this study as compared to the proportion of intracerebral haemorrhage observed in the typical clinical population (10% to 15%) indicated the presence of spectrum bias (highly selected patient sample) that was likely to have increased sensitivity and influenced specificity.

7.

MRI results for haemorrhagic stroke

Discussion

The emergency management of patients with stroke relies heavily on accurate and rapid diagnosis. Early identification of patients with stroke and the distinction between ischaemic and haemorrhagic stroke are crucial for therapeutic decision making and, in particular, for selecting patients for thrombolytic treatment. CT and MRI are both used in clinical practice to identify patients with acute stroke who might benefit from reperfusion therapy (which is absolutely contraindicated in patients with stroke due to intracerebral haemorrhage). We conducted a systematic review to compare the accuracy of these two imaging methods for detection of acute ischaemic and haemorrhagic stroke. Six studies for the detection of ischaemic stroke, one study for the detection of haemorrhagic stroke, and one study which assessed both haemorrhagic and ischaemic stroke fulfilled our inclusion criteria. For the assessment of ischaemic stroke we included only comparative studies that evaluated both imaging techniques in the same patients, as they provide the best evidence on which to judge the relative performance of CT and MRI for the detection of ischaemic stroke lesions. In this limited cohort of studies MRI had higher sensitivity than CT but similar specificity (see Figure). The two studies that contributed to the assessment of haemorrhagic stroke provide similar sensitivity and specificity estimates for MRI (see Table and Figure). We could not assess whether CT and MRI were equally good at identifying stroke mimics as the majority of studies excluded non‐stroke patients.

Findings on ischaemic stroke

The Table summarises the results of the seven included studies on ischaemic stroke. Our findings are in line with the well‐established claim in the literature that, in patients subsequently confirmed to have acute stroke, diffusion‐weighted MR sequences are more sensitive for detecting acute ischaemia than plain CT; especially in the first hours after symptoms onset and chiefly in patients with mild stroke (Fiebach 2001; Fiebach 2002; Jaillard 2002; Kucinski 2002; Mullins 2002). There are, however, important considerations to be made. The characteristics of the patient population varied between studies and all but one study included a very narrow spectrum of stroke patients. The study with a broader spectrum of patients (Chalela 2007) still only included mostly mild strokes (median NIHSS = 3) and, therefore, was not representative of the typical population being assessed for thrombolysis. The evaluation of mild cases may increase the sensitivity of DWI, which is known to be particularly useful in detecting small ischaemic lesions (Keir 2004). The exclusion of more severe cases may further disadvantage CT as more severely affected patients are more likely to have a CT‐visible lesion, which may help to explain the large difference between CT and DWI sensitivity estimates in this study (Chalela 2007). The severity of stroke was rarely reported in the remaining included studies. However, it is known that many patients with severe stroke do not tolerate MRI (Hand 2005). Information related to the patients who were excluded because they either could not tolerate MRI or had contraindications was provided in only three studies. Of these, the largest study reported that about 11% of the patients initially screened for inclusion were subsequently excluded due to MRI contraindications (Chalela 2007). Moreover, the majority of included studies enrolled patients with typical anterior circulation stroke. Negative DWI findings have been reported to occur more often in posterior circulation stroke during the first 24 hours (Lövblad 1998; Oppenheim 2000). Thus in an unselected population of stroke patients the inclusion of posterior circulation strokes may reduce the apparently greater diagnostic accuracy of DWI. In all but one study CT was performed about an hour before DWI. Thus the lesions on CT could have been less conspicuous and more difficult to detect than they would have been at a later stage (a more rigorous approach would have been to determine the sequence of tests by random allocation). This could in part have contributed to the variability in the observed CT sensitivity estimates. Studies were generally of very small sample sizes, which may have jeopardised blinding and had an effect on the estimates of accuracy, especially for sensitivity. Furthermore, CT and DWI were evaluated in a highly selected group of patients in all but one study. Most of the patients had a final diagnosis of ischaemic stroke or TIA. TIA cases were usually counted as 'stroke negative' cases, except in one study (Chalela 2007) where TIA cases with evidence of ischaemic lesions on DWI were reclassified as 'strokes' (true positive cases). This 'reclassification' might have added a negative effect on CT and did clearly switch the reference standard to an MRI diagnosis. The incorporation of DWI findings in the reference standard (incorporation bias) was likely to have inflated the observed DWI estimates of sensitivity.  Moreover, although it has been demonstrated that DWI may show an acute cerebral infarction in approximately half of patients with TIA (Kidwell 1999) the clinical significance of DWI‐positive TIAs remains uncertain. Specificity estimates were very high in all studies. Indeed, in most of the studies stroke mimics (for example cerebral neoplasms, systemic infections) or patients with other cerebrovascular lesions were not included in the spectrum of patients assessed. This renders the sample poorly representative of the acute patients typically seen in clinical practice, where 15% to 30% of patients with an initial clinical diagnosis of stroke are ultimately found to have stroke‐mimic pathologies (Libman 1995; Hacke 1998; Scott 2003; Hand 2006). In turn, this makes it difficult to be certain that these estimates of accuracy apply in routine clinical practice to a wider spectrum of patients and provides no information on the accuracy of CT and MRI in detecting mimics.

Findings on haemorrhagic stroke

CT is the imaging modality most commonly used to distinguish the acute presentation of intracerebral haemorrhage from ischaemic stroke in the evaluation of potential candidates for thrombolytic therapy. More recently it has been suggested that MRI, including diffusion‐weighted and gradient‐echo sequences, could detect haemorrhage in the first hours after stroke (Weingarten 1994; Patel 1996; Atlas 1998; Linfante 1999; Kidwell 2004). However, methodologically rigorous data on haemorrhagic stroke are scanty and even the two studies that met our pre‐defined inclusion criteria (Oppenheim 2005; Chalela 2007) suffered from major methodological biases and limitations. Hence, there is insufficient evidence on which to draw any sound conclusions on the accuracy of MRI for detection of haemorrhagic stroke in routine practice. In both included studies the reference standard was a hospital discharge diagnosis which incorporated all available clinical and imaging data (including acute imaging data) but without CT in many cases, thereby leading to a comparison of MRI with itself. The presence of this incorporation bias may have overestimated the reported MRI diagnostic accuracy. Furthermore, in both studies patients with non‐stroke lesions were not assessed and it would have been useful to know how well MRI (compared with CT) could distinguish haemorrhagic lesions from non‐stroke lesions (for example neoplasms). Thus, while the ability of CT to distinguish acute haemorrhagic lesions from non‐stroke lesions is well established the accuracy of MRI assessment of suspected acute stroke is still somewhat unclear.  

Summary of main results

In conclusion, we identified only a limited number of studies that directly compared MRI versus CT for the early detection of stroke lesions. The overall methodological quality of these studies was poor. Our results suggest that diffusion‐weighted MRI is probably more sensitive than CT, but not more specific, for the early detection of ischaemic stroke in highly selected patient populations. Our data do not allow any comments to be made on the merits of MRI for the detection of haemorrhagic stroke. Moreover, estimates of diagnostic accuracy of CT and MRI were obtained from well‐defined groups of patients with a final diagnosis of stroke so may be of limited clinical utility as they may not be applicable to the broad range of patients with suspected acute stroke usually seen in routine clinical practice. Neither practicality nor cost‐effectiveness was effectively taken into consideration in the included studies. Additional well‐designed studies are needed to estimate more reliably whether MRI can be used as the primary imaging modality for patients presenting with suspected acute stroke.

