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


     

Published online before print June 27, 2005, 10.1148/radiol.2361040741
This Article
Right arrow Abstract Freely available
Right arrow Figures Only
Right arrow Full Text (PDF)
Right arrow All Versions of this Article:
2361040741v1
236/1/31    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 Suryanarayanan, S.
Right arrow Articles by D'Orsi, C. J.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Suryanarayanan, S.
Right arrow Articles by D'Orsi, C. J.
(Radiology 2005;236:31-36.)
© RSNA, 2005

Breast Imaging

Detection of Simulated Lesions on Data-compressed Digital Mammograms1

Sankararaman Suryanarayanan, MS, MBA, Andrew Karellas, PhD, Srinivasan Vedantham, PhD, Sandra M. Waldrop, PhD and Carl J. D'Orsi, MD

1 From the Department of Radiology, Emory University School of Medicine, Winship Cancer Institute, 1701 Uppergate Dr, Bldg C, Suite 5018, Atlanta, GA 30322 (S.S., A.K., S. V., S.M.W., C.J.D.); and the Wallace H. Coulter Department of Biomedical Engineering, Georgia Institute of Technology and Emory University, Atlanta, Ga (S.S., A.K.). Received April 23, 2004; revision requested July 8; revision received August 13; accepted November 8. Supported in part by National Institutes of Health grants RO1-CA88792 from the National Cancer Institute and RO1-EB002123 from the National Institute of Biomedical Imaging and Bioengineering and by a Georgia Cancer Coalition infrastructure grant from the Cancer Scholars Program. Address correspondence to A.K. (e-mail: akarell{at}emory.edu).


   ABSTRACT
TOP
 ABSTRACT
INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 References
 
PURPOSE: To evaluate retrospectively the effect of a wavelet-based compression method on the detection of simulated masses of various sizes and clustered microcalcifications on data-compressed digital mammograms.

MATERIALS AND METHODS: The images used in this study were acquired with institutional review board approval and patient informed consent, both of which allowed subsequent image data analysis. Patient identification was removed from images, and the study complied with requirements of the Health Insurance Portability and Accountability Act. Masses 3, 6, and 8 mm in diameter were analytically simulated and added to clinical mammographic backgrounds. In addition, microcalcifications were extracted from a clinical mammogram and hybridized with simulated microcalcifications for use in this study. Image compression conditions of 1:1, 15:1, and 30:1 were investigated. Observer responses were recorded with a six-point rating scale, and receiver operating characteristic (ROC) analysis was performed. In addition, two well-established numeric observer models were used to study the effect of image compression under the same compression conditions as were used with human observers. Analysis of variance was performed after observer adjustment to compare the mean values for area under the ROC curve (Az) across the three compression levels for the masses and microcalcification clusters.

RESULTS: The results of the study indicated no significant differences in the Az values for masses with the compression conditions investigated. For images of microcalcifications, there were significant differences in Az values between compression ratios of 1:1 and 30:1 (P = .0005) and of 15:1 and 30:1 (P = .004); the difference between compression ratios of 1:1 and 15:1 was nonsignificant (P = .053). The observer models and human observers exhibited similar trends in detection of the masses investigated in this study.

CONCLUSION: Detection of simulated masses was not affected by the compression method with the conditions used in this study, while the detection of microcalcifications was significantly reduced with a compression ratio of more than 15:1.

© RSNA, 2005


   INTRODUCTION
TOP
 ABSTRACT
 INTRODUCTION
MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 References
 
Digital mammography is now a viable clinical method for breast cancer screening and diagnosis (1,2), and a number of solid-state detector technologies are currently used successfully for digital mammography (3–5). As more large-area digital detectors with high spatial resolution are used clinically for mammography, the need for data storage and transfer capabilities will increase dramatically. Moreover, recent developments in three-dimensional imaging techniques (6–9) are likely to necessitate high-volume image data storage and transfer capabilities. The key objective in medical image data compression is to maximize compression while maintaining diagnostic information content (10–14). Although progress is being made in data storage hardware, the increasing needs for image archival and communication can be met only by advanced image compression techniques. For example, in many institutions there is resistance to integrating digital mammograms with the rest of the digital image network because of the large storage capacity required.

