| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Experimental Studies |
1 From the Departments of Radiology (L.J.M.K., J.G., W.J.H.V., H.M.Z., A.d.R.) and Medical Statistics (B.J.A.M.), C2-S, Leiden University Medical Center, C2S, Albinusdreef 2, 2333 ZA Leiden, the Netherlands. Received February 7, 2003; revision requested April 24; revision received June 13; accepted August 8. Address correspondence to L.J.M.K. (e-mail: l.j.m.kroft@lumc.nl).
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMATERIALS AND METHODS RESULTS DISCUSSIONREFERENCES |
|---|
MATERIALS AND METHODS: Simulated nodules and interstitial nodular and interstitial linear lesions were attached onto an anthropomorphic chest phantom. One hundred sixty-eight lesions were distributed over 25 configurations. A posteroanterior chest radiograph of each configuration was obtained with each technique. The images were presented to six observers. Each lesion was assigned one of two outcome scores: "detected" or "not detected." False-positive readings were evaluated. Differences between the imaging methods were analyzed by using a semiparametric logistic regression model.
RESULTS: For simulated nodules and interstitial linear disease, no statistically significant difference was found in diagnostic performance between CCD digital radiography and AMBER. The detection of simulated interstitial nodular disease was better with CCD digital radiography than with AMBER: Sensitivity was 71% (77 of 108 interstitial nodular lesions) with CCD digital radiography but was 56% (60 of 108 lesions) with AMBER (P = .041). Better results for the detection of all lesion types in the mediastinum were observed with CCD digital radiography than with Bucky screen-film radiography: Sensitivity was 45% (227 of 504 total simulated lesions) with CCD digital radiography but was 24% (119 of 504 lesions) with Bucky radiography (P < .001). There were fewer false-positive observations with CCD digital radiography (35 [5.7%] of 609 observations) than with Bucky radiography (47 [9.5%] of 497 observations; P = .012).
CONCLUSION: Differences were in favor of the full-field slot-scan CCD digital radiographic technique. This technique provides a digital alternative to AMBER and Bucky screen-film radiography.
© RSNA, 2004
Index terms: Dosimetry • Phantoms • Radiography, comparative studies, 68.11, 68.1215 • Radiography, digital, 68.1215 • Thorax, radiography, 68.11, 68.1215
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODS RESULTS DISCUSSIONREFERENCES |
|---|
For chest imaging, Bucky screen-film radiography provides insufficient dynamic range for good visualization of the lungs and the mediastinal area simultaneously because of the wide difference in x-ray absorption. With advanced multiple beam equalization radiography (AMBER)—also a screen-film–based technique—this limitation is partially overcome by modulating the local exposure and thereby increasing the exposure in the mediastinum (4). The wide dynamic range and linear response (ie, the relationship between exposure and raw pixel values) of digital radiography systems are expected to be particularly advantageous for chest radiography, owing to the large difference in x-ray absorption between the lungs and the mediastinum.
Several technologies for digital chest imaging have been developed and introduced to the market. Among these technologies are photostimulable storage phosphors and direct-readout electronic matrix detectors such as thin-film transistors and scintillator charge-coupled devices (CCDs). Scintillator CCD–based systems are well suited to digital radiography because of their high spatial resolution, wide dynamic range, low electronic noise, and linear response (5).
Until recently, the major limitation of CCD digital radiography for chest imaging was demagnification. The size of CCDs is small compared with the required field for chest imaging. Optical coupling of the scintillator to the CCDs has been achieved with lens systems or fiber-optic tapers, resulting in a projection of the x-ray image on the CCDs. Demagnification is undesirable because it causes reduced dose efficiency and degradation of the image (2,6–8).
Recently, a new system for digital chest imaging (ThoraScan; Delft Imaging Systems, Veenendaal, the Netherlands) was introduced in our hospital. The ThoraScan unit contains a digital direct readout system that is based on a thallium-doped cesium iodide scintillator that is coupled to a linear array of CCDs. Images are acquired with slot-scan technology. With this combination, full-field chest images are obtained without demagnification. An important advantage of slot-scan technology is the limited number of detector elements that it requires (thereby enabling cost-effective use of CCDs as detectors), even as it yields images with high spatial resolution. Moreover, use of a combination of a collimated fan beam and a matching detector means that the system inherently rejects scatter in a highly efficient fashion, so there is no need to use a scatter grid (7).
