ORIGINAL CONTRIBUTION

Comparison of MRI and CT for Detection

of Acute Intracerebral Hemorrhage
Chelsea S. Kidwell, MD Context Noncontrast computed tomography (CT) is the standard brain imaging study
Julio A. Chalela, MD for the initial evaluation of patients with acute stroke symptoms. Multimodal mag-
netic resonance imaging (MRI) has been proposed as an alternative to CT in the emer-
Jeffrey L. Saver, MD
gency stroke setting. However, the accuracy of MRI relative to CT for the detection of
Sidney Starkman, MD hyperacute intracerebral hemorrhage has not been demonstrated.
Michael D. Hill, MD Objective To compare the accuracy of MRI and CT for detection of acute intrace-
Andrew M. Demchuk, MD rebral hemorrhage in patients presenting with acute focal stroke symptoms.
John A. Butman, MD, PhD Design, Setting, and Patients A prospective, multicenter study was performed
at 2 stroke centers (UCLA Medical Center and Suburban Hospital, Bethesda,
Nicholas Patronas, MD Md), between October 2000 and February 2003. Patients presenting with focal
Jeffry R. Alger, PhD stroke symptoms within 6 hours of onset underwent brain MRI followed by non-
contrast CT.
Lawrence L. Latour, PhD
Main Outcome Measures Acute intracerebral hemorrhage and any intracerebral
Marie L. Luby, MEng, MS hemorrhage diagnosed on gradient recalled echo (GRE) MRI and CT scans by a con-
Alison E. Baird, FRACP, PhD sensus of 4 blinded readers.
Megan C. Leary, MD Results The study was stopped early, after 200 patients were enrolled, when it be-
Margaret Tremwel, MD, PhD came apparent at the time of an unplanned interim analysis that MRI was detecting
cases of hemorrhagic transformation not detected by CT. For the diagnosis of any hem-
Bruce Ovbiagele, MD orrhage, MRI was positive in 71 patients with CT positive in 29 (P⬍.001). For the di-
Andre Fredieu, MD agnosis of acute hemorrhage, MRI and CT were equivalent (96% concordance). Acute
hemorrhage was diagnosed in 25 patients on both MRI and CT. In 4 other patients,
Shuichi Suzuki, MD, PhD acute hemorrhage was present on MRI but not on the corresponding CT—each of
J. Pablo Villablanca, MD these 4 cases was interpreted as hemorrhagic transformation of an ischemic infarct. In
Stephen Davis, MD 3 patients, regions interpreted as acute hemorrhage on CT were interpreted as chronic
hemorrhage on MRI. In 1 patient, subarachnoid hemorrhage was diagnosed on CT
Billy Dunn, MD but not on MRI. In 49 patients, chronic hemorrhage, most often microbleeds, was
Jason W. Todd, MD visualized on MRI but not on CT.
Mustapha A. Ezzeddine, MD Conclusion MRI may be as accurate as CT for the detection of acute hemorrhage in
patients presenting with acute focal stroke symptoms and is more accurate than CT
Joseph Haymore, MS, ACNP for the detection of chronic intracerebral hemorrhage.
John K. Lynch, DO, MPH JAMA. 2004;292:1823-1830 www.jama.com
Lisa Davis, MSN, RN
Steven Warach, MD, PhD
Author Affiliations: UCLA Stroke Center (Drs Kidwell, University of Calgary, Calgary, Alberta (Dr Hill); Di-

N
ONCONTRAST COMPUTED TO- Saver, Starkman, Alger, Tremwel, Ovbiagele, Fredieu, agnostic Radiology Department, Warren G. Magnu-
mography (CT) has been the Suzuki, and Villablanca), Department of Neurology (Drs son Clinical Center, National Institutes of Health,
Kidwell, Saver, Starkman, Tremwel, Ovbiagele, Fredieu, Bethesda, Md (Drs Butman, Patronas, and Alger); Di-
standard imaging modality and Suzuki), Department of Emergency Medicine (Dr vision of Stroke and Vascular Neurology, Beth Israel
for the initial evaluation of Starkman), Department of Radiology (Drs Fredieu, Su- Deaconess Medical Center, Boston, Mass (Dr Leary);
zuki, and Villablanca), Ahmanson-Lovelace Brain Map- Sparks Neurology Center, Fort Smith, Ark (Dr Trem-
patients presenting with acute stroke ping Center (Dr Alger), UCLA Medical Center, Los An- wel); Royal Melbourne Hospital, Victoria, Australia (Dr
symptoms.1,2 The primary diagnostic ad- geles, Calif; National Institute of Neurological Disorders Davis); Department of Neurology, Emory University,
vantage of CT in the hyperacute phase and Stroke, National Institutes of Health, Bethesda, Atlanta, Ga (Dr Ezzeddine); Center for Neuroscience
Md (Drs Chalela, Butman, Patronas, Latour, Baird, Services, Washington Adventist Hospital, Takoma Park,
(0 to 6 hours) is its ability to rule out Dunn, Todd, Lynch, and Warach, and Mss Luby, and Md (Mr Haymore).
Davis); Calgary Stroke Program, Department of Clini- Corresponding Author: Chelsea S. Kidwell, MD, Wash-
cal Neurosciences, University of Calgary, Foothills Hos- ington Hospital Center Stroke Center, 100 Irving St
See also p 1883 and Patient Page. pital, Calgary, Alberta (Drs Hill and Demchuk); De- NW, East Bldg Room 6126, Washington, DC 20010
partments of Medicine/Community Health Sciences, (ckidwell@ucla.edu).

