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


     

This Article
Right arrow Abstract Freely available
Right arrow Figures Only
Right arrow Full Text (PDF)
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 Vyborny, C. J.
Right arrow Articles by Stein, A. A.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Vyborny, C. J.
Right arrow Articles by Stein, A. A.
(Radiology. 2000;215:703-707.)
© RSNA, 2000

Computer Applications

Breast Cancer: Importance of Spiculation in Computer-aided Detection1

Carl J. Vyborny, MD, PhD, Takeshi Doi, MS, Kathryn F. O'Shaughnessy, PhD, Harlan M. Romsdahl, BS, Alexander C. Schneider, BS and Alan A. Stein, PhD

1 From the Department of Radiology, LaGrange Memorial Hospital, 5100 S Willow Springs Rd, LaGrange, IL 60525 (C.J.V.); University of Chicago Hospitals, Ill (C.J.V.); and R2 Technology, Los Altos, Calif (T.D., K.F.O., H.M.R., A.C.S., A.A.S.). From the 1999 RSNA scientific assembly. Received September 9; revision requested October 14; revision received November 11; accepted November 15. Address correspondence to C.J.V.


   Abstract
TOP
 Abstract
Introduction
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 References
 
PURPOSE: To determine the prevalence of spiculation in a large series of screening-detected breast cancers appearing as masses on mammograms and to assess the sensitivity of a computer-aided detection (CAD) algorithm that uses spiculation measures in the detection of such lesions.

MATERIALS AND METHODS: Six hundred seventy-seven consecutive cases of breast cancers detected as masses on mammograms were independently reviewed by three radiologists who determined if the lesions were spiculated. All cancers were then analyzed by the CAD system.

RESULTS: All three radiologists interpreted 375 (55%) of the 677 masses as being spiculated on at least one view. The CAD algorithm correctly marked 322 (86%) of the 375 clearly spiculated masses, with a mean of 0.24 additional mass mark per image. With a looser definition of spiculation, 585 (86%) of the 677 masses were called spiculated by at least one radiologist on one view. The algorithm correctly marked 464 (79%) of the 585 lesions that were spiculated or possibly spiculated.

CONCLUSION: Spiculation was clearly present in a majority (55%) of consecutive screening-detected breast cancer masses found on mammograms in a large clinical trial. Incorporation of spiculation measures is, therefore, an important strategy in the detection of breast cancer with CAD. A present-generation CAD algorithm correctly identified a large proportion (86%) of spiculated breast cancers.

Index terms: Breast neoplasms, 00.32 • Breast neoplasms, diagnosis, 00.32 • Cancer screening, 00.11, 00.32 • Computers, diagnostic aid


   Introduction
TOP
 Abstract
 Introduction
MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 References
 
The detection of mass lesions on mammograms can be a difficult task for human observers or machines. The potential variability and heterogeneity of normal breast tissue often produces a number of localized findings that may simulate mass lesions or, depending on the observer, create distractions during the search process. Strategies must, therefore, be developed to ensure a reasonable performance in detection. These strategies include the identification of findings that are focal and opaque, are asymmetric when compared with findings on the other side, and display a radiating or spiculated pattern (1–3).

Spiculation can be particularly valuable, since it provides a direct radiographic manifestation of the local aggressivity of invasive breast cancer (1–3). Although a number of benign lesions can possibly produce a spiculated pattern (4), biopsy of spiculated abnormalities of uncertain cause is invariably indicated (2,5). Even with this knowledge, findings with spiculation or architectural distortion are often missed during screening mammography (6–8).

Spiculated findings on mammograms are amenable to detection by use of the computer with a wide variety of analytic approaches. These approaches generally rely on the identification of radiating patterns of optical density (9–12). The output of such detection algorithms has been shown to identify lesions that are missed by human observers (9,13). It is, therefore, important to consider whether spiculation measures should be integral components of computer algorithms that are designed to detect masses or architectural distortions on mammograms.

The purpose of this study was to determine the relative occurrence of spiculated findings in a large series of biopsy-proved carcinomas detected at screening mammography and to determine the performance of a commercial computer-aided detection (CAD) algorithm in the identification of such findings.


