| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Original Research |
1 Department of Radiology, University of Cologne, Kerpenerstraße 62,
Cologne, NRW 50924, Germany.
2 Department of Medical Statistics, University of Cologne, Cologne, NRW 50924,
Germany.
Received November 15, 2005;
accepted after revision May 16, 2006.
Address correspondence to K. B. Krug
(Barbara.Krug{at}uk-koeln.de).
Abstract
 Top Abstract IntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
MATERIALS AND METHODS. Studies were performed with a digital mammography system (Selenia) and an analog mammography system (Mammomat 3). Sixty-five transparent films were used as test specimens. Randomly distributed round and heterogeneous silicate particles (diameter, 100-1,400 µm) and an anthropomorphic scatter body were applied to the films. All radiographs were taken at identical settings and exposures. Six radiologists rated the films and monitor-displayed images independently of each other in random order on a standardized electronic questionnaire.
RESULTS. Interpretations based on monitor reading produced superior results over those based on digital image reading and analog film reading. In 41.1% (95% CI, 38.7-43.5%) of all the monitor readings, 20.2% (18.2-22.2%) of all digital images, and 19.6% (17.6-21.6%) of all analog films, the number of detectable microcalcifications agreed with the gold standard method. The diameter of visible microcalcifications was interpreted correctly in 35.6% (33.2-38.0%) of monitor readings, 19.0% (17.1-21.0%) of digital images, and 21.0% (18.9-23.0%) of analog films; and microcalcification shape was interpreted correctly in 53.8% (51.4-56.3%) of monitor readings, 28.2% (26.0-30.4%) of digital images, and 28.3% (26.0-30.5%) of analog films. Microcalcification number and size were underestimated more frequently than overestimated. Regardless of display medium, accuracy increased proportionately with the diameter of the simulated microcalcifications for all evaluation variables.
CONCLUSION. Digital flat-panel mammography is superior to the analog screen-film method for the detection and morphologic characterization of microcalcifications larger than 200 µm in diameter when the display medium is a monitor.
Keywords: breast cancer • digital imaging • screen-film mammography • mammography • women's imaging • X-ray technology
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Digital flat-panel detector radiography with indirect conversion takes place in two steps. In the first step, a scintillator layer made of gadolinium oxysulfide or cesium iodide captures X-ray quanta and converts them to light. In the second step, the light is converted to electrical charges in photodiodes made of amorphous silicon. In digital flatpanel radiography using direct conversion, X-ray quanta are converted directly to electrical charges in a layer of amorphous selenium. Direct flat-panel detectors represent one of the more recent technologic approaches and, compared with the other methods, offer the advantages of a higher quantum efficiency because X-ray conversion takes place in just one step and, thanks to the chemical properties of amorphous selenium, larger-area imaging systems are possible.
The results of more recent experimental studies on digital mammography using stored-image and indirect flat-panel techniques have shown that advances in development have allowed the aforementioned technologies to be regarded as at least equivalent to analog screen-film mammography in terms of image quality and diagnostic accuracy [9-14]. A previous phantom study performed using direct flat-panel mammography suggested that direct flat-panel mammography was superior to the analog screen-film method, to digital stored-image radiography, and to indirect flat-panel detector radiography in detecting fibrous structures, although all methods detected microcalcifications and round densities equally well based on a comparably low number of interindividual observations [8]. For the study detailed in this article, we aimed to go beyond this by determining, on a larger sample and using an anthropomorphic model, the following: first, whether the detection rate and morphologic characterization of varyingly configured microcalcifications achievable with direct digital flat-panel mammography (test method) are comparable with the diagnostic profile of the analog screen-film technique (reference method); second, whether monitor interpretation is diagnostically equivalent or superior to film interpretation; and, third, whether within an interobserver comparison results are affected by the previous experience and expertise that the interpreting radiologists have in digital imaging techniques.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
The analog mammography unit (Mammomat 3, Siemens Medical Solutions) was equipped with a double-focus molybdenum-tungsten bimetal anode. The nominal focal spot size for survey images was 0.3 mm. The takes were done using a molybdenum filter (up to 40 kV, 0.03 mm). The focus-film distance was 60 cm. Min-R 2190 screens (Eastman Kodak) and single-coated MIN-R 2000 mammography films (Eastman Kodak) with a blue base support were used. The films had a format of 18 x 24 cm, equivalent to a nominal local resolution of 12 lp/mm, and were developed on a daylight load system (Miniloader 2 Plus, Eastman Kodak).