©2004 American Medical Association. All rights reserved. (Reprinted) JAMA, October 20, 2004—Vol 292, No. 15 1823

Downloaded from www.jama.com at Ucla, on July 23, 2005

MRI VS CT FOR HEMORRHAGE DETECTION

cal Center and National Institutes of

Figure 1. Acute Intraparenchymal Hematoma Imaged With Computed Tomography and
With Magnetic Resonance Imaging Health [NIH] Stroke Center at Subur-
ban Hospital, Bethesda, Md). Initially,
Computed Tomography Magnetic Resonance Imaging (GRE) 2 additional centers were involved but
A B
subsequently discontinued participa-
tion in the study because of inadequate
patient enrollment. Patients present-
ing with focal stroke symptoms within
6 hours of onset were screened for en-
rollment. Only symptomatic patients
with a definite last known well time
when initial imaging took place were eli-
gible. Patients were excluded if any of
the following were present: coma; pace-
maker or other contraindication to MRI;
symptoms suggestive of subarachnoid
hemorrhage (SAH); inability to obtain
MRI within 6 hours from last known
well time; initiation of thrombolytic
therapy, intravenous antithrombotics or
anticoagulants, or antithrombotic in-
Example of acute intraparenchymal hematoma imaged at 1 hour 54 minutes from symptom onset with com-
puted tomography (A, white arrowhead) and at 1 hour 12 minutes from symptom onset with magnetic reso- vestigational drug prior to completion
nance imaging (B, gradient recalled echo [GRE] image, black arrowhead). of both imaging studies; or cardiores-
piratory instability precluding MRI.
the presence of hemorrhage. Accurate ing the initial 6 hours after stroke symp- Site participation in the study was
early detection of blood is crucial since tom onset. A growing body of data have contingent upon the site’s current rou-
a history of intracerebral hemorrhage is suggested that hyperacute parenchy- tine clinical practice of obtaining MRI
a contraindication to the use of throm- mal blood can be accurately detected followed by CT for patients with po-
bolytic agents. However, a major disad- using gradient recalled echo (GRE) tential acute stroke. The institutional
vantage of conventional CT within the pulse sequences that are sensitive to review board (IRB) at each site gave ap-
first few hours of symptom onset is its static magnetic field inhomogeneity proval to prospectively collect and ana-
limited sensitivity for identifying early (ie, T2*-sensitive).5-9 These sequences lyze clinical and imaging data with iden-
evidence of cerebral ischemia. detect the paramagnetic effects of tifying information removed. At UCLA,
Conversely, multimodal magnetic deoxyhemoglobin and methemoglo- the IRB waived consent; at Suburban
resonance imaging (MRI), including dif- bin. The hyperacute lesion on GRE/ Hospital, the study was performed un-
fusion-weighted imaging (DWI), has T2* typically consists of a core of hetero- der an IRB approved natural history of
excellent capacity to delineate the pres- geneous signal intensity, reflecting the stroke protocol in which waiver was
ence, size, location, and extent of hy- most recently extravasated blood that permitted in individual cases if waiver
peracute ischemia,3 but unproven reli- may still contain significant amounts of of consent could not be obtained.
ability in identifying early parenchymal diamagnetic oxyhemoglobin, sur-
hemorrhage. The advent of thrombo- rounded by a rim of hypointensity, sig- Imaging Techniques
lytic therapy and other interventional nifying parenchymal blood that has had All patients underwent MRI followed
therapies for acute ischemic stroke has time to become more fully deoxygen- by CT. Imaging time goals were
led to increasing interest in using MRI ated and paramagnetic (FIGURE 1).7,10 completion of both MRI and CT within
to select and stratify candidates for treat- We undertook a prospective com- 90 minutes of presentation to the emer-
ments.4 Currently, many stroke cen- parison study of MRI vs CT in a large gency department, with no more than
ters obtain both CT and MRI in the ini- cohort of patients with acute stroke to 30 minutes between the end of MRI and
tial evaluation of patients with stroke. establish that GRE MRI sequences are the start of CT. Each site was required
The use of both modalities is time- sensitive to acute hemorrhage. to keep a monthly log of all patients pre-
consuming and expensive. senting within 6 hours of stroke symp-
While conventional T1- and METHODS tom onset to ensure that at least 50%
T2-weighted MRI pulse sequences are Patients and Settings of all fully eligible patients were being
sensitive for the detection of subacute The Hemorrhage and Early MRI Evalu- enrolled. To qualify for enrollment,
and chronic blood, they are less sensi- ation (HEME) study was performed at both GRE and DWI had to be
tive to parenchymal hemorrhage dur- 2 academic stroke centers (UCLA Medi- completed.
1824 JAMA, October 20, 2004—Vol 292, No. 15 (Reprinted) ©2004 American Medical Association. All rights reserved.