Although a variety of data compression techniques are available, the two popular compression standards that have been evaluated for medical applications are the Joint Photographic Experts Group (JPEG) and JPEG 2000 standards (11,15–20). The lossy JPEG compression standard is based on methods that use the discrete cosine transform to map the image as a series of 8 x 8-pixel segments and to transform each segment into spectral coefficients that are subsequently quantized and encoded (16). The encoded image can be decoded, dequantized, and reconstructed by applying an inverse discrete cosine transform (16).

The JPEG 2000 compression standard, on the other hand, uses the discrete wavelet transform (16,18), whereby a series of high- and low-pass digital filters are applied repeatedly to the entire image, thus decomposing the image into subimages or subbands, as described in previous studies (16,18). These subbands can be encoded independently, and the image can be reconstructed at a selectable resolution by decoding bit streams that contain the appropriate subbands (16,18). The main characteristic features of the JPEG 2000 standard include higher compression efficiency compared with that of JPEG images, region-of-interest (ROI) compression, lossless and lossy compression, and multiresolution support (16). From a perceptual standpoint, the wavelet basis functions in JPEG 2000 images tend to produce smooth artifacts, unlike the blocklike artifacts produced by JPEG images, which tend to blend in with the rest of the anatomy (16).

Observer-independent metrics, such as peak signal-to-noise-ratio, are frequently used to assess the performance of compression algorithms (21,22), but for clinical images it is imperative to evaluate the quality of image compression from the end user's perspective by incorporating the human visual system in the evaluation process. Since compression techniques such as JPEG 2000 compression alter the spatial frequency information content of the images, it is critical to understand their effects on lesion detection by human observers.

In a recent study, Good et al (23) evaluated JPEG image compression for scanned mammograms and showed that it did not affect the detection of masses but did affect the detection of microcalcifications at high compression ratios. Another study investigated the effect of a wavelet-based compression technique on the detection of microcalcifications in mammograms, and the authors demonstrated a threshold compression ratio of 40:1 for microcalcification detection (24). In a phantom-based study, the effect of JPEG 2000 compression on contrast-detail characteristics of digital mammograms indicated certain promising traits (25). Thus, the purpose of the current study was to evaluate retrospectively the effect of a wavelet-based compression method on the detection of simulated masses of various sizes and clustered microcalcifications on data-compressed digital mammograms.


   MATERIALS AND METHODS
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
RESULTS
 DISCUSSION
 References
 
The images used in this study were acquired with institutional review board approval and patient informed consent, both of which allowed for subsequent image data analysis. Patient identification was removed from images, and the study complied with requirements of the Health Insurance Portability and Accountability Act.

Image Preparation
The images used in this study were obtained with a clinical full-field digital mammographic system (GE Medical Systems, Milwaukee, Wis) (26). The clinical images used in this study were acquired in 2000 and 2001. For the purpose of this investigation, only craniocaudal mammograms were selected (both left and right views). Image ROIs of 256 x 256 pixels were cropped from a region of uniform thickness in each mammogram and served as backgrounds in this study. One of the authors (S.S.), who is not a radiologist but has more than 5 years of experience in mammographic imaging research, developed an automated technique to select and extract the ROIs. The technique selected backgrounds relatively close to the chest wall (within approximately 10–15 mm) while ensuring that they did not cover the pectoralis, as the effect of breast tissue was of interest. Backgrounds without artifacts, microcalcifications, masses, or architectural distortion were selected. The cropped images were checked again under different window and level conditions to ensure the absence of the aforementioned structures, and those that did not meet our selection criteria were discarded. To ensure that each background was unique, the mean value of each ROI was computed and checked to ensure that no two ROIs had identical mean values. We did not try to select backgrounds of a specific type of breast tissue, and 260 backgrounds encompassing a mixture of fatty and glandular tissue were collected.

