Radiation Protection Dosimetry Advance Access published April 5, 2015 Radiation Protection Dosimetry (2015), pp. 1–7
doi:10.1093/rpd/ncv078
DIGITAL BREAST TOMOSYNTHESIS IN ONE OR TWO VIEWS AS A REPLACEMENT OR ADJUNCT TECHNIQUE TO FULL-FIELD DIGITAL MAMMOGRAPHY T. M. Svahn* and N. Houssami School of Public Health, Sydney Medical School, University of Sydney, Sydney 2006, NSW, Australia *Corresponding author: [email protected]
INTRODUCTION Breast cancer is one of the most common cancer types among women in western countries. Risk factors include reproductive factors (e.g. non-parity, late first birth, early menarche and late menopause), hormone replacement therapy, genetic factors, ionising radiation and high breast density on mammography. There is also evidence that life style, such as high alcohol consumption and low physical activity, may be associated with increased risk. Mammography has gained wide acceptance because of its ability to visualise non-palpable early-stage changes in the breast, which at screening has shown to reduce mortality by 20–30 %(1). The evolution from 2D mammography to 3D (digital breast tomosynthesis, DBT) reduces the essential problems of over-projected anatomy, for example, effect of tissue overlap and the masking effect of tissue density. At present, DBT is increasingly being used as a diagnostic tool, though, not yet considered the standard of care for breast cancer screening. Recent clinical trials embedded in population screening programmes have shown their immense potential in screening when interpreted in two views together with full-field digital mammography (FFDM), but promising results have also been reported when used as a stand-alone technique. Reported breast cancer detection rates have shown 30– 50 % increases in relation to that of FFDM, usually at reduced recall rates(2, 3). In a randomised 2D screen-film mammography trial, a two-view protocol was found to detect 24 % more women with breast cancer at a reduced recall rate by 15 % when compared with a one-view
protocol(4). Significant technological advancements have been made since then, but the problem of superimposed tissues owing to the 2D nature of the technique (FFDM) remains. Because DBT is a 3D imaging device, it has been hypothesised that additional views to the mediolateral oblique (MLO) projection would not be needed, but as depth resolution is limited due to constraints present in the image sampling(5), the use of two views to acquire DBT might be advantageous. In clinical studies using mammogram sets with ascertained outcomes (often referred to as observer performance studies, or reader studies), a variety of imaging protocols have been examined. In this paper, a review of such studies was performed, in addition to preliminary data assessed from screening trials with respect to imaging protocol to determine whether performance-related trends or patterns are apparent. MATERIALS AND METHODS Review and applied accuracy measures A literature search was performed on clinical studies reporting on breast cancer detection that compared the diagnostic tests using DBT and FFDM (PubMed search: July 2008 to September 2014, in addition to two proceedings papers; IWDM, SPIE and two ECRpresentations; literature search was performed by T.S.). The literature was restricted to studies where interpretation was done blinded to if the patient had cancer or not, with case mixtures consisting of abnormal and normal/benign findings in a retrospective design. A
# The Author 2015. Published by Oxford University Press. All rights reserved. For Permissions, please email: [email protected]
Downloaded from http://rpd.oxfordjournals.org/ at University of Winnipeg on September 1, 2015
Clinical studies using different imaging protocols to perform digital breast tomosynthesis (DBT) were reviewed (2008–14) to assess interpretive accuracy. Descriptive pooled statistics were used to estimate and summarise accuracy measures for each type of imaging protocol in relation to that of two-view full-field digital mammography (FFDM). In studies comparing multiple DBT imaging protocols, a trend of increased performance was often seen when including both the mediolateral oblique and craniocaudal views for DBT alone and even more so for DBT adjunct to FFDM. Overall, the average DAUC (%; sd) across studies for stand-alone DBT (relative to FFDM), in one and in two views, were 2.2 (+ + 3.7) and 5.9 (+ + 4.6), and when used together with FFDM, 3.9 (+ + 2.0) and 6.7 (+ + 0.9). With respect to individual studies, improvements in accuracy using DBT were present for different types of imaging protocols although the magnitude of the impact varied between studies, and some studies did not show significant improvements in comparison with FFDM. The most consistent effect of improvement in breast cancer detection was seen across studies for two-view DBT with FFDM. These summary findings may depend on the sampling constraints present in tomosynthesis imaging and on other factors discussed in this paper. In order to investigate these effects more thoroughly and how they might impact outcomes, comparative or randomized-controlled trials are warranted.
