| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
1 Department of Radiology, Division of Neuroradiology, Massachusetts General
Hospital, 55 Fruit St., Boston, MA 02114.
2 Department of Radiology, Harvard Medical School, 25 Shattuck St., Boston, MA
02115.
3 Department of Neurology, Massachusetts General Hospital, Boston, MA
02114.
Received July 12, 2001;
accepted after revision January 10, 2002.
Address correspondence to M. H. Lev.
Abstract
 Top Abstract IntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
MATERIALS AND METHODS. The records of 733 consecutive patients with a clinically based admission diagnosis of early stroke were reviewed. Among the criteria for inclusion in our study were the availability of an unenhanced CT scan (561 cases) or diffusion-weighted MR imaging examination (409 cases) obtained at admission and a discharge diagnosis indicating whether a patient had actually had a stroke. The radiology requisition forms, available at the time of image interpretation, were classified as either indicating or not indicating a clinical suspicion of early stroke. Sensitivity, specificity, and accuracy of stroke detection were computed, stratified by the presence or absence of an available history indicating suspicion of stroke. Results were compared using the Fisher's exact two-tailed test.
RESULTS. Unenhanced CT sensitivity was 52% (specificity, 95%) for the suspicion-of-stroke group and 38% (specificity, 89%) for the no-suspicion-of-stroke group (p = 0.008). Diffusion-weighted MR imaging sensitivity was 95% (specificity, 94%) for the suspicion-of-stroke group and 94% (specificity, 98%) for the no-suspicion-of-stroke group (p = 0.822).
CONCLUSION. Availability of a clinical history indicating that early stroke is suspected significantly improves the sensitivity for detecting strokes on unenhanced CT without reducing specificity. In contradistinction, the availability of such a history did not significantly improve the sensitivity for detecting stroke using diffusion-weighted MR imaging. Whenever possible, relevant clinical history should be made available to physicians interpreting emergency CT scans of the head.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Perceptual misses may account for up to 60% of diagnostic errors in radiology, but perception can be influenced by expectations [9]: "[O]ne knows what one sees and one sees what one knows" [2]. Moreover, according to bayesian statistics, assuming that the findings on radiography are detected, the probability of disease (i.e., the posttest probability) is modified by the clinical suspicion of disease (i.e., by the pretest probability) [10]. An example of these principles is provided by Gomez-Hassan et al. [11] who, in 2000, reported the error rate in the MR imaging localization of seizure foci in patients for whom the clinical history and electroencephalographic results were unknown to the reviewing radiologists at the time of image interpretation. On retrospective reevaluation of the MR images of the brain and with the full clinical data available, a significant abnormality was discovered, which led to a modification of the treatment plan, in 42% of these patients [11].
The diagnosis of new-onset stroke remains a challenge for several reasons: the data provided by the patient or the family may be vague; commonly occurring conditons that mimic stroke, when present, may limit the physician's ability to conduct a detailed physical examination; and the ischemic hypodensities found on unenhanced CT scans that are characteristic of acute cytotoxic edema may be subtle [12,13,14,15,16]. The assessment of early stroke with unenhanced CT is complicated by a sometimes marked intra- and interobserver variability and remains imperfect, despite a small increase in sensitivity with the use of optimized, soft-copy CT window and level review settings [13,14,15,16]. In recent years, the advent of diffusion-weighted MR imaging has improved the sensitivity and accuracy of acute stroke detection to values approaching 100%, but the use of diffusion-weighted MR imaging requires advanced, high-gradient-field MR imaging scanners that are not universally available in hospital emergency departments [17]. Unenhanced CT, therefore, because of its widespread accessibility as well as its exquisite sensitivity for revealing intracranial hemorrhage, remains the first-line imaging test in urgent evaluations for stroke.
In our study, we investigated the effect that awareness by the interpreting radiologist of the clinical suspicion of early stroke had on the sensitivity for stroke detection using both unenhanced CT and diffusion-weighted MR imaging. We hypothesized that a knowledge of the clinical concern for stroke by the reviewing radiologist would facilitate the detection of abnormal imaging findings, regardless of the interpretation of such findings, resulting in a more accurate radiologic diagnosis.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Image Acquisition
All head CT scans and brain MR images were obtained for appropriate
clinical indications according to the following standard imaging protocols
with only minimal variation in scanning parameters: Unenhanced head CT scans
were obtained on either a HiSpeed Advantage helical CT scanner located in the
emergency room, or on a CT/i scanner (both scanners, General Electric Medical
Systems, Milwaukee, WI) located in the hospital radiology department, using
5-mm-thick contiguous axial slices, 140 kVp, 170 mA, and a scanning time of 2
sec (340 mAs).
