| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Original Research |
1 Department of Radiology, Boston Medical Center, Boston University, Boston, MA
02118.
2 Present address: Department of Radiology, New York-Presbyterian Hospital, New
York Weill Cornell Medical Center, 525 E 68th St., Starr Pavilion, Box 141,
New York, NY 10021.
3 Deceased.
Received July 25, 2004;
revised December 6, 2004;
Address correspondence to R. S. Title.
Abstract
 Top Abstract IntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
MATERIALS AND METHODS. One hundred two noncontrast thoracic CT examinations were qualitatively reviewed by three independent reviewers. Hounsfield unit measurements of the blood in the left ventricle were recorded by a fourth individual. Anemia was defined as a hemoglobin level of less than 10 g/dL. Receiver operating characteristic (ROC) analyses of expert reviewers were compared with measured Hounsfield units.
RESULTS. Hounsfield unit measurements performed significantly better than subjective reviewer analyses for differentiation of an anemic from a nonanemic state (area under ROC curve = 0.85 vs 0.72, 0.70, and 0.69; 95% confidence interval, 0.78–0.92 vs 0.63–0.81, 0.61–0.79, and 0.60–0.78, respectively; p < 0.05). With use of a CT density threshold of 35 H, the sensitivity for anemia was 76% and specificity was 81%, whereas the sensitivity of three reviewers was 40–72% with a specificity of 60–83%. Interobserver agreement was found to be poor by kappa statistic (0.0906–0.2128). The correlation coefficient for the analysis of Hounsfield unit versus hemoglobin level was 0.72. Separating data by patient sex revealed a correlation coefficient of 0.81 for men versus 0.52 for women, although the actual regression lines were not statistically different (p > 0.05).
CONCLUSION. Despite expert reviewer analyses, subjective evaluations of blood attenuation characteristics are prone to inaccuracy and show poor interobserver agreement. Quantitative measurements of CT density in Hounsfield units should be performed to accurately differentiate an anemic from a nonanemic state when serum hemoglobin levels are not readily available.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Hounsfield units (H) are the units of measurement of density on CT. In CT, a number (between –1,000 and 1,000 H) is assigned by computer to represent the difference in X-ray attenuation between a given material and water (where air is –1,000 H and pure water is 0 H). This number is then used by the computer to assign a gray-scale shade to the represented image. The exquisite contrast resolution of current CT scanners can detect differences in contrast of less than 5 H (< 0.5%).
|
|
|
|
|
The aims of this study were to evaluate the ability and reliability of expert reviewers to differentiate an anemic from a nonanemic state on the basis of a visual assessment of the attenuation of blood in the left ventricle on noncontrast thoracic CT images and to compare this subjective evaluation with density measured in Hounsfield units. We attempted to determine if subjective (visual assessment) or objective (measurement of Hounsfield units within the lumen of the left ventricle) evaluation was sufficiently accurate in differentiating anemic from nonanemic states and whether quantitative measurements of Hounsfield units are required for confident diagnosis.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Statistical Design
Using the method of Hanley and McNeil
[5] for the same population of
subjects and a power of 0.8, at least 25 abnormal and 65 normal cases would be
required to perform an adequate analysis at the 0.05 cutoff.
Demographics
A cohort of 102 patients was identified on the basis of these criteria.
There were 61 men and 41 women who ranged in age from 21 to 93 years (mean,
56.8 years). The hemoglobin levels ranged from 7.0 to 18.2 g/dL (mean, 11.6
g/dL; normal laboratory values: men, > 13.5 g/dL, and women, >12 g/dL).
Anemia was defined as a hemoglobin level of less than 10 g/dL. Twenty-five of
the 102 patients had a hemoglobin level of less than 10 g/dL.
CT examinations were performed on two different CT scanners. Fifty-six patients (41 men, 15 women) underwent CT on an MDCT scanner (MX 8000, Philips Medical Systems) using 3.2-mm thickness, 1.25 pitch, 200 mAs, and 140 kVp; the remaining 46 patients (20 men, 26 women) underwent CT on a helical scanner (PQ 5000, Philips Medical Systems) using 5-mm thickness, 1.5 pitch, 225 mAs, and 120 kVp.
Qualitative Image Analysis
One hundred two noncontrast thoracic CT examinations were visually reviewed
independently by three experienced thoracic CT reviewers blinded to each
other's readings and to the anemic state of the patient using soft-tissue
window settings (window, 400 H; level, 30 H) on a commercial PACS-based
workstation (Amicas, Version 3.5.5, Amicas). Based on visual assessment of the
perceived attenuation of blood in the left ventricle, observers were asked to
classify the patient as definitely anemic, probably anemic, indeterminate,
probably not anemic, or definitely not anemic using a 5-point scale, which is
similar to other subjective reviewer scales validated previously
[6]. The reviewers were shown
two sample cases representing the patients with the lowest and highest
hemoglobin levels to standardize reviewer interpretation.