Phantom and Simulated Microcalcifications
On 65 transparent universal laser printer films with a format of 21 x
30 cm (type A P/N 003R96019, Xerox), a circle of 10 cm in diameter was drawn
and the circle was divided into four quadrants (I-IV). Each universal laser
printer film was then labeled for radiologic visualization by gluing a wire to
the film (Fig. 1A,
1B,
1C,
1D,
1E,
1F). After this step, round or
heterogeneous silicate particles (composition: SiO2 [65.0%],
Al2O3 [0.5-2.0%], Fe2O2 [0.15%],
MgO [2.5%], CaO [8.0%], Na2O [14.0%]; diameter classes: 1, 100-199
µm; 2, 200-399 µm; 3, 400-599 µm; 4, 600-800 µm) were scattered
within the quadrants by random distribution and were fixed with adhesive tape
(Fig. 2). Single quadrants were
left out of this particle assignment. The shape and size of the particles on
the films were recorded for each quadrant. The number of simulated
microcalcifications located in each quadrant was counted under a surgical
microscope (Wild M680, Leica) (Tables
1 and
2).
|
|
|
|
|
|
|
|
|
The 65 films (phantoms) with four quadrants each (phantom quadrants) were imaged with a digital mammography system (Selenia) and an analog mammography system (Mammomat 3). A 1.5-cm-thick piece of acrylic plastic sheet (Plexiglas, Degussa) made of polymethylmethacrylate (PMMA) and a similarly thick layer of ground meat placed on the film before X-ray exposures were taken served as scattering bodies. Because the two X-ray systems were located in one spatial unit, the digital and analog images could be produced in direct succession without having to change the scatter-body configuration.
Imaging Technique
All images were obtained in the craniocaudal plane. The same compression
setting was used for both methods. Preliminary tests for defining the optimal
exposure settings were conducted in which the kilovoltage settings were
systematically changed between 20 and 32 kV using the automatic exposure
control and a molybdenum filter on the analog mammography system. After these
tests, all analog and digital scans were taken at a tube voltage of 29 kV and
an exposure tube current of 18 mAs. The half-value layer (HVL), entrance skin
exposure, and average glandular dose were 0.40, 1.925 mGy, and 0.593 mGy for
the digital system; and 0.41, 1.875 mGy, and 0.588 mGy for the analog system,
respectively. The entrance skin exposure was 1.875 mGy for the Mammomat 3
system and 1.925 mGy for the Selenia system, and the average glandular dose
was 0.588 mGy for the Mammomat 3 system and 0.593 mGy for the Selenia system.
The doses were measured with a calibrated dosimeter (Solidose 300, RTI
Electronics AB). The manufacturer states that the measuring error of the
dosimeter is ± 5%. This test proved that any minor differences observed
between the analog and digital methods with regard to the exposure doses are
within the range of error. Thus, the two machines are comparable in terms of
overall exposure levels.
Image Display
The digital imaging data sets were printed on a dry laser printer (DryView
8610, Eastman Kodak) that operates according to the direct thermography
principle. The geometric print resolution was 655 dots per inch (DPI), the
effective pixel edge length was 36 x 39 µm in a 5,025 x 6,200
pixel matrix, and the contrast resolution had a 12-bit gray-scale depth (4,096
gray levels). An infrared-sensitive photothermographic film with a blue film
base coating on one side was used in a 18 x 24 cm format (DryView laser
imaging film, Eastman Kodak).
