| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Experimental Studies |
1 From the Department of Clinical Radiology, University of Muenster, Albert-Schweitzer-Strasse 33, 48129 Muenster, Germany (U.R.B., T.M.B.); Department of Biometrics and Medical Informatics, Otto-von-Guericke University, Magdeburg, Germany (F.W.R.); Department of Radiology, Wesley Medical Center, Wichita, Kan (R.C.G.); ECO, Barleben, Germany (H.S.); and Institute for Diagnostic and Interventional Radiology, University of Essen, Germany (U.W.K.). From the 2000 RSNA scientific assembly. Received April 20, 2001; revision requested June 8; final revision received October 28, 2002; accepted November 22. Address correspondence to U.R.B. (e-mail: bernhart@uni-muenster.de).
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMATERIALS AND METHODS RESULTS DISCUSSIONREFERENCES |
|---|
MATERIALS AND METHODS: Ground-glass, linear, miliary, and reticular patterns; nodules; and catheters were superimposed over an anthropomorphic chest phantom. Hard copies were generated at different dose levels (speeds: 400, 800, and 1,600) with a flat-panel detector and were compared with copies generated with an asymmetric screen-film system (speed, 400). Detection performance of eight radiologists was compared with a receiver operating characteristic analysis of 19,200 observations per pattern. A difference was significant with a P value of .05.
RESULTS: There was no statistically significant difference between the flat-panel detector and the asymmetric screen-film system at the same speed (P > .05) and between the flat-panel detector at a speed of 800 and the asymmetric screen-film system at a speed of 400 (P > .05). The visibility of linear, miliary, and reticular patterns over lucent lung and of nodules smaller than 10 mm and catheters over obscured chest regions on copies generated at a speed of 1,600 with the flat-panel detector decreased, compared with the visibility of these features on copies generated with the asymmetric screen-film system (P < .05).
CONCLUSION: The diagnostic performance of the flat-panel detector is comparable to that of the asymmetric screen-film system for depiction of all simulated patterns of interstitial lung diseases, nodules, and catheters at the same speed and offers the potential of dose reduction to a speed of 800.
© RSNA, 2003
Index terms: Flat panel detector • Phantoms • Radiography, digital • Screens and films
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODS RESULTS DISCUSSIONREFERENCES |
|---|
|
MATERIALS AND METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODS RESULTS DISCUSSIONREFERENCES |
|---|
The 400-speed asymmetric screen-film combination (Insight VHC; Eastman Kodak, Rochester, NY) with a hard-copy size of 35 x 43 cm was used, and images were obtained with a generator (Polydoros 50S; Siemens) and Bucky unit (Multix CPH/Vertix 2E; Siemens Medical Systems, Forchheim, Germany). The phantom was placed on the table above the detector.
The asymmetric screen-film system used in this study has wide latitude, with improved assessment of the mediastinum, compared with the features of conventional screen-film systems. For these reasons, this screen-film system was chosen.
Acquisition Parameters
Equivalent acquisition parameters for the flat-panel detector system and the asymmetric screen-film system were used to obtain chest radiographs. The source-to-detector distance was 115 cm for both imaging modalities. With the flat-panel detector, images were obtained with 102 kVp and a stationary grid (ratio, 15:1; 80 lines per centimeter). This stationary grid is recommended for radiographs produced by using this system. The digital images were generated at three speeds, namely, 400, 800, and 1,600, with a focal spot size of 0.6 mm and a digital output of 12 bits per pixel.
With the asymmetric screen-film system, images were obtained with 102 kVp and a moving grid for chest radiographs produced by using this system (ratio, 12:1), with the detector dose according to a speed of 400 and automatic exposure. The focal spot size was 0.6 mm.
Processing for the Digital Detector System
The data were transferred to a workstation (Magic-View; Siemens Medical Systems) for optimization of window level (level, 1,900) and window width (width, 2,500). These parameters were optimized with the consensus of two radiologists (U.R.B., T.M.B.). Parameters were used in a postprocessing procedure as follows: All data were processed with one defined parameter set. The preprocessing procedure included a gray-scale look-up table for simulating a gradation curve, adaptive unsharp masking for edge enhancement, and dynamic range compression. The image postprocessing procedure is the routine image processing for chest radiographs produced by using this system. The absolute value of the length of the kernel is the number of pixels (n = 255) multiplied by the length of the pixel (length, 143 µm). The absolute value of the kernel length was therefore 3.6 cm. The gain of 0.15 corresponds to the enhancement factor.