Strengths and weaknesses of the review

For the detection of ischaemic stroke, we focused exclusively on comparative studies that evaluated both CT and MRI versus a reference standard of clinical diagnosis and imaging follow up in the same patients, which is known to provide the best evidence about the diagnostic accuracy of two different methods. We searched major electronic databases to identify all relevant studies. Three review authors with different expertise (a methodologist and two neurologists) independently selected studies and extracted data. Four review authors (a methodologist, two neurologists, and a radiologist) independently assessed the quality of the included studies.

Our review has some limitations. Overall, our findings are limited by the relatively small number of comparative studies available in the literature; incomplete reporting of studies' characteristics and results; limited methodological quality; and relatively small sample sizes. Diagnostic imaging studies seem to be particularly prone to these problems (Lijmer 1999; Rutjes 2006). Shortcomings in study design may affect the estimates of diagnostic accuracy resulting in an overestimation, particularly in studies including non‐representative samples of patients and invalid reference standards (Lijmer 1999; Rutjes 2006). Future studies should include an appropriate spectrum of patients; a consistent reference standard independent of the imaging modalities under investigation, to reduce incorporation bias; and blind interpretation of tests results. They should also comply with the Standards for Reporting of Diagnostic Accuracy (STARD) recommendations for improving the quality of reporting of diagnostic studies (Bossuyt 2003).

With regard to our literature searches to identify relevant studies to include in our review, a couple of points are worth raising. We included a methodological search filter in our MEDLINE and EMBASE searches to identify studies of diagnostic accuracy. The use of a search filter, even though it may have reduced the overall sensitivity of the MEDLINE search, was justified by the fact that a literature search combining MeSH terms and text words for the target condition with those for the diagnostic tests under evaluation (as for the current recommendation of the Screening and Diagnostic Tests Methods Group) would have retrieved an unmanageable number of hits; CT is an imaging test used very frequently in clinical practice and therefore referred to in many research papers. We did not search additional electronic databases, such as BIOSIS, LILACS, or Science Citation Index, firstly because the number and relevance of indexed journals in these databases are limited compared to those indexed in MEDLINE and EMBASE and secondly we are confident we have enhanced the sensitivity of our literature searches by searching 'specialised' databases (for example MEDION for systematic reviews of diagnostic accuracy) and professional bodies' websites (for example National Stroke Association, American Stroke Association, Royal College of Radiology), handsearching all conference proceedings of two major international stroke conferences for a 10‐year period, and contacting experts in the field. In this way, even though relying on limited resources we have maximised sensitivity and specificity for identifying comparative studies on the use of CT and MRI for detection of acute stroke lesions.

We could not use the currently recommended summary ROC curve methodology to compare the performance of CT and DWI for the early diagnosis of ischaemic stroke as our data were insufficient to fit this complex statistical model (Irwig 1995; Macaskill 2004; Harbord 2006). Similarly, due to the limited number of identified studies, we were not able to perform sensitivity analyses to assess which methodological aspects may have contributed to clinical heterogeneity (for example time of imaging, characteristics of patient population) or heterogeneity related to study design (for example prospective versus retrospective studies, presence of incorporation bias).

We were unable to address practicality and applicability issues as only three studies mentioned the number of patients who were excluded because they could not undergo MRI. However, practical difficulties in performing MRI, instead of CT, have been documented in about 20% of stroke patients and many patients with severe stroke do not tolerate MRI (Singer 2004; Barber 2005; Hand 2005). Similarly we did not assess the cost‐effectiveness of MRI compared with CT as a first‐line test for the early detection of stroke and we did not consider the relative impact on clinical outcomes of a policy of routine MRI versus routine CT. However, in deciding whether MRI could substitute for CT as the primary method for early imaging of patients with suspected ischaemic or haemorrhagic stroke it is important to consider the relative diagnostic accuracy of each imaging test together with practicality and cost‐effectiveness issues. In many countries CT is known to be rapid, easy to tolerate, and more readily available in most emergency settings. On the other hand MRI is not immediately available in many hospitals and is more expensive, contraindicated for patients with pacemakers and metal implants, and can be unpleasant or difficult to tolerate especially for patients with more severe strokes. A recent survey conducted in the UK showed that even though 78% of all acute hospitals that admitted patients with acute stroke had access to MRI facilities, MRI was rarely performed either at all or sufficiently quickly to be of value in the acute management of stroke (Kane 2008). A European survey suggests that similar problems exist in the rest of the EU and, in fact, placed the UK at the top of the table for comprehensive stroke centres (Leys 2007).

Applicability of findings to the review question

We reviewed the diagnostic accuracy of MRI compared with CT for acute ischaemic stroke, and the accuracy of MRI for early detection of haemorrhagic stroke. There is some evidence that MRI is more accurate than CT for the detection of mild ischaemic strokes. However, the use of MRI in the management of acute patients needs to take into consideration practicality and cost‐effectiveness. In many countries CT is quicker to perform, inexpensive, applicable to a higher proportion of acutely ill stroke patients, and more readily available in most emergency care settings. MRI is contraindicated in patients with pacemakers and some metal implants. In acutely ill stroke patients it may be difficult to monitor their condition while they are being MRI scanned (which increases the risk of any developing respiratory difficulty or cardiovascular compromise being detected during the scan and so may have adverse effects for the patient). If the patient is confused or restless as a result of the stroke, the patient may not be able to co‐operate for the longer scan times required for MRI. Furthermore, in clinical practice CT is the most used imaging technique for the diagnosis of acute intracerebral haemorrhage (and therefore for selecting patients for thrombolytic therapy). The role of MRI as the first choice modality for patients presenting with stroke symptoms requires further investigations.

Authors' conclusions

Implications for practice.

It is likely that, in the future, both CT and MRI techniques will be more widely available in many countries. Pending further evidence, both techniques should be used in a complementary way with CT for the majority of strokes pre‐thrombolysis and MRI for milder strokes, according to local, specific clinical needs.

Implications for research.

Future research should focus on a robust and objective cost‐effectiveness comparison of CT and MRI with particular attention to the evaluation of patients in broader and unselected patient populations more relevant to routine clinical practice in non‐specialist stroke centres. In particular, further studies are needed to provide clear evidence that MRI can be used as the imaging modality of first choice for patients with suspected acute stroke in routine practice, and that patients without evidence of acute intracerebral haemorrhage on MRI really do not have acute intracranial bleeding (and hence can be safely considered for thrombolytic treatments).

Appendices

Appendix 1. MEDLINE (Ovid) search strategy ‐ January 1995 to March 2009

1. cerebrovascular disorders/ or basal ganglia cerebrovascular disease/ or exp brain ischemia/ or carotid artery diseases/ or carotid artery thrombosis/ or cerebrovascular accident/ or exp brain infarction/ or exp hypoxia‐ischemia, brain/ or intracranial arterial diseases/ or cerebral arterial diseases/ or exp "intracranial embolism and thrombosis"/

2. ((brain or cerebr$ or cerebell$ or vertebrobasil$ or hemispher$ or intracran$ or intracerebral or infratentorial or supratentorial or middle cerebr$ or mca$ or anterior circulation) adj5 (isch?emi$ or infarct$ or thrombo$ or emboli$ or occlus$ or hypoxi$)).tw.