The purpose of our study was to evaluate the diagnostic performance of and the radiation exposure involved in the use of a digital full-field slot-scan CCD chest radiography unit, in comparison with those of AMBER and Bucky screen-film radiography systems, for the detection of simulated chest disease in clinical conditions.
|
MATERIALS AND METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODS RESULTS DISCUSSIONREFERENCES |
|---|
|
Radiation Exposure
Radiation exposure was measured (as dose-area product) for the CCD digital radiography and the AMBER and Bucky screen-film radiography systems with a Diamentor M4 dosimeter (PTW-Freiburg, Freiburg, Germany). Effective dose was calculated by using PCXMC software (9). This is a dedicated computer program for Monte Carlo simulation of radiation transport in x-ray projection radiology. We calculated the effective dose (with the arms kept outside the x-ray beam) for an average patient (height, 170 cm; weight, 70 kg), who was represented by an anthropomorphic chest phantom (Model PBU-S-3; Kaguku, Kyoto, Japan). Specific exposure conditions such as the actual tube voltage, beam filtration, and anode angle were taken into account, as was the beam modulation of the AMBER unit. Since dose-area product depends linearly on field size, the same anatomic area was imaged during the dose measurements for the different systems. For CCD digital radiography, dose during the prescan was included in the effective dose.
Data Acquisition
Simulation and distribution of lesions.—The anthropomorphic chest phantom used for the study consisted of a skeleton and lungs that were simulated by synthetic materials and the heart and blood vessels of a pig. These structures were embedded in soft-tissue–equivalent plastic. The following chest diseases were simulated: chest tumors, interstitial nodular disease, and interstitial linear disease (Fig 2). Chest tumors were simulated by nodules of synthetic modeling clay (SES, Enschede, the Netherlands). These nodules varied in size and were either moderately opaque (ie, flat) or opaque (ie, round). Moderately opaque nodules were prepared in five sizes (0.7, 1.0, 1.5, 2.0, and 2.5 cm), and opaque nodules were prepared in three sizes (1.0, 1.5, and 2.0 cm). Three representations of interstitial nodular disease were simulated by using three kinds of birdseed. For each type of seed, 20 seeds, each of which was about 1–3 mm in size, were used per simulated lesion. Three manifestations of interstitial linear disease were simulated by using two types of untwisted rope soaked in iodinated contrast agent and one type of thin copper wire.
|
|
Image acquisition.—One posteroanterior radiograph of each of the 25 configurations was obtained with the three different chest units, yielding a total of 75 radiographs. Images were acquired with the ThoraScan CCD digital chest radiography system, an AMBER chest unit (Nucletron-Oldelft, Veenendaal, the Netherlands), and an Optimus S/F Bucky screen-film radiography wall stand (Philips Medical Systems, Best, the Netherlands). Care was taken to ensure equal positioning of the phantom at each of the three units. After each configuration was imaged, metal objects that encoded for the lesion type were placed onto the lesions, and imaging was repeated with all three units to provide for standard-of-reference images. The imaging parameters of the CCD digital radiography and the AMBER and Bucky screen-film radiography systems are shown in Table 1. Antiscatter grids were used for the screen-film systems. For AMBER, a grid with a ratio of 12:1 and 36 lines per centimeter was used. For Bucky screen-film radiography, a grid with a ratio of 12:1 and 36 lines per centimeter was used. AMBER and Bucky images were obtained with a 400-speed screen-film combination (Lanex regular screens and T-mat-G 65500 film; Kodak, Rochester, NY).
|
In each session, all 25 images obtained with one system (CCD digital radiography or AMBER or Bucky screen-film radiography) were shown, and each series of 25 images was presented in a different but standard sequence. AMBER and Bucky radiographs were read by using a film viewing box (Rotolux Planilux; Philips Medical Systems) with a light output of 2,925 candelas per square meter and good homogeneity. Care was taken to ensure standard and optimal environmental light conditions. The observers were allowed to use a bright light (Cool Brite illuminator, model CB2; RADX, Houston, Tex) for viewing details in the screen-film radiographs. The digital CCD radiographs were presented as soft copies at a 2,048 x 1,530-pixel workstation (type 3C workstation; Dome Imaging Systems, Waltham, Mass). The observers read all soft copies in a 1:1 pixel mode by roaming in such a way that all parts of the image were reviewed. The observers were allowed to alter the window width and window level.