Downloaded from www.jama.com at Ucla, on July 23, 2005

 MRI VS CT FOR HEMORRHAGE DETECTION

All MRIs were performed on 1.5-T

Table 1. Clinical Characteristics of HEME Cohort
scanners equipped with echo-planar im-
Age, mean (range), y 75 (25-99)
aging capability: UCLA, Siemens Vi-
Sex, No. (%) women 200 (55)
sion (Siemens Medical System, Iselin,
Time to MRI, median (range)* 2 h 13 min (23 min to 5 h 58 min)
NJ); and Suburban Hospital, GE Signa
Time to CT, median (range) 3 h 3 min (56 min to 6 h 20 min)
scanner (General Electric Medical Sys-
Baseline NIHSS† score, median (range) 6 (0-33)
tems, Milwaukee, Wis). Computed to- Final discharge diagnosis, No. (%)
mographic scans were performed on AICS 169 (84.5)
1 of the following fourth-generation Acute intracerebral hemorrhage 27 (13.5)
scanners: Somatom Plus scanner (Sie- Both 4 (2)
mens), High Speed Advantage scan- Abbreviations: AICS, acute ischemic cerebrovascular syndrome15; CT, computed tomography; HEME, Hemorrhage
and Early MRI Evaluation; MRI, magnetic resonance imaging.
ner (General Electric), or Lightspeed *MRI had to be performed within 6 h of symptom onset. Goal time from MRI to CT was 30 min. In 4 patients, time from
scanner (General Electric). Images were MRI to CT time exceeded this goal (range for these 4 patients was 42-63 min).
†National Institutes of Health Stroke Scale (NIHSS), an examination technique used by clinicians to assess and grade
acquired following the orbito-meatal level of consciousness, best gaze, visual, facial palsy, motor arm, motor leg, limb ataxia, sensory, best language,
dysarthria, and extinction and inattention capabilities in suspected stroke patients. NIHSS is a 0-42 point scale with
plane with 5 mm thickness for the en- higher score indicating more severe neurologic deficit.
tire examination. Both scanners used
the following pulse sequence param-
eters: slice thickness, 7 mm (GRE and type(s) (subarachnoid, subdural, epi- terpretation was reached by group con-
DWI); repetition time (TR), 800 ms dural, intraventricular, intraparenchy- sensus discussion.Final hospital
(GRE); flip angle 30° (GRE); acquisi- mal), location (cortical, subcortical white discharge diagnosis incorporating all
tion matrix, 256⫻192 (GRE) and matter, basal ganglia, brainstem, cer- available clinical, laboratory, and imag-
128⫻128 (DWI). Pulse parameters at ebellum, thalamus), and number (single ing data was made at the time of dis-
UCLA and at Suburban Hospital, re- or multiple). For MRI interpretations, charge by the attending physician.
spectively, were: field of view, 24 cm readers had access to DWI b0, trace DWI
and 22 cm (GRE and DWI); echo time b1000, and GRE images. Statistical Analysis
(TE), 20 ms and 15 ms (GRE); TR, 6000 Intraparenchymal hemorrhage was The primary objective of the study was
ms and 60 000 ms (DWI) (20 con- further classified as hematoma, hemor- to compare the accuracy of MRI vs CT
tiguous slices, interleaved, and co- rhagic transformation, or microbleed. for the detection of acute hemorrhage.
localized); and TE, 100 ms and 72 ms Microbleeds were defined as rounded, Secondary objectives were to compare
(DWI). punctate, homogeneous hypointensi- the accuracy of MRI vs CT for any hem-
ties generally less than 0.5 cm in size orrhage (acute or chronic) and for
Outcome Measures within the parenchyma, visualized on chronic hemorrhage alone.
A panel of 4 readers (2 neuroradiolo- GRE MRI scans, and thought to repre- Initial sample size calculations as-
gists and 2 stroke neurologists) inde- sent regions of chronic hemosiderin sumed that CT was 100% accurate for
pendently evaluated each scan blinded deposition.11-13 Hemorrhagic transfor- hemorrhage and sought to demon-
to the clinical information and all pa- mation (petechial hemorrhage) was de- strate that MRI was also 100% accu-
tient identifiers. None of the 4 readers fined as a region of hyperdensity (CT) rate. In this noninferiority design, the
was involved in the clinical care or evalu- or hypointensity (GRE MRI) occurring sample size required to narrow the dif-
ation of the enrolled patients. Before per- within an acute, subacute, or chronic is- ference in the 95% confidence interval
forming study interpretations, the read- chemic lesion. Chronic hematoma was (CI) between MRI and CT to less than
ers were given examples (compiled from defined as a slit-like region of hypoden- 5% was exact concordance between
an independent data set) of each hem- sity (CT) or hypointensity (GRE MRI) MRI and CT on 55 hemorrhages. The
orrhage type to ensure consistency of in- thought to be due to hemosiderin depo- selected software was Microsoft Ex-
terpretation to a common standard. In- sition from a remote hematoma. Com- cel, using binomial theory. The a priori
terpretations for each imaging modality puted tomographic acute hemorrhage confidence level is 95%; however, an a
(CT vs MRI) for a single patient were volumes were subsequently calculated priori significance level is unavailable
performed on different days to avoid (by C.S.K.) using a volumetric imaging since we are making a confidence
reader recognition or recall of findings analysis program. bound, not significance testing.
from the other modality. The order of If unanimous agreement regarding the In early 2003, an unplanned interim
presentation of the films was random- presence and acuity of hemorrhage on analysis was performed when prelimi-
ized and differed for each modality. The an individual scan was not achieved by nary results of a complementary study
following data were recorded by each each of the 4 readers, the interpretation became available.14 During the interim
reader for each scan: hemorrhage pres- of the majority of readers was used as the analysis, it became apparent that MRI was
ent or absent; if hemorrhage present, final imaging diagnosis. In evenly dis- detecting acute hemorrhages not visu-
hemorrhage age (acute or chronic), tributed disagreements (2 vs 2), final in- alized on CT and, therefore, the initial
©2004 American Medical Association. All rights reserved. (Reprinted) JAMA, October 20, 2004—Vol 292, No. 15 1825