T. M. SVAHN AND N. HOUSSAMI
false positives (Figure 1). All five radiologists performed best in the combined setting. Similarly, another study found that DBT MLO þ FFDM improved accuracy significantly (relative FFDM). However, this study also evaluated two-view DBT þ FFDM, which doubled the increase in sensitivity and in overall accuracy (relative FFDM), while further reducing the recall rate(14). The difference between the two DBT imaging protocols was statistically significant. The benefits were found ‘not’ to correlate with radiologists´ experience levels, hence suggested to apply to a vast majority of radiologists, which has been validated in screening trials. In the study by Thibault et al. (15) various imaging protocols, including conventional FFDM, resulted in a similar accuracy. Sensitivity markedly increased for protocols including ultrasonography but was counterbalanced by a loss in specificity. A high proportion of cancers were not detected by single imaging techniques(15). Pooled accuracy across studies by DBT imaging protocol For stand-alone DBT, the average DAUC (%; sd) across studies performed in one and in two views, respectively, were 2.2 (+3.7) and 5.9 (+4.6), and when used adjunct to FFDM in one and in two views, 3.9 (+2.0) and 6.7 (+0.9) (Table 1A,B). The average improvement in sensitivity and specificity increased for DBT of both views used adjunct to FFDM relative to
RESULTS Studies reported in 2008–14 Multiple imaging protocols in paired study designs In many of the studies, there is a gradual increase in performance present with additional imaging information. Yet, a number of studies only found significant improvements in breast cancer detection for the DBT imaging protocol associated with ‘most’ imaging information (relative FFDM), covering ‘both’ MLO and CC views, stand-alone or adjunct to FFDM(7, 12 – 14). In one of the studies, performance was highest for a combined setting of DBT MLO þ FFDM CC when compared with stand-alone one-view DBT, and FFDM(7), associated with more breast cancers detected at fewer
Figure 1. The accuracy (area under the curve) was highest for DBT MLO þ FFDM CC, 5 % higher than one-view DBT and 11 % higher than FFDM alone, with more breast cancers localised and identified (LLF) at a fewer false positives (FPF) as described by the endpoints of the solid lines of the AFROC plot (7).
Page 2 of 7
Downloaded from http://rpd.oxfordjournals.org/ at University of Winnipeg on September 1, 2015
common study type used is where several radiologists individually interpret all images for each modality under investigation, the multiple-reader multiple-case (MRMC) design, which is done to maximise the sensitivity of detecting differences between imaging modalities (statistical power). Unlike other studies included, the TOMMY trial is population based using consecutive sampling of FFDM-recalled cases at screening. To assess whether there are trends or consistent patterns apparent, information was extracted on sensitivity and specificity, diagnostic accuracy and recall rate for assessment. As sensitivity and specificity are anti-correlated and depend on the decision threshold of the radiologist, it is valuable to use measures that combine the effects seen on both, denoted diagnostic accuracy. Diagnostic accuracy was estimated by the area under the ROC curve (AUC) or by area under the alternative free-response ROC curve (JAFROC)(6). The main difference between the methods is that ROC considers the cases as a whole (e.g. if they are abnormal or normal/benign), whereas the free-response method also considers locations of individual breast cancers, thereby providing increased realism and power(6). The AUC was here used as a unified measure across studies (inferred AUC was estimated from freeresponse ROC data(6 – 10)). AUC can be defined as the average sensitivity value for all possible values of specificities(11). Note that, as a stand-alone technique, FFDM refers to interpretation of both MLO and CC views and is hence ‘always’ the baseline modality that the DBT modality is compared with. The estimated accuracy changes were obtained by subtracting results from the DBT modality with that of FFDM (for instance, DAUC ¼ AUCDBTmodality AUCFFDM ), which were pooled across studies by imaging protocol. Cancer detection rates and recall or false-positive rates were assessed from screening trials. At first, results obtained from studies where several different imaging protocols were investigated based on paired study designs, mostly the MRMC type, are summarised.