Images were hybridized with the addition of simulated masses and microcalcifications to the clinical mammographic data backgrounds. Two-dimensional masslike lesions were simulated by using an analytic equation approach to generate "designer nodules" (27). Masses 3, 6, and 8 mm in diameter were simulated for this study, and the peak amplitude of the mass was scaled in terms of digital units to achieve an area under the receiver operating characteristic (ROC) curve (Az) value well above 0.5. The same amplitude value was used for all three masses, to investigate lesion size effects. A single cluster of microcalcifications was extracted from a digital mammogram by filtering the original image to suppress the microcalcifications and subtracting the filtered image from the original. Care was taken, when filtering the original image, to avoid excessive smoothing of the background and thus to facilitate the effective removal of background structures at subtraction. Any undesirable structures that were present in the subtracted image were then removed by setting those pixel values at 0. An algorithm was developed to add simulated microcalcifications to the extracted microcalcifications from the clinical mammogram. A number of locations in the vicinity of the extracted microcalcifications were chosen, and pixels with amplitudes in the same range as the amplitudes of the extracted microcalcifications were inserted. A number of these inserted pixels were randomly grouped to mimic clinical microcalcifications, to produce a reasonable range of calcification sizes. The sizes of the individual calcifications were approximately 100–300 mm. As in the case of the masses, the amplitude of the calcification cluster was scaled so as to ensure an Az value well above 0.5.

One of the authors (S.S.) prepared image data sets by randomly selecting 100 mammographic backgrounds and adding a single mass or microcalcification cluster to the central portion of half the images, producing 50 backgrounds with a lesion and 50 background-only images. Four primary image data sets (100 images in each) were prepared for each of three mass sizes (3, 6, and 8 mm) and one microcalcification cluster size range (0.1–0.3 mm). In addition, a separate set of 60 backgrounds was used for observer training. Each of the data sets was compressed with commercially available JPEG 2000 image compression software (Aware, Bedford, Mass). The software allowed custom scripting to perform batch compression of images. Three data compression conditions—ratios of 1:1 (no compression), 15:1, and 30:1—were investigated with a lossy compression method, which allowed image compression at higher ratios. Examples of a simulated mass and microcalcification cluster are shown in Figure 1 and Figure 2, respectively.



View larger version (160K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 1. Sample image of the 6-mm simulated mass embedded in a digital mammographic background. The amplitude of the mass has been scaled for visualization purposes.

 


View larger version (152K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 2. Sample image of the hybrid extracted and simulated microcalcification cluster embedded in a digital mammographic background. The amplitude of the microcalcification cluster has been scaled for visualization purposes.

 
Human Observer Study
An interface graphics program was developed with "widget programming" techniques by using Interactive Data Language (Research Systems, Boulder, Colo) for the display of images to the observer in random order. The images were displayed on a high-resolution gray-scale clinical monitor (Clinton Electronics, Loves Park, Ill), and the window width and level were fixed for the entire course of the study. The monitor settings were adjusted to approximate the Barten tone scale (Digital Imaging and Communications in Medicine standard, part 14) (28,29). The backgrounds were positioned in the central region of the monitor, and a visual cue in the form of crosshairs was provided to indicate the location of the lesion to the observer, thereby facilitating a signal-known-exactly and location-known-exactly study design (27). The amplitude of the crosshairs was scaled to be slightly greater than that of the background so that the crosshairs did not affect the dynamic range of the image, and the spacing of the crosshairs was set to accommodate lesions of various sizes. We designed preprogrammed clickable buttons describing a six-point ROC scale with annotations to register the following observer responses: (a) high confidence that a lesion is present, (b) moderate confidence that a lesion is present, (c) low confidence that a lesion is present, (d) low confidence that a lesion is absent, (e) moderate confidence that a lesion is absent, and (f) high confidence that a lesion is absent. The region around the displayed image was masked, and the observations were conducted in a darkened room.

The observer study was conducted with radiologists who had approximately 5–32 years of experience in mammography and a 1st-year resident who had some training in clinical mammography. Five observers (four radiologists and one resident) participated in the study with the masses, and four (three radiologists and one resident) participated in the study with the microcalcification cluster.

One of the authors (C.J.D.) participated as an observer in the experiments with the masses and microcalcifications. The study was conducted in multiple sessions, in which each observer independently reviewed the displayed images and assigned confidence responses. The experiment was first conducted with simulated masses and then with the microcalcification cluster. Before each session, the observers were trained with 60 images for a specific mass size that included the three compression conditions. Similar training was provided for the microcalcification cluster. Feedback was provided to the observer in the form of a high-contrast version (reference standard) of the test image, to enable observer orientation within the experiment.