3. (isch?emi$ adj6 (stroke$ or apoplex$ or cerebral vasc$ or cerebrovasc$ or cva or attack$)).tw.

4. 1 or 2 or 3

5. exp Magnetic Resonance Imaging/

6. ((magnetic resonance or MR or NMR or diffusion weighted or T2‐weighted) adj2 imag$).tw.

7. ((MR or NMR) adj2 tomograph$).tw.

8. (MRI or DWI).tw.

9. 5 or 6 or 7 or 8

10. exp Tomography, X‐Ray Computed/

11. (CT or CAT).tw.

12. (comput$ adj3 tomograph$).tw.

13. 10 or 11 or 12

14. 4 and 9 and 13

15. ("1995$" or "1996$" or "1997$" or "1998$" or "1999$" or "200$").ed.

16. 14 and 15 (most sensitive search ‐ ischaemic stroke)

17. cerebrovascular disorders/di or basal ganglia cerebrovascular disease/di or exp brain ischemia/di or carotid artery diseases/di or carotid artery thrombosis/di or cerebrovascular accident/di or exp brain infarction/di or exp hypoxia‐ischemia, brain/di or intracranial arterial diseases/di or cerebral arterial diseases/di or exp "intracranial embolism and thrombosis"/di

18. exp *Magnetic Resonance Imaging/

19. exp *Tomography, X‐Ray Computed/

20. 17 and 18 and 19 and 15 (SET DOWNLOADED 1)

21. exp "Sensitivity and Specificity"/

22. false negative reactions/ or false positive reactions/ or diagnostic errors/

23. (sensitiv$ or specificity or distinguish$ or differentiat$ or enhancement or identif$ or detect$ or diagnos$ or accur$).tw.

24. (predictive adj4 value$).tw.

25. (false adj (positive$ or negative$)).tw.

26. (receiver operat$ adj (characteristic$ or curve or analysis)).tw.

27. (ROC or SROC).tw.

28. comparative study/

29. (compared or comparison or correlat$ or versus).tw.

30. or/21‐29

31. (16 and 30) not 20 (ischaemia + MRI + CT + years + diagnostic filter ‐ set downloaded) (SET DOWNLOADED 2)

32. exp basal ganglia hemorrhage/ or exp intracranial hemorrhages/

33. ((brain$ or cerebr$ or cerebell$ or intracerebral or intracran$ or parenchymal or intraventricular or infratentorial or supratentorial or basal gangli$ or putaminal or putamen or posterior fossa) adj10 (haemorrhage$ or hemorrhage$ or haematoma$ or hematoma$ or bleed$)).tw.

34. 32 or 33

35. 9 and 34 and 15 (MRI + haemorrhage + years) (most sensitive search ‐ haemorrhagic stroke)

36. exp basal ganglia hemorrhage/di or exp intracranial hemorrhages/di

37. 36 and 18 and 15 (haemorrhage diagnosis/di + exp MRI + years)

38. or/21‐27 (diagnostic filter)

39. 35 and 38 (MRI + haemorrhage + years + diagnostic filter)

40. 37 not (20 or 31) (haemorrhage diagnosis/di + exp MRI + years ‐ downloaded sets 1 and 2) (SET DOWNLOADED 3)

41. 39 not (20 or 31 or 37) (MRI (not exp MRI) + haemorrhage (not haemorrhage diagnosis/di) + years + diagnostic filter ‐ downloaded sets 1 and 2) (SET DOWNLOADED 4)

The above search strategy has been designed to cover both MRI and CT for the detection of ischaemic stroke and MRI for the detection of haemorrhagic stroke. In theory two separate search strategies could have been designed but we preferred to combine the searches to avoid looking at duplicate references.

Summary of the MEDLINE search strategy

• Line 16 of the search strategy identifies all records related to ischaemic stroke (broad terms) and the use of both MRI and CT for the period 1995 – 2009;

• Line 20 of the search strategy identifies records that focus specifically on MRI and CT for the diagnosis of ischaemic stroke (stroke terms searched with the subheading /di) for the period 1995 – 2009;

• Line 31 of the search strategy employs a diagnostic filter to identify  records related to ischaemic stroke (broad terms) and the use of both MRI and CT for the period 1995 – 2009;

• Line 35 of the search strategy identifies all records related to haemorrhagic stroke (broad terms) and the use of MRI for the period 1995 – 2009;

• Line 37 of the search strategy identifies records that focus specifically on MRI for the diagnosis of haemorrhagic stroke (stroke terms searched with the subheading /di) for the period 1995 – 2009;

• Line 39 of the search strategy employs a diagnostic filter to identify  records related to haemorrhagic stroke (broad terms) and the use of MRI for the period 1995 – 2009.

The most sensitive search for ischaemic stroke would have been to assess all the references at line 16 (approximately 2800 hits from 1995). However, as CT and MRI are routinely used in clinical practice and the terms occur very frequently in abstracts, we tried to limit the search specifically to imaging diagnostic studies in two ways: (a) by using the subheading diagnosis (/di) on the stroke MeSH terms and the focused imaging MeSH terms, and (b) by developing a search filter for diagnostic studies to increase precision. In order to test this approach we intended to scan the remaining references from line 16 to see if any relevant papers were missed and not identified by (a) or (b). However, due to the limited resources available this approach proved unfeasible.

The above comments apply to the search section on haemorrhagic stroke with line 35 being the most sensitive search.

EMBASE (Ovid) search strategy ‐ January 1995 to March 2009

Adapted from the MEDlINE search strategy

1. cerebrovascular disease/ or cerebral artery disease/ or cerebrovascular accident/ or stroke/ or vertebrobasilar insufficiency/ or carotid artery disease/ or exp carotid artery obstruction/ or exp brain infarction/ or exp brain ischemia/ or exp occlusive cerebrovascular disease/

2. ((brain or cerebr$ or cerebell$ or vertebrobasil$ or hemispher$ or intracran$ or intracerebral or infratentorial or supratentorial or middle cerebr$ or mca$ or anterior circulation) adj5 (isch?emi$ or infarct$ or thrombo$ or emboli$ or occlus$ or hypoxi$)).tw.

3. (isch?emi$ adj6 (stroke$ or apoplex$ or cerebral vasc$ or cerebrovasc$ or cva or attack$)).tw.

4. 1 or 2 or 3

5. exp nuclear magnetic resonance imaging/

6. ((magnetic resonance or MR or NMR or diffusion weighted or T2‐weighted) adj2 imag$).tw.

7. ((MR or NMR) adj2 tomography).tw.

8. (MRI or DWI).tw.

9. 5 or 6 or 7 or 8

10. exp computer assisted tomography/

11. (CT or CAT).tw.

12. (comput$ adj3 tomograph$).tw.

13. 10 or 11 or 12

14. 4 and 9 and 13

15. ("1995$" or "1996$" or "1997$" or "1998$" or "1999$" or "200$").em.