The digital system was connected to a viewing workstation (Selenia Softcopy Workstation, Lorad/Hologic) that was equipped with soft-copy reading software (MeVis BreastCare, MeVis Bremen). The graphics controllers (BarcoMed 5MP1H, Barco NV) operated with an internal 12-bit gray-scale depth equivalent to 4,096 levels of gray and a 10-bit digital-to-analog converter (DAC) that converts the signal for a display depth equivalent to 1,024 levels of gray. The two monitors (Barco, Barco NV) had an image field size of 30 x 40 cm in diameter with a line resolution of 2,048 x 2,560 pixels (effective pixel edge length, 147 x 156 µm). Whenever a scan with a format of 18 x 24 cm was completely displayed on the monitor, a reduction in the geometric resolution of approximately a factor of 0.8 resulted, which is equivalent to 7 lp/mm. For a 1:1 reproduction of the digital image data set, the image was cut off slightly on one edge, producing an effective display resolution of approximately 8 lp/mm. Through bicubic interpolation, the optical magnifying glass option allowed factor-2 zooming of the digital image data set.
Image Evaluation
A team of six board-certified radiologists rated the analog films, digital
scans, and monitor-displayed digital images in randomized order and
independently of each other. Each observer specializes in diagnostic radiology
and at the time of the study had 3-15 years' experience in interpreting
routine diagnostic analog mammograms and 1.5 years' experience interpreting
digital mammograms. Observers 1 and 2 each had completed many years of
scientific research in digital radiography. Observers 1 and 3 had the most
experience with routine diagnostic mammography.
The observers' notations were entered on structured questionnaires. The
observers were required to rank the number of simulated microcalcifications
visible as follows: class 0, no microcalcifications; 1, 1-4
microcalcifications; 2, 5-9 microcalcifications; 3, 10-19 microcalcifications;
4, 20-39 microcalcifications; or 5, > 39 microcalcifications. Size was
ranked according to class as follows: 0, not applicable (i.e., no
microcalcifications in the phantom quadrant); 1, 100-199 µm; 2, 200-399
µm; 3, 400-599 µm; or 4, 
600 µm. The size rankings were based on
the maximum size of simulated microcalcifications observed for each cluster.
Shape was defined as round, heterogeneous, or not applicable (NA) if no
microcalcifications were visible in the phantom quadrant. Data regarding
diameter and shape were related to the respectively largest silicate particle
visible in one quadrant. For orientation, each observer was given templates
produced by the analog and digital techniques and showing a perfect depiction
of all shapes and sizes without the scatter layer of ground meat
(Fig. 2). The images were
evaluated in darkened rooms. While reviewing images on viewboxes, the
observers were allowed to use illumination lamps and magnifying glasses.
Statistical Analysis
The notations made by the six observers for the ranked variables' number
and size and for the qualitative variable shape of detected
microcalcifications, separated according to the three display media, were
compared with the experimentally predefined constellation in the phantom
quadrants. In a first step, the correct interpretations of the six observers
were crosstabulated in tables to summarize the results. Kappa statistics were
used to indicate the degree of agreement between the gold standard and the
ratings. Second, the differences between the ranks of the observers'
interpretations and those of the predefined experimental constellations of
microcalcifications were used to describe the direction and magnitude of the
error (deviation from reference) for each quadrant for the first two
variables. Assuming that each of the four quadrants of a film counted as an
independent rating, each observer produced 4 x 65 interpretations, or a
total of 260 interpretations, per display medium that were available for
deviation analysis. With regard to the variables' number and size, the sign of
the error allowed us to distinguish between observers' underestimations
(negative sign) and overestimations (positive sign) with regard to the
predefined phantom constellation on the scale of the respective rating
entity.
For better graphic illustration of interobserver variability, the deviation distributions were presented in box plots (number and size) or contingency tables (shape) separated according to observers and display medium and independently of the simulated microcalcification size. To describe the effect that simulated microcalcification size had on the quality and accuracy of microcalcification detectability, the analyses of the variables' number and shape were divided into classes for size of the actual experimentally predefined constellations and were then repeated.
