Radiology
HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
QUICK SEARCH:   [advanced]


     

Published online before print March 13, 2007, 10.1148/radiol.2432061065
This Article
Right arrow Abstract Freely available
Right arrow Figures Only
Right arrow Full Text (PDF)
Right arrow All Versions of this Article:
2432061065v1
243/2/396    most recent
Right arrow Submit a response
Right arrow Alert me when this article is cited
Right arrow Alert me when eLetters are posted
Right arrow Alert me if a correction is posted
Right arrow Citation Map
Services
Right arrow Email this article to a friend
Right arrow Similar articles in this journal
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Right arrow reprints & permissions
Citing Articles
Right arrow Citing Articles via HighWire
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Samei, E.
Right arrow Articles by Delong, D. M.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Samei, E.
Right arrow Articles by Delong, D. M.
(Radiology 2007;243:396-404.)
© RSNA, 2007

Experimental Studies

Digital Mammography: Effects of Reduced Radiation Dose on Diagnostic Performance1

Ehsan Samei, PhD, Robert S. Saunders, Jr, PhD, Jay A. Baker, MD, and David M. Delong, PhD

1 From the Duke Advanced Imaging Laboratories, Department of Radiology (E.S., R.S.S., J.A.B.), and Division of Breast Imaging, Department of Radiology (J.A.B.), Duke University Medical Center, 2424 Erwin Rd, Suite 302, Durham, NC 27705; and Departments of Physics (E.S., R.S.S.), Biomedical Engineering (E.S.), and Biostatistics and Bioinformatics (D.M.D.), and Medical Physics Graduate Program (E.S.), Duke University, Durham, NC. Received June 19, 2006; revision requested August 21; revision received November 3; final version accepted December 11. Supported in part by grants from the National Institutes of Health (R21-CA95308) and U.S. Army Medical Research and Materiel Command (W81XWH-04-1-0323). Address correspondence to E.S. (e-mail: samei{at}deckard.duhs.duke.edu).


   ABSTRACT
TOP
 ABSTRACT
INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 APPENDIX
 IMPLICATIONS FOR PATIENT CARE
 References
 
Purpose: To experimentally determine the relationship between radiation dose and observer accuracy in the detection and discrimination of simulated lesions for digital mammography.

Materials and Methods: This HIPAA-compliant study received institutional review board approval; the informed consent requirement was waived. Three hundred normal craniocaudal images were selected from an existing database of digital mammograms. Simulated mammographic lesions that mimicked benign and malignant masses and clusters of microcalcifications (3.3–7.4 cm in diameter) were then superimposed on images. Images were rendered without and with added radiographic noise to simulate effects of reducing the radiation dose to one half and one quarter of the clinical dose. Images were read by five experienced breast imaging radiologists. Results were analyzed to determine effects of reduced dose on overall interpretation accuracy, detection of microcalcifications and masses, discrimination between benign and malignant masses, and interpretation time.

Results: Overall accuracy decreased from 0.83 with full dose to 0.78 and 0.62 with half and quarter doses, respectively. The decrease associated with transition from full dose to quarter dose was significant (P < .01), primarily because of an effect on detection of microcalcifications (P < .01) and discrimination of masses (P < .05). The level of dose reduction did not significantly affect detection of malignant masses (P > .5). However, reduced dose resulted in an increased mean interpretation time per image by 28% (P < .0001).

Conclusion: These findings suggest that dose reduction in digital mammography has a measurable but modest effect on diagnostic accuracy. The small magnitude of the effect in response to the drastic reduction of dose suggests potential for modest dose reductions in digital mammography.

© RSNA, 2007


   INTRODUCTION
TOP
 ABSTRACT
 INTRODUCTION
MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 APPENDIX
 IMPLICATIONS FOR PATIENT CARE
 References
 
Digital mammography differs from screen-film mammography in several important ways (1–4). The foremost difference is that digital mammography images are captured with a digital sensor (5). In conventional mammography the analog film serves as both the detector and the display medium, whereas in digital mammography the use of a digital sensor enables dissociation between the detection and display functions. An important consequence of this dissociation is the independence of display contrast from subject contrast so that the image quality of a digital mammogram is not limited by display contrast, which can be manipulated after images have been acquired, but rather by image noise dictated by the number of photons used to form the image.

The shift from contrast-limited imaging to noise-limited imaging has a fundamental implication for radiation dose in digital mammography. In the clinical implementation of digital mammography, it is imperative to use the appropriate level of radiation (no more and no less) for the diagnostic task at hand. On one hand, a higher radiation dose will lower the noise level but may impart radiation doses to the patient that are higher than necessary (6). On the other hand, a lower radiation dose will lower the signal-to-noise ratio of the image and consequently negatively affect the presentation of information and thus the diagnosis. The proper radiation dose for mammography should be dictated by the amount of radiation required to achieve an adequate signal-to-noise ratio to depict image details needed to render an accurate diagnosis.