16. 14 and 15 (most sensitive search ‐ ischaemic stroke)

17. cerebrovascular disease/di or cerebral artery disease/di or cerebrovascular accident/di or stroke/di or vertebrobasilar insufficiency/di or carotid artery disease/di or exp carotid artery obstruction/di or exp brain infarction/di or exp brain ischemia/di or exp occlusive cerebrovascular disease/di

18. exp *nuclear magnetic resonance imaging/

19. exp *computer assisted tomography/

20. 17 and 18 and 19 and 15 (SET DOWNLOADED 1)

21. "sensitivity and specificity"/

22. laboratory diagnosis/

23. prediction/

24. "prediction and forecasting"/

25. receiver operating characteristic/ or roc curve/

26. diagnostic accuracy/

27. diagnostic value/

28. reliability/

29. (sensitiv$ or specificity or distinguish$ or differentiat$ or enhancement or identif$ or detect$ or diagnos$ or accur$).tw.

30. (predictive adj4 value$).tw.

31. (false adj (positive$ or negative$)).tw.

32. (receiver operat$ adj (characteristic$ or curve or analysis)).tw.

33. (ROC or SROC).tw.

34. comparative study/

35. exp controlled study/

36. intermethod comparison/

37. correlation analysis/

38. (compared or comparison or correlat$ or versus).tw.

39. or/21‐38

40. (16 and 39) not 20 (ischaemia + MRI + CT + years + diagnostic filter ‐ set downloaded 1) (SET DOWNLOADED 2)

41. basal ganglion hemorrhage/ or brain hemorrhage/ or brain ventricle hemorrhage/ or cerebellum hemorrhage/

42. ((brain$ or cerebr$ or cerebell$ or intracerebral or intracran$ or parenchymal or intraventricular or infratentorial or supratentorial or basal gangli$ or putaminal or putamen or posterior fossa) adj10 (haemorrhage$ or hemorrhage$ or haematoma$ or hematoma$ or bleed$)).tw.

43. 41 or 42

44. 9 and 15 and 43 (MRI + years + haemorrhage) (most sensitive search ‐ haemorrhagic stroke)

45. basal ganglion hemorrhage/di or brain hemorrhage/di or brain ventricle hemorrhage/di or cerebellum hemorrhage/di

46. 45 and 18 and 15 (haemorrhage diagnosis/di + exp MRI + years)

47. 46 not (20 or 40) (haemorrhage diagnosis/di + exp MRI + years ‐ downloaded sets 1 and 2) (SET DOWNLOADED 3)

48. or/21‐33 (diagnostic filter)

49. 44 and 48 (MRI + years + haemorrhage + diagnostic filter)

50. 49 not (20 or 40 or 47) (MRI + years + haemorrhage + diagnostic filter ‐ downloaded sets 1,2 and 3) (SET DOWNLOADED 4)

Data

Presented below are all the data for all of the tests entered into the review.

Tests. Data tables by test.

Test	No. of studies	No. of participants
1 DWI ‐ ischaemic stroke	7	226
2 CT ‐ ischaemic stroke	7	226
3 GRE/DWI	1	90
4 DWI	1	82
5 GRE	1	82

1. Test.

DWI ‐ ischaemic stroke.

2. Test.

CT ‐ ischaemic stroke.

3. Test.

GRE/DWI.

4. Test.

DWI.

5. Test.

GRE.

Characteristics of studies

Characteristics of included studies [ordered by study ID]

Barber 1999.

Clinical features and settings	Patients with suspected acute ischaemic stroke in the middle cerebral artery territory who were studied with both DWI and CT within 6 hours of symptom onset
Participants	17 patients (53% men) presenting with stroke symptoms Mean age: 68.5 years Mean Canadian Neurological Scale score: 5.8 (range 1.5 to 11)
Study design	Prospective
Target condition and reference standard(s)	Ischaemic stroke Clinical diagnosis and imaging follow up
Index and comparator tests	DWI versus CT
Follow‐up	T2‐weighted imaging performed at 90 days
Notes	None of the patients were treated with thrombolysis One patient was unable to tolerate MRI and was not included
Table of Methodological Quality
Item	Authors' judgement	Description
Representative spectrum? All tests	No	Selected sample of patients Only patients with signs consistent with acute ischaemic stroke were included Patients with previous history of stroke and non‐ischaemic neurological deficits were excluded
Acceptable reference standard? All tests	Yes	Standard clinical criteria and imaging follow up (T2‐WI)
Acceptable delay between tests? All tests	Yes	Mean time between index test (DWI) and comparator (CT) = 1 hour Mean time between index tests and reference standard = 90 days
Partial verification avoided? All tests	Yes	All patients were verified by the reference criteria
Differential verification avoided? All tests	Yes	All patients were verified by the same reference standard
Incorporation avoided? All tests	Yes	Final diagnosis of acute stroke did not include the acute DW images and CT scans
Reference standard results blinded? All tests	Unclear	Unclear whether the follow up T2‐W images were read blind to the acute images
Index test results blinded? All tests	Yes	Acute CT and DWI images were read separately by 2 neuroradiologists who were blinded to patients' clinical data and results of the other imaging study For the CT and DWI studies in which the readers disagreed, the scans were jointly re‐analysed and a final decision was reached by consensus
Relevant clinical information? All tests	Yes	No clinical information was provided to the clinicians who interpreted CT and DW images
Uninterpretable results reported? All tests	Yes	1 patient could not tolerate MR
Withdrawals explained? All tests	Yes	No withdrawals
Prospective study?	Yes	Prospective recruitment of consecutive patients
Expertise of imaging tests readers reported? All tests	Yes	2 neuroradiologists
Sequence of tests determined at random? All tests	No	The order of tests was not randomised CT was performed before DWI
Scans read blind to clinical information? All tests	Yes	The neuroradiologists who interpreted CT and DWI results were blinded to patients' clinical details

Bozzao 1999.

Clinical features and settings	Patients with suspected acute ischaemic stroke who underwent imaging within 12 hours of symptom onset
Participants	15 stroke patients (40% men) Mean age 67.6 years (range: 54 to 81 years)
Study design	Prospective
Target condition and reference standard(s)	Acute ischaemic stroke Clinical diagnosis and imaging follow up
Index and comparator tests	DWI versus CT
Follow‐up	CT at 8 days
Notes	Haemorrhage excluded Severity of stroke not reported Stroke vascular territory not specified 1 patient could not undergo MRI because of agitation
Table of Methodological Quality
Item	Authors' judgement	Description
Representative spectrum? All tests	No	Selected sample of patients Only patients with signs consistent with acute ischaemic stroke were included Patients with intracerebral haemorrhage were excluded Severity of stroke not given
Acceptable reference standard? All tests	Unclear	Probably standard clinical criteria and imaging follow up Time of follow up imaging not clearly reported
Acceptable delay between tests? All tests	Yes	Time between CT and DWI = 1 hour Time between index tests and reference standard not reported
Partial verification avoided? All tests	Yes	All patients were verified by the reference criteria
Differential verification avoided? All tests	Yes	All patients were verified by the same reference criteria
Incorporation avoided? All tests	Yes	Final diagnosis did not include the results of acute CT and DWI
Reference standard results blinded? All tests	Unclear	Not clearly reported
Index test results blinded? All tests	Unclear	Unclear whether the acute CT and DW images were read blind to patients clinical details and final diagnosis
Relevant clinical information? All tests	Unclear	Unclear whether the clinicians who interpreted the CT and DW images were provided with patients' clinical details
Uninterpretable results reported? All tests	Unclear	Unclear whether patients with uninterpretable results were excluded
Withdrawals explained? All tests	Yes	No withdrawals
Prospective study?	Yes	Prospective recruitment of consecutive patients
Expertise of imaging tests readers reported? All tests	No	Not reported
Sequence of tests determined at random? All tests	No	The order of tests was not randomised CT was performed before DWI
Scans read blind to clinical information? All tests	Unclear	Unclear whether the clinicians who interpreted CT and DW images were blinded to patients' clinical information

Chalela 2007.

