| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Original Research |
1 Department of Radiology, Dartmouth-Hitchcock Medical Center, One Medical
Center Dr., Lebanon, NH 03756.
2 Department of Community and Family Medicine, Dartmouth Medical School,
Hanover, NH.
3 Norris Cotton Cancer Center, Lebanon, NH.
Received January 31, 2007;
accepted after revision April 19, 2007.
Address correspondence to S. P. Poplack
(steven.p.poplack{at}hitchcock.org).
Abstract
 Top Abstract IntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
MATERIALS AND METHODS. Women with an abnormal screening mammography were recruited sequentially. Consenting women underwent tomosynthesis of the affected breast corresponding to the views obtained with diagnostic mammography. The study radiologist compared the image quality, including lesion conspicuity and feature analysis, of tomosynthesis with diagnostic film-screen mammography and assessed the need for recall when tomosynthesis was added to digital screening mammography. Screening recalls were considered unnecessary when tomosynthesis did not show a corresponding abnormality or allowed definitely benign lesion characterization. Fisher's exact test was used to determine the association of equivalence and recall status with mammographic finding type.
RESULTS. There were 99 digital screening recalls in 98 women. The image quality of tomosynthesis was equivalent (n = 51) or superior (n = 37) to diagnostic mammography in 89% (88/99). Finding type was significantly (p < 0.001) associated with equivalence. Approximately half—52/99 (52%)—of the findings would not have been recalled when digital screening mammography was supplemented with tomosynthesis. When adjusting for confounding conditions, the recall reduction was 40% (37/92). The likelihood of recall was also dependent on finding type (p = 0.004).
CONCLUSION. Subjectively, tomosynthesis has comparable or superior image quality to that of film-screen mammography in the diagnostic setting, and it has the potential to decrease the recall rate when used adjunctively with digital screening mammography.
Keywords: breast • breast cancer • digital mammography • screening • tomosynthesis
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Tomosynthesis is a tomographic application of digital mammography. The physical principles of this technology have been described [4, 5]. In brief, the tomosynthesis acquisition mimics conventional mammography with regard to breast positioning and compression, but unlike conventional mammography, the X-ray tube takes multiple low-dose exposures as it moves through a limited (e.g., 30°) arc of motion. The resulting digital data set is reconstructed into tomographic sections through the breast in the orientation of acquisition—that is, craniocaudal, oblique, or 90° lateral.
We report an initial clinical experience with tomosynthesis in a cohort of 98 women with an abnormal digital screening mammography. The primary aim of our study was to assess the image quality of tomosynthesis compared with diagnostic film-screen mammography to learn more about the diagnostic imaging potential of this technology. We also provide initial estimates of the recall rate of tomosynthesis when used adjunctively with digital screening mammography.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
|
At the time of the diagnostic imaging evaluation, each woman underwent a tomosynthesis examination using a research prototype tomosynthesis unit (Genesis, Hologic) (Fig. 1) in up to three projections matched in orientation to the diagnostic mammography views. For example, if the diagnostic mammography evaluation included focal compression craniocaudal (CC) and mediolateral (ML) views, then tomosynthesis examinations were acquired in the CC and ML projections. No tomosynthesis equipment failure occurred. Screening mammography was acquired with full-field digital technology (Selenia, Hologic) and was interpreted in a soft-copy format, whereas diagnostic mammography was performed with film-screen technology (Lorad MIV, Hologic). This reflected the imaging algorithm of our clinical practice at the time of the study.
Core needle biopsy was performed with stereotactic guidance (Lorad Multicare Breast Biopsy System, Hologic) or sonographic guidance (HDI 5000, Philips Medical Systems), depending on lesion conspicuity. All stereotactic biopsies used vacuum-assisted needles (either Mammotome, Ethicon Endo-Surgery of Johnson-Johnson; or Suros, Pearl, or Sapphire, Suros Surgical Systems, Hologic); sonographic biopsies used either 14-gauge cutting needles (Max*Core, C. R. Bard) or handheld vacuum-assisted devices (InterV handheld needle, Medical Device Technologies [formerly Meditech]).
Seven certified mammography technologists were trained to perform tomosynthesis examinations by an application specialist before the initiation of the study. The same mammography technologist performed both diagnostic mammography and tomosynthesis. The tomosynthesis unit was equipped with a molybdenum target and molybdenum and rhodium filters. The tomosynthesis examination was acquired with a degree of breast compression similar to that of conventional mammography. Compression was obtained with a manual technique using standard compression paddles because of the prototypical nature of the tomosynthesis unit.