The radiation dose used for current clinical digital mammographic applications has generally been the same as that used with equivalent analog systems. This may be due in part to following prior conventions with analog systems, as well as the fact that the relationship between noise and diagnostic accuracy has not been well established for digital mammography. Analog radiation doses have been used despite the fact that (a) digital systems are not limited by the contrast-limited constraints of analog systems and (b) the improved detective quantum efficiency of most digital systems offers the potential for dose reduction (5,7). Thus, the purpose of our study was to experimentally determine the relationship between radiation dose and observer accuracy in the detection and discrimination of simulated lesions for digital mammography.


   MATERIALS AND METHODS
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
RESULTS
 DISCUSSION
 APPENDIX
 IMPLICATIONS FOR PATIENT CARE
 References
 
Our study was compliant with the Health Insurance Portability and Accountability Act and was approved by the institutional review board at our institution. The informed consent requirement was waived.

Image Selection
Three hundred normal craniocaudal images were randomly selected from an existing database of digital mammograms. All images were originally acquired by using a commercially available indirect flat-panel mammographic system (Senographe; GE Medical Solutions, Waukesha, Wis) with a tube voltage of 25–30 kVp (average, 27.6 kVp), a molybdenum anode, and molybdenum or rhodium filtration. The selected images, which were considered normal according to the radiologists' interpretation of the images in the routine clinical procedure, reflected a compressed breast thickness of 2.7–7.4 cm (average, 5.1 cm) and a full range of breast densities from fatty to extremely dense. As a requirement of subsequent steps of the study, the images were used in their native raw format, with no additional postprocessing except that for signal gain and bad pixel corrections implemented by the system.

Simulation of Mammographic Lesions
Simulated mammographic lesions, which served as the reference standard, were placed on the selected images by using a lesion simulation program (8). Three common categories of lesions were simulated: (a) benign-appearing masses (modeled after oval circumscribed and oval obscured lesions), (b) malignant-appearing masses (modeled after irregular ill-defined and irregular spiculated lesions), and (c) microcalcifications (modeled after clustered pleomorphic and fine linear branching lesions) (9). All simulated lesions had contrast, contrast profile, shape, border characteristics, and distributions similar to those of the lesion type being simulated. In a prior study, researchers confirmed that radiologists could not differentiate these simulated lesions from actual lesions (8).

The original set of 300 images was divided into two groups of 150 images. One group was used to generate 150 images with benign masses and 150 images with malignant masses; the other group was similarly used to generate 150 images with microcalcifications and 150 images without lesions. This scheme was designed to enable matching backgrounds for mass discrimination and microcalcification detection tasks to improve the associated statistics while minimizing the number of times a particular background was viewed by the observers.

The sizes of the simulated lesions were determined with pilot experiments designed to target detection accuracy in the neighborhood of 80% for the experimental condition. The diameters of the simulated masses ranged from 3.3 to 4.1 mm (average, 3.7 mm). Individual calcifications had mean major and minor axis lengths of 0.37 and 0.25 mm, respectively. The diameters of pleomorphic lesions ranged from 4.0 to 7.0 mm. Distributions for fine linear branching lesions had lines of microcalcifications with lengths of 4.0–9.0 mm. The overall contrast magnitudes of simulated lesions were determined on the basis of the characteristics of image formation and real lesions (10) and the level of scattered radiation on each mammogram (11). The simulated lesions were added to the mammographic images in a logarithmic scale to result in contrast magnitudes that would be independent of the image background at the location of the insertion (12).

Noise Addition
Current flat-panel mammographic systems are limited by quantum noise (13,14); therefore, the main consequence of dose reduction is a proportionate increase in the level of quantum noise on the image. Thus, to create images with an image noise similar to that caused by a reduction in radiation dose, a noise modification routine was used to add radiographic noise to the images. To do so, each group of 150 images described previously was divided into three subgroups corresponding to full dose (no added noise), half dose, and quarter dose of the original clinical dose conditions.

With use of the noise addition routine, which was previously described in detail (15), one was capable of adding noise according to an a priori magnitude and texture (16–19). The desired radiographic noise magnitude was ascertained with the aid of the measured relationship between noise variance and exposure for the imaging system used (Appendix). At each dose level, the noise magnitude was adjusted on the basis of the pixel value to properly account for the effects of breast attenuation on image noise (15). The noise texture was similarly based on the measured noise power spectrum for the mammographic system (13–15).

Image Postprocessing
The lesion and noise simulation processes described previously were performed for images obtained in the raw format to properly emulate the subject contrast of real mammographic lesions and the noise properties of low-dose mammograms. However, images obtained in the raw format were not suitable for interpretation by radiologists. Since the proprietary algorithms of GE Medical Solutions could not be applied in the postprocessing environment, two generic postprocessing steps were applied to the images to make the image appearance representative of images in clinical practice. In the first step, unsharp masking and contrast equalization techniques (20) were used to enhance the depiction of smaller structures and equalize broad signal intensity variations between the center and borders of the breast. The associated parameters for this operation were determined subjectively by means of visual analysis of processed mammograms (J.A.B., 7 years of experience as an attending mammographer and 5000 cases reviewed per year). Identical processing was then applied to all images.