To compensate for learning bias, the order of reading the CCD digital and the AMBER and Bucky screen-film radiographs was unique for each of the six observers. In other words, each of the six possible permutations of the three imaging techniques was presented to one of the six observers. At least 1 week elapsed between the reading sessions. Before each session, a sequence of three radiographs—obtained with the imaging technique that was being evaluated at that session—that depicted the different lesion types and their matching standard-of-reference radiographs were shown to the observers in a training session, allowing the observers to become accustomed to the look of the anthropomorphic phantom and the visibility of the lesions. These training examples were not among the images being evaluated.
The observers drew the outer contours of each lesion they detected precisely on the radiographs to produce annotated images for comparison with the standard-of-reference images. For CCD digital radiographs, this was performed on laser-print hard copies that were presented next to the workstation on the same film viewing box used to view the screen-film radiographs. We assumed that the soft-copy readings were of better quality than the hard-copy readings and that the hard copies did not aid the observers. Only one image was observed at a time, and each image was viewed only once per session. Observers were not allowed to review an image they had seen previously.
Statistical Analysis
The images annotated by the observers were evaluated in consensus by two "outcome" readers (L.J.M.K. and W.J.H.V.) by placing the matching standard-of-reference images over the annotated images. The two natural probabilities for measuring the efficiency of an imaging system are (a) the probability that a lesion will be detected by the observer and (b) the probability that the judgment of the observer is correct. Accordingly, each lesion was given an outcome score of "detected" (representing a true-positive observation) or "not detected" (representing a false-negative observation). Annotated areas that did not contain a simulated lesion were scored as falsely positive.
To evaluate possible systematic differences between the CCD digital radiography and the AMBER and Bucky screen-film radiography systems with respect to the probability of detecting simulated lesions, an analysis was performed by using a semiparametric logistic model for binary outcome with normally distributed random effects and taking the observer variability into account by using the PROC NLMIXED function in the SAS/STAT software, version 8 (SAS Institute, Cary, NC).
In the analysis we performed, the first action is to fit a simple "base" (logistic regression) model. We then sequentially expanded (refined) the model if the data provided evidence of differences between lesion detection probabilities that were not explained by the previous model. This procedure, although conservative, does guarantee that the most simple explanation supported by the data will be maintained and that no overly detailed hypotheses for explaining observed differences in detection probability will be put forward owing to unwarranted assumptions.
The semiparametric logistic regression analysis models the odds of detecting lesions as expressed by the following equation:
|
is the so-called intercept term for the final model (this refers to the reference category, which in our study was "detection of a moderately opaque nodule in the lungs with the CCD digital radiography technique at the first reading session"). In the equation, tech represents technique and eff refers to all terms required by the final model to, for example, distinguish the different odds of detection not only between the modalities and according to lesion location and lesion type, but also according to reading session number and other study parameters. The eff terms are then so defined as to describe deviations in detection odds from the reference condition . The term 
j is a random effect that is included to differentiate between the possibly different scoring habits of the respective jth observer. We assumed that observers were randomly drawn from a normally distributed population of experienced radiologists. The probability of detection of a lesion (prob) can be calculated from the odds ratio (OR) by using the following equation: prob = OR/OR + 1. P values of less than .05 were considered to indicate statistically significant differences.
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODS RESULTS DISCUSSIONREFERENCES |
|---|
|
|
Additional Information
Additional statistically significant information derived from the semiparametric logistic regression model (Table 3) was as follows: Opaque nodules (i = 4), interstitial nodular lesions (i = 5), and interstitial linear lesions (i = 6) were detected more frequently than were moderately opaque nodules. Lesions were detected less frequently in the mediastinum (i = 7) than in the lungs. In the mediastinum, interstitial nodular lesions (i = 19) were detected more frequently than were moderately opaque nodules and interstitial linear lesions (i = 20) were detected less frequently than were moderately opaque nodules.