The tomosynthesis acquisition took approximately 19 seconds and included 11 low-dose exposures as the X-ray source moved in a 28° arc. The total radiation dose per acquisition was approximately 400 mrem (4 mSv) for a breast of average thickness, which is approximately twice the current dose per exposure of digital mammography, and historically is the maximum allowable dose per exposure of xeroradiography. Reconstruction into 1-mm-thick slices was performed offline (i.e., at a different time from the image acquisition) and took 30-60 seconds, depending on breast size. Tomosynthesis examinations were not available to the clinical radiologist at the time of diagnostic evaluation to avoid affecting clinical decision making. Technique factors for both diagnostic mammography and tomosynthesis were recorded for each acquisition by the technologist performing the examination.
Interpretation
The screening mammography interpretation that led to recall and also
determined eligibility for study inclusion was based on soft-copy evaluation
of digital mammography by one of seven clinical radiologists. Film-screen
diagnostic mammography was interpreted by one of six clinical radiologists.
Interpretation of the tomosynthesis and the diagnostic examinations was
performed by one of two study radiologists. The study radiologist reviewed the
tomosynthesis examinations, including the 11 source images and the 1-mm
reconstructions, on a prototype digital workstation. The first step in the
review process was a direct comparison of the tomosynthesis examination with
the film-screen diagnostic mammography. The study radiologist then viewed the
tomosynthesis examination in the context of screening and made a determination
of recall, both for the site that had been recalled clinically and for the
remainder of the breast. Tomosynthesis reconstructions were reviewed in cine
or manual scroll modes on a dedicated soft-copy workstation. No special
procedures were adopted to blind the interpreter to the clinical outcome.
The interpreter and coordinator recorded the subject's identifying information, date of screening and diagnostic examinations, breast side, breast composition, type and location of the screening abnormality, number of tomosynthesis acquisitions, and BI-RADS assessments, from the clinical radiologist's interpretation and the study radiologist's interpretation. Whereas diagnostic mammography assessments reflected the actual clinical assessment rendered at the time of diagnostic imaging, including input from comparison mammography and breast sonography, the tomosynthesis assessments were based solely on the tomosynthesis examination.
The study mammographer subjectively rated the equivalence (i.e., better, equivalent, or worse) of the image quality of the tomosynthesis examination compared with the diagnostic mammography examination. Equivalence was based on lesion conspicuity and feature analysis specific to the finding type. For example, if the margins of a mass were clearly more visible with tomosynthesis, then tomosynthesis was rated superior to diagnostic mammography; if magnification mammography showed more calcifications in a cluster, then diagnostic mammography was rated superior to tomosynthesis. Image quality was rated equivalent if the comparative benefit was questionable or marginal. This comparison was restricted to the diagnostic mammography component of the diagnostic imaging evaluation and did not include an evaluation of sonography.
The study radiologist also assessed whether the tomosynthesis examination would have prompted a recall of the primary screening abnormality and whether a recall would have been indicated for additional findings detected solely with tomosynthesis. When "no recall" from tomosynthesis was determined, the study radiologist provided justification, including "no abnormality seen," "threshold differences," "positioning," or "better lesion characterization." "Positioning" refers to differences in breast positioning between the tomosynthesis examination and the screening mammography examination. In some instances, the study radiologist did not think a recall was warranted on the basis of the screening mammogram, reflecting a difference in interpretation between the clinical and study radiologists rather than a difference between technologies. This variation in the threshold for recall between the study radiologist and the clinical radiologist was termed "recall threshold differences."
|
|
|
|
Statistical Analysis
Data were subsequently entered via electronic forms into a relational
database (FileMaker Pro 7, FileMaker) and reformatted for statistical analysis
using S-PLUS (version 6.2.1, Insightful).
Preliminary analysis included descriptive statistics summarizing the demographic and clinical characteristics of the patient group. Cross-tabulations were formed to investigate possible differences in equivalence ratings and recall frequency by finding type. Due to the small cell counts in the cross-tabulations, exact methods were used to evaluate p values for Fisher's exact test for association using the S-PLUS function "Fisher.test." A p value of less than 0.05 was deemed statistically significant.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Comparison with Diagnostic Mammography
The comparison analysis of tomosynthesis with diagnostic mammography is
shown according to finding type in Table
1. The image quality of tomosynthesis was subjectively rated as
equivalent (52%, 51/99) or superior (37%, 37/99) to diagnostic mammography in
89% (88/99) of cases; the test that this proportion was greater than 50% was
highly significant (p < 0.001). Overall, the test for the
association of equivalence ratings with finding type in
Table 1 was highly significant
(p < 0.001). For masses, tomosynthesis image quality was rated as
equivalent in 26% (5/19) or superior (to diagnostic mammography) in 68%
(13/19) of cases. Masses constituted 19% (19/99) of screening-detected
findings but were 35% (13/37) of findings in which tomosynthesis had superior
image quality. In contrast, calcifications usually showed better image quality
on diagnostic mammography, including focal magnification mammography, than on
tomosynthesis.