In the second processing step, we established window and level settings appropriate for optimum viewing of each image. The window and level settings were determined by means of histogram analysis of full mammograms, with the goal of establishing clinically representative contrast levels in the central breast while maintaining adequate contrast along the breast boundary. A sigmoid function was then fitted to all window and level settings to provide a smooth transition at the extremes of the gray-scale range. A breast imaging radiologist (J.A.B., 7 years of experience interpreting mammograms) who did not participate in the subsequent observer study reviewed all images after window and level processing to ensure the image appearance matched that seen in common clinical practice.

Observer Performance Experiment
An observer performance experiment was conducted to assess the effect of reduced dose on lesion detection and discrimination. A 5.12 x 5.12-cm (512 x 512-pixel) region centered at the location of the lesion was extracted from each image (Fig 1). Five breast imaging radiologists with 3–17 years (average, 9.8 years) of experience reading 4000–15 000 screening mammograms per year (average, 8000 mammograms per year) used a location-known-exactly experimental paradigm to score all images. A custom graphic user interface allowed the observers to indicate whether an image appeared to contain a benign or malignant mass, a microcalcification cluster, or no lesion. Observers chose only one answer for each image, and a rating scale was not used. The interface encouraged observers to indicate their choices with the keyboard and thus substantially shortened the time required for image interpretation. In addition, a modified version of the graphic user interface was used for a supplemental experiment in which the observers were able to score images only in terms of a specific diagnostic task (ie, whether a microcalcification was present).


Figure 1
View larger version (187K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 1: Mammograms obtained with full dose (left column), half dose (center column), and quarter dose (right column) in the observer experiment show microcalcification distributions (arrows in top row) and malignant (arrows in middle row) and benign (arrows in bottom row) masses.

 
All images were viewed on a 5-megapixel liquid crystal monitor (Nova V; National Display Systems, San Jose, Calif) equipped with a 10-bit display controller (MD5mp; RealVision, Kanagawa, Japan). The device was calibrated to the Digital Imaging and Communications in Medicine gray-scale standard display function (21), within a luminance range of 0.52–371.00 candelas per square meter (22). All readings were performed in our display laboratory with controlled low ambient lighting.

Before the actual reading sessions, each observer read a different set of 100 images and received immediate feedback to make him or her familiar with the rating interface and appearance of lesions. Each observer then scored a fixed number of images in two sessions, with 5-minute breaks between sessions to reduce observer fatigue. Viewing sessions lasted 20–30 minutes. Two observers (11 and 6 years of experience, respectively) performed additional repeat reading of images on the modified graphic user interface with reduced scoring functionality, which provided scoring options for only the task at hand, to obtain necessary data used to assess the magnitude of potential bias in our categorical scoring scheme.

Images of different dose levels were displayed in a random order and in one of six random orientations (four orthogonal rotations with horizontal and vertical flips) to minimize the effects of reading order and memory.

Statistical Analysis
Observer results were analyzed to assess the effects of varying dose and noise levels on overall accuracy. Variances were estimated by applying the bootstrap technique to mammograms, and t tests were used to compute the statistical significance of estimated differences with use of the Bonferroni correction to preserve type I error (23,24). The outcome analyzed was the overall accuracy across all diagnostic tasks, as presented in two-dimensional contingency tables. Overall accuracy was computed as the percentage of images correctly rated by each observer. A combined overall accuracy was computed as the average over all observers.

While overall accuracy analysis combined data for all tasks into one number, the data analysis involved further examination of the statistical effect of reduced dose on four specific clinical tasks (ie, detection of microcalcifications, detection of benign masses, detection of malignant masses, and discrimination between benign and malignant masses). For each task, a task-specific metric of accuracy was computed as the average of sensitivity and specificity (Fig 2). This metric was approximately equal to the area under the binormal receiver operating characteristic curve (25). The task performance data were averaged across observers, and these results were compared for different dose levels by using a t test for statistical significance with Bonferroni correction (23). To assess the presence of any potential bias associated with our scoring method, scores from the repeat single-task study were used to adjust the multiple-task data. Data on the standard and adjusted performances of microcalcification detection and mass discrimination were compared to test for any potential bias.


Figure 2
View larger version (15K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 2: Contingency table used to deduce performance results. This table was used to detect malignant masses. B = benign mass, C = microcalcification, M = malignant mass, N = normal.

 
Data analysis included the reading time associated with each dose subgroup of images for each observer and the average across observers. Standard errors were calculated by using bootstrap analysis. A reduced signal-to-noise ratio might have a detrimental effect on observer confidence, which might be reflected in terms of reading time; therefore, data were also analyzed by using survival curves and proportional hazard analysis to determine whether the reduced-dose images required a longer time for interpretation (26,27).