   MATERIALS AND METHODS
TOP
 Abstract
 Introduction
 MATERIALS AND METHODS
RESULTS
 DISCUSSION
 References
 
Case Selection
In a previous study (14), 1,083 consecutive cases of biopsy-proved cancers detected at screening mammography were obtained from 13 institutions. A reference radiologist from each site (the site radiologist) reviewed the standard radiographs, any additional diagnostic radiographs, and radiology and pathology reports to characterize the radiographic findings of the lesion in each patient and to create a standard description. The location of the lesion in each applicable view was noted and identified on a transparent overlay. The primary diagnostic feature of each abnormality was classified as microcalcifications or as being masslike, which included masses, architectural distortions, and other findings. The 677 cases of cancer examined in this study comprise those in which the primary radiographic feature was masslike.

The mean age of the patients in the series was 64 years; 631 (93%) of 677 cases were invasive carcinomas. Median lesion size was 12 mm; lesion size was unreported by the site radiologist in 86 (13%) of the cases. The standard four screening views were evaluated in this study—craniocaudal and mediolateral oblique views of each breast. Forty-nine cases had either unilateral mammograms (n = 40) or were cases in which only three (n = 6), two (n = 2), or one (n = 1) radiograph was retrieved from the patient's file, yielding a total of 2,615 radiographs for study. In 37 cases, the lesion of concern was visible in only one view (craniocaudal in six, mediolateral oblique in 31).

Radiologist Review Sessions
The 677 cases were assigned to sets of approximately 120 cases each for the observer study. Nine radiologists participated in the study (some in multiple sessions); three independently reviewed each set of cases. All radiologists were qualified to read screening mammograms according to the criteria of the Mammography Quality and Standards Act. Each had a minimum of 6 years of experience in the interpretation of screening mammograms (median, 13 years).

For the observer study, radiologists were shown high-quality digital copies of the screening mammograms printed with 100-µm pixelation and a gray-scale look-up table that matched the appearance of the original mammograms. Side-by-side comparisons of the 20 original mammograms and 20 digital images showed no degradation in the rendering of pertinent detail in the latter. The observers knew that each case had a detectable cancer and that the location of the lesion was indicated by the overlay prepared by the site radiologist. No restriction was placed on the amount of time taken to examine the images.

A training session for the radiologists preceded each reading session. In the actual observer test, the radiologists were asked to rate the subtlety of the lesion by using the following four-point scale: 0 was not visible, 1 was subtle, 2 was medium (visible but not conspicuous), and 3 was obvious. They were also asked to classify each lesion as 0, which indicated no spiculation, or 1, which indicated, in their opinion, the clear presence of spiculation. If the lesion was clearly not spiculated or if the findings of spiculation were borderline (eg, possibly related to overlapping structures), the case was given a rating of 0.

During training, examples of each of the six possible combinations of subtlety and spiculation were shown to the radiologists to help them match their clinical experience with the categories used in the study. Examples are shown in Figure 1. The radiologists were also asked to rate the density of the breast tissue by using the following four-point BI-RADS (Breast Imaging Reporting and Data System) scale from the American College of Radiology (15): 1 was fatty, 2 was scattered fibroglandular, 3 was heterogeneously dense, and 4 was dense.



View larger version (105K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 1a. Radiographs depict examples of cases used in the training set. (a) Left mediolateral oblique view shows a subtle spiculated mass (arrow). (b) Left craniocaudal view shows a subtle nonspiculated mass (arrow). (c) Left mediolateral view shows an obvious spiculated mass (arrow). (d) Right mediolateral view shows an obvious nonspiculated mass (arrow).

 


View larger version (96K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 1b. Radiographs depict examples of cases used in the training set. (a) Left mediolateral oblique view shows a subtle spiculated mass (arrow). (b) Left craniocaudal view shows a subtle nonspiculated mass (arrow). (c) Left mediolateral view shows an obvious spiculated mass (arrow). (d) Right mediolateral view shows an obvious nonspiculated mass (arrow).