|
A representative case depicting superior image quality of focal magnification film-screen mammography compared with tomosynthesis is shown in Figure 2A, 2B, 2C, 2D. Calcifications accounted for 73% (8/11) of the cases in which tomosynthesis had inferior image quality. In the one patient with two screening abnormalities in the same breast, one of which was excluded from the analysis, the excluded finding was deemed equivalent. Both tomosynthesis and diagnostic imaging assessed the excluded site as negative (BI-RADS category 1).
The diagnostic mammography and tomosynthesis assessments of the 99 findings recalled from screening mammography are displayed in Table 2. Although all assessments from diagnostic mammography were complete, some of these assessments were based on contributory data from breast sonography and comparison with prior mammography. Of the 26 findings that were assessed as incomplete (BI-RADS category 0) at tomosynthesis, sonography was recommended in half (13/26), magnification mammography was recommended in two, and comparison with prior mammography was requested in one. The remaining 10 cases had recommendations for multiple ancillary tests including some combination of sonography, additional mammography, magnification views, and comparison mammography.
|
Recall Status
The use of tomosynthesis as an adjunct to digital screening mammography led
to a decrease in the recall rate by 52% (52/99). However, in seven subjects
recalled by the clinical radiologist, the study radiologist would not have
recalled the subject on the basis of the screening mammogram. This was
attributed to recall threshold differences between interpreters rather than to
technologic differences. Eight additional recalls of a different site stemmed
from the tomosynthesis examination. Because of the investigative nature of the
tomosynthesis technique, recalls originating from the tomosynthesis
examination were evaluated with additional imaging only if also noted in
retrospect on conventional mammography and deemed worthy of evaluation. There
were no instances of abnormalities that were suspicious or highly suggestive
of malignancy (BI-RADS category 4 or 5) seen only on the tomosynthesis
examination. When adjusting for the fewer overall number of recalls due to
reviewer threshold differences and the additional tomosynthesis-prompted
recalls, the recall rate reduction was 40% (37/92).
The status and clinical outcomes of adjunctive tomosynthesis recall recommendations are listed by finding type in Table 3. The likelihood of a tomosynthesis recall was finding-dependent (p = 0.004). Tomosynthesis was less likely to indicate recall for screening-detected findings of asymmetry, focal asymmetry, and possible architectural distortion than for masses and calcifications. Figure 3 diagrams the conditions for no tomosynthesis recall for the 52 screening recalls. The absence of an abnormality with tomosynthesis (i.e., "no abnormality seen" [Fig. 3]) was the most common reason for no recall and reflected tissue overlap in most cases. In one case, a difference in breast positioning that allowed the inclusion of farther-posterior tissue in the tomosynthesis examination accounted for the lack of recall. Improved characterization of definitely benign findings was the second most common reason for the lack of tomosynthesis recall.
|
|
Clinical and Pathology Outcomes
Of the 98 participants, 83% (81/98) had pathologic or imaging follow-up,
and 17% (17/98) were lost to follow-up (Tables
2 and
3). Sixty-two subjects had
imaging follow-up within a median of 15 months (mean, 12.7 months; range, 6-24
months). None of the follow-up imaging was reported as suspicious or highly
suggestive of malignancy, and no biopsy recommendations were made in the
follow-up group. Of the 17 subjects lost to follow-up, one died from ovarian
cancer and one moved out of state. One of the subjects lost to follow-up had a
probably benign diagnostic mammography assessment, whereas the remaining 16
had negative or benign diagnostic mammography assessments.