A P value of less than .05 was considered to indicate a statistically significant difference. All statistical analyses were performed with Matlab, version 7, release 14 (Mathworks, Natick, Mass), and JMP 6 (SAS Institute, Cary, NC) software.


   RESULTS
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
DISCUSSION
 APPENDIX
 IMPLICATIONS FOR PATIENT CARE
 References
 
Overall Accuracy
There was a reduction in overall accuracy with reduced radiation dose. While accuracies of individual observers varied, they all exhibited similar trends with a reduced radiation dose (Figs 3, 4). Reductions in accuracy were significant for the transition from a full dose to a quarter dose (P < .05) and notable but not significant for the transition from a full dose to a half dose (Table).


Figure 3A
View larger version (24K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 3a: Contingency tables at (a) full dose, (b) half dose, and (c) quarter dose averaged across observers indicate the fraction of which the observers scored the images of a given class. The scale to the right of each table shows the fraction of images from a given truth category that were rated in a given category. B = benign mass, C = microcalcification, M = malignant mass, N = normal.

 

Figure 3B
View larger version (22K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 3b: Contingency tables at (a) full dose, (b) half dose, and (c) quarter dose averaged across observers indicate the fraction of which the observers scored the images of a given class. The scale to the right of each table shows the fraction of images from a given truth category that were rated in a given category. B = benign mass, C = microcalcification, M = malignant mass, N = normal.

 

Figure 3C
View larger version (19K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 3c: Contingency tables at (a) full dose, (b) half dose, and (c) quarter dose averaged across observers indicate the fraction of which the observers scored the images of a given class. The scale to the right of each table shows the fraction of images from a given truth category that were rated in a given category. B = benign mass, C = microcalcification, M = malignant mass, N = normal.

 

Figure 4
View larger version (31K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 4: Bar graph shows variation in the overall accuracy, representing an average over all diagnostic tasks involved, as a function of dose level for individual observers and the average across observers. The variance for each observer was calculated by using bootstrap analysis, with error bars representing one standard error. This graph shows that for each observer and across all observers, overall accuracy was reduced as radiation dose was decreased.

 

View this table:
[in this window]
[in a new window]

  		Summary Statistics of the Results Indicating the Reduction in Accuracy in Average Observer Performance

 
Task-specific Accuracy
For task-specific average accuracy, there was a clear decrease in the detection of microcalcifications and the discrimination between masses with reduced dose at both dose reduction levels. This decrease was statistically significant for the transition from full dose to quarter dose for calcification detection and mass discrimination. However, detection of malignant masses did not appear to be affected much by dose reduction, and the detection of benign masses was affected only when the dose was reduced to a quarter of the normal level (Table, Fig 5).


Figure 5
View larger version (31K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 5: Bar graph shows the effect of dose level on the detection of microcalcifications and the detection and discrimination of malignant and benign masses. Data correspond to averages for all observers, with standard errors (error bars) calculated in a fashion similar to the manner in which data were calculated in Figure 4. With the reduction from a full dose to a quarter dose, there was a significant decrease in microcalcification detection (P < .01) and mass discrimination (P < .05). Detection of malignant masses was reduced only with a quarter dose, and detection of benign masses changed little when radiation dose was reduced.

 
Effect of Bias
We used a categorical scoring scheme in the observer performance experiment. To assess how this scheme might have biased our task-based results, we also repeated part of the experiment and gave observers only binary choices (lesion present vs lesion absent or benign lesion vs malignant lesion). The categorical scoring scheme introduced minimal or no bias, with the results of the two schemes being essentially the same (P > .5) (Fig 6).


Figure 6
View larger version (45K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 6: Bar graph shows the potential effect of bias in the detection of microcalcifications. Bias could be potentially due to the multiplicity of observer grading tasks; therefore, this graph shows the raw detection results and the detection results adjusted to remove any potential bias. Error bars represent one standard error. Nearly identical results were found for the two scoring schemes (ie, categorical and two task) and thus showed that categorical scoring introduced minimal or no bias.

 
Timing Performance
Each observer took approximately 2.5 hours to read the entire set of images, including training and breaks. The overall timing results (Fig 7) indicated a discernable effect of dose reduction on interpretation time. The median interpretation time per image increased from 2.38 seconds ± 0.07 (standard error) for images obtained with a full dose to 2.42 seconds ± 0.09 and 3.04 seconds ± 0.09 for images obtained with half and quarter doses, respectively. The differences were found to be statistically significant (P < .0001). The individual observer results enabled us to confirm the same behavior, with the average timing performance of the observers appearing to be grossly correlated with reader experience and current reading volume.


Figure 7
View larger version (17K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 7: Graph shows the number of unrated images at a given time. The three lines show the reading times for images with signal-to-noise ratios that reflect full, half, and quarter doses. A significant relationship was found between radiation dose and observer interpretation time.