Simulated Patterns
Patterns of simulated interstitial lung disease, nodules, and catheters were used. The ground-glass pattern was simulated with layers of cellulose soaked with contrast medium (iopamidol, Solutrast 300, and ioxithalamic acid, Telebrix; Byk Gulden, Konstanz, Germany). The linear pattern was simulated with silk thread soaked with contrast medium (iopamidol). The miliary pattern was simulated with clusters of bird seed (approximate size of each grain, 2 mm). The reticular interstitial lung disease pattern was simulated with gauze soaked with contrast medium (iopamidol).
Nodules were simulated with balls of beeswax that were 5–15 mm in diameter; 50% of the nodules were smaller than 10 mm. The venous catheters (Cavafix Certo 255 G18; B. Braun, Melsungen, Germany) were 30–50 mm long with an 18-gauge lumen.
Images of an anthropomorphic chest phantom (Humanoid Systems, Carson, Calif) that had 12 areas (ie, six pulmonary [lucent lung], three mediastinal, and three subdiaphragmatic areas [obscured chest regions]) were acquired.
Fifty templates, with each template having 12 areas, were used. Each of these 50 templates was superimposed over the phantom. A value of n = 60 was used for the ground-glass, linear, miliary, and reticular patterns; a value of n = 85 was used for nodules smaller than 10 mm or 10 mm or larger; and a value of n = 60 was used for catheters in 50 templates. The simulated patterns were represented in the 12 areas per template (total areas, 600 [12 areas multiplied by 50 templates]) randomly so that dependencies between the templates were prevented. Fifty percent of the areas were covered with the simulated patterns and 50% were not. The areas could be totally empty or contain one or more types of patterns.
Evaluation
A total of 200 images (50 templates multiplied by four modalities) obtained with the asymmetric screen-film system at a speed of 400 and with the flat-panel detector at speeds of 400, 800, and 1,600 were prepared randomly at different sittings. The asymmetric screen-film and digital images of each template and speed were presented to the readers in different sittings to avoid learning effects. Eight readers with 2–8 years of experience in chest radiology participated in this study. For each area of the 50 templates and for each pattern, the readers were asked to state whether the simulated pattern was present or absent and to rank their level of confidence with a five-point scale, as follows: score 1, definitely present; score 2, probably present; score 3, uncertain; score 4, probably not present; score 5, definitely not present.
A standard film viewer was used for assessing the hard-copy images.
Data and Statistical Analysis
The resulting 19,200 observations per pattern (50 templates multiplied by 12 areas used as cases in the receiver operating characteristic [ROC] analysis) per template multiplied by eight readers multiplied by four modalities (ie, the asymmetric screen-film system at a speed of 400 and the flat-panel detector at speeds of 400, 800, and 1,600) were analyzed according to ROC analysis to evaluate the depiction of different patterns. Factors of interest were the differences between the readers, the localization (ie, obscured chest regions, lucent lung), and the corresponding interactions. Evaluation was performed separately for the simulated patterns by using ROC analysis. ROC curves were created with a maximum-likelihood curve-fitting algorithm. Pattern depiction was estimated by means of the area under the ROC curve (Az). For the calculation of the Az, a statistical program was used (LABMRMC 1.0B, BETA version 3; Charles E. Metz, PhD, and Benjamin A. Herman, University of Chicago, Ill) (13). The subroutines were taken from programs developed with the supervision of Donald D. Dorfman, PhD, Department of Psychology, University of Iowa, Iowa City. The base of the program was a three-factorial analysis of variance (ANOVA). This program calculates a maximum-likelihood estimate of the discrete-rating data from as many as five potentially correlated diagnostic tests and as many as 10 readers and estimates the binormal ROC curves implied by those data. To estimate the Az for the covered areas, the same number of areas that were not covered were assigned. The statistical program can be used to calculate the statistical significance of the differences between the averaged Az by using a three-way factorial ANOVA method with the factors treatment (flat-panel detector at speeds of 400, 800, and 1,600) and readers as repeated factors, cases (areas) as independent factors, and the two-factorial interactions. Only the 50 templates were fully stochastically independent, but on some templates, more than one area was covered with the same pattern. To determine whether simplification of the model with independent areas in the statistical program was empirically justified, we used the system (SAS 8.02; SAS Institute, Cary, NC) to analyze the original data (the 1.0 version of the statistical software had no possibility for a four-factorial ANOVA with templates as random factors). The comparison of the results of ANOVA, with templates as random factors and the results of ANOVA with Az values as independent cases, by using the original data provided similar results.