Downloaded from www.jama.com at Ucla, on July 23, 2005

MRI VS CT FOR HEMORRHAGE DETECTION

sample size, based on the assumption of clusion of the 191 patients who were not patients, the region of acute hemorrhage
using CT as the criterion standard, was enrolled include: pacemaker or other apparent on CT was also apparent on
not valid. Accordingly, the primary contraindication to MRI (43); medical in- MRI but was interpreted as “chronic
analysis plan was changed to bidirec- stability for MRI such as vomiting, coma, hemorrhage” rather than acute. In the
tional comparison of CT vs MRI with- or cardiorespiratory instability (10); non- fourth patient, a region of hyperdensity
out assuming that one technique was in- availability of both imaging techniques on CT in the left frontal lobe was inter-
herently a criterion standard. In addition, within the time window (99); initiation preted as subarachnoid blood by 2 of 4
at this juncture, the study was stopped of thrombolytic or anticoagulant therapy readers on CT; on MRI this abnormal-
early after 200 patients were enrolled, as before or between scans (9); and other ity was clearly apparent as a serpiginous,
the investigators believed it would be im- reasons (30). hypointense lesion in a sulcus on GRE
portant to expedite, complete, and re- Characteristics of enrolled patients images. Although 1 of the readers did ini-
port the analysis of these patients be- are summarized in TABLE 1. The com- tially interpret this lesion as acute SAH
cause of the potential major impact the parisons between CT and MRI perfor- on MRI, 3 did not. The final discharge
findings could have on current acute mance for any hemorrhage, acute hem- diagnosis for this patient was acute is-
stroke management. orrhage, and chronic hemorrhage are chemic stroke with SAH.
Interrater reliability was calculated shown in TABLE 2 and TABLE 3. Ranges For the 26 primary intraparenchy-
for paired observers of both CT and for interrater reliability based on the ␬ mal hematomas visualized on CT, me-
MRI interpretations using the kappa (␬) statistic for paired observers were: 0.75 dian hematoma volume was 20.8 mL
statistic. The McNemar test for paired to 0.82 for acute hemorrhage on MRI (range, 0.2-157.2 mL). Subarachnoid
proportions was used to determine if and 0.87 to 0.94 for acute hemorrhage blood was visualized in 2 cases, includ-
one imaging modality diagnosed hem- on CT; 0.42 to 0.66 for chronic hem- ing 1 isolated SAH associated with acute
orrhage more frequently than the other. orrhage on MRI (not applicable for CT); ischemic stroke (case above) and 1 SAH
0.58 to 0.80 for any hemorrhage on MRI with an intracerebral and intraventricu-
RESULTS and 0.85 to 0.92 for any hemorrhage lar hemorrhage. In the latter, the sub-
Between October 2000 and February on CT. arachnoid blood component was noted
2003, 391 consecutive patients present- The panel read acute hemorrhage in on CT only, although both CT and MRI
ing with focal stroke symptoms within 25 patients on both CT and MRI. In detected the intraparenchymal and in-
6 hours of onset were screened for en- 4 additional patients, acute hemor- traventricular components. Intraven-
rollment in the study and a total of 200 rhage was interpreted as being present tricular blood was interpreted as being
patients were enrolled. Reasons for ex- on MRI but not on the corresponding present in 16 patients, all with intra-
CT (FIGURE 2). In all 4 of these pa- parenchymal hematomas. The intra-
tients, regions of hypointensity were ventricular blood was apparent on both
Table 2. MRI vs CT Panel Results (n=200) seen on the GRE images within an CT and MRI in 11 cases, on MRI only
CT+ CT– P Value ischemic field (identified by DWI). in 1, and on CT only in 4. Subdural
Any hemorrhage Each of these cases was interpreted as hemorrhage was seen in only 1 pa-
MRI+ 28 43 tient and was identified on both MRI
MRI− 1 128 .001 hemorrhagic transformation of an is-
Acute hemorrhage chemic infarct by the treating physi- and CT. No epidural hematomas were
MRI+ 25 4 identified.
MRI− 4 167 ⬎.99 cians based on all clinical and radio-
Chronic hemorrhage logic data. Chronic hemorrhage was seen in 52
MRI+ 0 52 In 4 patients, acute hemorrhage was patients on MRI and in no patients on
MRI− 0 148 ⬍.001
read by the panel on CT but not the cor- CT. Of these 52 MRI patients, 4 were in-
Abbreviations: CT, computed tomography; MRI, mag-
netic resonance imaging. responding MRI (FIGURE 3). In 3 of these terpreted as regions of chronic hemor-
rhagic transformation, 9 as chronic he-
Table 3. Comparison of Individual and Consensus Interpretations for Acute Hemorrhage matomas, 34 as 1 or more microbleeds
for MRI vs CT (FIGURE 4), and 7 as both one or more
CT MRI microbleeds and one or more hemato-
No. of Readers mas. Three of the cases interpreted as
Diagnosing Acute Consensus Read, Consensus Read,
Hemorrhage No. (%) % Positive No. (%) % Positive chronic hematoma on MRI were visu-
All 4 Yes 26 (13) 100 20 (10) 100 alized as acute blood on CT. Of the 41
3 Yes 4 (2) 100 7 (3.5) 100 patients with MRI-evident micro-
2 Yes 0 0 4 (2) 50 bleeds, 10 had single lesions and 31
1 Yes 5 (2.5) 0 10 (5) 0 multiple lesions—with none visual-
None Yes 165 (82.5) 0 159 (79.5) 0 ized on CT.
Total 200 (100) 200 (100) Compared with final discharge di-
Abbreviations: CT, computed tomography; MRI, magnetic resonance imaging. agnosis, which incorporated informa-
1826 JAMA, October 20, 2004—Vol 292, No. 15 (Reprinted) ©2004 American Medical Association. All rights reserved.