Quantitative Image Analysis
The density in Hounsfield units of the blood in the left ventricle was
measured by a fourth observer. An oval or circular region ranging in area from
3.0 to 3.9 cm2 (mean, 3.5 cm2) located within the
confines of the lumen of the left ventricle was selected. The Hounsfield unit
measurements were entered into a computer database.
Statistical Analysis
True-positive cases were defined as anemic patients correctly identified.
True-negative cases were defined as nonanemic patients that were correctly
identified. Interobserver agreement was calculated using the kappa statistic
[7]. Receiver operating
characteristic (ROC) curves were calculated to compare the results of the
three reviewers with the density measurements in Hounsfield units
[8–10].
The diagnostic accuracy was determined by calculating the area under each ROC
curve (Az)
[11]. The
Az values for the three reviewers and for density
measurements in Hounsfield units were compared using the method of proportions
analysis. All statistical analyses were performed using STATA software
(Stata).
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Statistical analyses using STATA software for different hemoglobin levels established that the greatest area under the ROC curve (Az) was obtained for a hemoglobin level of 12 g/dL (Az = 0.87). Using this ROC curve, we determined the Hounsfield unit value closest to a sensitivity of 1.0 and specificity of 1.0. The ROC curve for density measured in Hounsfield units (Fig. 4) revealed that the use of a Hounsfield unit value of 35 H to define anemia produces the most accurate results in terms of maximizing true-positives and minimizing false-negatives. With the use of a Hounsfield unit threshold of 35 H, the sensitivity of Hounsfield unit measurements for the anemic state was 76% and specificity was 81%.
The sensitivity of the three reviewers for determining the anemic state ranged from 40% to 72% and specificity ranged from 60% to 83%. Interobserver variability was found to be very high by kappa analysis. The interobserver kappa values for the three reviewers were 0.0906, 0.0914, and 0.2128 (poor), where a kappa value of 1 represents complete agreement between reviewers. Seven of 25 anemic patients were correctly classified by all three reviewers as definitely or probably anemic, and 24 of 77 nonanemic patients were correctly classified by all three reviewers as definitely or probably not anemic.
ROC curves for measured Hounsfield units and for the three independent reviewers for distinguishing anemic from nonanemic states are shown in Figures 4 and 5. The mean area under the ROC curve (Az) for measured Hounsfield units was significantly higher than that for the individual reviewer analysis (0.85 vs 0.72, 0.70, and 0.69; 95% confidence intervals, 0.78–0.92 vs 0.63–0.81, 0.61–0.79, and 0.60–0.78 for reviewers 1, 2, and 3, respectively; p < 0.05).
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
There are several potential limitations of this study. Patients were studied on two different types of CT scanners, and Hounsfield unit measurements are known to vary among scanners because of maladjustments, shortand long-term drift, X-ray fan, and detector asymmetry, which could lead to nonuniformity of data. However, a study looking at interscanner conformity in CT densitometry of the lungs concluded that after correction for poor air calibration, scanner conformity was acceptable [13], and correlation of data between the two scanners used in this study was high.
Another limitation to the study is that it is difficult to definitively distinguish intracardiac structures in nonanemic patients because of the lack of a differential between cavitary and myocardial densities [1]. Thus, one cannot be certain that the oval region within the confines of the lumen of the left ventricle selected for measurement of Hounsfield units did not contain any myocardium. Finally, a subjective difference exists in the way that CT images are visually interpreted in clinical practice and the necessary method of data collection for this study. The three observers in this study examined each CT scan to specifically analyze the images of the heart to evaluate for anemia and then rated the severity or certainty of the finding. In clinical practice, the examination is usually performed for an indication completely unrelated to anemia, and the question of anemia might arise only if one observes a striking contrast between luminal blood and adjacent myocardium. Therefore, a random observer's threshold for making the diagnosis of anemia may differ from that of the observers in this study, and more subtle cases might be ignored in clinical practice.
The fact that the correlation of Hounsfield unit values and hemoglobin levels proved greater in men than in women is a curious finding that may be attributable to breast attenuation or body fat distribution and merits further investigation. Although this was not the goal of our study, the different correlations are significant and because the same hemoglobin cutoff levels were used for men and women, in contrast to other studies [4], a confounding variable could be menstruation or other causes of active bleeding. At our site, unstable patients are unstable predominantly as a result of trauma, which indicates that these cases were automatically excluded because unstable patients are usually given IV contrast material.
In conclusion, despite expert reviewer analyses, subjective evaluations of blood attenuation characteristics are prone to inaccuracy and show high interobserver variability (i.e., lack certainty and consistency). If a diagnosis of anemia is questioned clinically or due to subjective visual analysis of the CT images, quantitative measurements of density in Hounsfield units should be performed to accurately differentiate an anemic from a nonanemic state when serum hemoglobin information may be lacking. In the measurement of the CT attenuation of blood in the lumen of the left ventricle, a threshold value of 35 H was found to best separate anemic from nonanemic states.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
CiteULike 
Complore 
Connotea 
Del.icio.us 
Digg 
Reddit 
Technorati What's this?
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