False-Positive Observations
Data regarding true-positive and false-positive observations are shown in Table 4. No difference was found in the frequency of false-positive observations between the CCD digital and the AMBER screen-film systems. With CCD digital radiography, there were fewer false-positive observations than with Bucky screen-film radiography (P = .012).
|
|
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODS RESULTS DISCUSSIONREFERENCES |
|---|
However, reported DQEs for digital systems are usually for the use of these systems without a grid, even though a grid is used in the clinical setting. Measurements with the grid in place should result in a degradation of the DQE by 40% at all frequencies, although the loss of DQE might be offset by the improved scatter properties achieved with the grid (10). The CCD digital radiography system used in the present study, which has a DQE of 60% at 0 line pairs per millimeter, is purported to be among the group of digital systems that have the highest DQEs—DQEs that ensure optimal image quality even when a grid is not used (11).
Until relatively recently, small-field-of-view CCD digital radiography systems have been commercially available only for such applications as dental radiography, spot-compression mammography, and guidance of stereotactic breast biopsy (12–14). For intraoral radiography, the diagnostic accuracy of small-field-of-view CCD systems has been shown to be equivalent to that of Bucky screen-film radiography (14). Results of experiments with digital spot-compression mammography have shown that the contrast-detail detectability achievable with CCD digital radiography is significantly superior to that achievable with Bucky screen-film radiography at the same radiation dose (12). In the past few years, a full-field digital mammography slot-scan CCD digital radiography system has been developed (15). With the introduction of the ThoraScan unit, it has become possible to image the entire chest by using a CCD detector without demagnification.
To obtain an unbiased evaluation of the diagnostic performance of the CCD digital radiography chest system as compared with that of screen-film chest radiography systems, we used a statistical model that allowed for distinct detection probabilities according to modality, lesion location, lesion type, and reading session. We also incorporated an interaction between modality and lesion location into the model to account for differential detection probabilities of the techniques according to lesion location. All these parameters were always included in the model.
After we created this base interaction model, we were able to expand the model to account for significant differences in lesion detection between modalities according to lesion type, as well as according to lesion type and location in lungs or mediastinum. For each expansion in the model, the complexity of the model increased. We determined the point at which adding more parameters to the model did not reduce the deviance significantly further. At that point, the detection proportions observed in the data were optimally fitted by the model. No further major differences in lesion detection probabilities that could have supported further refinement of our model should exist. In the model, we always corrected for a possible random observer effect. The data modeling clearly revealed close agreement between the "theoretical" fitted proportions from our final model and the "crude" estimates of proportions detected by the observers.
This sequential modeling procedure was crucial because our experiment was factorial in nature. An advantage of this statistical modeling technique is that all effective study parameters (and data) are taken into account simultaneously. For example, restricting analysis to specific subgroups (eg, mediastinum, interstitial nodular disease) a priori implies by definition that one will no longer be able to interpret any possible difference between imaging techniques with explicit reference to the specific subgroup one is interested in. Thus, such a procedure would achieve the opposite of what researchers are interested in and would almost inevitably lead to overly restrictive and unwarranted interpretation of results. Interpretation of the P values in our study that indicated significance in detection of lesions in the mediastinum provides an example of the errors that could occur. These P values were all small, owing to the generally poor performance of conventional screen-film radiography as compared with the performance of CCD digital radiography (P < .001), irrespective of lesion type.
In the present study, interstitial nodular lesions were better detected with the CCD digital radiography system than with the AMBER screen-film system. These interstitial nodular lesions were composed of small seeds with inherently low contrast. Improved visualization of especially small low-contrast objects with digital imaging has previously been reported by other researchers, who used a selenium detector or a scintillator-photodiode thin-film transistor–based flat-panel detector (16,17). This improved visualization has been attributed to the higher DQE of the digital system as compared with the DQE of screen-film radiography (17). Except for differences in the detection of small low-contrast interstitial nodular lesions, we found no differences between CCD digital radiography and the AMBER screen-film system. CCD digital radiography can be considered a digital alternative to AMBER screen-film techniques.
To our knowledge, high-DQE digital chest radiography systems have not been compared with AMBER screen-film systems before. There are some similarities between the ThoraScan CCD digital radiography system we used in our study and AMBER screen-film systems. First, both ThoraScan and AMBER involve the use of slot-scan technology for building up an image. Second, AMBER incorporates a feedback mechanism to control local x-ray exposure during scanning to ensure increased exposure in the mediastinum (4). The exposure control improves the dynamic range of AMBER screen-film systems, thereby improving the detectability of lesions in the mediastinum with these systems in comparison with the detectability of such lesions with Bucky screen-film radiography (18). Additionally, the wide dynamic range and linear response of the CCD digital radiography system results in improved detection of lesions in the mediastinum as compared with detection of such lesions with Bucky screen-film radiography.
