| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Original Research |
1 Department of Radiology, Stanford University School of Medicine, 300 Pasteur Dr., S-072B, Stanford, CA 94305-5105.
Received August 25, 2004;
accepted after revision December 10, 2004.
Address correspondence to G. D. Rubin
(grubin{at}stanford.edu).
Abstract
 Top Abstract IntroductionSubjects and MethodsResultsDiscussionReferences |
|---|
SUBJECTS AND METHODS. Fifty patients (18 females, 32 males; age range, 15-93 years; mean age, 63.6 years) underwent 4-MDCT of the chest (detector width, 1 mm; beam pitch, 1.5). Contrast material was administered in 20 of the 50 patients. Coronal and transverse sections were reformatted into 5-mm-thick sections at 3.5-mm intervals. All available image and clinical data consensually reviewed by two thoracic radiologists served as the reference standard. Subsequently, three other thoracic radiologists independently evaluated reformatted coronal and transverse images at two separate review sessions. Each image set was assessed in 58 categories for abnormalities of the lungs, mediastinum, pleura, chest wall, diaphragm, abdomen, and skeleton. Interpretation times and number of images assessed were recorded. Sensitivity, specificity, and interobserver concordance were calculated. Differences in mean sensitivities and specificities were evaluated with Wilcoxon's signed rank test.
RESULTS. The most common findings identified were pulmonary nodules (n = 73, transverse images; n = 72, coronal images) and emphysema (n = 45, transverse; n = 40, coronal). The mean detection sensitivity of all lesions was significantly (p = 0.001) lower on coronal (44% ± 26% [SD]) than on transverse (51% ± 22%) images, whereas the mean detection specificity was significantly (p = 0.005) higher (96% ± 5% vs 95% ± 6%, respectively). Reporting findings for significantly (p < 0.001) fewer coronal images (mean, 63.0 ± 4.6 images) than transverse images (mean, 91.9 ± 8.8 images) took significantly (p = 0.025) longer (mean, 263 ± 56 sec vs 238 ± 45 sec, respectively).
CONCLUSION. Primary interpretation of thoracic MDCT is less sensitive and more time-consuming using 5-mm-thick coronal reformations as compared with transverse images.
|
TopAbstractIntroductionSubjects and MethodsResultsDiscussionReferences |
|---|
A concomitant result of MDCT technology and the ability to scan the entire thorax with thin collimation is the massive amount of data produced. New paradigms are needed to facilitate the interpretation, transfer, and storage of these data [4]. Because the sagittal dimension of the thorax is almost invariably smaller than the craniocaudal dimension, we hypothesized that replacing traditional axial scans with coronal reformatted images would result in a fewer number of images requiring less time to interpret.
Several studies have shown that multiplanar reconstruction images can improve diagnostic accuracy when used in conjunction with axial images for the assessment of certain chest abnormalities, such as airway stenosis and pulmonary embolism [1, 2]. Multiplanar reconstruction images are also an efficient method of communicating with referring physicians [5].
Presumably, the more views available to the interpreter, the greater the diagnostic confidence will be. Indeed, interactive 3D rendering is a rapidly evolving technology. Nonetheless, at most institutions, the limited number of workstations with such capability poses a barrier to the routine use of real-time postprocessing of data. Moreover, CT manufacturers recently have introduced protocols that allow the prospective selection of multiplanar reconstructions for creation after scan acquisition, allowing radiologists to primarily interpret alternative reformations.
High concordance between axial and coronal CT images for the primary evaluation of lung disease has been shown [6, 7]. However, to our knowledge, no previous study has evaluated the use of coronal reformatted thoracic CT images for primary presentation and evaluation in subjects who were not preselected for a specified disease or disease process. There is also no previous study to our knowledge that has compared the diagnostic accuracy of axial versus coronal thoracic CT images against a reference standard. The goal of our study was to evaluate the accuracy and efficiency of primary interpretation of thoracic MDCT using coronal reformations as compared with transverse images.
|
TopAbstractIntroductionSubjects and MethodsResultsDiscussionReferences |
|---|
CT Technique
The following standard protocol was implemented for all routine chest CT
examinations other than a nonhelical, dedicated diffuse lung disease protocol
and CT angiograms of the thorax: All scans were obtained on a 4-MDCT scanner
(Volume Zoom, Siemens Medical Solutions). Patients were scanned craniocaudally
from the thoracic in-let through the lung bases within a single breath-hold.