All 19 subjects recommended for biopsy underwent imaging-guided core needle biopsy, which showed malignancy in four subjects and benign results in 15 (Table 4). Two of the 15 benign pathology results had atypical histology, including one patient with a papillary neoplasm with cytologic atypia and one patient with atypical ductal hyperplasia. Both subjects underwent subsequent surgical excisional biopsy that showed no evidence of malignancy and no residual atypia. One subject recalled for asymmetry at digital screening mammography, who proceeded to biopsy on the basis of diagnostic mammography findings, was assessed with tomosynthesis as BI-RADS category 1 and would not have been recalled or undergone biopsy on the basis of tomosynthesis.
|
There were five malignancies in four women. Three of the five malignancies were detected with both screening mammography and tomosynthesis, and two were occult on both techniques. In two of the three malignancies visible with both techniques, tomosynthesis and diagnostic mammography had identical assessments (BI-RADS categories 4 and 5) and were rated equivalent. Figure 4A, 4B, 4C shows representative diagnostic mammography, tomosynthesis, and pathology images of one of these cases, an 11-mm invasive lobular carcinoma. The third detected cancer was considered suspicious (BI-RADS category 4) at diagnostic mammography but was assessed on tomosynthesis as BI-RADS category 0, with a recommendation for additional evaluation and comparison with prior mammography. The additional evaluation recommended by tomosynthesis was performed at the time of diagnostic mammography and contributed to the diagnostic mammography assessment.
|
|
|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
In contrast, the image quality of tomosynthesis was inferior to diagnostic mammography primarily in the characterization of calcifications. Diagnostic mammography (i.e., standard and focal magnification views) out-performed tomosynthesis on the basis of better conspicuity of calcifications and better discrimination of particle number and morphology. Two limitations of the tomosynthesis system used in this study may account for these deficits. The lengthy exposure time (19 seconds) of the tomosynthesis acquisition may have introduced motion-related blur, obscuring additional particles and particle morphology. (Technologic modifications on subsequent tomosynthesis prototypes have reduced acquisition times by approximately half, which may mitigate this effect.) Second, in this study, tomosynthesis acquisitions were reconstructed at a 1-mm-equivalent slice thickness, which may have been too thin to show the clustered distribution of calcifications. Reconstruction thickness can be varied and may ultimately be tailored to the preference of the clinical radiologist or optimized for specific functions, such as a thicker "slab mode" reconstruction for the detection and characterization of calcifications.
Recall Status
When used adjunctively with digital screening in this cohort of 98 women,
tomosynthesis would have decreased the recall rate by nearly half. In an
attempt to mitigate bias, we also calculated the recall reduction after
adjusting for additional recalls from tomosynthesis (i.e., second site) and
for recall threshold differences between the study radiologist and the
clinical radiologist. Although some practices may use independent
double-interpretation, one of the study aims was to estimate the effect of
tomosynthesis on recall rate when used adjunctively with digital screening
mammography by a single observer. After these adjustments, the recall
reduction was 40%.
The observed decrease in recall rate was mostly due to recognition of tissue overlap, because no abnormality was seen with tomosynthesis in those cases. This result is reflected by the statistically significant association of finding type with the likelihood of tomosynthesis recall; vague findings such as asymmetry, focal asymmetry, and questionable architectural distortion were less likely to be recalled by tomosynthesis than discrete findings such as calcifications or masses. Less commonly, recalls were avoided with tomosynthesis because the recall "abnormality" could be definitively characterized as benign (as in usual mammography practice), implying that overlapping tissue had obscured a characteristically benign feature—for example, fat in an intramammary lymph node. However, if we consider the additional recalls generated by tomosynthesis, and if flaws in the study design (discussed in the following text) account for some of the recall reduction, then tomosynthesis may have a negligible effect or even increase the recall rate. Additional and more rigorous studies are needed to confirm a genuine recall reduction from the adjunctive use of tomosynthesis.
Cancer Evaluation
The effect of tomosynthesis on cancer detection and characterization cannot
be quantitatively determined because the study was not designed or empowered
to evaluate this objective. On the basis of BI-RADS assessment categorization,
mammography appeared to outperform tomosynthesis by detecting and correctly
characterizing a 15-mm mucinous carcinoma that was not visible with
tomosynthesis (Table 4). In
reality, the initial (screening mammography) recalled finding in this subject
was a benign cyst. The malignancy was detected serendipitously by sonography
of another site in the breast, and diagnostic mammography was
noncontributory.
In two malignant cases—an approximately 1-cm invasive ductal carcinoma (BI-RADS category 5) and a 1.1-cm invasive lobular carcinoma (BI-RADS category 4) that had identical mammography and tomosynthesis assessments—margin spiculation was shown more conspicuously with tomosynthesis than with diagnostic mammography, but the magnitude of the improvement did not warrant a superior image quality rating. Although tomosynthesis may have enabled a more confident diagnosis in these cases, mammography showed the key features (i.e., margin characteristics and mass shape) adequately. In the case of the invasive lobular carcinoma, tomosynthesis also showed fat in the malignant mass, which was later confirmed at histo-pathologic analysis (Fig. 4A, 4B, 4C).