 

   DISCUSSION
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
APPENDIX
 IMPLICATIONS FOR PATIENT CARE
 References
 
The radiation dose associated with mammographic screening procedures has been a common concern to the radiology community (28–35). In fact, it was in response to such concerns that the U.S. federal government regulated mammographic examinations with the Mammography Quality Standards Act of 1992 (36–38). There is an opportunity to potentially reduce the mammographic dose in the transition from analog to digital mammography. Such reductions have been explored in a few studies (31,39–41). Multiple reports have noted the limitations that anatomic noise imposes on mammographic tasks (42–44); these findings further support such dose reductions. However, concerns about the potential loss of image quality and the resultant effect on diagnostic accuracy have prevented any notable dose reduction in clinical procedures, with clinical implementations still aiming to mostly maintain the dose used with digital systems at a level similar to that used with analog systems.

We found that decreasing the dose in digital mammography by as much as half had a minimal effect on the detection of malignant masses but a notable effect on the detection of microcalcifications, the discrimination between benign and malignant masses, and the interpretation time. These findings imply that the influence of reduced dose and the associated increased noise is mostly on the perception of the high-frequency components of the lesion signal, as these components represent the defining features of microcalcifications and the distinguishing features that indicate the differences between malignant and benign masses.

The most important clinical implication of our findings is that the mammographic dose, even for digital mammography with a potentially higher detective quantum efficiency, has an effect on diagnostic accuracy; thus, proper setup and control of radiation exposure are essential requirements for digital mammographic procedures. However, the small magnitude of the effect on diagnostic accuracy in relation to the notable reduction in dose suggests that dose may potentially be decreased with limited effect on clinical utility. This potential is perhaps better appreciated for certain uses, such as extra views for images to confirm placement of clips or wires during or after biopsies (45). However, our results imply that there might be potential for modest dose reduction in screening applications as well. This implication should not be confirmed until future studies in which accuracy is evaluated at multiple incremental dose levels have been performed.

The results of our study are consistent with the findings of previous research related to dose reduction in mammography. Dance et al (40) and Huda et al (39) used physical measurements to assess the effects of mammographic beam quality and dose reduction on the detection of simple simulated lesions. They found that dose could be reduced by using optimum beam qualities while maintaining a constant signal difference–to-noise ratio. In two additional studies that were based more on clinical practice, researchers examined whether reduced dose affected lesion detection by radiologists: Obenauer et al (41) used an indirect flat-panel detector to obtain images in an anthropomorphic breast phantom that contained simulated calcifications. Similarly, Hemdal et al (31) conducted a human performance experiment with 28 real mammograms acquired with full and half doses. In both studies, researchers found there was potential for a substantial dose reduction in digital mammography. These prior studies were hindered by either the physical measures of image quality (ie, signal difference–to-noise ratio) or the limited number of clinical tasks and lesion types examined. In contrast, in our study we explored a larger number of clinical tasks and included a greater number of images and lesions, which allowed our results to be more generalizable to a larger patient population.

Most diagnostic observer performance experiments currently involve rating images for the presence of one type of abnormality and assigning each image a grade ranging from definitely absent to definitely present, with multiple grades in between. The number of gradations range from four to 100 (46–48). This approach is essential for receiver operating characteristic curve analysis (49,50), which is the current de facto standard for the evaluation of diagnostic systems. However, this approach falls short of reflecting many of the diagnostic tasks performed in the clinical setting, where examiners need to make binary decisions about the presence of a lesion or the need for biopsy for multiple abnormal findings that might appear on an image. In our study, we asked observers to rate images for the presence of different types of abnormalities without confidence ratings. This categorical approach closely emulated the clinical paradigm. It also substantially shortened the time needed to rate an individual image.

While the previously mentioned approach has a strong appeal in terms of clinical relevance, it might be prone to potential bias problems. Bias might be introduced if the assessment of a given diagnosis is affected by the inclusion of a rating that is not relevant to the task at hand. For example, if an observer is assessing the discrimination of benign and malignant masses on an image, he or she might change his or her natural score if a scoring option is provided for the presence of microcalcifications (ie, an unrelated option). If the score is changed more or less often for malignant mass images than for benign mass images, it will create bias in the results. We recognized that this might have an effect on our results; therefore, we performed a supplemental study in which two observers were provided with only those scoring options that were relevant to the task at hand. A comparison of the results with and without the multiple scoring option indicated that a potential effect of bias was nonexistent at best and minimal at worst. The findings encourage the use of categorical methods in future observer performance experiments.