The scans were obtained with 4 x 1 mm collimation and a table feed of 6
mm per 500-msec scanner rotation, which resulted in a beam pitch of 1.5.
Scanning was performed at 120 kV and 100-150 mAs. Ninety milliliters of IV
contrast material (iohexol [Omnipaque 300, Amersham]) was administered in 20
of the 50 patients. Images were then reconstructed with a
high-spatial-frequency reconstruction kernel into contiguous 1.5-mm-thick
transverse sections. These sections were reformatted into 5-mm-thick sections
at an interval of 3.5 mm in the transverse and coronal planes.
Image Analysis
All images were evaluated on a cathode-ray tube monitor using a stacked
off-center ratio cine mode and displayed at standard window settings for lung
(level, -700 H; width, 1,500 H), mediastinum (level, 40 H; width, 400 H), and
bone (level, 600 H; width, 2,000 H). Observers were free to alter these window
and level settings. Before evaluating the 50 study scans, all observers
evaluated the chest CT scans of the practice interpretation group by
simultaneously evaluating transverse and coronal displays to familiarize
themselves with the coronal image display and the reporting worksheet.
To establish a reference standard, two fellowship-trained thoracic radiologists with five and 10 years of experience interpreting chest CT scans interpreted by consensus the 50 CT scans of the study group. Clinical history, online pathology reports, and all available imaging data were used, including the 1.25-mm CT transverse images and all reconstructed images. The full spectrum of clinically relevant observations was assessed.
The CT scans were evaluated on the basis of the following guidelines, using definitions as presented by the Nomenclature Committee of the Fleischner Society [8]: Pulmonary nodules and masses were evaluated for presence, number, lobe location, calcification, invasion of adjacent structures, and cavitation. Nodules smaller than 3 mm were not included unless they were part of a diffuse disease; in that case, they were included in the category "interstitial abnormalities." Interstitial abnormalities were evaluated for predominant type (nodular, reticular, ground-glass, or cystic), secondary lobular distribution (peripheral, peribronchovascular, or neither), and zonal predominance (upper lung, lower lung, or neither). The presence or absence of emphysema was determined. Air-space opacities were subdivided into two categories: consolidation and lobar atelectasis. Lobar atelectasis was defined as loss of more than 50% of the expected lobar volume. Consolidation was evaluated for location (unilobar or multilobar) and predominant distribution (peripheral, perihilar, or random). The location of lobar atelectasis, if present, was determined. Airway abnormalities were subdivided into three categories: bronchiectasis, intraluminal lesions, and wall thickening. Bronchiectasis was evaluated for distribution (unilobar or multilobar). Intraluminal airway lesions and wall thickening were evaluated for location (trachea, mainstem and lobar bronchi, segmental bronchi, or peripheral airways). Lymphadenopathy (nodes > 1 cm in short-axis diameter) was evaluated for location and accessibility by mediastinoscopy or mediastinotomy. Mediastinal masses were evaluated for location (anterior, middle, or posterior mediastinum), encasement or invasion of adjacent structures, and content (fat, calcification, or neither). Cardiovascular structures were evaluated for the presence or absence of valvular calcification, aortic aneurysm, venous obstruction, pericardial effusion, or pulmonary embolus. Pleural abnormalities were subdivided into effusions, thickening, and pneumothorax. Effusions were evaluated for size (small, < 10%; moderate, 11-50%; or large, > 50%) and loculation. Pleural thickening was evaluated for the presence of calcification. Furthermore, the presence or absence of a chest wall mass or a diaphragmatic hernia was assessed. Abdominal organs were assessed for abnormalities of the liver, adrenals, or spleen. Location (ribs, spine, sternum, or scapula) and type (fracture, mass, lytic lesion, or sclerotic lesion) of skeletal abnormalities were assessed.