DIGITAL BREAST TOMOSYNTHESIS IN ONE OR TWO VIEWS Table 1. Clinical studies using digital breast tomosynthesis alone (DBT; A), in one and in two views, and as adjunct to two-view digital mammography (FFDM; B) in comparison with standard two-view FFDM. Study
Subjects Radiologists (abnormal)
DAUC (%, sd)
D (sensitivity/ specificity)
Benign/normals (false recalls)
Cancer cases
(211.2)
(210.5)
212.5 210.7 269.7
21.2 219.3 252.5
231
224
22.6 210
þ3.3 þ5.7
(211.2)
(0)
217.5 25 28.8
21.2 21.2 þ2.9
272.7 238.5*
268.3 23.9
255.6
234.2
242.6* 230*
þ5.7
269.7* 238.3* 245.7*
27.8 þ1.1 þ1.0
245.3
0
The results of the studies are presented in reader-averaged difference, increase (þ) or decrease (2), in tomosynthesis performance relative to two-view FFDM. The change in recall rate (%) is relative to two-view FFDM; (RRDBT modality 2 RRFFDM)/RRFFDM. Significance is indicated by an asterisk (*). a Pooled average estimates are applied (e.g. not adjusted/weighted for study size). b The studies by Wallis et al. and Zanca et al. are overlapping; hence, data points were only included in the pooled avg. estimates from the study by Zanca et al. c Screen-film mammography (SFM) was also included in this comparison. The results are thus presented from when the readers had read SFM and FFDM, in comparison with those from when they had read SFM, FFDM and DBT. d The studies by Svahn et al.(7), Gennaro et al.(20), Thibault et al.(21) and Shin et al.(22) evaluated DBT MLO view plus FFDM CC, whereas Rafferty et al.(14) and Waldherr et al.(19) evaluated DBT MLO plus two-view FFDM.
Page 3 of 7
Downloaded from http://rpd.oxfordjournals.org/ at University of Winnipeg on September 1, 2015
A. Studies comparing stand-alone DBT (as a replacement to FFDM) versus FFDM One-view DBT 376 (63) 6 þ1.5 24.5/þ4.1 Gennaro et al.(16) Svane et al.(17) 144 (76) 2 20.07 23.9/þ5.1 (8) 185 (89) 5 þ9.4* þ10.8*/20.8 Svahn et al. Wallis et al. (12) 130 (40) 10 (þ0.1)b Zanca et al.(18) Exp. readers 130 (40) 5 þ1 Inexp. readers 5 21 144 (86) 2 (in con.) þ14*/þ2.1 Waldherr et al. (19) 130 (55) 7 þ2.3 27/þ11 Thibault et al.(15) a Average 208 (68) 6 þ2.2 (+3.7) þ1.9/þ4.3 Two-view DBT 30 (25) 9 þ2 Good et al.(10) 125 (35) 6 þ5/þ4 Gur et al.(13) (9) 501 (111) 1 (21.5; 0/21.7 Teertstra et al. classical metric) 130 (40) 10 (þ7.9*)b Wallis et al.(12) (18) Zanca et al. Exp. readers 130 (40) 5 þ4.7 Inexp. readers 5 þ11.0* a Average 197 (53) 6 þ5.9 (+4.6) þ2.5/þ1.2 B. Studies comparing DBT adjunct with FFDM versus FFDM d One-view DBT Svahn et al.(7) 50 (25) 5 þ7.1 þ9.6/22 469 (68) 6 þ2.1 þ3.4/þ1.9 Gennaro et al. (20) (21) 130 (55) 7 þ2.4 25/þ11 Thibault et al. 144 (86) 2 (in con.) þ17.5*/21.4 Waldherr et al.(19) Rafferty et al.(14) 310 (51) 15 þ3.6* þ8.7*/20.2 149 (102) 3 þ4.1* þ7.1*/23.4 Shin et al.(22) a Average 209 (65) 6 þ3.9 (+2.0) þ6.9/þ0.4 Two-view DBT 316 (48) 12 þ7.1* Smith et al.(23) 125 (30) 8 þ5/þ12 Gur et al.(13) (24) c 738 (204) 6 þ7.2* Michell et al. (25) 312 (48) 12 þ7.2* þ10.7/þ5.1 Rafferty et al. Rafferty et al. (25) 312 (51) 15 þ7.1* þ16*/21.7 310 (51) 15 þ6.8* þ16*/21.7 Rafferty et al.(14) (3) 5224 (1087) 3 þ5* þ3*/þ12* Gilbert et al. TOMMY trial a Average 1048 (217) 10 þ6.7 (+0.9) þ10.1/þ6.5
Relative change in recall rate (%)
T. M. SVAHN AND N. HOUSSAMI