Downloaded from www.jama.com at Ucla, on July 23, 2005

 MRI VS CT FOR HEMORRHAGE DETECTION

tion from both imaging studies as well rhage and 6 of the 8 discordant hem- COMMENT
as additional laboratory, pathologic, and orrhage cases (3 each for CT negative, Neuroimaging plays a crucial role in the
clinical data, CT and MRI performed and MRI negative) were within this evaluation of patients presenting with
equally well with no significant differ- cohort. Thirty-four patients were acute stroke symptoms. While patient
ence in the accuracy of the scans ob- treated with intravenous tissue plas- symptoms and clinical examinations
tained from UCLA Medical Center vs minogen activator (tPA) within 3 may suggest the diagnosis, only brain
Suburban Hospital (TABLE 4). hours of onset. The remaining patients imaging studies can confirm the diag-
A first imaging study was performed were not treated due to rapidly resolv- nosis and differentiate hemorrhage from
within 3 hours of onset for 129 ing or nondisabling deficits, or other ischemia with high accuracy. This dif-
patients. Nineteen of the cases with a contraindication to thrombolytic ferentiation is critical in making acute
final discharge diagnosis of hemor- therapy. treatment decisions, including patient

Figure 2. Hemorrhage Visualized on Magnetic Resonance Imaging but Not on Computed Tomography

A Patient 1 Magnetic Resonance Imaging

Computed Tomography Gradient Recalled Echo Diffusion-Weighted Imaging

B Patient 2 Magnetic Resonance Imaging

Computed Tomography Gradient Recalled Echo Diffusion-Weighted Imaging

Computed tomography (CT) and magnetic resonance (MR) images from representative axial slices from 2 patients (A,B) in whom hemorrhage was visualized on MRI,
but not CT, by our consensus panel. For each patient, the left panel shows the CT image, the middle panel shows the corresponding gradient recalled echo (GRE)
image, and the far right panel shows the diffusion-weighted images (DWI). In each case, hemorrhagic transformation was visualized on GRE (black arrowheads) oc-
curring within regions of ischemia (yellow arrowheads) visualized on DWI scan. Hypointensity on GRE indicates susceptibility induced signal loss due to hemorrhage.

©2004 American Medical Association. All rights reserved. (Reprinted) JAMA, October 20, 2004—Vol 292, No. 15 1827

Downloaded from www.jama.com at Ucla, on July 23, 2005

MRI VS CT FOR HEMORRHAGE DETECTION

eligibility for thrombolytic therapy. Al- on MRI in all cases of CT positive acute ated the accuracy of CT vs MRI in dis-
though noncontrast CT has been con- intraparenchymal hematomas; in 25 tinguishing acute intracerebral hemor-
sidered the criterion standard for as- cases, the blood was interpreted as acute rhage (50 patients) from acute ischemic
sessing intracerebral hemorrhage, and in 3 cases as chronic. stroke (50 patients) using a design in
formal studies have never been per- The HEME study provides comple- which patients were randomized to
formed to validate the accuracy of this mentary data to that of a recently pub- either CT or MRI first. The HEME study
technique compared to the true crite- lished study performed by the Ger- enrolled all eligible patients, rather than
rion standard, pathology. In our study, man Stroke Competence Network (B5 simply an equal number of patients with
hemorrhage was accurately identified Hemorrhage Study).14 This group evalu- hemorrhagic and ischemic stroke.