There have been some clinical studies in which digital flat-panel detector systems were compared with Bucky screen-film chest radiography systems. In two observer preference studies that included 115 and 100 patients, respectively, digital flat-panel detector systems, as compared with Bucky screen-film systems, enabled significantly superior visualization of anatomic chest structures (19,20). In a study involving 80 patients in which computed tomographic findings were used as a standard of reference, use of a digital flat-panel detector technique was found to result in improved detection of mediastinal disease as compared with detection of such disease with Bucky screen-film radiography, whereas no difference was found between the two techniques in terms of lesion detection in the lungs (21). In the present study, a variety of lesions were simulated to approach the diversity of clinical situations as much as possible. We found that the improved detection with CCD digital radiography, as compared with the detection with Bucky screen-film radiography, was also striking for lesions in the mediastinum, and detection was improved for all lesion types. The surplus value of digital imaging appears mainly advantageous for imaging the mediastinal region.
The radiation exposure that was chosen for the CCD digital chest radiography system in our study is comparable to that of AMBER. Dose-reduction studies with the CCD digital radiography system are currently being performed at our institution.
There were some limitations to our study. The frequency with which lesions were detected by the observers increased with each reading session because of the learning effect. This was expected because the phantom model remained the same while the lesions changed in each configuration. However, this disturbing factor was corrected for in the study design and in the statistical model used for analysis.
Practical application: Full-field slot-scan CCD digital radiography constitutes an additional concept for digital chest imaging. Results of this experimental study show that the differences we observed between CCD digital radiography and AMBER and Bucky screen-film radiography in the detection of simulated lesions are in favor of the use of CCD digital radiography for chest imaging. The main benefit with CCD digital radiography, as compared with Bucky screen-film radiography, is improved depiction of lesions in the mediastinum. CCD digital radiography provides a digital alternative to AMBER and Bucky screen-film radiography. The CCD digital chest radiography system (ThoraScan) evaluated in this study is currently being used for routine chest radiography at our hospital.
|
ACKNOWLEDGMENTS |
|---|
|
FOOTNOTES |
|---|
Author contributions: Guarantor of integrity of entire study, L.J.M.K.; study concepts and design, J.G., L.J.M.K.; literature research, L.J.M.K., J.G.; experimental studies, J.G., L.J.M.K., W.J.H.V.; data acquisition, J.G., L.J.M.K., W.J.H.V., H.M.Z., A.d.R.; data analysis/interpretation, J.G., L.J.M.K., B.J.A.M., W.J.H.V.; statistical analysis, B.J.A.M., J.G.; manuscript preparation and definition of intellectual content, L.J.M.K., J.G., B.J.A.M.; manuscript editing, L.J.M.K., J.G.; manuscript revision/review and final version approval, all authors
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODS RESULTS DISCUSSIONREFERENCES |
|---|
This article has been cited by other articles:
|
|
 |
|
  M. Korner, C. H. Weber, S. Wirth, K.-J. Pfeifer, M. F. Reiser, and M. Treitl Advances in Digital Radiography: Physical Principles and System Overview RadioGraphics, May 1, 2007; 27(3): 675 - 686. [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]
|
|
|
|
 |
|
 L. J. M. Kroft, W. J. H. Veldkamp, B. J. A. Mertens, M. V. Boot, and J. Geleijns Comparison of Eight Different Digital Chest Radiography Systems: Variation in Detection of Simulated Chest Disease Am. J. Roentgenol., August 1, 2005; 185(2): 339 - 346. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
W. J. H. Veldkamp, L. J. M. Kroft, B. J. A. Mertens, and J. Geleijns Digital Slot-Scan Charge-coupled Device Radiography versus AMBER and Bucky Screen-Film Radiography: Comparison of Image Quality in a Phantom Study Radiology, June 1, 2005; 235(3): 857 - 866. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| RADIOLOGY | RADIOGRAPHICS | RSNA JOURNALS ONLINE |