Our study had several limitations. First, while our results indicate the relative effect of dose reduction on various diagnostic tasks in digital mammography, the direct relationship between breast dose and diagnosis could only be inferred, as the dose reduction was applied only in a relative sense: For a given image acquired with a specific radiographic technique, breast dose was directly related to exposure and noise; thus, a relative reduction in dose can be achieved with a linear reduction in exposure and a corresponding increase in radiographic noise. However, the relationship between exposure, dose, and noise is dependent on the tube voltage, beam filtration, and breast composition and thickness, which vary from image to image. Thus, while our results can tell us what would happen if a lower than standard tube current, exposure time, or exposure setting were used in a given breast, they cannot tell us the specific quantitative relationship between glandular breast dose and accuracy. Second, in our study, we investigated the effect of dose reduction by using a signal-known-exactly paradigm in which the observers knew the approximate location of a lesion. While this strategy eliminated visual searching, it was implemented to keep other sources of variability under control. However, if we had used full images and incorporated visual searching, we might have possibly observed bigger differences as a function of dose, a prospect that cannot be substantiated with our results. Finally, our study was based on simulated lesions and dose levels. While the simulations were realistic, there are always differences between real and simulated situations, which might have a bearing on the findings.

In summary, the findings of our experimental study suggest that a reduction in radiation dose by as much as half can have a measurable but modest effect on diagnostic accuracy in digital mammography, particularly in the detection of microcalcifications and the discrimination of malignant and benign masses. Dose reduction also appears to lengthen interpretation time.

Practical application: The results suggest that, given the small magnitude of the effect on accuracy in response to the drastic reduction in dose, there may be potential for modest dose reductions in digital mammography. While this potential awaits confirmation in a follow-up clinical trial, one should pay careful attention to radiation dose and associated image quality when setting up and operating digital mammography units.


   APPENDIX
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 APPENDIX
IMPLICATIONS FOR PATIENT CARE
 References
 
To accurately simulate dose reduction on a mammogram, the magnitude of the added noise needs to correspond with a proportionate exposure reduction. In our study, we maintained the mean pixel value of the images but altered the image signal-to-noise ratio to simulate the effects of reduced exposure. The actual signal-to-noise ratio (SNRA) of an image was related to the detective quantum efficiency (DQE) and the ideal signal-to-noise ratio (SNRI), as follows:

Formula A1(A1)
where q represents the ideal signal-to-noise ratio squared per unit exposure, f stands for the spatial frequency, and {xi} represents the exposure (51). We used measured values of the detective quantum efficiency for the mammographic detector (13,14) and estimated values for the ideal signal-to-noise ratio, which were calculated by using an x-ray modeling program (xSpect; Henry Ford Health System, Detroit, Mich) (10). This equation was solved to determine the actual signal-to-noise ratio at full (SNRAF) and reduced (SNRAR) exposure levels. The scalar magnitude of the noise was then determined from the computed signal-to-noise ratio values with the following equation:

Formula A2(A2)
where {sigma}AN indicates the standard error of the additional noise and {xi}in is the exposure associated with the input image being modified.


   IMPLICATIONS FOR PATIENT CARE
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 APPENDIX
 IMPLICATIONS FOR PATIENT CARE
References
 


   ACKNOWLEDGMENTS
 
The authors thank Cherie Kuzmiak, DO, Joseph Lo, PhD, Dag Pavic, MD, Etta Pisano, MD, Mary Scott Soo, MD, Georgia Tourassi, PhD, and Ruth Walsh, MD, for participating in the observer studies and Andrew Karellas, PhD, and Sankar Suryanarayanan, MS, for providing the mammographic background images used in the study.


   FOOTNOTES
 
Authors stated no financial relationship to disclose.

Author contributions: Guarantor of integrity of entire study, E.S.; study concepts/study design or data acquisition or data analysis/interpretation, all authors; manuscript drafting or manuscript revision for important intellectual content, all authors; manuscript final version approval, all authors; literature research, E.S., R.S.S., J.A.B.; experimental studies, E.S., R.S.S., J.A.B.; statistical analysis, E.S., R.S.S., D.M.D.; and manuscript editing, all authors


   References
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 APPENDIX
 IMPLICATIONS FOR PATIENT CARE
 References
 