First, the statistical software was used across the four systems and the eight readers (ie, asymmetric screen-film system and flat-panel detector system at speeds of 400, 800, and 1,600). Second, it was used across the three systems (ie, asymmetric screen-film system at a speed of 400 and flat-panel detector system at speeds of 400 and 800). In order to receive the CIs of the Az values and the pairwise comparisons for two modalities, one must repeat the analyses for each of the two systems. For an adequate assessment of the significance, a Bonferroni adjustment was performed. The 95% CIs of the mean Az values were calculated. This resulted in nine patterns (ie, the ground-glass, linear, miliary, and reticular patterns and the nodules [<10 mm over lucent lung, <10 mm over obscured chest regions, 
10 mm over lucent lung, 10 mm over obscured chest regions]) multiplied by four flat-panel detector speeds to yield 36 CIs of means of the Az values. The statistical program also can be used to calculate the Az values for each reader and each flat-panel detector speed.
All tests were performed with a .05 probability error.
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODS RESULTS DISCUSSIONREFERENCES |
|---|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
Catheters
There were no statistically significant differences between the asymmetric screen-film system and the 400-speed flat-panel detector in the depiction of catheters (P = .39). The ability to depict catheters on the images generated at the speed of 800 showed no statistically significant differences, compared with results with the asymmetric screen-film system at a speed of 400 over obscured chest regions. However, the visibility of catheters on digital images at a speed of 1,600 indicated a statistically significant difference (P < .001) and was poorer compared with that of the asymmetric screen-film system (Fig 8).
|
|
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODS RESULTS DISCUSSIONREFERENCES |
|---|
Good performance of the flat-panel detector is a result of both high contrast resolution and high DQE. The efficiency of the flat-panel detector exceeds the performance of storage phosphor plates and conventional screen-film systems and is comparable with the performance of the selenium-drum detector (9,14). Pixel size for a thoracic imaging system based on selenium is 200 µm. Spatial resolution requirements in radiography vary, depending on the subject of interest. MacMahon et al (15) showed that a pixel size of 200 µm led to a lower spatial resolution than that of conventional screen-film systems. The optimal pixel size will vary with the detector and the radiation dose level. Further factors in a digital radiograph that influence the visibility of subtle findings are markedly improved scatter rejection and image-processing techniques. The pixel size of 143 µm for the flat-panel detector is considered to be sufficient for chest imaging applications. Strotzer et al (9) indicated that the flat-panel detector was statistically superior when compared with the conventional screen-film system with respect to the depiction of linear and miliary opacities at equivalent radiation exposures. However, in their study, the images were acquired with 125 kVp and a film-focus distance of 170 cm, and their transparent hard copies were printed on laser-generated films (Ektascan DHG; Eastman Kodak, Rochester, NY) by using a laser printer (Ektascan XLP; Eastman Kodak). Moreover, they used different simulation materials.
Voelk et al (7) demonstrated that a flat-panel detector at a digital speed of 400 was significantly superior to a conventional screen-film system at a speed of 400 with respect to the detection of foreign bodies. In contrast to the systems we used in our study, they used a symmetric screen-film combination (Lanex Regular/T-MAT Plus DG film; Kodak, Stuttgart, Germany) instead of an asymmetric screen-film system with a speed of 400, as used in this study. The advantages of asymmetric screen-film systems include statistically significant superior diagnostic performance in depiction of patterns over the obscured lung without any disadvantage in the depiction of patterns over the lucent lung in a study in phantoms (11). Hence, we compared the flat-panel detector with the asymmetric screen-film system.