MR images were obtained on a 1.5-T Signa system (General Electric Medical Systems) with echoplanar capability. Our unenhanced MR imaging protocol for an urgent brain examination consists of sagittal T1-weighted images (TR/TE, 650/16; field of view, 20 cm; matrix, 256 x 192 pixels; slice thickness, 5 mm with a 1-mm gap; and number of excitations, 1); followed by either fast spin-echo proton density-weighted images (2500/18; field of view, 20 cm; matrix, 256 x 256 pixels; slice thickness, 5 mm with a 1-mm gap; and number of excitations, 1) or axial fluid-attenuated inversion recovery (FLAIR)-weighted images (10002/141; inversion time, 2200 msec; field of view, 24 cm; matrix, 256 x 192; slice thickness, 5 mm with a 1-mm gap, and number of excitations, 1); followed by axial fast spin-echo T2-weighted images (4200/102; field of view, 20 cm; matrix, 256 x 256 pixels; slice thickness, 5 mm with a 1-mm gap; and number of excitations, 1); and finally, axial diffusion-weighted images with associated apparent diffusion coefficient maps.
The diffusion-weighted MR images were obtained using single-shot echoplanar imaging (6000/118; field of view, 40 x 20 cm; matrix, 256 x 128 pixels; slice thickness, 6 mm with a 1-mm gap; and 20 axial slices). The effective gradient strength was 14 mT/m, and b values were 1221 sec/mm2 and 4 sec/mm2 in six gradient directions with three signal averages and an image acquisition time of 126 sec. Axial diffusion-weighted MR imaging sequences are routinely obtained for all patients who receive MR imaging at our hospital.
Data Collection
Hospital records were reviewed to determine the patient's age, sex, time of
presentation, admitting diagnosis assigned in the emergency department, and
final discharge diagnosis assigned by the attending neurologists on the basis
of all available imaging and clinical data. The discharge diagnoses considered
positive for early infarction were stroke, intracranial hemorrhage (including
intraparenchymal and petechial hemorrhage), lacunar stroke, and cerebral
infarction with transient symptoms. Other discharge diagnoses were considered
negative for early infarction, including subarachnoid and extraaxial
intracranial hemorrhage. The precise time of stroke ictus was not available
for all patients.
The exact time of CT scanning or MR imaging, the clinical history stated on the radiology requisition form (and therefore available to the neuroradiologists at the time of image interpretation), and the radiologist's signed reports indicating the scanning findings were obtained from the IDXrad version 9.7.1 (IDX Systems, Burlington, VT) electronic radiologic record system maintained by our radiology department. The patients' clinical histories available to the radiologists at the time of image interpretation, as recorded on the digital requisition forms, were categorized as either indicating or not indicating a clinical suspicion of early stroke. Before CT scanning or MR imaging readout, all neuroradiologists in our department routinely review the electronic requisition form. The words or phrases on the requisition form considered to be positive for suspicion of early stroke were "stroke," "cerebrovascular accident," "infarct," "ischemia," or "brain attack." A specific description of localizing or lateralizing signs or symptoms suggestive of early infarction, such as "aphasia" or "hemiparesis," was also considered to indicate a suspicion of early stroke (Table 1). The words or phrases not considered to indicate an explicit suspicion of early stroke were "rule out" or "question" used in conjunction with "mass lesion," "hydrocephalus," "traumatic injury," or "bleed" (with "bleed" not otherwise specified and anecdotally referring typically to aneurysmal subarachnoid hemorrhage in the emergency setting). Notations of nonlocalizing or nonlateralizing symptoms, such as "headache," "seizure," or "change in mental status," were not considered to indicate a clinical suspicion of early stroke.