Limitations
Several limitations in the study design may have biased these results. The
tomosynthesis platform was this vendor's first clinical prototype, and some
technologic enhancements, such as automated compression or differential
compression paddles, were not available with tomosynthesis, which may have
degraded the quality of the tomosynthesis acquisition. The tomosynthesis
assessment did not have the full complement of data that were available to
diagnostic mammography, such as comparison mammography information and
sonography results.
Diagnostic mammography was performed with the film-screen technique, in accordance with our clinical practice at that time. It thereby remains to be studied how the image quality of tomosynthesis compares with that of digital diagnostic mammography. In addition, although tomosynthesis acquisitions were oriented similarly with respect to diagnostic views, they were not identical. For example, if a focal spot compression in the CC view was obtained diagnostically, a full CC tomosynthesis view was obtained for comparison.
Interpretation was performed by one of two study radiologists in an unblinded fashion. Although most interpretations were rendered without knowledge of the diagnostic or pathology outcome, two of the interpretations may have been biased by this information. This potential source of bias is unlikely to have had a major effect on the diagnostic equivalence results because diagnostic mammography and tomosynthesis were considered equivalent in both of these cases. The image quality comparison between tomosynthesis and diagnostic mammography was a subjective judgment that may have been influenced by the reviewers' bias either for or against tomosynthesis.
The reduction in recall rate from superim-position may reflect a flaw in the study design that is related to examination sequencing and is due to a repositioning effect rather than a technologic advancement. In this study, tomosynthesis was performed after digital screening and can be viewed as a repeat examination. In clinical practice, repeating the identical mammography view that prompted recall from screening will frequently resolve asymmetry due to tissue overlap. If this repeat-view phenomenon was responsible for the lack of tomosynthesis recall in the 32 cases in which no abnormality was seen with tomosynthesis, then tomosynthesis may not provide the magnitude of benefit in recall reduction that these data estimate.
It also remains to be seen how these data will generalize to other patient populations and clinical practice circumstances. Tomosynthesis may perform differently in a general screening population than in the study group of women who were recruited because of a screening abnormality. In this cohort, only 5% (5/98) of women had fatty breast composition; the remainder, 95% (93/98), had higher density compositions. In a population with a greater proportion of fatty breast composition, tomosynthesis may have a lesser effect [4].
Conclusion
In conclusion, the results of this study are promising but must be viewed
cautiously because of limitations of the study design. Based on a subjective
analysis of film quality, tomosynthesis appears to offer advantages over
diagnostic film-screen mammography: It showed superior image quality compared
with diagnostic mammography for all finding types in aggregate, and especially
in the characterization of masses and asymmetry. It was less effective in the
evaluation of calcifications, an outcome that may be mitigated by future
technologic enhancements.
Tomosynthesis may also find application as an adjunct to screening mammography because of its potential to decrease recall rate. The capacity of tomosynthesis for cancer detection—that is, sensitivity—requires additional study. Further investigation with a more rigorous study design, including larger accrual targets with more cancer outcomes, alternating examination acquisition between screening mammography and tomosynthesis, and blinded interpretations by multiple observers, is needed to confirm the ultimate role of tomosynthesis in breast imaging.
Acknowledgments
Special thanks are extended to Andy Smith of Hologic and Wendy A. Wells of
Dartmouth-Hitchcock Medical Center (DHMC) for providing images, to Loren
Niklason of Hologic for assistance with the technical specifications of
tomosynthesis, to Trina L. Janeczek of DHMC for database entry and management,
and to Robyn E. Mosher of DHMC for editorial assistance.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
CiteULike 
Complore 
Connotea 
Del.icio.us 
Digg 
Reddit 
Technorati What's this?
This article has been cited by other articles:
|
|
 |
|
  P. BALDWIN Digital Breast Tomosynthesis Radiol. Technol., September 1, 2009; 81(1): 57M - 74M. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 D. Gur, G. S. Abrams, D. M. Chough, M. A. Ganott, C. M. Hakim, R. L. Perrin, G. Y. Rathfon, J. H. Sumkin, M. L. Zuley, and A. I. Bandos Digital Breast Tomosynthesis: Observer Performance Study Am. J. Roentgenol., August 1, 2009; 193(2): 586 - 591. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
W. F. Good, G. S. Abrams, V. J. Catullo, D. M. Chough, M. A. Ganott, C. M. Hakim, and D. Gur Digital Breast Tomosynthesis: A Pilot Observer Study Am. J. Roentgenol., April 1, 2008; 190(4): 865 - 869. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
D. Gur Tomosynthesis: Potential Clinical Role in Breast Imaging Am. J. Roentgenol., September 1, 2007; 189(3): 614 - 615. [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