  1. Fischer U, Hermann KP, Baum F. Digital mammography: current state and future aspects. Eur Radiol 2006;16:38–44.[CrossRef][Medline]
  2. Conant EF, Maidment AD. Update on digital mammography. Breast Dis 2001;13:109–124.[Medline]
  3. James JJ. The current status of digital mammography. Clin Radiol 2004;59:1–10.[CrossRef][Medline]
  4. Pisano ED, Gatsonis C, Hendrick E, et al. Diagnostic performance of digital versus film mammography for breast-cancer screening. N Engl J Med 2005;353:1773–1783.[Abstract/Free Full Text]
  5. Yaffe MJ, Mainprize JG. Detectors for digital mammography. Technol Cancer Res Treat 2004;3:309–324.[Medline]
  6. National Academies of Science. BEIR: health risks from exposures to low levels of ionizing radiation. Washington, DC: National Academies of Science, 2005.
  7. Haus AG, Yaffe MJ. Screen-film and digital mammography: image quality and radiation dose considerations. Radiol Clin North Am 2000;38:871–898.[CrossRef][Medline]
  8. Saunders R, Samei E, Baker J. Simulation of mammographic lesions. Acad Radiol 2006;13:860–870.[CrossRef][Medline]
  9. D'Orsi CJ, Bassett LW, Feig SA, et al. Illustrated breast imaging reporting and data system (BI-RADS). 3rd ed. Reston, Va: American College of Radiology, 1998.
  10. Samei E, Flynn MJ. An experimental comparison of detector performance for direct and indirect digital radiography systems. Med Phys 2003;30:608–622.[CrossRef][Medline]
  11. Boone JM, Lindfors KK, Cooper VN 3rd, Seibert JA. Scatter/primary in mammography: comprehensive results. Med Phys 2000;27:2408–2416.[CrossRef][Medline]
  12. Samei E, Flynn MJ, Eyler WR. Detection of subtle lung nodules: relative influence of quantum and anatomic noise on chest radiographs. Radiology 1999;213:727–734.[Abstract/Free Full Text]
  13. Suryanarayanan S, Karellas A, Vedantham S. Physical characteristics of a full-field digital mammography system. Nucl Instrum Methods Phys Res 2004;533:560–570.[CrossRef]
  14. Vedantham S, Karellas A, Suryanarayanan S, et al. Full breast digital mammography with an amorphous silicon-based flat panel detector: physical characteristics of a clinical prototype. Med Phys 2000;27:558–567.[CrossRef][Medline]
  15. Saunders RS, Samei E. A method for modifying the image quality parameters of digital radiographic images. Med Phys 2003;30:3006–3017.[CrossRef][Medline]
  16. Giger ML, Doi K, Fujita H. Investigation of basic imaging properties in digital radiography. VII. Noise Wiener spectra of II-TV digital imaging systems. Med Phys 1986;13:131–138.
  17. Giger ML, Doi K, Fujita H. Analysis of noise Wiener spectra in digital II/TV imaging systems. Med Phys 1984;11:385.
  18. Giger ML, Doi K, Metz CE. Investigation of basic imaging properties in digital radiography. II. Noise Wiener spectrum. Med Phys 1984;11:797–805.
  19. Saunders RS Jr, Samei E, Jesneck JL, Lo JY. Physical characterization of a prototype selenium-based full field digital mammography detector. Med Phys 2005;32:588–599.[CrossRef][Medline]
  20. Stahl M, Aach T, Dippel S. Digital radiography enhancement by nonlinear multiscale processing. Med Phys 2000;27:56–65.[CrossRef][Medline]
  21. Digital imaging and communications in medicine (DICOM) Part 14: grayscale standard display function—PS 3.14. Rosslyn, Va: National Electrical Manufacturers Association, 2000.
  22. Samei E, Badano A, Chakraborty D, et al. Assessment of display performance for medical imaging systems: executive summary of AAPM TG18 report. Med Phys 2005;32:1205–1225.[CrossRef][Medline]
  23. Bender R, Lange S. Adjusting for multiple testing: when and how? J Clin Epidemiol 2001;54:343.
  24. Casella G, Berger RL. Statistical inference. Pacific Grove, Calif: Thomson Learning, 2002.
  25. Swets JA, Pickett RM. Evaluation of diagnostic systems: methods from signal detection theory. New York, NY: Academic Press, 1982.
  26. Cox DR. Regression models and life-tables. J R Stat Soc 1972;34:187–220.
  27. Lawless JF. Statistical models and methods for lifetime data. New York, NY: Wiley, 1982.
  28. Parker MS, Hui FK, Camacho MA, Chung JK, Broga DW, Sethi NN. Female breast radiation exposure during CT pulmonary angiography. AJR Am J Roentgenol 2005;185:1228–1233.[Abstract/Free Full Text]
  29. Morin Doody M, Lonstein JE, Stovall M, Hacker DG, Luckyanov N, Land CE. Breast cancer mortality after diagnostic radiography: findings from the U.S. Scoliosis Cohort Study. Spine 2000;25:2052–2063.[CrossRef][Medline]
  30. Sigurdson AJ, Doody MM, Rao RS, et al. Cancer incidence in the US radiologic technologists health study, 1983–1998. Cancer 2003;97:3080–3089.[CrossRef][Medline]
  31. Hemdal B, Andersson I, Grahn A, et al. Can the average glandular dose in routine digital mammography screening be reduced? a pilot study using revised image quality criteria. Radiat Prot Dosimetry 2005;114:383–388.[Abstract/Free Full Text]
  32. Ramos M, Ferrer S, Villaescusa JI, Verdu G, Salas MD, Cuevas MD. Use of risk projection models to estimate mortality and incidence from radiation-induced breast cancer in screening programs. Phys Med Biol 2005;50:505–520.[CrossRef][Medline]
  33. Law J, Faulkner K. Two-view screening and extending the age range: the balance of benefit and risk. Br J Radiol 2002;75:889–894.[Abstract/Free Full Text]
  34. Law J, Faulkner K. Concerning the relationship between benefit and radiation risk, and cancers detected and induced, in a breast screening programme. Br J Radiol 2002;75:678–684.[Abstract/Free Full Text]
  35. Huda W, Sourkes AM, Bews JA, Kowaluk R. Radiation doses due to breast imaging in Manitoba: 1978–1988. Radiology 1990;177:813–816.[Abstract/Free Full Text]
  36. Suleiman OH, Spelic DC, McCrohan JL, Symonds GR, Houn F. Mammography in the 1990s: the United States and Canada. Radiology 1999;210:345–351.[Abstract/Free Full Text]
  37. Pisano ED, Schell M, Rollins J, et al. Has the Mammography Quality Standards Act affected the mammography quality in North Carolina? AJR Am J Roentgenol 2000;174:1089–1091.
  38. Kneece J. Breast imaging: why MQSA (Mammography Quality Standards Act). Adm Radiol 1994;13:33–34.[Medline]
  39. Huda W, Sajewicz AM, Ogden KM, Dance DR. Experimental investigation of the dose and image quality characteristics of a digital mammography imaging system. Med Phys 2003;30:442–448.[CrossRef][Medline]
  40. Dance DR, Thilander AK, Sandborg M, Skinner CL, Castellano IA, Carlsson GA. Influence of anode/filter material and tube potential on contrast, signal-to-noise ratio and average absorbed dose in mammography: a Monte Carlo study. Br J Radiol 2000;73:1056–1067.[Abstract]
  41. Obenauer S, Hermann KP, Grabbe E. Dose reduction in full-field digital mammography: an anthropomorphic breast phantom study. Br J Radiol 2003;76:478–482.[Abstract/Free Full Text]
  42. Bochud FO, Valley JF, Verdun FR, Hessler C, Schnyder P. Estimation of the noisy component of anatomical backgrounds. Med Phys 1999;26:1365–1370.[CrossRef][Medline]
  43. Bochud FO, Verdun FR, Valley JF, Hessler C, Moeckli R. Importance of anatomical noise in mammography. In: Kundel HL, ed. Proceedings of SPIE: medical imaging 1997—image perception. Vol 3036. Bellingham, Wash: International Society for Optical Engineering, 1997; 74.
  44. Burgess AE, Jacobson FL, Judy PF. Human observer detection experiments with mammograms and power-law noise. Med Phys 2001;28:419–437.[CrossRef][Medline]
  45. Riedl CC, Jaromi S, Floery D, Pfarl G, Fuchsjaeger MH, Helbich TH. Potential of dose reduction after marker placement with full-field digital mammography. Invest Radiol 2005;40:343–348.[CrossRef][Medline]
  46. Metz CE. Fundamental ROC analysis. In: Beutel J, Kundel HL, Van Metter RL, eds. Handbook of medical imaging. Bellingham, Wash: International Society for Optical Engineering, 2000; 751–770.
  47. Chesters MS. Human visual perception and ROC methodology in medical imaging. Phys Med Biol 1992;37:1433–1476.[CrossRef][Medline]
  48. Turner SR, Samei E, Hertzberg BS, et al. Sonography of fetal choroid plexus cysts: detection depends on cyst size and gestational age. J Ultrasound Med 2003;22:1219–1227.[Abstract/Free Full Text]
  49. Metz CE. Some practical issues of experimental design and data analysis in radiological ROC studies. Invest Radiol 1989;24:234–245.[Medline]
  50. Metz CE. ROC methodology in radiologic imaging. Invest Radiol 1986;21:720–733.[Medline]
  51. Dobbins J. Image quality metrics for digital systems. In: Beutel J, Kundel HL, Van Metter RL, eds. Handbook of medical imaging. Bellingham, Wash: International Society for Optical Engineering, 2000; 163–222.