Each session was divided into three sections for the three compression conditions (after training), and only masses of a specific size were reviewed in a given session. The selection and order of the displayed backgrounds were randomized between sections to prevent observer familiarization with the images. Further, the order of mass size and compression condition were varied across observers to prevent reader order effects (30). For the microcalcification cluster, only the order of the compression condition was varied across observers, as only a single cluster was evaluated.

No restrictions were placed on the observers in terms of viewing distances and time, but on average the observers completed each section in about 15 minutes. The use of a magnifier was restricted for visualization of masses but was allowed to all observers for visualization of microcalcifications. One of the observers, who wore contact lenses for vision correction, was permitted to use a magnifier also for visualization of masses, as this observer used a magnifier for this purpose in the clinic. Data with regard to image order and observer confidence responses were recorded and later analyzed.

Observer Models
Two numeric observer models—namely, nonprewhitening matched filter with an eye filter (NPWE) (27,31) and Laguerre-Gauss channelized Hotelling observer (LG-CHO) (32)—were used to study the effect of image compression on lesion detection. The models were applied to study the detection of masses under the conditions investigated. The general process for both observer models included template generation and detection index computation from test statistics. Six Laguerre-Gaussian channels were used for the LG-CHO model. A detailed description of these numeric observer models is provided elsewhere (27,31–33). Five hundred twelve digital mammographic backgrounds were used to generate the templates, and a different set of 512 backgrounds were used to compute the percentage of correct detections for each mass size and compression condition by randomly selecting backgrounds with and without the signal (33). The percentages of correct detections were transformed into indexes of detection and later transformed into an Az value according to the relation derived by Barrett et al (34).

Statistical Analysis
The recorded human observer data for masses and microcalcifications for all the compression conditions were imported into a software program (ROCKIT; Charles E. Metz, University of Chicago, Chicago, Ill), and corresponding Az values were computed. Forty-five Az values (five observers x three mass sizes x three compression conditions) were obtained for the masses, and 12 Az values (four observers x one cluster x three compression conditions) were obtained for the microcalcification cluster. Statistical analysis was performed with analysis of variance after observer adjustment to compare the mean Az values across the three compression levels for each mass diameter (3, 6, and 8 mm). Similar analysis was performed for the microcalcification cluster. When the overall test for comparison of the Az estimated values at each compression ratio showed statistically significant differences (P < .05), pairwise comparisons were performed between the compression ratios (1:1 vs 15:1, 1:1 vs 30:1, and 15:1 vs 30:1). All statistical tests were two sided and unadjusted for multiple comparisons.


   RESULTS
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
DISCUSSION
 References
 
The Az values obtained from the ROC study are summarized in Tables 1 and 2 for masses and microcalcifications. Statistical analysis indicated no significant differences in the Az values for each mass size between the different compression conditions studied (P = .58 for 3 mm, P = .60 for 6 mm, and P = .33 for 8 mm). However, for images of microcalcifications, there was a significant difference in Az values between images with compression ratios of 1:1 and 30:1 (P = .0005) and 15:1 and 30:1 (P = .004) but no significant difference between images with compression ratios of 1:1 and 15:1 (P = .053). A marginal decrease in Az values with increasing mass size was observed. The performance data for the numeric observer models in general indicated good agreement in trends for detection of masses, and the Az values obtained with the NPWE observer model were in good agreement with human observer results for all mass diameters investigated in this study. A one-sided t test indicated that the NPWE-predicted Az values did not differ significantly (P > .05) from human observer Az values with any of the conditions investigated (Fig 3).


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

  		TABLE 1. Az Values for Detection of Masses of Different Sizes under Different Data Compression Conditions

 

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

  		TABLE 2. Az Values for Detection of Clustered Microcalcifications under Different Data Compression Conditions

 


View larger version (34K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 3a. Graphs show mean Az values (± standard errors) for human observers and two numeric observer models for (a) 3-mm, (b) 6-mm, and (c) 8-mm masses with the compression conditions investigated in this study. Detection of masses was not affected by the image compression ratio, but detection of microcalcifications was decreased at compression ratios of more than 15:1. There was good agreement in detection trends between the numeric observer models and the human observers.