other imaging protocols (was rarely reported for twoview DBT; Table 1A). The reduction of false recall rates was highest when DBT was combined with FFDM. All screening trials evaluating two-view DBT þ FFDM were associated with improvements over FFDM both in cancer detection rates (10 –53 %) and in reduced recall rates or false-positive rates (15– 38 %). Stand-alone one-view DBT (MBTST) was associated with an increase in cancer detection rate by 43 % at an equally increased recall rate by 43 %. DISCUSSION
Page 4 of 7
Downloaded from http://rpd.oxfordjournals.org/ at University of Winnipeg on September 1, 2015
The clinical application of DBT is currently evolving. Therefore, it seems timely to present a review of the findings from clinical studies using DBT in various imaging protocols. As indicated, results vary across studies but some findings are evident; in particular, increased accuracy was generally seen in studies that added DBT to FFDM and more so when two-view DBT was used (as shown in the summary evidence tables). For instance, the improvement in avg. sensitivity was about five times higher for two-view DBT adjunct to FFDM in comparison with that of stand-alone one-view DBT at a slight increase in avg. specificity. In screening trials, two-view DBT þ FFDM was always associated with an increase in cancer detection rate at a reduced recall rate or FP rate, whereas at one-view DBT cancer detection rates increased as much as the recall rate, relative to that of FFDM. There are several possible explanations for the results. DBT images are collected within a limited angular range and reconstructed into pseudo-tomographic images with partial blurring outside a selected plane and, hence, effects from overlapping tissues (anatomical noise) are not completely eliminated(5). In addition, an additional 2D examination may yield a different compression clearing some possible tissue superimpositions, as well as another dose delivery, and hence a different quantum noise pattern. However, this problem was usually not treated by the papers reviewed, but may have limited impact if acquisitions are performed under the same compression (i.e. 2D & 3D - same view)(31, 32). Owing to the incomplete elimination of anatomical noise, a cancer can be undetected or misinterpreted. Breast cancers such as small spiculated tumours might preferentially be oriented in one plane and difficult to identify in other planes. As known in 2D mammography, lobular carcinomas are usually best visualised in the CC projection(33), which seem to apply for DBT as well(34). In the same manner, the presence of anatomical noise might adversely affect the radiologists’ ability to confirm (and hence dismiss) normal/ benign cases. Another reason for the greater stability (sd) seen for DBT adjunct to FFDM might be in calcification detection. FFDM is the current reference standard for detection of calcifications, and studies have reported lower image quality or partially poorer
interpretation for them at DBT(8, 35, 36). This might depend on motion artefacts caused by the longer DBT acquisition time compared with FFDM; for example, due to patient movement between projections, blur associated with the focal spot movement (in continuous movement machines) or the reduced spatial resolution at acquisition (e.g. from binning of pixels)(37). Effects from these factors such as unsharpness of details may influence morphology-based decisions but can be solved by technical progress(38). In addition, tomosynthesis reduces tissue overlap, structural noise and thus helps to better show breast masses. Microcalcifications are, on the other hand, less affected by it and may in that way be more appropriately