 


View larger version (126K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 1c. Radiographs depict examples of cases used in the training set. (a) Left mediolateral oblique view shows a subtle spiculated mass (arrow). (b) Left craniocaudal view shows a subtle nonspiculated mass (arrow). (c) Left mediolateral view shows an obvious spiculated mass (arrow). (d) Right mediolateral view shows an obvious nonspiculated mass (arrow).

 


View larger version (122K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 1d. Radiographs depict examples of cases used in the training set. (a) Left mediolateral oblique view shows a subtle spiculated mass (arrow). (b) Left craniocaudal view shows a subtle nonspiculated mass (arrow). (c) Left mediolateral view shows an obvious spiculated mass (arrow). (d) Right mediolateral view shows an obvious nonspiculated mass (arrow).

 
CAD System
A commercial CAD system (ImageChecker M1000 V2.0; R2 Technology, Los Altos, Calif) was used to analyze the cases in the study after the radiologists reviewed them. The system includes a precision image digitizer (50-µm resolution, 12 bits of depth) and a computer that runs the proprietary signal processing and analysis algorithms. The algorithms in the commercial device are designed to identify and locate or mark regions of interest that manifest features associated with masses and/or distortions or microcalcifications. It is important to emphasize that only the mass marker was relevant to this study and that all results reported here refer to only those markers.

The detection algorithm includes in its design an ability to identify features that are commonly associated with masses, namely, an area with central opacity and radiating lines. When no central opacity is detected, the radiating lines must be more pronounced to be marked, as illustrated in Figure 2.



View larger version (36K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 2. Schematic representation depicts the relative effects of the central opacity and radiating lines in the determination of whether a commerical CAD algorithm marks a lesion. (Reprinted with permission from R2 Technology, Los Altos, Calif.)

 
The images for each of the 677 cases were analyzed with the system software, and the resultant output detection locations were compared against the standard overlays created by the site radiologist. If the location designated by the computer was within the area indicated by the site radiologist, the view was considered to be correctly marked.

Data Analysis
The individual ratings of the three radiologists were combined to give a case-specific rating for spiculation and subtlety as follows. For spiculation, a score for each craniocaudal or mediolateral oblique view was first determined by summing the scores of each radiologist. The score for a lesion on a given view could, therefore, range between 0 and 3. For the purposes of detection, the more relevant score is the greater one; therefore, the consensus rating of spiculation for the case was taken to be the larger of the spiculation scores of the two views (or the single view score if the lesion was visible in only one view).

The subtlety score for the view was defined as the sum of ratings of the three radiologists; the score for a single view could range from 0 to 9. The larger single-view score was defined as the subtlety score for the case. Three consensus ratings of subtlety for the case were then defined as follows: subtle (scores of 0 to 5), medium (scores of 6 or 7), and obvious (scores of 8 or 9). The overall breast density rating was chosen to be the median of the ratings of the three radiologists.

If the CAD system correctly designated the location of the cancer in any view on which it was visible, the case was considered to be correctly marked. CAD sensitivity was taken as the number of cases correctly marked divided by the total number of cases. In addition, the total number of markers designated by the mass algorithm at other (ie, incorrect) locations on the images was summed, and the mean for all images was calculated.

A {chi}2 test of independence for a contingency table was used to determine the statistical significance of any differences measured in the data. A P value of less than .05 indicated a statistically significant difference.


   RESULTS
TOP
 Abstract
 Introduction
 MATERIALS AND METHODS
 RESULTS
DISCUSSION
 References
 
Table 1 shows the consensus ratings of spiculation for the 677 cases. In 375 (55%) of the cases, all three radiologists described the lesion as spiculated on at least one view. The radiologists completely agreed on the assessment of the presence (n = 375) or absence (n = 92) of spiculation in a substantial majority (467 cases [69%]) of the cases. The distribution of cases by consensus subtlety rating is shown in Table 1. The more spiculated abnormalities tended to be the more obvious in this series.