 


View larger version (32K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 3b. Graphs show mean Az values (± standard errors) for human observers and two numeric observer models for (a) 3-mm, (b) 6-mm, and (c) 8-mm masses with the compression conditions investigated in this study. Detection of masses was not affected by the image compression ratio, but detection of microcalcifications was decreased at compression ratios of more than 15:1. There was good agreement in detection trends between the numeric observer models and the human observers.

 


View larger version (33K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 3c. Graphs show mean Az values (± standard errors) for human observers and two numeric observer models for (a) 3-mm, (b) 6-mm, and (c) 8-mm masses with the compression conditions investigated in this study. Detection of masses was not affected by the image compression ratio, but detection of microcalcifications was decreased at compression ratios of more than 15:1. There was good agreement in detection trends between the numeric observer models and the human observers.

 

   DISCUSSION
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
References
 
The motivation for this study was to understand the clinical effect of wavelet-based compression on digital mammograms, especially because high-resolution digital mammography is gaining prominence. Mammograms in general depict complex and variable tissue patterns that make lesion detection a challenging task even without data compression. It becomes much more challenging with the use of lossy compression techniques such as the one investigated in this study, which alter the spatial frequency content of the image. Nevertheless, we believe it is important to investigate advanced compression methods for digital mammographic data without compromising the diagnostic value of the information. The use of human observers takes into account the characteristics of the human visual system in evaluating compression methods, as the response of that system to different spatial frequency information plays an integral role in detection.

The results of this study suggest that image compression with use of the JPEG 2000 standard does not affect the detection of subtle simulated masses in digital mammographic backgrounds but does affect the detection of microcalcifications at compression ratios of more than 15:1 and with the conditions investigated. Our results agree with those reported by Good et al (23), who demonstrated the effect of JPEG image compression for digitized mammographic film images. Many compression techniques, such as JPEG 2000, use frequency decomposition of the image data with visual weighting that could potentially lead to a loss of high-frequency information, especially at higher compression ratios (16).

Typical mammographic masses and the simulated masses investigated in this study predominantly have low-frequency content with smooth edges and tend to blend in with the surrounding tissue structures. This property may well have played a role in preserving their detectability. On the other hand, microcalcifications are rich in high-spatial-frequency content because of their small size and relatively abrupt edges, and compression at higher ratios could have filtered the high-frequency information from the signal (microcalcification), thereby affecting its detection.

The decrease in Az values for masses with increasing size and constant amplitude could be attributed to the background spatial statistics of mammograms. The variation in image noise as a frequency (noise power spectrum) has an effect on detection trends. For constant-amplitude lesions like those in this study, a reduction in signal-to-noise ratio has been shown to occur in power law–related noise as lesion size is increased (27). Mammograms have been shown to exhibit a power law–dependent noise power spectrum in which the shape of the spectrum has a spatial frequency dependence of 1/f3 (27), and it was not surprising that we observed a general degradation in the detection of the constant-amplitude masses as their size increased. It is interesting to note that such a trend appears to be valid even for data-compressed mammograms. In another study, Lucier et al (35) found morphologic distortion of microcalcifications caused by a specific type of wavelet compression at high compression levels, which led to higher variability in diagnosis.

Contrary to our observations with JPEG image compression, with JPEG 2000 compression we did not notice any block or discrete artifacts, and none of the observers noticed any differences between the uncompressed and compressed images. Compression was limited to a ratio of 30:1. Beyond this ratio, a visual smoothing effect was noticeable with the ROI images used in this study. The study was conducted with extracted ROIs from clinical digital mammograms, since we sought only a relative comparison of detection with respect to lesion size and compression ratio rather than absolute detection rates. The general trend of our results agrees with those of Good et al (23) and Kocsis et al (24), who conducted studies with clinical mammograms but with a different image compression technique.

This study was limited in that only ROIs were used and the signal location was known a priori to the observer. In a clinical situation, the radiologist would not have this luxury and would have to go through a search-and-find process to identify lesions. Further, the study did not incorporate a variety of mass and microcalcification types, as might be observed clinically. In addition, it was designed to evaluate the effect of image compression only on lesion detection. Evaluation of lesion characterization was beyond the scope of this investigation, and the effect of our compression technique on the morphologic features of malignant lesions needs to be investigated.