perceived by 2D views, which enable an overview of calcifications that are spread out in the breast at various depths. The use of synthetic 2D views, thicker slice thicknesses, for example, ‘slabbing’ and other developed methods,(39) can help the radiologist to maintain the accuracy of calcification detection and may ease the comparison with prior years’ screening images. Besides the influence of image protocol used, there are other possible explanations for different results in studies, as discussed by Houssami et al. (40) An important aspect is case difficulty. Cases that are more likely detectable by only one modality are essential to show differences between medical imaging devices, if any, but having higher occurrences of them in relation to the general population could result in enhanced effects(41, 42). Spectrum effects(42) may be pronounced when comparing a per-view more powerful technique (DBT) with another one (FFDM); and relevant to consider if DBT is acquired (for one-view DBT studies) only in the view where the abnormality was most visible at FFDM, or if case selection is mainly determined by difficulty level at FFDM. In order to obtain generalisable results, it is crucial that the case sampling mirror the usual spectrum of findings from the target population, hence allowing both comparative techniques to demonstrate their full detection potential. The pooled accuracy across studies can even out such effects (Table 1A,B), if present, and thus provide a more reliable average, though system-specific differences are not explicitly being addressed. In general, the fact that so many studies have found substantial benefits should be regarded as highly promising and would justify larger service-based trials. The publication of several large screening trials has in recent years provided insight of the clinical impact of DBT (Table 2). These have been conducted using different designs, with and without double reading and arbitration management. With regard to imaging protocols, only two different approaches have been investigated: two-view DBT plus FFDM(3, 26 – 28, 43, 44) and stand-alone one-view DBT (MBTST)(30). All trials show substantial improvements in breast cancer detection (Table 2). Stand-alone one-view DBT (MBTST) was, however, also associated with a 43 % significant
DIGITAL BREAST TOMOSYNTHESIS IN ONE OR TWO VIEWS Table 2. The results of screening trials in breast cancer detection rate, recall rate or false-positive rate. Studya
12501 7292 13174 (7058/6116) 23355 (13856/9499) (281187/173663) 7500
6.1 5.3 5.2 4 4.2 6.3
Recall rate
4.4 11.7 8.7 10.7 2.6
b Falsepositive rate per 1000 women
Detection per 1000 women
61
8.0 (þ27*) 8.1 (þ53*) 5.7 (þ9.6) 5.4 (þ35) 5.4 (þ29) 8.9 (þ43*)
Recall rate
b Falsepositive rate per 1000 women
53 (215*) 3.5 (220*) 8.4 (228.2*) 5.5 (237.5*) 9.1 (215) 3.8 (þ43*)
The Malmo¨ (MBTST) trial compares stand-alone one-view DBT relative FFDM, whereas the other trials compare two-view DBT adjunct to FFDM relative FFDM. The numbers in the parentheses show the percentage increase or decrease in performance measure relative to that of FFDM. Statistical significance is indicated with an asterisk. a The trials performed at Yale University and at Houston Breast Center are based on a retrospective study design using nonpaired data with single independent reading. The OTST, STORM and MBTST trials are prospective using paired data. b False-positive rate is presented before arbitration management.