|
All the CT and MR imaging findings included in our study were prospectively reported by an on-call neuroradiologist at the time of the patient's admission. The findings of unenhanced CT associated with acute stroke include parenchymal hypodensity (in an appropriate vascular distribution), sulcal effacement, and dense vessel sign [13]; the MR imaging finding most strongly associated with acute stroke is hyperintensity on diffusion-weighted imaging (with a corresponding restricted signal on apparent diffusion coefficient maps) [17]. The archived reports from these CT and MR imaging examinations were retrospectively rated as either indicating or not indicating the presence of an early infarction. We considered the modifiers "probable" or "likely" to be positive for early infarction when used in conjunction with "new infarct," "stroke," "intraparenchymal hemorrhage," "lacune," or "ischemia." If any of the following modifiers were present in the report, however, the examination was considered negative for new-onset infarction: "possible," "rule out," "cannot exclude," or "no evidence of" as well as "old," "chronic," "prior," or "remote" infarct, stroke, hemorrhage, lacune, or ischemia. For a small number (n = 17) of MR imaging reports for which the final written impression was equivocal or confusing, the final categorization was determined by consensus review (of both the report and the original MR images) by three neuroradiologists who were unaware of the final discharge diagnosis but were aware of the clinical history stated on the requisition form.
Statistics
Admission CT and MR imaging report results, stratified by the presence or
absence of a history indicating a clinical suspicion of early stroke on the
digital radiology requisition form, were compared with the final neurology
service discharge diagnosis of "stroke" or "no
stroke." If the reported CT or MR imaging results agreed with the final
discharge diagnoses, they were rated as true-positive (stroke present) or
true-negative (stroke absent). If the results conflicted with the final
discharge diagnosis, they were rated as false-positive (stroke incorrectly
suspected on basis of imaging) or false-negative (stroke incorrectly excluded
on basis of imaging). Sensitivity, specificity, positive predictive value,
negative predictive value, and accuracy were calculated. The 95% confidence
intervals for sensitivity and specificity were determined according to
standard statistical methodology. Fisher's exact two-tailed test was applied
to all groups to assess statistical significance.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Detection of Early Stroke on Unenhanced CT Stratified by Presence or
Absence of Suspicion of Stroke on the Radiology Requisition Form
Initial head CT scans (n = 561) were obtained within 48 hr of
admission (range, 0-48 hr; mean, 2 hr 19 min ± 2 hr 14 min). Patients
were stratified into two groups, "clinical suspicion of stroke" or
"no clinical suspicion of stroke," on the basis of the histories
available on the electronic requisition form at the time of image
interpretation (Table 2). The
men-to-women ratios (1.0:1.0 for the stroke group and 1.2:1.0 for the
no-stroke group) and mean ages (67.5 ± 16.3 years for the stroke group
and 69.3 ± 14.9 years for the no-stroke group) were similar for each
group (Table 2). Sensitivity
for stroke detection was greatest (52%) for the stroke group and lowest (38%)
for the no-stroke group (p = 0.008). Specificity was greater for the
stroke group (96%) than for the no-stroke group (89%), but this difference did
not reach statistical significance (p = 0.680). Positive predictive
value was similarly high (99% and 94%, respectively) for both the stroke and
no-stroke groups. Negative predictive value was similarly low (26% and 23%,
respectively) for both groups. Overall diagnostic accuracy for stroke
detection was higher in the stroke group (59% vs 47% in the no-stroke
group).
|
Detection of Early Stroke on Diffusion-Weighted Imaging, Stratified
by Presence or Absence of Suspicion of Stroke on the Radiology Requisition
Form
Initial diffusion-weighted MR imaging of the brain (n = 409) was
performed within 5 days of admission (mean time interval, 8 hr 3 min ±
5 hr 34 min; range, 0-5 days). Such imaging was performed before initial CT in
fewer than 1% of patients. Patients were stratified into two groups,
"clinical suspicion of stroke" or "no clinical suspicion of
stroke," on the basis of the histories available on the electronic
requisition form at the time of image interpretation
(Table 2). The men-to-women
ratios (1.4:1.0 for the stroke group and 1.1:1.0 for the no-stroke group) and
mean ages (65.5 ± 17.3 years for the stroke group and 67.1 ±
17.1 years for the no-stroke group) were similar for each group
(Table 2). Sensitivity for
stroke detection was statistically similar for the both groups (95% vs 94%;
p = 0.82). Specificity was also statistically similar for both groups
(95% vs 98%; p = 0.528). Positive predictive value was similarly high
(99% and 100%, respectively) for both groups. Negative predictive value was
similarly lower (68% and 75%, respectively) for both groups. Overall
diagnostic accuracy for stroke detection was statistically identical for the
stroke and no-stroke groups (95% and 94%, respectively).