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

  		TABLE 1. Consensus Spiculation Ratings versus Consensus Subtlety Ratings
 
The CAD results for the different spiculation and subtlety ratings are shown in Table 2. The sensitivity of the CAD system in marking clearly spiculated lesions (consensus spiculation rating of 3) was 322 (86%) of 375 cases; the algorithm marked 255 (94%) of 271 clearly spiculated cases that the reviewing radiologists considered to be obvious. For lesions that were considered to be spiculated by at least one radiologist (ie, consensus spiculation rating of 1, 2, or 3), the CAD sensitivity was 464 (79%) of 585. The number of additional marks made by the system was, on average, 0.24 mark per image, or approximately one extra mark for a four-image examination.


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

  		TABLE 2. CAD Results as a Function of Consensus Spiculation and Subtlety Ratings
 
The distribution of breast density ratings versus consensus ratings of case subtlety is given in Table 3. The relative proportion of subtle cases in each breast density category increased in breasts that were more dense. For breasts with a BI-RADS rating of 1 (fatty), subtle cases made up nine (5.4%) of 167 cases; for breasts with a BI-RADS rating of 4 (dense), subtle cases made up a significantly larger proportion of the total, or 17 (17%) of 100 cases (P = .002). The performance of the CAD system showed no statistically significant dependence on breast density in this set of cancers (P = .53), as shown in Table 3.


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

  		TABLE 3. BI-RADS Breast Density Rating versus Consensus Subtlety Rating
 

   DISCUSSION
TOP
 Abstract
 Introduction
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
References
 
Our results show that only a small proportion, 92 (14%) of 677, of consecutive cancers detected at screening mammography (which did not appear primarily as microcalcifications) showed no evidence of spiculation on mammograms that were independently reviewed by three radiologists. Rather, 498 (74%) of such lesions were determined to be spiculated on at least one view by two or three of the three radiologists. This value is larger than (16) or comparable to (13,17) those reported in some series. Considerable variability in the characterization of mammographic abnormalities, including spiculation, has been reported (18). For this reason and in the absence of a quantitatively precise definition of spiculation, it is not surprising that observations regarding spiculation vary in the literature and that observers in this study did not entirely agree. Nevertheless, the importance of spiculation as a mammographic finding in breast cancer is evident in this large series.

Given the frequent occurrence of spiculated features as a sign of malignancy, their identification is a desirable if not essential component of computer algorithms that are used to detect mass lesions or distortions. When these features are combined with features of central opacity (Fig 2), a large proportion of spiculated abnormalities, 322 (86%) of 375, were detected by use of a commercial computer algorithm, with a low false-positive rate of less than 0.25 mark per image. In the actual clinical application of such information, the radiologist would still be required to decide if the area marked by the algorithm was truly abnormal. The previous result does not imply that a radiologist who prospectively uses CAD would always take appropriate action for the true cancers marked by the device and correctly dismiss all false marks (14).

The subtlety ratings in individual cases given by the radiologists in this study are also of interest. In particular, these radiologists, in aggregate, indicated that only 88 (13%) of the 677 screening-detected cancers were subtle, whereas 435 (64%) were obvious. The degree to which such ratings were biased by prior knowledge is difficult to judge, but given the fact that nine radiologists participated in the study and that each case was reviewed three times, the basic trend seems clear. If one assumes that undetected subtle lesions eventually progress to more obvious ones, this study is consistent with the reported high incidence rate of breast cancers that are visible at retrospective review of prior mammograms (6,14,19). Late-generation CAD algorithms are known to have the capability to detect many such retrospectively visible carcinomas (13,14,20).

One indication of the possible value of CAD information is given by the data related to breast density presented here. The radiologists who retrospectively reviewed the cancers clearly believed that a greater proportion of the subtle lesions occurred in patients with dense breasts (Table 3). Study findings have shown that the sensitivity of screening mammography is lower in women with primarily dense breasts (21); these findings are compatible with ours. The performance of the computer algorithm, on the other hand, was independent of breast density (Table 3), as might be expected given the geometric basis on which spiculation is detected (9–12). This result suggests that CAD has the capability to supplement the performance of radiologists in detection as they review mammograms of dense or heterogeneously dense breasts.