increase in recall rate, which results in a slightly better positive predictive value for FFDM (PPV; cancers detected/100 recalls), and a comparable TP:FP ratio between the techniques as often seen for that imaging protocol (Table 1A). Other screening trials have often been associated with substantial PPV increases of 50 to 110 % for two-view DBT plus FFDM. It should be noted that the presented findings (Table 2) partially comprise interim analyses, or preliminary results presented at international conferences. Hence, the final results of some of the studies have not yet been published. The authors’ pooled analysis was not adjusted/ weighted for population size; hence, it is not a studyadjusted meta-analysis. A meta-analysis would generally use sensitivity and specificity pairs without including the rating data. By using the DAUC, the impact of rating data is being acknowledged. Variable thresholds were partially used for reporting sensitivity and specificity, which could even out between the different groups but is avoided in the DAUC. In studies, where several imaging protocols were investigated based on the same populations of cases and readers (MRMC), the comparisons were consistent. Radiation dose typically increased with the number of views and is described more elsewhere(45), but was not addressed in relation to performance. MRMC studies are sensitive at detecting differences between the performances of modalities. In general, they subject only a limited number of patients to radiation dose. However, they do not necessarily declare where the clinical operating point is, but (AUCs) have however shown to be robust to effects of cancer prevalence(46 – 48). Another potential limitation is that the studies are performed under experimental settings with no direct link between radiologist decision and patients health or outcome. Interestingly, the operating points of the TOMMY
trial were within the range of those from other studies of the same imaging protocol, with the same ‘total’ accuracy increments in sensitivity and specificity (Sn þ Sp 14–17 %), indicating that valuable estimates have been obtained from reader studies. Moreover, differences in radiologist detection threshold should not substantially affect the relative differences found between imaging protocols. Considering that these have been relatively large (Table 1A,B) and present in DBT systems of various designs(7, 12, 14), a more comprehensive evaluation of the effects and their impact on outcomes would be desirable. Preferably performed in comparative or randomized-controlled trials and perhaps for a couple of different types of DBT units, as system-specific factors vary. Although variable factors are present, and there may be a large proportion of experience involved, the 3D representation of the breast in DBT is incomplete, which is true for any DBT system. Therefore, analogous to 2D mammography(4), a relevant difference could be present between a oneand a two-view DBT setting even at higher performance levels (alone and/or adjunct to FFDM), as shown in the authors’ findings of individual studies and as an overall pattern of studies performed over 6 y (2008–14). In future practice, the increasing use of tomosynthesis with and without additional views needs to be considered in relation to the potential increase in radiation dose, and the additional resources needed such time to review the cases and additional costs that may have logistical implications for breast screening programmes.
ACKNOWLEDGEMENTS The authors thank So¨ren Mattsson, Dev Chakraborty, Juan Min Chang, Nancy Obuchowski, Jennifer Bullen,
Page 5 of 7
Downloaded from http://rpd.oxfordjournals.org/ at University of Winnipeg on September 1, 2015
Oslo (OTST)(26) Trento/Verona (STORM)(3) Yale(27) Houston, Breast Center(28) Friedewald et al.(29) Malmo¨ (MBTST)(30)
Detection per 1000 women
T. M. SVAHN AND N. HOUSSAMI
Fiona Gilbert and Federica Zanca for their valuable contributions. FUNDING N.H. receives research support via a National Breast Cancer Foundation (NBCF Australia) Practitioner Fellowship. T.M.S. was partially supported by the Franke & Margareta Bergqvist foundation. REFERENCES
Page 6 of 7