Subsequently, three other board-certified, fellowship-trained thoracic radiologists with 6, 9, and 10 years of experience independently evaluated the 5-mm coronal and the 5-mm transverse image sets in random order during two review sessions separated by a 2-month interval using the same evaluation guidelines as described. The observers never saw the primary 1.5-mm reconstructions, the patient identifiers, or the results of the other medical records available for the reference standard interpretation. The randomization was structured so that each patient (coronal or transverse section set) was reviewed once during each of the two sessions. The review time for each data set and the number of images reconstructed through the transverse and coronal planes were recorded.
Statistical Analysis
For the 58 main categories and subcategories, the sensitivity and
specificity of the transverse plane and coronal plane interpretations were
averaged across the three observers. Differences in averaged transverse plane
versus coronal plane sensitivity and specificity values were statistically
evaluated using Wilcoxon's matched-pair signed rank test. The more traditional
McNemar test was not used because sample size requirements for powerful
inferential statistical evaluation of these differences exceed the available
sample size.
For unordered evaluation categories, we assessed the interobserver
concordance with Cohen's kappa coefficient (
)
[9]; for rank-ordered
evaluation categories, Cohen's weighted kappa coefficient
(w) was used
[10]. Exact 95% confidence
intervals for kappa and weighted kappa were computed using StatXact software
(Cytel Software Corporation) for Windows (Microsoft)
[11]. For kappa coefficients,
we used the bench-marks proposed by Landis and Koch
[12] for evaluation, with a
kappa value of less than 0 indicating agreement below chance; 0-0.19, poor
agreement; 0.20-0.39, fair agreement; 0.40-0.59, moderate agreement;
0.60-0.79, substantial agreement; and 0.80 or higher, almost perfect
agreement.
The distributional properties for differences between correlated (same-sample) kappa coefficients are not known; thus, a closed-form inferential test statistic is not readily available [13]. We therefore assessed between-observer differences in kappa coefficients for observers' evaluations with the reference standard with a resampling technique, as implemented by KAPCOM software (Monash University Department of Psychological Medicine) [14].
Differences in reporting times across observers and across review sessions were analyzed with a two-way repeated-measures analysis of variance, with the three observers as the first within-subjects factor and the plane of interpretation (transverse vs coronal) as the second within-subjects factor.
|
TopAbstractIntroductionSubjects and MethodsResultsDiscussionReferences |
|---|
Reference Standard
In 48 of 50 patients, abnormalities of the lungs (e.g., pulmonary nodules,
interstitial opacities, emphysema, consolidation, atelectasis, pleural
effusion and thickening, bronchiectasis, intraluminal airway lesions, and wall
thickening) were present. The majority of these patients presented with
multiple findings (mean number of lung abnormalities, 2.65 ± 1.39).
Interstitial opacities were classified as predominantly nodular (n = 3), reticular (n = 7), ground-glass opacification (n = 10), or cystic (n = 1). The secondary lobular distribution of opacities was assessed as peripheral (n = 8), peribronchovascular (n = 8), or random (n = 5). The zonal distribution of opacities was assessed as located predominantly in the upper portions of the lung (n = 7), the lower portions (n = 5), or random (n = 9). The distribution of air-space consolidations was assessed as unilobar (n = 5) or multilobar (n = 2), with zonal predominance in the lung periphery (n = 3), perihilar region (n = 1), or neither (n = 3). Lobar atelectasis was located in the right middle lobe (n = 2) and right lower lobe (n = 2). Bronchiectasis was unilobar (n = 2) or multilobar (n = 3). Pneumomediastinum was not found in any patients. Mediastinal masses were located in the anterior (n = 1), middle (n = 6), or posterior (n = 3) mediastinum with sizes ranging from 1.7 to 8 cm in maximal diameter. In one patient the mass revealed encasement of the aorta, and in another patient the mass was calcified. Pleural effusions were classified as small (n = 4) or moderate (n = 2) and as located on the right (n = 3) or left (n = 5) side. Pleural thickening was seen on the right (n = 7) or left (n = 10) side. No pneumothorax was seen. Skeletal abnormalities were classified as fracture (n = 9) or sclerotic lesion (n = 3).