Spiculation was a common feature (375 [55%] of 677) in a large series of carcinomas that appeared primarily as masses on mammograms. Incorporation of spiculation measures, therefore, represents an important strategy in the automated detection of breast cancer with use of the computer. A present generation CAD algorithm correctly identified a large proportion (322 [86%] of 375) of spiculated breast cancers.


   Acknowledgments
 
The authors are grateful to Jimmy R. Roehrig, PhD, who has directed the development of the mass detection algorithm at R2 Technology since the inception of the company.


   Footnotes
 
C.J.V. is a stockholder in R2 Technology. T.D., K.F.O., H.M.R., A.C.S., and A.A.S. are employees of R2 Technology

Abbreviations: BI-RADS = Breast Imaging Reporting and Data System, CAD = computer-aided detection

Author contributions: Guarantors of integrity of entire study, C.J.V., K.F.O., A.A.S.; study concepts, C.J.V., A.A.S.; study design, A.A.S., K.F.O.; definition of intellectual content, C.J.V., A.A.S.; literature research, C.J.V., K.F.O.; clinical studies, K.F.O.; experimental studies, C.J.V., T.D., K.F.O., H.M.R., A.C.S.; data acquisition, T.D., K.F.O., H.M.R., A.C.S.; data analysis, C.J.V., T.D., K.F.O., H.M.R., A.C.S.; statistical analysis, K.F.O.; manuscript preparation and editing, C.J.V., K.F.O.; manuscript review, all authors.


   References
TOP
 Abstract
 Introduction
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 References
 

  1. Sickles EA. Breast masses: mammographic evaluation. Radiology 1989; 173:297-303.[Abstract/Free Full Text]
  2. D'Orsi CJ, Kopans DB. Mammographic feature analysis. Semin Roentgenol 1993; 28:204-230.[Medline]
  3. Tabar L, Dean PB. Teaching atlas of mammography 2nd ed. Stuttgart, Germany: Georg Thieme Verlag, 1985; 15-136.
  4. Franquet T. Spiculated lesions of the breast: mammographic-pathologic correlation. RadioGraphics 1993; 13:841-852.[Abstract]
  5. Kopans DB. Analyzing the mammogram. Breast imaging 2nd ed. Philadelphia, Pa: Lippincott–Raven, 1998; 302-304.
  6. Bird RE, Wallace TW, Yankaskas BC. Analysis of cancers missed at screening mammography. Radiology 1992; 184:613-617.[Abstract/Free Full Text]
  7. Burrell HC, Sibbering DM, Wilson ARM, et al. Screening interval breast cancers: mammographic features and prognostic factors. Radiology 1996; 199:811-817.[Abstract/Free Full Text]
  8. Birdwell RL, Ikeda DM, O'Shaughnessy KF, Sickles EA. Mammographic characteristics of 111 missed cancers later detected by screening mammography (abstr). Radiology 1999; 213(P):240.
  9. Kegelmeyer WP, Pruneda JM, Bourland PD, Hillis A, Riggs MW, Nipper ML. Computer-aided mammographic screening for spiculated lesions. Radiology 1994; 191:331-337.[Abstract/Free Full Text]
  10. Parr T, Astley S, Boggis C. The detection of stellate lesions in digital mammograms. In: Gale AG, Astley SM, Dance DR, Cairns AY, eds. Proceedings of the Second International Workshop on Digital Mammography. Amsterdam, the Netherlands: Elsevier Science, 1994; 231-239.
  11. Karssemeijer N. Recognition of stellate lesions in digital mammograms. In: Gale AG, Astley SM, Dance DR, Cairns AY, eds. Proceedings of the Second International Workshop on Digital Mammography. Amsterdam, the Netherlands: Elsevier Science, 1994; 211-219.
  12. Zhang M, Giger ML. Automated detection of spiculated lesions and architectural distortions in digitized mammograms. In: Loew MH, eds. Proceedings of medical imaging 1995: image processing. Vol 2434. Bellingham, Wash: SPIE, 1995; 846-854.
  13. te Brake GM, Karssemeijer N, Hendriks JHCL. Automated detection of breast carcinomas not detected in a screening program. Radiology 1998; 207:465-471.[Abstract/Free Full Text]
  14. Warren Burhenne LJ, Wood SA, D'Orsi CJ, et al. Potential contribution of computer-aided detection in the sensitivity of screening mammography. Radiology 2000; 215:554-562.[Abstract/Free Full Text]
  15. American College of Radiology. Illustrated Breast Imaging Reporting and Data System 3rd ed. Reston, Va: American College of Radiology, 1998; 167-175.
  16. Sickles EA. Mammographic features of 300 consecutive nonpalpable breast cancers. AJR Am J Roentgenol 1986; 146:661-663.[Abstract/Free Full Text]
  17. Orel SG, Kay N, Reynolds C, Sullivan DC. BI-RADS categorization as a predictor of malignancy. Radiology 1999; 211:845-850.[Abstract/Free Full Text]
  18. Baker JA, Kornguth PJ, Floyd CE. Breast Imaging Reporting and Data System standardized mammography lexicon: observer variability in lesion description. AJR Am J Roentgenol 1996; 166:773-778.[Abstract/Free Full Text]
  19. Harvey JA, Fajardo LL, Innis CA. Previous mammograms in patients with impalpable breast carcinoma: retrospective vs blinded interpretation. AJR Am J Roentgenol 1993; 161:1167-1172.[Abstract/Free Full Text]
  20. Schmidt RA, Nishikawa RM, Osnis RB, Schreibman KL, Giger ML, Doi K. Computerized detection of lesions missed by mammography. In: Doi K, Giger ML, Nishikawa RM, Schmidt RA, eds. Digital mammography ‘96. Amsterdam, the Netherlands: Elsevier Science, 1997; 105-110.
  21. Kerlikowske K, Grady D, Barclay J, Sickles EA, Ernster V. Effect of age, breast density, and family history on the sensitivity of first screening mammography. JAMA 1996; 276:33-38.[Abstract]