This article has been cited by other articles:


Home page
 Radiat Prot DosimetryHome page
C. Van Ongeval, A. Van Steen, C. Geniets, F. Dekeyzer, H. Bosmans, and G. Marchal
CLINICAL IMAGE QUALITY CRITERIA FOR FULL FIELD DIGITAL MAMMOGRAPHY: A FIRST PRACTICAL APPLICATION
Radiat Prot Dosimetry, March 4, 2008; (2008) ncn029v1.
[Abstract] [Full Text] [PDF]

This Article
Right arrow Abstract Freely available
Right arrow Figures Only
Right arrow Full Text (PDF)
Right arrow All Versions of this Article:
2432061065v1
243/2/396    most recent
Right arrow Submit a response
Right arrow Alert me when this article is cited
Right arrow Alert me when eLetters are posted
Right arrow Alert me if a correction is posted
Right arrow Citation Map
Services
Right arrow Email this article to a friend
Right arrow Similar articles in this journal
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Right arrow reprints & permissions
Citing Articles
Right arrow Citing Articles via HighWire
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Samei, E.
Right arrow Articles by Delong, D. M.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Samei, E.
Right arrow Articles by Delong, D. M.

HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
RADIOLOGY RADIOGRAPHICS RSNA JOURNALS ONLINE