Clinical features and settings	Patients with suspected acute stroke who underwent both DWI and CT within 3 hours of symptom onset Patients selection was not restricted to MCA strokes
Participants	90 patients presenting with stroke symptoms Median age 76 years (range 21 to 100 years) Median score at NIHSS = 3 (range 0 to 37)
Study design	Prospective
Target condition and reference standard(s)	Acute stroke Final diagnosis based on all available evidence including acute and follow‐up imaging
Index and comparator tests	DWI versus CT for detection of ischaemic stroke MRI sequences for detection of haemorrhagic stroke
Follow‐up	Imaging
Notes	None of the patients were treated with thrombolysis TIAs with imaging evidence of infarction were counted as true positive cases The distribution of patients was skewed towards mild cases Patients who could not tolerate MRI or with uninterpretable imaging results were excluded
Table of Methodological Quality
Item	Authors' judgement	Description
Representative spectrum? All tests	Yes	A consecutive series of patients referred to the hospital's stroke team because of suspicion of acute stroke; irrespective of time from onset, symptom severity, or ultimate clinical diagnosis
Acceptable reference standard? All tests	Yes	All available clinical information and follow up brain imaging
Acceptable delay between tests? All tests	Unclear	Time between DWI and CT for detection of ischaemic stroke: 120 minutes Time of follow‐up images not given
Partial verification avoided? All tests	Yes	All patients were verified by the reference criteria
Differential verification avoided? All tests	Unclear	Unclear whether all patients were verified by the same reference criteria
Incorporation avoided? All tests	No	Final diagnosis included both acute and follow‐up brain images
Reference standard results blinded? All tests	No	Follow‐up images were not read blind to the findings of acute images
Index test results blinded? All tests	Yes	Acute images were read blind to patients clinical details and final diagnosis
Relevant clinical information? All tests	Yes	No clinical information were provided to the clinicians who interpret CT and DW images
Uninterpretable results reported? All tests	No	Uninterpretable brain images were excluded
Withdrawals explained? All tests	Yes	No withdrawals
Prospective study?	Yes	Prospective recruitment of a series of consecutive patients
Expertise of imaging tests readers reported? All tests	Yes	Images were analysed by 2 expert neuroradiologists and 2 expert stroke neurologists
Sequence of tests determined at random? All tests	No	The order of tests was not randomised For the assessment of ischaemic stroke MRI was performed before CT
Scans read blind to clinical information? All tests	Yes	The clinicians who read the acute images were blinded to patients' clinical information and final diagnosis

Gonzalez 1999.

Clinical features and settings	Patients with suspected ischaemic stroke and with a negative or inconclusive CT scan and for whom MRI was deemed essential for establishing proper management Imaging was performed within 6 hours of symptom onset Most of the patients had a stroke in the MCA territory
Participants	22 patients (55% men) with acute stroke Mean age: 66.2 years
Study design	Retrospective Original scans were re‐examined de novo by study investigators
Target condition and reference standard(s)	Acute ischaemic stroke Clinical and imaging follow up
Index and comparator tests	DWI versus CT
Follow‐up	Clinical assessment and imaging
Notes	Haemorrhage excluded Severity of stroke not reported 3 patients were excluded because they did not undergo CT
Table of Methodological Quality
Item	Authors' judgement	Description
Representative spectrum? All tests	No	Selected sample Patients with acute stroke‐like symptoms and a negative or inconclusive CT Patients with intracerebral haemorrhage were excluded Severity of stroke not given
Acceptable reference standard? All tests	Yes	Clinical criteria and imaging follow up (CT or MRI)
Acceptable delay between tests? All tests	Yes	Mean time between CT and DWI = 4.2 hours Imaging follow up performed 24 hours or more after the onset of stroke
Partial verification avoided? All tests	Yes	All patients were verified by the reference criteria
Differential verification avoided? All tests	Yes	All patients were verified by the same reference criteria
Incorporation avoided? All tests	Yes	Final diagnosis did not include the acute CT and DW images
Reference standard results blinded? All tests	Unclear	Unclear whether the physicians who confirmed the final diagnosis were aware of the acute CT and DWI findings
Index test results blinded? All tests	Unclear	The neuroradiologists who interpreted the acute CT and DW images were not blind to patients' clinical history but they were reported to be blinded to final diagnosis
Relevant clinical information? All tests	No	A brief description of patients' clinical symptoms was provided to the neuroradiologists who interpreted CT and DW images
Uninterpretable results reported? All tests	Yes	No uninterpretable results
Withdrawals explained? All tests	Yes	No withdrawals
Prospective study?	No	Retrospective study The authors reviewed the patients' hospital records
Expertise of imaging tests readers reported? All tests	Yes	Images were reviewed by a neuroradiology fellow and a staff neuroradiologist
Sequence of tests determined at random? All tests	No	The order of tests was not randomised CT was performed before DWI
Scans read blind to clinical information? All tests	No	The 2 neuroradiologists who reviewed CT and DW images were provided with a short clinical history for each study

Oppenheim 2005.

Clinical features and settings	Patient details were retrospectively extracted from the acute stroke database of 2 university hospitals which used MRI as the first imaging modality for patients reaching hospital within 6 hours of symptoms onset Only patients with a stroke severity of ≥ 3 points on the NIHSS were deemed suitable for inclusion
Participants	86 patients (64%) with and without haemorrhagic stroke Mean age: 68.8 years
Study design	Retrospective study Original scans were re‐examined de novo by study investigators
Target condition and reference standard(s)	Acute haemorrhagic stroke Clinical and imaging follow up
Index and comparator tests	DWI and GRE MR sequences
Follow‐up	Clinical assessment and imaging
Notes	Only a minority of patients had a CT scan and the diagnosis of acute intracerebral haemorrhage was based on multisequence MRI (incorporation bias)
Table of Methodological Quality
Item	Authors' judgement	Description
Representative spectrum? All tests	Unclear	Retrospecitve study Information extracted from the acute stroke database of 2 university hospitals Patient characteristics not clearly reported
Acceptable reference standard? All tests	Unclear	Only a minority of patients had a CT scan and the diagnosis of acute intracerebral haemorrhage was based on MR sequences
Acceptable delay between tests? All tests	Unclear	Not clearly reported
Partial verification avoided? All tests	Yes	All patients were verified by the reference criteria
Differential verification avoided? All tests	Unclear	Only a minority of patients were verified by CT (numbers not provided) MRI findings were used to verify patients' clinical condition Unclear whether all patients were verified by the same MR sequences
Incorporation avoided? All tests	No	Multisequence MRI findings contributed to final diagnosis
Reference standard results blinded? All tests	No	The final diagnosis was based on all imaging information included the acute MR images
Index test results blinded? All tests	Yes	Clinicians who reviewed acute MR images were blinded to final diagnosis
Relevant clinical information? All tests	No	The clinicians who read the MR images were aware that all patients were initially referred for a suspicion of acute stroke and that they had been imaged within 6 hours of stroke onset
Uninterpretable results reported? All tests	No	All MR images not considered to be of diagnostic quality were excluded
Withdrawals explained? All tests	Yes	No withdrawals were reported
Prospective study?	No	Retrospective study Patients information extracted from the acute stroke database of 2 university hospitals
Expertise of imaging tests readers reported? All tests	Yes	Acute MR sequences were analysed by 2 expert radiologists and 1 neurologist
Sequence of tests determined at random? All tests	Yes	MR examinations were randomly numbered The 5 MR sequences were then sorted and archived separately
Scans read blind to clinical information? All tests	Yes	The clinicians who read the acute MR sequences were blinded to patients' clinical data and final diagnosis

Saur 2003.