Figure 3. Hemorrhage Visualized on Computed Tomography but Not Interpreted as Acute Blood on Magnetic Resonance Imaging

A Patient 3 Magnetic Resonance Imaging

Computed Tomography Gradient Recalled Echo Diffusion-Weighted Imaging

B Patient 4 Magnetic Resonance Imaging

Computed Tomography Gradient Recalled Echo Diffusion-Weighted Imaging

Computed tomography (CT) and magnetic resonance (MR) images from representative axial slices from 2 patients (A,B) in whom hemorrhage was visualized on CT,
but not interpreted as acute blood on MRI, by our consensus panel. For each patient, the left panel shows the CT image, the middle panel shows the corresponding
gradient recalled echo (GRE) image, and the right panel shows the diffusion-weighted image (DWI). In patient 3, the hemorrhage is apparent on CT as a hyperdense
lesion (white arrowhead). A corresponding hypointensity is marked on the GRE (black arrowhead) and on the DWI image (yellow arrowhead). In this patient, the MRI
lesion was recognized as blood but was interpreted as chronic, not acute, hemorrhage. In patient 4, a left frontal lesion is interpreted as subarachnoid blood on CT
(white arrowhead). This lesion is apparent on the GRE sequence (black arrowhead) but was interpreted as blood by only 1 of 4 members of our panel. The corre-
sponding DWI image shows hyperintensity indicative of acute ischemia within the left anterior cerebral artery territory (yellow arrowheads).

1828 JAMA, October 20, 2004—Vol 292, No. 15 (Reprinted) ©2004 American Medical Association. All rights reserved.

Downloaded from www.jama.com at Ucla, on July 23, 2005

 MRI VS CT FOR HEMORRHAGE DETECTION

Figure 4. Multiple Subcortical Microbleeds With Acute Cerebellar Hemorrhage

Magnetic Resonance Imaging

Computed Tomography Gradient Recalled Echo Gradient Recalled Echo

A B C

Multiple subcortical microbleeds are not identified as old hemorrhage on computed tomography (CT) (panel A). Representative axial slices from gradient recalled echo
(GRE) sequence demonstrating the microbleeds (panel B, black arrowheads); in addition to an acute cerebellar hemorrhage (panel C, black arrowhead). Acute intra-
ventricular hemorrhage is seen in the left occipital horn on both MRI and CT.

Our panel of readers identified acute fied as chronic microbleeds. In these

Table 4. MRI and CT Panel Results for
hemorrhage on MRI in 4 cases in which cases, the typical pattern of acute hem- Acute Hemorrhage vs Final Discharge
hemorrhage was not apparent on CT. orrhage on GRE was more difficult for Diagnosis (n=200)*
Each of these was interpreted as a region reviewers to appreciate (2 patients) or not Positive Negative
of hemorrhagic transformation (pete- present (1 patient). Physicians should be Discharge Discharge
Diagnosis Diagnosis
chial blood) within an acute ischemic aware that in cases of small hemor-
Positive MRI 28 1
infarct field. These results are sup- rhages (non-microbleeds), it may be dif- Negative MRI 3 168
ported by recent case reports of “CT- ficult to make an exact distinction be- Positive CT 28 1
Negative CT 3 168
negative intracerebral hemorrhages.”16-18 tween acute and chronic hemorrhage
Abbreviations: CT, computed tomography; MRI, mag-
It is possible that hyperacute hemor- based on GRE images alone. A noncon- netic resonance imaging.
*P values for both MRI and CT panel results were .63.
rhagic transformation of ischemic infarcts trast CT may be necessary in these cases
is an underrecognized phenomenon. to determine hemorrhage age. With acute
The implication of this finding for the medium-large hemorrhages, the char- silent, small, punctate hemosiderin
neuroimaging evaluation of acute stroke acteristic appearance of mixed signal in- lesions appearing hypointense on GRE
patients who are candidates for throm- tensity and the surrounding hyperinten- sequences.13 The role of microbleeds in
bolytic therapy is unclear. In the National sity due to edema is very specific and will determining patient eligibility for throm-
Institute of Neurological Disorders and make the age of the hemorrhage appar- bolytic therapy remains unknown. How-
Stroke (NINDS) trial, intravenous tPA ent.7 However, small hemorrhages may ever, prior studies suggest that the
was shown to be effective based on CT have similar characteristics to calcifica- presence of microbleeds may be an in-
enrollment criteria.19 While it may be tions and intravascular thrombus and dependent risk factor for hemorrhage in
hypothesized that patients with MRI evi- have minimal edema making the deter- patients treated with antithrombotic or
dence of hemorrhagic transformation are mination of hemorrhage age as well as thrombolytic therapy.11,21-23
at higher risk of developing symptom- the distinction of hemorrhage vs non- Our study has several limitations. We
atic hemorrhage if treated with throm- hemorrhage more difficult.10,20 initiated the study using CT as the cri-
bolytics, it is also possible that overall this The HEME study confirms the supe- terion standard for diagnosis of hemor-
group of patients may receive net ben- riority of GRE MRI sequences for the rhage. However, following the un-
efit from therapy. A prospective study will identification of chronic hemorrhage. In planned interim analysis indicating that
be needed to answer this question. 49 patients, GRE demonstrated chronic GRE sequences were detecting hemor-
All 3 acute hemorrhages that the panel blood not apparent on CT. The major- rhage not seen on CT, we switched to a
misclassified as chronic on MRI were ity of these chronic hemorrhages were 2-sided analysis based on the assump-
relatively small, but none were classi- categorized as microbleeds—clinically tion that these MRI findings repre-
©2004 American Medical Association. All rights reserved. (Reprinted) JAMA, October 20, 2004—Vol 292, No. 15 1829