Task-specific optimization of the compression algorithm was beyond the scope of this investigation. However, the similarity in detection trends between human and numeric observers for the mass-type lesions is encouraging, and such models or their variants may offer a viable method for evaluating image compression techniques. The LG-CHO model could not be applied to the microcalcification data, as stable template estimation was constrained with nonsymmetric signals. The performance of the NPWE observer model also was not a good predictor of human detection of microcalcifications, because of restrictions in template estimation, especially after image compression. However, numeric observer models have been investigated to analyze the performance of compression algorithms for other applications with objective metrics (36,37). Optimization of compression transformation and quantization strategies could improve the performance of data compression, especially for high-frequency objects such as microcalcifications (37,38). Furthermore, the effect of image compression on computer-aided diagnosis needs to be studied in great detail (39). Much work is needed to clarify the intricate mechanisms by which image compression could affect lesion visualization and detection on mammograms. Further, similar studies with other types of digital mammography systems will provide useful insights into system-specific image compression performance.

On the basis of the tested conditions and results in this study, it appears that (a) image data compression at ratios up to 30:1 might be applicable for digital mammograms with masses; (b) it may become feasible to use a compression ratio of as much as 15:1 for mammograms with microcalcifications, without degrading the diagnostic value of the image; (c) numeric observer models can be an efficient method to optimize compression techniques; and (d) this study serves as a precursor for comprehensive clinical studies to evaluate JPEG 2000 image compression for digital mammography. It should be noted, however, that the results of this study are limited to lesion detection, and lower compression ratios might be required for lesion characterization to be preserved. By optimizing compression parameters, it may be possible to improve observer performance substantially. It may be appropriate to evaluate whether it is feasible to store uncompressed mammograms for a specific period (eg, 2 years), after which a compression of those mammograms would be performed with a modest compression ratio. Greater image compression may be appropriate for mammograms older than 5 years. Overall, the results of this study suggest that JPEG 2000 compression is a promising technique for the compression of clinical digital mammographic data.


   ACKNOWLEDGMENTS
 
The authors thank Kirk A. Easley, MS, Department of Biostatistics, Rollins School of Public Health, Emory University, for assistance with the statistical analysis.


   FOOTNOTES
 

Abbreviations: Az = area under the ROC curve • JPEG = Joint Photographic Experts Group • LG-CHO = Laguerre-Gauss channelized Hotelling observer • NPWE = nonprewhitening matched filter with an eye filter • ROC = receiver operating characteristic • ROI = region of interest

The contents are solely the responsibility of the authors and do not necessarily represent the official views of the National Cancer Institute, the National Institute of Biomedical Imaging and Bioengineering, or the National Institutes of Health.

Authors stated no financial relationship to disclose.

Author contributions: Guarantor of integrity of entire study, S.S.; study concepts, S.S., A.K., S.V., C.J.D.; study design, S.S., A.K., C.J.D.; literature research, S.S., A.K., S.V.; experimental studies, S.S., A.K., S.V., C.J.D.; data acquisition, S.S., C.J.D.; data analysis/interpretation, S.S., S.M.W.; statistical analysis, S.S., S.M.W.; manuscript preparation, S.S., A.K., S.V., S.M.W.; manuscript definition of intellectual content, S.S., A.K.; manuscript editing, all authors; manuscript revision/review and final version approval, A.K., C.J.D.