Using global comparisons of all three display media (Kruskal-Wallis test), the distribution of the differences in the observers' notations from the experimentally predefined actual constellation (gold standard) with respect to the display medium were analyzed on the basis of the aforementioned errors in direction and size and were supplemented by explorative paired comparisons between the different types of scans (Mann-Whitney tests). The p values were not corrected for multiple test scenarios. Therefore, the interpretation of any differences observed was only of an explorative nature.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
), the difference between perfect
agreement and agreement expected by chance for monitor interpretation ratings
was approximately 25.3% and has been accounted for in the agreement between
observers' interpretations and the gold standard; for digital images and
analog films, kappa was less than 1%. When all observers and quadrants were
considered, the diameter of the microcalcifications was correctly identified
in 35.6% (33.2-38.0%) using monitor interpretation ( = 0.193), in 19.0%
(17.1-21.0%) on digital images ( = -0.010), and in 21.0% (18.9-23.0%)
on analog films ( =0.013) (Table
4). Microcalcification shape was classified correctly using
monitor interpretation in 53.8% (51.4-56.3%), digital images in 28.2%
(26.0-30.4%), and analog films in 28.3% (26.0-30.5%); the measurement of
agreement calculated to kappa values of 0.338, -0.028, and -0.014,
respectively.
|
|
Figures 3A and 3B present the absolute magnitudes of the observer classification deviations (error) from the gold standard. Observers 1 and 2, the two radiologists most experienced in digital techniques, produced better results for monitor interpretations and digital image interpretations than the other four observers. The number of microcalcifications was underestimated more frequently than overestimated—a finding that particularly applied to the comparatively inexperienced observers 4-6. Regardless of the display medium, less experienced observers tended to underestimate microcalcification size, whereas the more experienced observers 1, 2, and 3 tended to overestimate microcalcification size.
|
|
600 µm). An advantage
of monitor interpretation was observed for the variable number of detectable
microcalcifications starting from a real diameter of 200 µm and for the
variable size starting from a real diameter 400 µm.
|
|
|
|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
The quality of a mammography method is determined by the capability of the imaging system to convert the attenuation of the X-ray beam in the breast as dimensionally accurately as possible into a perceivable image. The most important variables affecting image quality are sharpness, contrast, and noise. Local resolution and contrast resolution of an imaging system are interdependent: An improvement in local resolution causes the lower number of X-ray quanta received per pixel to lead to a poorer contrast-to-noise ratio and thus to poorer contrast resolution and vice versa. The mutual dependency of these two factors is expressed in terms of digital quantum efficiency (DQE). The DQE, which is 50-80% for digital mammography systems, indicates how many photons hitting the detection plane are used for imaging and is determined by comparing the local dose measured behind different filters with the exposure dose and includes every step in the process—from radiation exposure to when the detection system receives the radiation pattern. What are not accounted for are process steps downline from the signal acquisition, such as the analog-to-digital conversion of the signals, technical noise, image processing, and the performance of the image reproduction system (film, monitor).
Digital mammography must be measured against the gold standard of analog screen-film mammography, which has proved its merits in clinical routine over decades. Its advantages include low investment and operating costs, a high local resolution of up to 20 lp/mm for high-contrast objects, and simple presentation options offered by the viewbox [2, 17]. Disadvantages involve the fact that a trade-off has to be made between image contrast and radiation exposure. As a result of the sigmoidal gradation curve, under- or overexposure can produce inadequate image contrast. The DQE of screen-film systems is comparably low, between 20% and 40%, as a result of the low absorption of the X rays on intensifier films [2].
|
The present study showed that direct flat-panel mammography (film interpretation) is equivalent to analog screen-film mammography in the detection and morphologic characterization of microcalcifications, thus confirming the results of more recent phantom studies on digital storage-plate and indirect flat-panel mammography [9-14, 16, 17]. In addition to this finding, the following were shown for direct flat-panel mammography: First, monitor interpretation of digital image data sets was diagnostically superior to the analog and digital film interpretations for microcalcifications greater than 200 µm in diameter. Second, not only was the number of simulated microcalcifications as accurately estimated as with analog and digital methods of film interpretation, but the description of their shape and size was also comparably reliable. Third, a positive correlation existed between the diagnostic accuracy of imaging systems and size of simulated microcalcifications. Finally, methodologic expertise in digital radiography was more crucial for the correct perception of microcalcifications than experience in routine clinical diagnostics. The latter unquestionably plays a role in differential diagnostics with regard to the interpretation of once-detected microcalcifications.