Downloaded from www.jama.com at Ucla, on July 23, 2005

MRI VS CT FOR HEMORRHAGE DETECTION

sented genuine acute hemorrhage. We to confirm the diagnosis in these cases. the early management of patients with ischemic stroke.
Stroke. 2003;34:1056-1083.
also specifically excluded any patient Our study suggests that GRE MRI may 3. Baird AE, Warach S. Magnetic resonance imaging of
with symptoms suggestive of SAH. Al- be able to detect regions of hemorrhagic acute stroke. J Cereb Blood Flow Metab. 1998;18:583-
609.
though prior studies have suggested that transformation of an acute ischemic 4. Albers GW. Expanding the window for thrombo-
both GRE MRI and fluid-attenuated in- stroke not evident on CT. Our study con- lytic therapy in acute stroke. Stroke. 1999;30:2230-2237.
5. Edelman RR, Johnson K, Buxton R, et al. MR of hem-
version recovery images may be accu- firms the superiority of MRI for detec- orrhage: a new approach. AJNR Am J Neuroradiol. 1986;
rate in identifying subarachnoid blood, tion of chronic hemorrhage, particularly 7:751-756.
6. Patel MR, Edelman RR, Warach S. Detection of hy-
this will need to be prospectively con- microbleeds. The role of these findings peracute primary intraparenchymal hemorrhage by mag-
firmed in a future study.24-26 Because nei- in the decision-making process for treat- netic resonance imaging. Stroke. 1996;27:2321-2324.
7. Linfante I, Llinas RH, Caplan LR, Warach S. MRI fea-
ther CT nor MRI can exclude SAH with ment of patients who are candidates for tures of intracerebral hemorrhage within 2 hours from
100% reliability, the clinician should thrombolytic therapy is currently un- symptom onset. Stroke. 1999;30:2263-2267.
8. Schellinger PD, Jansen O, Fiebach JB, Hacke W, Sar-
pursue an extensive evaluation in any known. Due to its advantages in delin- tor K. A standardized MRI stroke protocol. Stroke. 1999;
patient with whom SAH is contem- eating ischemic pathophysiology, in 30:765-768.
plated, including CT as well as lumbar combination with the findings suggest- 9. Kidwell CS, Saver JL, Mattiello J, et al. Diffusion-
perfusion MR evaluation of perihematomal injury in hy-
puncture if CT is negative. ing equivalency to CT for detecting acute peracute intracerebral hemorrhage. Neurology. 2001;57:
Interreader reliability (␬ statistic) for hemorrhage, MRI may be acceptable as 1611-1617.
10. Atlas SW, Thulborn KR. MR detection of hyper-
detection of hemorrhage was better for the sole imaging technique for acute acute parenchymal hemorrhage of the brain. AJNR Am
CT than for MRI. This is likely due to sev- stroke at centers with expertise in inter- J Neuroradiol. 1998;19:1471-1477.
11. Kidwell CS, Saver JL, Villablanca JP, et al. Mag-
eral factors, including less experience of preting these findings. netic resonance imaging detection of microbleeds be-
the readers in interpreting acute MRI for fore thrombolysis. Stroke. 2002;33:95-98.
Author Contributions: Dr Kidwell had full access to 12. Roob G, Fazekas F. Magnetic resonance imaging of
hemorrhage and differences in the intrin- all of the data in the study and takes full responsibil- cerebral microbleeds. Curr Opin Neurol. 2000;13:69-73.
sicconspicuityofhemorrhageappearance ity for the integrity of the data and the accuracy of 13. Fazekas F, Kleinert R, Roob G, et al. Histopathologic
the data analysis. analysis of foci of signal loss on gradient-echo T2*-
on CT and MRI. Therefore, a comprehen- Study concept and design: Kidwell, Chalela, Saver, weighted MR images in patients with spontaneous in-
sive educational program should be un- Starkman, Warach. tracerebral hemorrhage. AJNR Am J Neuroradiol. 1999;
Acquisition of data: Kidwell, Chalela, Saver, Starkman, 20:637-642.
dertaken at any institution choosing to Hill, Demchuk, Patronas, Alger, Latour, Baird, Leary, 14. Fiebach JB, Schellinger PD, Gass A, et al. Stroke mag-
perform only MRI and not CT for the Tremwel, Ovbiagele, Fredieu, Suzuki, Villablanca, Davis, netic resonance imaging is accurate in hyperacute intra-
Dunn, Todd, Ezzeddine, Haymore, Lynch, Davis, Warach. cerebral hemorrhage. Stroke. 2004;35:502-506.
evaluation of acute stroke patients. Analysis and interpretation of data: Kidwell, Chalela, 15. Kidwell CS, Warach S. Acute ischemic cerebrovas-
Recent reports have indicated wide- Saver, Butman, Patronas, Alger, Latour, Luby, cular syndrome. Stroke. 2003;34:2995-2998.
Villablanca, Davis, Warach. 16. Packard AS, Kase CS, Aly AS, Barest GD. “Computed
spread availability of advanced MRI tech- Drafting of the manuscript: Kidwell, Chalela, Fredieu, tomography-negative” intracerebral hemorrhage. Arch