   References
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 References
 

  1. Lewin JM, Hendrick ER, D'Orsi CJ, et al. Comparison of full-field digital mammography with screen-film mammography for cancer detection: results of 4,945 paired examinations. Radiology 2001; 218:873–880.[Abstract/Free Full Text]
  2. Lewin JM, D'Orsi CJ, Hendrick RE, et al. Clinical comparison of full-field digital mammography and screen-film mammography for detection of breast cancer. AJR Am J Roentgenol 2002; 179:671–677.[Abstract/Free Full Text]
  3. Pisano ED, Yaffe MJ, Hemminger BM, et al. Current status of full-field digital mammography. Acad Radiol 2000; 7:266–280.[CrossRef][Medline]
  4. 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]
  5. Karellas A, Vedantham S, Suryanarayanan S, Belanger PL, D'Orsi CJ. Digital mammography: current state-of-the-art and accreditation issues. In: Dixon RL, Butler PF, Sobol WT, eds. Accreditation programs and the medical physicist. American Association of Physicists in Medicine medical physics monographs, no. 27. Madison, Wis: Medical Physics Publishing, 2001; 295–317.
  6. Niklason LT, Christian BT, Niklason LE, et al. Digital tomosynthesis in breast imaging. Radiology 1997; 205:399–406.[Abstract/Free Full Text]
  7. Suryanarayanan S, Karellas A, Vedantham S, et al. Comparison of tomosynthesis methods used with digital mammography. Acad Radiol 2000; 7:1085–1097.[CrossRef][Medline]
  8. Boone JM, Nelson TR, Lindfors K, et al. Dedicated breast CT: radiation dose and image quality evaluation. Radiology 2001; 221:657–667.[Abstract/Free Full Text]
  9. Chen B, Ning R. Cone-beam volume CT breast imaging: feasibility study. Med Phys 2002; 29:755–770.[CrossRef][Medline]
  10. Lo SC, Huang HK. Radiological image compression: full-frame bit-allocation technique. Radiology 1985; 155:811–817.[Abstract/Free Full Text]
  11. Agarwal A, Rowberg AH, Kim Y. Fast JPEG 2000 decoder and its use in medical imaging. IEEE Trans Inf Technol Biomed 2003; 7:184–190.[CrossRef][Medline]
  12. Egashira K, Nakata H, Watanabe H, et al. Clinical evaluation of irreversible data compression for computed radiography of the chest. J Digit Imaging 1998; 11:176–181.[Medline]
  13. Schreiber D. Swelling archives warrant closer look at compression. Radiol Manage 2003; 25:36–39.
  14. Clunie DA. Lossless compressions of grayscale medical images: effectiveness of traditional and state-of-the-art approaches. In: Blaine GJ, Siegel EL, eds. Medical imaging 2000: PACS design and evaluation—engineering and clinical issues. Proceedings of SPIE, vol 3980. Bellingham, Wash: SPIE–The International Society for Optical Engineering, 2000; 74–84.
  15. Joint Photographic Experts Group and Joint Bi-level Image Experts Group. JPEG 2000 requirements and profiles. V.6.3. Joint Photographic Experts Group Web site. http://www.jpeg.org. Published July 2000. Accessed April 21, 2005.
  16. Jones PW, Rabbani M. JPEG compression in medical imaging. In: Kim Y, Horii SC, eds. Display and PACS. Handbook of medical imaging, vol 3. Bellingham, Wash: SPIE–The International Society for Optical Engineering, 2000; 221–275.
  17. Slone RM, Muka E, Pilgram TK. Irreversible JPEG compression of digital chest radiographs for primary interpretation: assessment of visually lossless threshold. Radiology 2003; 228:425–429.[Abstract/Free Full Text]
  18. Foos DH, Muka E, Slone RM, et al. JPEG 2000 compression of medical imagery. In: Blaine GJ, Siegel EL, eds. Medical imaging 2000: PACS design and evaluation—engineering and clinical issues. Proceedings of SPIE, vol 3980. Bellingham, Wash: SPIE–The International Society for Optical Engineering, 2000; 85–96.
  19. Kalyanpur A, Neklesa VP, Taylor CR, Daftary AR, Brink JA. Evaluation of JPEG and wavelet compression of body CT images for direct digital teleradiologic transmission. Radiology 2000; 217:772–779.[Abstract/Free Full Text]
  20. Ko JP, Rusinek H, Naidich DP, et al. Wavelet compression of low-dose chest CT data: effect on lung nodule detection. Radiology 2003; 228:70–75.[Abstract/Free Full Text]
  21. Iyriboz TA, Zukoski MJ, Hopper KD, et al. A comparison of wavelet and Joint Photographic Experts Group lossy compression methods applied to medical images. J Digit Imaging 1999; 12(2 suppl 1):14–17.[Medline]