Clinical features and settings	Patients with acute ischaemic stroke in the middle cerebral artery territory for whom DWI and CT were performed within 6 hours of stroke onset and with a time interval of less than 45 minutes
Participants	46 stroke patients (67% men) Mean age: 62.8 years (range: 35 to 89 years) Mean NIHSS score: 13.3 (range: 3 to 23)
Study design	Retrospective Original scans were re‐examined de novo by study investigator
Target condition and reference standard(s)	Ischaemic stroke (middle cerebral artery territory) Clinical diagnosis and imaging follow up
Index and comparator tests	DWI versus CT
Follow‐up	Clinical assessment and imaging
Notes	Haemorrhage excluded
Table of Methodological Quality
Item	Authors' judgement	Description
Representative spectrum? All tests	No	Selected sample Patients with acute ischaemic signs for whom imaging with CT and MRI was performed within 6 hours of stroke onset Patients with intracerebral haemorrhage were excluded No information on previous strokes
Acceptable reference standard? All tests	Yes	Clinical criteria and imaging follow up (CT or MRI)
Acceptable delay between tests? All tests	Yes	Time interval between CT and DWI = less than 45 minutes Follow up imaging 1‐12 days after stroke onset
Partial verification avoided? All tests	Yes	All patients were verified by the reference criteria
Differential verification avoided? All tests	Yes	All patients were verified by the same reference criteria
Incorporation avoided? All tests	Yes	Final diagnosis did not include the acute CT and DW images
Reference standard results blinded? All tests	No	The 2 authors who analysed the follow‐up CT or MR images to verify the site and extent of the stroke lesion were the same who interpreted the acute images
Index test results blinded? All tests	Yes	The 3 neuroradiologists and the 3 radiologists who interpreted the acute CT and DW images were not aware of the number of patients with an ischaemic stroke
Relevant clinical information? All tests	Yes	The neuroradiologists and neurologists who reviewed the acute images were blinded to patients' clinical history
Uninterpretable results reported? All tests	Yes	No uninterpretable results
Withdrawals explained? All tests	Yes	No withdrawals
Prospective study?	No	Retrospective study Authors reviewed records of patients with acute ischaemic stroke
Expertise of imaging tests readers reported? All tests	Yes	3 neuroradiologists and 3 neurologists
Sequence of tests determined at random? All tests	No	The order of tests was not randomised CT was performed before DWI
Scans read blind to clinical information? All tests	Yes	Both the neuroradiologists and neurologists who read the CT and DW images were blinded to patients' clinical details

Sorensen 1996.

Clinical features and settings	Patients with suspected stroke for whom imaging was performed within 12 hours of symptoms onset Patients with intracerebral haemorrhage were excluded
Participants	11 patients (73% men) with acute ischaemic stroke Mean age: 64 years (range 47 to 91 years)
Study design	Prospective
Target condition and reference standard(s)	Ischaemic stroke Clinical diagnosis and imaging follow up
Index and comparator tests	DWI versus CT
Follow‐up	MR imaging
Notes	Haemorrhage excluded Severity of stroke not reported Stroke vascular territory not specified
Table of Methodological Quality
Item	Authors' judgement	Description
Representative spectrum? All tests	No	Selected sample Patients presenting with symptoms of acute stroke Patients with a non‐ischaemic cause of stroke were excluded No information on severity of stroke and previous history of stroke.
Acceptable reference standard? All tests	Yes	Clinical criteria and imaging follow up (CT or MRI)
Acceptable delay between tests? All tests	Yes	DWI performed within 90 minutes of CT Time of follow up images (CT or MR) not given
Partial verification avoided? All tests	Yes	All patients were verified by the reference criteria
Differential verification avoided? All tests	Yes	All patients were verified by the same reference criteria
Incorporation avoided? All tests	Yes	Final diagnosis did not include the acute CT and DW images
Reference standard results blinded? All tests	Unclear	Unclear whether the physicians who interpreted follow up images and confirmed final diagnosis were aware of the acute imaging findings
Index test results blinded? All tests	Unclear	Each image was reviewed by a technologist, who was blind to the date and time of image acquisition, and subsequently analysed by a radiologist Unclear whether both the technologist and the radiologist were blinded to patients' clinical details and diagnosis
Relevant clinical information? All tests	Unclear	Unclear whether the clinicians who interpreted CT and DWI were provided with patients' clinical details
Uninterpretable results reported? All tests	Yes	No uninterpretable results
Withdrawals explained? All tests	Yes	No withdrawals
Prospective study?	Yes	Prospective recruitment of a consecutive series of patients
Expertise of imaging tests readers reported? All tests	Yes	A technologist and a radiologist
Sequence of tests determined at random? All tests	No	The order of tests was not randomised CT was performed before DWI
Scans read blind to clinical information? All tests	Unclear	Not clearly reported

Urbach 2000.

Clinical features and settings	Patients with acute ischaemic stroke in the middle cerebral artery territory for whom DWI and CT were performed within 6 hours of stroke onset
Participants	30 patients (60% men) with acute ischaemic stroke Mean age: 52 years (range 18 to 76 years)
Study design	Retrospective
Target condition and reference standard(s)	Acute ischaemic stroke Clinical diagnosis and imaging follow up
Index and comparator tests	DWI versus CT
Follow‐up	Clinical assessment and imaging
Notes	Severity of stroke not reported
Table of Methodological Quality
Item	Authors' judgement	Description
Representative spectrum? All tests	No	Selected sample Patients with acute cerebral hemispheric symptoms Patients with intracerebral haemorrhage were excluded No information on severity of stroke and previous history of stroke
Acceptable reference standard? All tests	Yes	Clinical criteria and imaging follow up (CT or MRI)
Acceptable delay between tests? All tests	Yes	Mean time between CT and DWI: 1 hour Time of imaging follow up not clearly reported
Partial verification avoided? All tests	Yes	All patients were verified by the reference criteria
Differential verification avoided? All tests	Yes	All patients were verified by the same reference criteria
Incorporation avoided? All tests	Yes	Final diagnosis did not include the acute CT and DW images
Reference standard results blinded? All tests	Unclear	Presence of ischaemic stroke was determined on follow‐up images by a consensus panel of all 5 neuroradiologists Unclear whether the follow up images were read blind to the acute CT and DWI findings
Index test results blinded? All tests	Unclear	The neuroradiologists who interpreted the acute CT and DW images were not blind to patients clinical details Unclear whether they were aware of the number of people with ischaemic stroke
Relevant clinical information? All tests	No	Patients clinical details were available to the neuroradiologists who reviewed acute CT and DWI
Uninterpretable results reported? All tests	Yes	No uninterpretable results
Withdrawals explained? All tests	Yes	No withdrawals
Prospective study?	No	Retrospective study
Expertise of imaging tests readers reported? All tests	Yes	5 neuroradiologists
Sequence of tests determined at random? All tests	No	The order of tests was not randomised CT was the initial examination in 25 patients, MRI in 5 patients
Scans read blind to clinical information? All tests	No	The 5 neuroradiologists who interpreted the acute images were aware of patients' symptoms