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Our results suggest a statistically significant improvement in both the sensitivity and accuracy of unenhanced CT stroke detection for cases in which the interpreting radiologist is aware of the clinical suspicion of early stroke. These findings support the hypothesis that the availability of a relevant clinical history is a correctable source of variability in the interpretation of emergency unenhanced head CT scans.
Four components of a radiologic examination have been described: image acquisition, lesion detection, image interpretation, and communication of test results to the ordering physicians [5]. Quality control assumes different forms for each component; for example, training and experience contribute to optimal lesion detection. Although we hypothesize that the availability of relevant clinical history results in both improved lesion detection and more accurate image interpretation, our study design does not permit us to distinguish between these two effects.
Perception, and therefore lesion detection, can be influenced by expectations [9, 19]. For example, Figure 1A,1B is a CT scan that was incorrectly interpreted as showing no evidence of early stroke. In retrospect, the faint punctate hypodensity present in the patient's left anterior thalamus might have been detected had a relevant clinical history indicating suspicion of thalamic infarct been available.
|
|
Moreover, interpretation of imaging findings is contingent not only on lesion detection but also on the clinical context: What meaning does the finding have in relation to the current clinical scenario? Without a clinical context, image interpretation may be limited, in keeping with simple bayesian statistics [10]. For example, Segnan et al. [20] in 1992 showed that agreement among raters evaluating possible lung cancer improved with increasing information, and Song et al. [6] showed that a knowledge of the patient's clinical history improved the accuracy of interpretation of conventional radiographs. Similarly, Doubilet and Herman [7] reported that the frequency of true-positive chest radiographic interpretations by radiology residents increased when suggestive clinical histories were supplied [7]. An "appropriate clinical context" should include a comparison with prior radiographs, when available. Aideyan et al. [8] reported that prior information significantly increased interpreters' confidence, facilitated new observations, and permitted greater diagnostic specificity. The accuracy of a diagnostic test also depends on an objective, quantifiable, and mathematic way of determining the prevalence of disease in the population being evaluated.
Unenhanced CT, despite having a lower sensitivity for stroke detection than diffusion-weighted MR imaging, remains the first-line imaging technique in acute stroke evaluation because of its widespread availability and its ability to accurately detect intracranial hemorhage [17]. Unenhanced CT, therefore, typically plays a central role in the triage of stroke patients to appropriate, but potentially risky, treatments, such as IV or intraarterial thrombolysis. However, the unenhanced CT findings of acute stroke—most notably, parenchymal hypodensity—can be subtle and can give rise to significant intra- and interobserver variability [14,15,16]. Interestingly, von Kummer et al. [21] studied interobserver agreement in detecting the classic unenhanced CT findings of middle cerebral artery stroke (including parenchymal hypodensity, brain swelling, and the dense vessel sign) and found that the results were not significantly altered by whether their investigators were aware or unaware of the patient's clinical symptoms. The results of that study, however, do not contradict those of our own because although the observers in that study were unaware of the symptoms of any individual patient, they were aware of the general clinical suspicion of early stroke in their study population.
Devising the optimal unenhanced CT evaluation for stroke is clinically important. Our reported sensitivity of 52% (for cases in which the interpreting radiologist was aware of the clinical suspicion of stroke) may underestimate the maximal achievable sensitivity for unenhanced CT for detection of stroke. Recent articles [13, 22] have suggested that further improvements in sensitivity can be achieved by providing brief periods of training, as well as by using optimized window width and center level settings during soft-copy review of CT.
Important potential limitations of our investigation include its reliance on the clinical discharge diagnosis as the gold standard for the presence or absence of stroke, the average 6-hr delay between the performance of the initial CT and that of the diffusion-weighted MR imaging, and the possibility that the interpreting radiologists knew more history than was communicated on the electronic radiology requisition form at the time of image interpretation. These limitations cannot be discounted. However, given our retrospective study design, any biases are likely to be distributed equally across all cases—both patients in the group with a clinical suspicion of stroke and those in the group with no such suspicion—because the reviewing radiologists were unaware of the parameters being evaluated at the time of their case reporting [23]. Moreover, the p value for our measured improvement in unenhanced CT sensitivity may be a conservative estimate; the availability of additional clinical history not noted on the requisition forms, if distributed across all cases, would have only served to weaken the statistical strength of our results. In addition, the final discharge diagnosis may have been influenced by the radiology report.