The influence of varying body diameter and scatter fractions was not studied. Our reading task was more comparable to a contrast-detail experiment. All 50 templates with simulated patterns were superimposed over the same slim anthropomorphic chest phantom.
In our study, the Az values for linear pattern and nodules demonstrated a large SD. Multiple factors may have contributed to this finding. One important factor may be the differences in the chest pattern simulations. The linear pattern and nodules were different in size, form, sharpness, and opacity, whereas the reticular and miliary patterns and the catheters were more uniform.
The large Az values for the nodules that were 10 mm or larger over lucent lung suggest that these patterns were too easy to depict. However, the results with this pattern did not differ from those with the other simulated patterns in this study.
The small Az values resulted from extremely subtle patterns (linear pattern over lucent lung), which were very difficult to depict. The results of this study indicated that the degree of subtlety correlated with the Az values and that extremely subtle patterns produce areas near 0.5, as was shown for a wide range of nodules with different sizes and degrees of subtlety by Shiraishi et al (16). In our study, the linear pattern was simulated more subtly than were the other patterns, and the readers were confused because of blurring.
There was a potential source of bias in our experiment: The film appearance was sufficiently different to allow a reader to distinguish the digital images from the asymmetric screen-film images. Thus, this bias may have influenced the results if one reader had a prejudiced opinion regarding one system.
The statistically significant factor regarding the reader for three patterns (Table) based on the experience in chest radiology because one of the readers with 3 years experience in chest radiology showed, for the asymmetric screen-film system and the flat-panel detector at all digital speeds, significantly decreased results in the depiction of catheters, nodules smaller than 10 mm, as well as miliary and reticular patterns, compared with all other readers.
Study findings have shown that dose reduction in general is possible by using digital systems, such as a selenium-based detector (11,17). Dose reduction for the flat-panel detector was shown for skeletal radiology and simulated chest lesions. Our results confirm that the flat-panel detector produces images of high quality in an anthropomorphic setting for the depiction of simulated interstitial lung diseases, nodules, and catheters, with no statistically significant differences when compared with the asymmetric screen-film system at the same speed and the digital speed of 800.
Dose reduction is most important in pediatric radiology and in repeat follow-up imaging. A reduction in radiation exposure by as much as 50% may be possible and does not impair depiction of fractures and cortical lesions with the flat-panel detector (6,12). In another study in phantoms (9), the authors evaluated simulated lung lesions at a speed of 800 with a digital system and did not find a statistically significant difference in the results as well. Voelk et al (7) also found results with a flat-panel detector at digital speeds of 800 and 1,200 to be equivalent to those with conventional radiography in the depiction of foreign bodies. Their results indicated a potential dose reduction of as much as 67% (7). In this respect, we were also able to demonstrate that the flat-panel detector at a digital speed of 800 yielded no statistically significant differences concerning the depiction of all patterns. However, in our study, the Az values were smaller for miliary pattern, reticular pattern, nodules over obscured chest regions, and catheters at a speed of 800 in a comparison of the asymmetric screen-film system and the flat-panel detector at a speed of 400, which indicated a loss of diagnostic accuracy. In a future study, dose reduction to a speed of 800 should be evaluated to confirm that the dose reduction does not significantly decrease diagnostic performance when compared with a speed of 400.
The improved DQE produces images of better diagnostic value with digital systems (11). The good absorption properties of the flat-panel detector are reflected in the high DQE of 65% at 70 kVp and 21 mm of aluminum prefiltering, which is superior to that of both conventional radiography and storage phosphor systems (1). The DQE for the flat-panel detector is superior to that of the asymmetric screen-film system and the storage phosphor system, with a factor of 2 (6). Moreover, high spatial and contrast resolution of the flat-panel detector result in a possible reduction of radiation dose (7). The dose reduction at a digital speed of 1,600 is not considered to be reasonable because a statistically significant difference was indicated, with inferior results in the depiction of miliary, linear, and reticular patterns over lucent lung, as well as in the depiction of catheters and small nodules over obscured chest regions. The decreasing signal-to-noise ratio as a result of the reduced dose at a digital speed of 1,600 led to a statistically significant difference with inferior assessment of these simulated patterns. For these subtle patterns, the readers were so confused, because of the blurring of the images, that a higher false-positive rate was produced that resulted in ROC curves with an Az value less than 0.5. However, for obvious patterns such as nodules over lung, this effect did not occur at the digital speed of 1,600. These patterns could be depicted even on images with blurring.