  22. Penedo M, Pearlman WA, Tahoces PG, Souto M, Vidal JJ. Region-based wavelet coding methods for digital mammography. IEEE Trans Med Imaging 2003; 22:1288–1296.[CrossRef][Medline]
  23. Good WF, Sumkin JH, Ganott M, et al. Detection of masses and clustered microcalcifications on data compressed mammograms: an observer performance study. AJR Am J Roentgenol 2000; 175:1573–1576.[Abstract/Free Full Text]
  24. Kocsis O, Costaridou L, Varaki L, et al. Visually lossless threshold determination for microcalcification detection in wavelet compressed mammograms. Eur Radiol 2003; 13:2390–2396.[CrossRef][Medline]
  25. Suryanarayanan S, Karellas A, Vedantham S, Waldrop SM, D'Orsi CJ. A perceptual evaluation of JPEG 2000 image compression for digital mammography: contrast-detail characteristics. J Digit Imaging 2004; 17:64–70.[CrossRef][Medline]
  26. Suryanarayanan S, Karellas A, Vedantham S, Ved H, Baker SP, D'Orsi CJ. Flat-panel digital mammography system contrast-detail comparison between screen-film radiographs and hard-copy images. Radiology 2002; 225:801–807.[Abstract/Free Full Text]
  27. Burgess AE, Jacobson FL, Judy PF. Human observer detection experiments with mammograms and power-law noise. Med Phys 2001; 28:419–437.[CrossRef][Medline]
  28. National Electrical Manufacturers Association. Digital Imaging and Communications in Medicine (DICOM). Part 14: Gray scale standard display function. Rosslyn, Va: National Electrical Manufacturers Association, 2004.
  29. Krupinski EA, Roehrig H. The influence of a perceptually linearized display on observer performance and visual search. Acad Radiol 2000; 7:8–13.[CrossRef][Medline]
  30. Metz CE. Some practical issues of experimental design and data analysis in radiological ROC studies. Invest Radiol 1989; 24:234–245.[Medline]
  31. Ishida M, Doi K, Loo LN, Metz CE, Lehr J. Digital image processing: effect on detectability of simulated low-contrast radiographic patterns. Radiology 1984; 150:569–575.[Abstract/Free Full Text]
  32. Barrett HH, Abbey CK, Gallas BD, Eckstein MP. Stabilized estimates of Hotelling-observer detection performance in patient-structured noise. In: Kundel HL, ed. Medical imaging 1998: image perception. Proceedings of SPIE, vol 3340. Bellingham, Wash: SPIE–The International Society for Optical Engineering, 1998; 7–43.
  33. Eckstein MP, Abbey CK, Bochud FO. A practical guide to model observers for visual detection in synthetic and natural noisy images. In: Van Metter RL, Beutel J, Kundel HL, eds. Physics and psychophysics. Handbook of medical imaging, vol 1. Bellingham, Wash: SPIE-The International Society of Optical Engineering, 2000; 593–628.
  34. Barrett HH, Abbey CK, Clarkson E. Objective assessment of image quality. III. ROC metrics, ideal observers, and likelihood-generating functions. J Opt Soc Am A Opt Image Sci Vis 1998; 15:1520–1535.[Medline]
  35. Lucier BJ, Kallergi M, Qian W, et al. Wavelet compression and segmentation of digital mammograms. J Digit Imaging 1994; 7:27–38.[Medline]
  36. Zhao B, Schwarz LH, Kijewski P. Effects of lossy compression on lesion detection: predictions of a nonprewhitening matched filter. Med Phys 1998; 25:1621–1624.[CrossRef][Medline]
  37. Zhang Y, Pham BT, Eckstein MP. Automated optimization of JPEG 2000 encoder options based on model observer performance for detecting variable signals in x-ray coronary angiograms. IEEE Trans Med Imaging 2004; 23:459–474.[CrossRef][Medline]
  38. Lo SC, Li H, Freedman MT. Optimization of wavelet decomposition for image compression and feature preservation. IEEE Trans Med Imaging 2003; 22:1141–1151.[CrossRef][Medline]
  39. Chan HP, Lo SC, Niklason LT, Ikeda DM, Lam KL. Image compression in digital mammography: effects on computerized detection of subtle microcalcifications. Med Phys 1996; 23:1325–1336.[CrossRef][Medline]



This article has been cited by other articles:


Home page
 RadiologyHome page
B. F. Branstetter IV
Basics of Imaging Informatics: Part 1
Radiology, June 1, 2007; 243(3): 656 - 667.
[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:
2361040741v1
236/1/31    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 Suryanarayanan, S.
Right arrow Articles by D'Orsi, C. J.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Suryanarayanan, S.
Right arrow Articles by D'Orsi, C. J.

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