Because of the lack of pathologic proof of diagnosis in such a clinical cohort, follow-up imaging with both CT and diffusion-weighted imaging would also have been more helpful in assessing the presence and extent of an infarct than the initial examinations. Follow-up imaging was not performed for most of our patients and could prove problematic in this era of cost efficiency. Moreover, the comparison between a CT scan obtained an average of 2 hr after the patient's presentation to the emergency department and a diffusion-weighted MR image obtained an average of 8 hr after the presentation introduces a potential bias that skews the results in favor of diffusion-weighted imaging. The infarctions could have evolved in the interval, making them easier to identify, and additional clinical history not recorded on the requisition sheet may have become available. This delay should not, however, have affected the CT results. Also, the aim of our study was not to directly compare CT and MR imaging but rather to determine the effect of the availability of relevant clinical history on the interpretation of each of these examinations in its own right.
In the future, a more controlled study would be helpful to eliminate any potential biases inherent in our study design, such as comparing results of CT and diffusion-weighted MR imaging performed 6 hr later. In our practice, diffusion-weighted MR imaging is not as readily accessible as is CT.
In summary, the availability of relevant clinical history is an important factor in the detection of early stroke using unenhanced CT. The knowledge that a patient's clinical history indicates a suspicion of early stroke significantly improves the sensitivity of unenhanced CT for detection of stroke by interpreting neuroradiologists, without reducing specificity. Whenever possible, relevant clinical history should be made available to physicians interpreting emergency CT scans of the head.
Acknowledgments
We thank Elkan Halpern for his excellent statistical guidance.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
CiteULike 
Complore 
Connotea 
Del.icio.us 
Digg 
Reddit 
Technorati What's this?
This article has been cited by other articles:
|
|
 |
|
  G. V. Goldmakher, E. C.S. Camargo, K. L Furie, A. B. Singhal, L. Roccatagliata, E. F. Halpern, M. J. Chou, T. Biagini, W. S. Smith, G. J. Harris, et al. Hyperdense Basilar Artery Sign on Unenhanced CT Predicts Thrombus and Outcome in Acute Posterior Circulation Stroke Stroke, January 1, 2009; 40(1): 134 - 139. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 R.I. Aviv, J. Mandelcorn, S. Chakraborty, D. Gladstone, S. Malham, G. Tomlinson, A.J. Fox, and S. Symons Alberta Stroke Program Early CT Scoring of CT Perfusion in Early Stroke Visualization and Assessment AJNR Am. J. Neuroradiol., November 1, 2007; 28(10): 1975 - 1980. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 E. C. S. Camargo, K. L. Furie, A. B. Singhal, L. Roccatagliata, M. E. Cunnane, E. F. Halpern, G. J. Harris, W. S. Smith, R. G. Gonzalez, W. J. Koroshetz, et al. Acute Brain Infarct: Detection and Delineation with CT Angiographic Source Images versus Nonenhanced CT Scans Radiology, August 1, 2007; 244(2): 541 - 548. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
A. M. Riddell and K. Khalili Assessment of Acute Abdominal Pain: Utility of a Second Cross-sectional Imaging Examination Radiology, February 1, 2006; 238(2): 570 - 577. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
M. E. Mullins, M. H. Lev, D. Schellingerhout, R. G. Gonzalez, and P. W. Schaefer Intracranial Hemorrhage Complicating Acute Stroke: How Common Is Hemorrhagic Stroke on Initial Head CT Scan and How Often Is Initial Clinical Diagnosis of Acute Stroke Eventually Confirmed? AJNR Am. J. Neuroradiol., October 1, 2005; 26(9): 2207 - 2212. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
R. D. Zimmerman Stroke Wars: Episode IV CT Strikes Back AJNR Am. J. Neuroradiol., September 1, 2004; 25(8): 1304 - 1309. [Full Text] [PDF]
|
|
|
|
|
|
D. Saur, T. Kucinski, U. Grzyska, B. Eckert, C. Eggers, W. Niesen, V. Schoder, H. Zeumer, C. Weiller, and J. Rother Sensitivity and Interrater Agreement of CT and Diffusion-Weighted MR Imaging in Hyperacute Stroke AJNR Am. J. Neuroradiol., May 1, 2003; 24(5): 878 - 885. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