This article has been cited by other articles:


Home page
 RadiologyHome page
R. L. Ellis, A. A. Meade, M. A. Mathiason, K. M. Willison, and W. Logan-Young
Evaluation of Computer-aided Detection Systems in the Detection of Small Invasive Breast Carcinoma
Radiology, October 1, 2007; 245(1): 88 - 94.
[Abstract] [Full Text] [PDF]

Home page
 RadiologyHome page
H. Hricak, P. L. Choyke, S. C. Eberhardt, S. A. Leibel, and P. T. Scardino
Imaging Prostate Cancer: A Multidisciplinary Perspective
Radiology, April 1, 2007; 243(1): 28 - 53.
[Abstract] [Full Text] [PDF]

Home page
 RadiologyHome page
V. R. Pai, N. E. Gregory, A. E. Swinford, and M. Rebner
Ductal Carcinoma in Situ: Computer-aided Detection in Screening Mammography
Radiology, December 1, 2006; 241(3): 689 - 694.
[Abstract] [Full Text] [PDF]

Home page
 RadiologyHome page
M. J. Morton, D. H. Whaley, K. R. Brandt, and K. K. Amrami
Screening Mammograms: Interpretation with Computer-aided Detection--Prospective Evaluation
Radiology, May 1, 2006; 239(2): 375 - 383.
[Abstract] [Full Text] [PDF]

Home page
 Am. J. Roentgenol.Home page
J. A. Baker, E. L. Rosen, J. Y. Lo, E. I. Gimenez, R. Walsh, and M. S. Soo
Computer-Aided Detection (CAD) in Screening Mammography: Sensitivity of Commercial CAD Systems for Detecting Architectural Distortion
Am. J. Roentgenol., October 1, 2003; 181(4): 1083 - 1088.
[Abstract] [Full Text] [PDF]

This Article
Right arrow Abstract Freely available
Right arrow Figures Only
Right arrow Full Text (PDF)
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 Vyborny, C. J.
Right arrow Articles by Stein, A. A.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Vyborny, C. J.
Right arrow Articles by Stein, A. A.

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