CT: computed tomography 
 DWI: diffusion‐weighted magnetic resonance imaging 
 GRE: gradient‐echo 
 MCA: middle cerebral artery 
 MR or MRI: magnetic resonance imaging 
 NIHSS: National Institute of Health Stroke Scale 
 TIA: transient ischaemic attack

Characteristics of excluded studies [ordered by study ID]

Study	Reason for exclusion
Allkemper 2004	Comparison of MRI at 1.5 and 3.0 T. No direct comparison MRI with CT.
Arenillas 2002	Focus on ‘early neurological deterioration’ in patients with proven MCA and ICA occlusion. Beyond the scope of this review.
Arnould 2004	Focus on haemorrhagic transformation in hyperacute ischaemic stroke. Beyond the scope of this review.
Ba‐Ssalamaha 2000	Heterogeneous sample. Vascular lesions in only 9 patients. No suitable diagnostic accuracy data.
Barber 2005	Comparison of CT and DWI in acute ischaemic stroke using the Alberta Stroke Programme Early Computed Tomography Score (ASPECTS criteria). No suitable diagnostic accuracy data.
Bartylla 1997	German study. No direct comparison of CT with DWI. Only DWI and T2WI assessed.
Brant‐Zawadzki 1996	Focus on FLAIR images ‐ not on DWI. No suitable imaging test.
Buckley 2003	No suitable time of imaging.
Chung 2002	Four single cases of ischaemic stroke assessed by DWI.
Chung 2003	Five single cases of haemorrhagic stroke assessed by DWI .
Dorenbeck 2005	Assessment of ADC values obtained using DWI. No suitable diagnostic accuracy data.
Dylewski 2000	Use of MRI in acute intracerebral haemorrhage. Heterogeneous etiologies of haemorrhage. No suitable patient population.
Eastwood 2003	Correlation of dynamic CT perfusion imaging and MR diffusion and perfusion imaging in acute stroke. No direct comparison of MRI with non‐contrast CT.
Ebisu 1997	No suitable diagnostic accuracy data.
Egelhof 1998	German study. MRI to detect acute ischaemic cerebral infarcts. Imaging performed within 48 hours of stroke onset (and within 24 hours in a subgroup of patients). No suitable time of imaging.
Eliasziw 2005	Letter/comment with no suitable diagnostic accuracy data.
Etgen 2004	Study looking at stroke in one anatomical region (brainstem infarcts). Only DWI assessed. 62% of patients were scanned outside 24 hours.
Fazekas 1996	Frequency and type of TIA‐related infarcts shown by MRI. Beyond the scope of this review.
Fiebach 2001	No enough data to allow construction of a 2x2 contengency table.
Fiebach 2002	Only sensitivity and specificity estimates reported. No enough data to construct a 2x2 contengency table
Fiebach 2004	No enough data to allow construction of a 2x2 contengency table.
Fitzek 1998	Comparison of CT with DWI for detection of acute ischaemic stroke. Imaging performed outside 12 hours of stroke onset.
Flacke 1998	German study. Assessment of diffusion‐weighted and perfusion imaging in addition to FLAIR‐TSE and T2W‐GraSE and MR angiography for the diagnosis of acute stroke. No direct comparison of MRI with CT.
Flacke 2000	MCA susceptibility sign compare with hyperdense MCA sign on CT. No suitable test comparison.
Girot 2003	Focus on inter‐ and intra‐observer reproducibility. No suitable diagnostic accuracy data.
Greer 2004	Evaluation of DWI versus CT for the detection of haemorrhage after thrombolysis. Beyond the scope of this review.
Griffiths 2000	Imaging performed within 18 hours of stroke onset. No suitable time of imaging.
Hacke 2000	Letter/comment with no suitable diagnostic accuracy data.
Haraguchi 2000	Japanese study. No suitable diagnostic accuracy data.
Heidenreich 2008	MRI in addition to CT for the diagnosis of hyoperacute stroke. MRI protocol included T2‐W, DWI, PWI, and MRA. No direct comparison of DWI with CT.
Hermier 2001	DWI for the detection of post‐ischaemic haemorrhage. Beyond the scope of this review
Jager 2000	Narrative review of the literature. No suitable diagnostic accuracy data.
Jaillard 2002	Early CT signs in acute stroke. Beyond the scope of this review.
Kamal 2003	Tissue response of the brain to intracranial haemorrhage as shown by DWI. Beyond the scope of this review.
Keir 2000	Systematic review of diffusion and perfusion imaging in acute ischemic stroke. No suitable diagnostic accuracy data.
Kidwell 2008	Duscussion paper on neuroimaging for the diagnosis of intracranial haemorrhage. No suitable diagnostic accuracy data.
Kimura 1999	Duration of symptoms in TIA. Beyond the scope of this review
Kloska 2004	No direct comparison of CT with DWI. Only CT assessed.
Koennecke 2001	Not suitable diagnostic data. Only positive cases on DWI analysed.
Krasnianski 2001	German study. MRI findings in patients with brainstem infarctions. Imaging performed within 7 days of stroke onset. No suitable time of imaging.
Krasnianski 2002	Brainstem infarctions in patients with normal MRI. Beyond the scope of this review.
Köhrmann 2007	Discussion paper on acute stroke imaging for thrombolytic therapy. No suitable diagnostic accuracy data
Laloux 1995	Mean interval of MRI: 11 days. No suitable time of imaging.
Lam 2003	CT and DWI for the detection of haemorrhagic stroke. Imaging performed within 40 hours of stroke onset. No enough data to allow construction of a 2X2 contingency table.
Lam 2005	Use of B0 echo planar imaging (EPI) for the detection of intracerebral bleeds. Imaging performed within 48 hours. Beyond the scope of this review.
Lansberg 2000	Comparison of DWI with CT for the detection of ischaemic stroke. No enough data to allow construction of a 2x2 contingency table.
Lansberg 2000a	Conventional MRI versus DWI. No direct comparison of DWI with CT.
Lee 2000	No direct comparison of MRI with CT.
Lee 2001	Assessment of the Yonsei Stroke Registry. No suitable diagnostic accuracy data.
Lev 2000	Focus on CTA. No suitable test comparison.
Lin 2001	Only a non‐random subset of patients undergo MRI and CT. Imaging performed within 4 days after stroke. No suitable time of imaging.
Linfante 1999	Description of five cases with intracerebral haemorrhage.
Linfante 2001	DWI in acure posterior circulation stroke. No suitable diagnostic accuracy data.
Linfante 2004	Letter/comment with no suitable diagnostic accuracy data.
Lövblad 1998	No direct comparison of CT with DWI. Only DWI assessed. Imaging performed within 24 of stroke onset.
Lövblad 1998a	Comparison of diffusion‐weighted spin‐echo with diffusion‐weighted HASTE sequences in ischaemic stroke. No data on CT.
Marx 2004	German study. DWI in vertebrobasilar ischaemia. Only posterior circulation strokes included. Imaging performed within 24 hours of stroke onset. No suitable time of imaging.
Masdeu 2006	Guideline on neuroimaging in acute stroke. No suitable diagnostic accuracy data.
Mayer 2000	Focus on haemorrhagic transformation. Beyond the scope of this review.
Melhem 1998	Dual‐echo gradient‐ and spin‐echo and fast spin‐echo MRI for haemorrhagic lesions. Heterogeneous patient population. Causes of haemorrhage included ischemia, trauma, vascular malformations, hypertension, and brain tumors.
Mohr 1995	T1‐T2 versus CT. No suitable test comparison
Mullins 2002	Retrospective studies on CT and DWI for detection of acute stroke. Ischaemic and haemorrhagic cases were not reported separately. No direct comparison of CT and DWI.
Mullins 2002a	Retrospective studies on CT and DWI for detection of acute stroke. Ischaemic and haemorrhagic cases were not reported separately. No direct comparison of CT and DWI. Same data as in Mullins 2002.
Na 1998	Evaluation of MCA occlusion using triphasic helical CT. Beyond the scope of this review.
Nighoghossian 2001	Focus on haemorrhagic transformations. Beyond the scope of this review.
Olszycki 2007	CT and MRI in patients with acute stroke. Imaging performed between 3 and 15 hours of stroke onset. No suitable time of imaging.
Oppenheim 2000	Assessment of DWI and FLAIR sequences for the diagnosis of ischaemic stroke. No data on CT.
Patel 1996	MRI for the detection of intraparenchimal haemorrhage. Description of five cases.
Poniatowska 2007	No direct comparison of DWI with CT. DWI performed only on negative cases. No suitable test comparison.
Powers 2000	Letter/comment with no suitable diagnostic accuracy data.
Rajajee 2008	Clinical and CT criteria versus MRI for the diagnosis of small deep infarcts. DW and MRA imaging used to exclude large‐vessel stenosis or occlusion (reference standard). No suitable test comparison.
Razumovsky 1999	TCD, MRA, and MRI in acute cerebral ischemia. No data on DWI.
Read 1998	CT at admission and DWI for the diagnosis of ischaemic stroke. Delay between CT and DWI varied from 11 to 36 hours. No suitable time of imaging.
Restrepo 2004	Assessment of TIA with diffusion and perfusion MRI. No suitable test comparison.
Rincon 2004	Dynamic CT perfusion for acute ischemia. No suitable imaging test.
Roberts 2001	Focus on CT perfusion. No suitable imaging test.
Rovira 2000	No direct comparison of CT with DWI. Imaging performed within 48 hours of stroke onset. No suitable time of imaging.
Rovira 2002	DWI in acute TIA. Patients studied with MRI within 10 days. No suitable time of imaging.
Schellinger 1999	Selected sample (9 patients with ICH). Assessment of hematoma size on CT and MRI. Beyond the scope of this review.
Schellinger 2000	Practicality of MRI in acute ischemia. Beyond the scope of this review.
Schellinger 2001	PWI and DWI lesion volumes in hyperacute ischaemia. Beyond the scope of this review.
Schramm 2002	Focus on CTA versus MRA. Assessment of blood volumes. No suitable test comparison.
Singer 1998	No direct comparison of CT with DWI. Only DWI assessed. Mean time from stroke onset to imaging: 48.1 hours (range 7 hours ‐ 4 days). No suitable test comparison.
Smajlovic 2004	DWI and CT in acute ischaemic stroke. DWI performed 48 hours after stroke onset. No suitable time of imaging.
Stapf 2000	No direct comparison of CT with DWI. Only CT assessed.
Sunshine 2001	No direct comparison of CT with DWI. Only DWI assessed.
Sunshine 2004	Discussion paper on the use of CT, MRI and MRA in the evaluation of acute stroke. No suitable diagnostic accuracy data.
Tei 1997	Japanese study. No direct comparison of DWI with CT.
Toyoda 2001	Use of FLAIR for detecting intra‐arterial signal of ischaemia. Beyond the scope of this review.
van Everdingen 1998	No direct comparison of CT with DWI. Only DWI assessed.
Verro 2002	Focus on CT angiography. Beyond the scope of this review.
Von Kummer 2000	Letter/comment with no suitable diagnostic accuracy data.
von Kummer 2001	No direct comparison of CT with DWI. Only CT assessed.
Von Kummer 2002	Letter/comment with no suitable diagnostic accuracy data.
Wang 1997	Chinese study. Not a diagnostic accuracy study.
Warach 1995	No direct comparison of CT with DWI. Only DWI assessed. Imaging performed within 48 of stroke onset.
Warach 1996	No direct comparison of CT with DWI. Only DWI assessed. Imaging performed within 48 of stroke onset. Same data as in Warach 1995.
Wardlaw 2003	Impact of delays in CT of the brain on the accuracy of stroke diagnosis. Beyond the scope of this review.
Watanabe 2000	Japanese study. No suitable test comparisons. No CT data.
Weber 2003	German study. No direct comparison of DWI with CT.
Wintermark 2005	Accuracy of dynamic perfusion CT. No suitable imaging test.
Wycliffe 2004	MRI for detection of haemorrhagic transformations. Beyond the scope of this review.
Zivin 1997	Letter/comment with no suitable diagnostic accuracy data.