Transverse and Coronal Interpretations
The mean sensitivity and specificity for all findings evaluated are
displayed in Tables 1,
2,
3. In 38 of the 58 main
categories and subcategories, the sensitivity value was higher for the
transverse plane interpretations than for the coronal plane interpretations.
In 12 categories, the results were reversed, and in seven categories the
results were tied. One category, mediastinal encasement, was omitted from
analysis because no cases were identified by the observers on coronal
sections.
|
|
|
The mean sensitivity for the interpretation of transverse images was higher than that for coronal reformatted images (51% ± 22% vs 44% ± 26%, respectively; p = 0.001, Wilcoxon's matched-pair signed rank test).
In 30 of the 58 categories, the specificity value was higher for the coronal plane interpretations as compared with the transverse plane interpretations. For 10 categories, the results were reversed, and for 18 categories the results were tied. This pattern was statistically significant (p = 0.005). The mean specificity ± SD across the 58 categories was 95% ± 6% for transverse plane interpretations versus 96% ± 5% for coronal plane interpretations.
The following findings of the observers are not presented in Tables 1, 2, 3 because they were not present according to the reference standard interpretations. Pneumomediastinum was falsely assessed in one patient by one observer on the basis of the transverse plane image. Each of two observers falsely assessed one venous obstruction, one on the basis of the transverse image and the other, the coronal plane image.
Categories Evaluated for Interobserver Concordance
Table 4 displays the mean
kappa values for concordance to the reference standard in transverse or
coronal interpretations for the following categories: number of pulmonary
nodules, type and distribution of interstitial opacities, distribution and
location of consolidations, distribution of bronchiectasis, quantification of
pleural effusions, and classification of skeletal lesions.
|
The concordance of the three observers' interpretations with the reference
standard was fair to moderate for most of the categories using either coronal
or transverse images, and the single-observer difference in kappa values for
coronal versus transverse images was not significant in 24 of 27, or 89%, of
the cases (3 observerx 9 categories). For location of consolidation, in
one observer the concordance to the reference standard was significantly
better on coronal images ( = 0.55) compared with transverse images
( = 0.20) (p < 0.01), whereas for another observer it was
significantly lower on coronal ( = 0.40) than on transverse ( =
0.66) images (p = 0.01). For distribution of bronchiectasis, the
concordance of one observer's interpretations with the reference standard was
significantly higher (p < 0.01) for transverse
(w = 0.67) compared with coronal (w =
0.25) images.
|
TopAbstractIntroductionSubjects and MethodsResultsDiscussionReferences |
|---|
Previous studies and review articles have indicated the value of multiplanar reconstruction chest images for the evaluation of airways, vessels, diaphragm, and preoperative evaluation of masses [1, 2, 15-18]. The excellent depiction of airway stenoses and of pulmonary emboli was achieved by tailored reformation in a paracoronal plane for display of the trachea or in a double oblique plane along the pulmonary arteries for visualization of pulmonary emboli [1, 2]. In the present study, the assessment of coronal display was based on "true" coronal reformatted images created without user interaction. Consequently, the detection rates for coronal display were low in some categories for which paracoronal oblique multiplanar reconstructions might have been helpful (e.g., airway abnormalities), and this study design may have contributed to the relatively low value for the sensitivity averaged across all 58 categories evaluated.