niques in the United States for the evalu- Davis Warach, Saver. Neurol. 2003;60:1156-1159.
17. Sandercock P, Foy PM, Brock LG. Negative comput-
ation of patients with acute stroke.27,28 Critical revision of the manuscript for important intel-
erised tomographic scan in patient with intracerebral
lectual content: Kidwell, Chalela, Saver, Starkman, Hill,
However, concerns have been raised re- Demchuk, Butman, Patronas, Alger, Latour, Luby, Baird, haematoma. Lancet. 1986;1:506-507.
garding the logistical aspects of acquir- Leary, Tremwel, Ovbiagele, Suzuki, Villablanca, Davis, 18. NakadaT,KweeIL.Computedtomography-negative
Dunn, Todd, Ezzeddine, Haymore, Lynch, Davis, Warach. acute thalamic hematoma. J Neuroimaging. 1996;6:119-
ing multimodal MRI in the acute stroke Statistical analysis: Kidwell, Chalela, Saver, Luby, Warach. 121.
setting, particularly with regard to im- Obtained funding: Kidwell, Saver, Warach. 19. NINDS rt-PA Stroke Group. Tissue plasminogen ac-
Administrative, technical, or material support: Kidwell, tivator for acute ischemic stroke. N Engl J Med. 1995;333:
age acquisition times (and potential de- Saver, Starkman, Hill, Alger, Latour, Luby, Leary, 1581-1587.
lays in initiating thrombolytic therapy). Fredieu, Davis, Todd, Haymore, Warach. 20. Hermier M, Nighoghossian N. Contribution of
susceptibility-weightedimagingtoacutestrokeassessment.
Based on our overall experience, the Study supervision: Kidwell, Chalela, Starkman, Baird,
Stroke. 2004;35:1989-1994.
Ezzeddine, Warach.
comprehensive MRI stroke protocol we 21. Chalela JA, Kang DW, Warach S. Multiple cerebral
Funding/Support: This study was supported in part by
microbleeds. J Neuroimaging. 2004;14:54-57.
used generally takes 10 to 15 minutes. the Division of Intramural Research, National Institute
22. Fan YH, Zhang L, Lam WW, Mok VC, Wong KS. Ce-
of Neurological Disorders and Stroke (NINDS) and grants rebral microbleeds as a risk factor for subsequent intra-
An abbreviated protocol, including DWI, from the American Heart Association (0170033N, Dr cerebral hemorrhages among patients with acute ische-
GRE, and perfusion weighted imaging Kidwell; AHA Western States Affiliate Fellowship Award, mic stroke. Stroke. 2003;34:2459-2462.
Dr Leary) and NINDS (K23 NS 02088, Dr Kidwell; NS
(PWI), takes less than 5 minutes and still 39498/EB 002087, Dr Alger; K24 NS 02092, Dr Saver).
23. Wong KS, Chan YL, Liu JY, Gao S, Lam WW. Asymp-
tomatic microbleeds as a risk factor for aspirin-associated
provides substantially more informa- Dr Hill was supported in part by the Heart & Stroke intracerebralhemorrhages.Neurology.2003;60:511-513.
tion than a noncontrast CT. Foundation of Alberta/NWT/NU and the Canadian In- 24. WiesmannM,MayerTE,YousryI,MedeleR,Hamann
stitutes for Health Research. GF, Bruckmann H. Detection of hyperacute subarach-
Our study may have implications for Role of the Sponsor: The study was wholly designed, noid hemorrhage of the brain by using magnetic reso-
the imaging evaluation of patients with conducted, analyzed, and reported by the authors with- nance imaging. J Neurosurg. 2002;96:684-689.
out any input from industrial sponsors. 25. Singer MB, Atlas SW, Drayer BP. Subarachnoid space
acute stroke symptoms. Our findings Acknowledgment: We would like to acknowledge the disease. Radiology. 1998;208:417-422.
supportpriorstudiessuggestingthatMRI invaluable assistance provided by Patricia Lyall, BA, 26. Mitchell P, Wilkinson ID, Hoggard N, et al. Detec-
Vickie Hyneman, Elisa Landis, BA, and Sarah Hilton, tion of subarachnoid haemorrhage with magnetic reso-
is as accurate as CT for the detection of BS, for the completion of this project. nance imaging. J Neurol Neurosurg Psychiatry. 2001;70:
hyperacute hemorrhage.14 One impor- 205-211.
27. Liebeskind DS, Yang CK, Sayre J, Bakshi R. Neuro-
tant caveat is that with small hemor- REFERENCES imaging of cerebral ischemia in clinical practice [abstract].
rhages, blood that appears as acute on 1. Broderick JP, Adams HP Jr, Barsan W, et al. Guide- Stroke. 2003;34:255.
lines for the management of spontaneous intracere- 28. Ruland S, Gorelick PB, Schneck M, Kim D, Moore
CT may appear as chronic on GRE MRI bral hemorrhage. Stroke. 1999;30:905-915. CG, Leurgans S. Acute stroke care in Illinois. Stroke. 2002;
and a noncontrast CT may be required 2. Adams HP Jr, Adams RJ, Brott T, et al. Guidelines for 33:1334-1339.

1830 JAMA, October 20, 2004—Vol 292, No. 15 (Reprinted) ©2004 American Medical Association. All rights reserved.

Downloaded from www.jama.com at Ucla, on July 23, 2005