Differences between protocol and review

One author (S Lewis) contributed to the protocol but not to the review.

The non‐English articles for which a translation could not be obtained were not noted in the Appendix, as stated in the protocol under the 'Methods' section, but were listed amongst the excluded studies in the Characteristics of excluded studies table.

In the protocol, in the 'Data collection and analysis' section, we stated that two authors would independently review all relevant full‐text reports and assess the methodological quality of included studies. Instead, three authors reviewed the full‐text reports and four authors independently assessed the methodological quality of all included studies.

We were not able to use the hierarchical SROC methods (HSROC), as described in the 'Statistical analysis and data synthesis' in the protocol, as our data were too sparse to estimate the correlation between sensitivity and specificity for DWI and CT.

Contributions of authors

MB, PS, and JW were responsible for the overall planning and conducting of the systematic review. MB, MGC, ER, and NA contributed to the quality assessment of included studies. MB, MGC, and ER extracted data from primary studies and interpreted results. MB, FC, and JD contributed to the statistical analyses. MB wrote the first draft of the review with additional input from PS, JW, and JD. MB supervised subsequent revisions based on comments from all authors. All authors have seen and approved the final version of the review.

Sources of support

Internal sources

• Chief Scientist Office of the Scottish Government Health Directorates (http://www.sehd.scot.nhs.uk), UK.

External sources

• No sources of support supplied

Declarations of interest

All authors declared no conflicts of interest.

New