|
|
|
|
|
|
|
|
|
Eibel et al. [20] reported that image quality in the paracardiac region of coronal reformatted chest CT scans was partially compromised due to pulsation artifacts, which were also seen in the present study. We could not show that a single lesion was obscured by that artifact, although the shape of some paracardiac nodules was altered due to transmitted pulsations (Figs. 4A and 4B). However, to overcome that problem, Flohr et al. [21] have recently introduced a modified reconstruction technique of retrospectively ECG-gated MDCT with extended volume coverage allowing suppression of motion artifacts of the heart in cardiothoracic imaging. It is also likely that pulsation artifacts would be reduced with a higher number of detector rows, because the table speed would be increased proportionally at a fixed pitch and pulsation artifacts would be spread over a greater longitudinal distance. Theoretically, another possible reason for lower sensitivity of coronal images as compared with transverse images is the occurrence of stair-step artifacts. Characteristically, stair-step artifacts are associated with surfaces or object borders inclined relative to the table translation direction [22]. However, Fleischmann et al. [23] have shown that these artifacts are quantitatively and subjectively smaller with 4-MDCT than with single-detector helical CT. Consistent with this, we did not show detection errors due to stair-step artifacts in the present study.
|
|
Previous studies have evaluated the effect of MDCT technology on time management, workflow, and scanner productivity [25, 26]. Jhaveri et al. [25] reported that the use of an MDCT unit leads to an increase in CT examinations, even though MDCT studies are performed using more complicated protocols than are used on a single-detector CT scanner. Because data acquisition time was no longer a problem, Roos et al. [26] showed that patient management, data reconstruction, and data storage are the most time-intensive tasks; therefore, well-trained technicians, state-of-the-art workstations, and fast networking are the most important factors for the improvement of MDCT workflow. On the other hand, to our knowledge, no data are available in the literature about how the reporting times of radiologists for interpreting MDCT scans could be improved.
MDCT offers unparalleled speed of acquisition, spatial resolution, and anatomic coverage. A challenge presented by these advantages is the substantial increase in the number of reconstructed cross-sections that are rapidly created and in need of analysis [27]. As a result, the efficiency of the interpreter might be adversely impacted; this has already been shown for reporting findings on conventional radiographs [28]. Potentially, the primary assessment of chest CT scans using fewer coronal as compared with transverse images could improve the efficiency of the workflow. Nevertheless, data of the present study show that although coronal images sets consisted of significantly fewer images than the transverse image sets, interpretations times for coronal image sets were significantly longer. The longer interpretations times likely relate to greater observer familiarity with the transverse display. However, one can assume that the time for assessment for coronal reformatted images would decrease with routine use. An alternative explanation is that with the ability to rapidly scroll through multiple images using cine viewing, the absolute number of images is less of a factor in determining total interpretation time.
Our pilot study was limited by the relatively small sample size and the low disease prevalence in some diagnostic subcategories. Thus, the statistical power was too low to establish statistically reliable significance tests for every abnormality. However, our conclusion was drawn from the statistically significant differences of averaged sensitivity and specificity across all categories and observers. Further, we cannot exclude that a bias due to subjectivity inherent in our reference standard assessment might have influenced values of sensitivity and specificity. Another limitation may be seen in the fact that the CT examination protocol was not optimized for some diagnostic categories (e.g., emphysema, wall thickening of airways, abdominal abnormalities). However, this fact does not alter our conclusion because this was true for both coronal and transverse images.
In conclusion, primary interpretation of thoracic MDCT using routine 5-mm-thick coronal reformations was found to be less sensitive and more time-consuming as compared with transverse images.
|
TopAbstractIntroductionSubjects and MethodsResultsDiscussionReferences |
|---|
CiteULike 
Complore 
Connotea 
Del.icio.us 
Digg 
Reddit 
Technorati What's this?
This article has been cited by other articles:
|
|
 |
|
  T. A. Jaffe, L. C. Martin, C. M. Miller, K. M. Franklin, E. M. Merkle, W. M. Thompson, R. C. Nelson, D. M. DeLong, and E. K. Paulson Abdominal Pain: Coronal Reformations from Isotropic Voxels with 16-Section CT--Reader Lesion Detection and Interpretation Time Radiology, January 1, 2007; 242(1): 175 - 181. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