Given the technologic characteristics discussed earlier in this article, the superiority of monitor interpretation versus analog film interpretation and digital film interpretation of large microcalcifications might be attributable to the type of digital image data processing and the light intensity of the monitors. The decline in accuracy of monitor-viewed interpretations when microcalcifications smaller than 200 µm in diameter were involved can be explained by the pixel size of the monitors, an observation also described by Rong et al. [16] in their experimental study of simulated microcalcifications with diameters between 112 and 160 µm imaged with digital screen-film, charge-coupled device (CCD), and indirect flat-panel systems. The superior diagnostic results for microcalcifications with a diameter greater than 200 µm achieved by monitor interpretation are clinically relevant insofar as this finding proves that radiologic malignancy criteria—that is, grouping trends and polymorphisms—can be interpreted more reliably when interpretations are performed using a high-resolution monitor rather than analog or digital films.
The power of the present study is partially limited by the lack of data on detector dose, considering that image quality and radiation dose are interdependent [18]. Even if the same anode filter combination is used, the same tube voltage and same milliampereseconds (mAs) settings will not yield a matching detector dose, particularly given the difference in focus-film distance of 5 cm, the different construction of the tubes, and the different grids.
The present research investigated the diagnostic performance profiles of only mammography systems used in the high-contrast range. A similarly designed study on the low-contrast range is currently in progress. More recently, standardized test phantoms have become available for digital mammography that allow quantification of the interplay of the characteristic physical variables of contrast, sharpness, and noise in a single test sequence. Test specimens, such as the Mammo-152 phantom (Wellhöfer) or the PAS1054 phantom (Artimis), provide robust tools for the testing and medicophysical expert evaluation of digital mammography systems. Nevertheless, they appear unsuitable for interobserver comparisons by virtue of the uniformity of their investigative items.
In conclusion, an anthropomorphic model showed that the detection rate and morphologic characterization of microcalcifications achieved with direct digital flat-panel mammography are comparably as effective as the diagnostic profile of the analog screen-film technique, that monitor interpretation is diagnostically superior to film interpretation for detecting large microcalcifications, and that the evaluating radiologists' experience with digital radiography affects diagnostic accuracy. Consequently, digital mammography in conjunction with monitor interpretation should be the preferred method for clinical diagnostics and breast cancer screening in the future.
Acknowledgments
We thank Claudia Morgenroth and Daniele Steinhaus-Wittig for their
evaluations of the analog films, digital scans, and monitor-displayed digital
images. We also thank Gerald Haupt (formerly of the Clinic and Policlinic for
Urology of the University of Cologne and currently of the Department of
Urology, St. Vincentius-Krankenhaus, Speyer) for counting the number of
simulated microcalcifications located in each quadrant.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
CiteULike 
Complore 
Connotea 
Del.icio.us 
Digg 
Reddit 
Technorati What's this?
This article has been cited by other articles:
|
|
 |
|
  K. B. Krug, H. Stutzer, R. Schroder, J. Boecker, J. Poggenborg, and K. Lackner Image Quality of Digital Direct Flat-Panel Mammography Versus an Analog Screen-Film Technique Using a Low-Contrast Phantom Am. J. Roentgenol., September 1, 2008; 191(3): W80 - W88. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
H. P. Forman Back to the Beginning Am. J. Roentgenol., February 1, 2007; 188(2): 295 - 296. [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
). Global analysis of
all 65 films showed that monitor reading produced more realistic estimation of
number, size, and shape of any microcalcifications present than when digital
film documentations or analog films were interpreted. Data for each display
medium are shown as follows: dark gray bars = analog films, striped bars =
digital films, and light gray bars = monitor-displayed images. Criterion is
number of detectable microcalcifications.