Our results suggest that a dose reduction with a system up to the digital speed of 800 can be promising.
Practical application: The flat-panel detector is suitable for the depiction of various patterns of simulated interstitial lung diseases, nodules, and catheters and offers the potential of a dose reduction to a speed of 800 because of its high DQE, as noted in this study in phantoms. The results of this study in phantoms suggest that a study in patients seems worthwhile to undertake.
|
FOOTNOTES |
|---|
Author contributions: Guarantors of integrity of entire study, U.R.B., T.M.B.; study concepts and design, U.R.B., T.M.B.; literature research, U.R.B., T.M.B.; experimental studies, U.R.B., T.M.B., U.W.K., R.C.G.; data acquisition, U.R.B., H.S., T.M.B.; data analysis/interpretation, U.R.B., F.W.R., H.S., T.M.B.; statistical analysis, F.W.R.; manuscript preparation, U.R.B., U.W.K., R.C.G., T.M.B.; manuscript revision/review, T.M.B.; manuscript definition of intellectual content, editing, and final version approval, U.R.B., T.M.B.
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODS RESULTS DISCUSSIONREFERENCES |
|---|
This article has been cited by other articles:
|
|
 |
|
  L J M Kroft, W J H Veldkamp, B J A Mertens, J-P A van Delft, and J Geleijns Dose reduction in digital chest radiography and perceived image quality Br. J. Radiol., December 1, 2007; 80(960): 984 - 988. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 L. J. M. Kroft, W. J. H. Veldkamp, B. J. A. Mertens, J. P. A. van Delft, and J. Geleijns Detection of simulated nodules on clinical radiographs: dose reduction at digital posteroanterior chest radiography. Radiology, November 1, 2006; 241(2): 392 - 398. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 K. Bacher, P. Smeets, L. Vereecken, A. De Hauwere, P. Duyck, R. De Man, K. Verstraete, and H. Thierens Image quality and radiation dose on digital chest imaging: comparison of amorphous silicon and amorphous selenium flat-panel systems. Am. J. Roentgenol., September 1, 2006; 187(3): 630 - 637. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 V. Neofotistou, V. Tsapaki, S. Kottou, A. Schreiner-Karoussou, and E. Vano Does digital imaging decrease patient dose? A pilot study and review of the literature Radiat Prot Dosimetry, December 1, 2005; 117(1-3): 204 - 210. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
U. Rapp-Bernhardt, T. M. Bernhardt, H. Lenzen, R. Esseling, F. W. Roehl, M. Schiborr, I. Theobald-Hormann, and W. Heindel Experimental Evaluation of a Portable Indirect Flat-Panel Detector for the Pediatric Chest: Comparison with Storage Phosphor Radiography at Different Exposures by Using a Chest Phantom Radiology, November 1, 2005; 237(2): 485 - 491. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
O. W. Hamer, C. B. Sirlin, M. Strotzer, I. Borisch, N. Zorger, S. Feuerbach, and M. Volk Chest Radiography with a Flat-Panel Detector: Image Quality with Dose Reduction after Copper Filtration Radiology, November 1, 2005; 237(2): 691 - 700. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
E. Norrman and J. Persliden A factorial experiment on image quality and radiation dose Radiat Prot Dosimetry, May 17, 2005; 114(1-3): 246 - 252. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
T. M. Bernhardt, U. Rapp-Bernhardt, H. Lenzen, F. W. Roehl, S. Diederich, K. Papke, K. Ludwig, and W. Heindel Low-Voltage Digital Selenium Radiography: Detection of Simulated Interstitial Lung Disease, Nodules, and Catheters--A Phantom Study Radiology, September 1, 2004; 232(3): 693 - 700. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| RADIOLOGY | RADIOGRAPHICS | RSNA JOURNALS ONLINE |