Downloaded from http://rpd.oxfordjournals.org/ at University of Winnipeg on September 1, 2015
1. Glasziou, P. and Houssami, N. The evidence base for breast cancer screening. Prevent. Med. 53(3), 100–102 (2011). 2. Skaane, P. et al. Prospective trial comparing full-field digital mammography (FFDM) versus combined FFDM and tomosynthesis in a population-based screening programme using independent double reading with arbitration. Eur.Radiol. 23(8), 2061–2071 (2013). 3. Ciatto, S. et al. Integration of 3D digital mammography with tomosynthesis for population breast-cancer screening (STORM): a prospective comparison study. Lancet Oncol. 14(7), 583–589 (2013). 4. Wald, N. J., Murphy, P., Major, P., Parkes, C., Townsend, J. and Frost, C. UKCCCR multicentre randomised controlled trial of one and two view mammography in breast cancer screening. BMJ 311(7014), 1189–1193 (1995). 5. Reiser, I. and Sechopoulos, I. A review of digital breast tomosynthesis. Med. Phys. Int. J., in press (2014). 6. Chakraborty, D. P. New developments in observer performance methodology in medical imaging. Semin. Nucl. Med. 41(6), 401– 418 (2011). 7. Svahn, T., Andersson, I., Chakraborty, D., Svensson, S., Ikeda, D., Fornvik, D., Mattsson, S., Tingberg, A. and Zackrisson, S. The diagnostic accuracy of dual-view digital mammography, single-view breast tomosynthesis and a dual-view combination of breast tomosynthesis and digital mammography in a free-response observer performance study. Radiat. Prot. Dosim. 139(1– 3), 113– 117 (2010). 8. Svahn, T. M., Chakraborty, D. P., Ikeda, D., Zackrisson, S., Do, Y., Mattsson, S. and Andersson, I. Breast tomosynthesis and digital mammography: a comparison of diagnostic accuracy. Br. J. Radiol. 85(1019), e1074–e1082 (2012). 9. Teertstra, H. J., Loo, C. E., van den Bosch, M. A., van Tinteren, H., Rutgers, E. J., Muller, S. H. and Gilhuijs, K. G. Breast tomosynthesis in clinical practice: initial results. Eur. Radiol. 20(1), 16– 24 (2010). 10. Good, W. F., Abrams, G. S., Catullo, V. J., Chough, D. M., Ganott, M. A., Hakim, C. M. and Gur, D. Digital breast tomosynthesis: a pilot observer study. AJR 190(4), 865– 869 (2008). 11. Hanley, J. A. and McNeil, B. J. The meaning and use of the area under a receiver operating characteristic (ROC) curve. Radiology 143, 29– 36 (1982). 12. Wallis, M. G., Moa, E., Zanca, F., Leifland, K. and Danielsson, M. Two-view and single-view tomosynthesis versus full-field digital mammography: high-resolution Xray imaging observer study. Radiology 262(3), 788– 796 (2012).
13. Gur, D., Abrams, G. S., Chough, D. M., Ganott, M. A., Hakim, C. M., Perrin, R. L., Rathfon, G. Y., Sumkin, J. H., Zuley, M. L. and Bandos, A. I. Digital breast tomosynthesis: observer performance study. AJR 193(2), 586– 591 (2009). 14. Rafferty, E. A., Park, J. M., Philpotts, L. E., Poplack, S. P., Sumkin, J. H., Halpern, E. F. and Niklason, L. T. Diagnostic accuracy and recall rates for digital mammography and digital mammography combined with one-view and two-view tomosynthesis: results of an enriched reader study. AJR 202(2), 273–281 (2014). 15. Thibault, F., Dromain, C., Breucq, C., Balleyguier, C. S., Malhaire, C., Steyaert, L., Tardivon, A., Baldan, E. and Drevon, H. Digital breast tomosynthesis versus mammography and breast ultrasound: a multireader performance study. Eur. Radiol. 23(9), 2441–2449 (2013). 16. Gennaro, G. et al. Digital breast tomosynthesis versus digital mammography: a clinical performance study. Eur. Radiol. 20(7), 1545– 1553 (2010). 17. Svane, G., Azavedo, E., Lindman, K., Urech, M., Nilsson, J., Weber, N., Lindqvist, L. and Ullberg, C. Clinical experience of photon counting breast tomosynthesis: comparison with traditional mammography. Acta Radiol. 52(2), 134–142 (2011). 18. Zanca, F., Wallis, M., Moa, E., Leifland, K., Danielsson, M., Oyen, R. and Bosmans, H. Diagnostic accuracy of digital mammography versus tomosynthesis: effect of radiologists’ experience. In: Proc. SPIE 8318, Medical Imaging 2012: Image Perception, Observer Performance, and Technology Assessment, 83180W (February 23, 2012). doi:10.1117/12905276 (2012). 19. Waldherr, C., Cerny, P., Altermatt, H. J., Berclaz, G., Ciriolo, M., Buser, K. and Sonnenschein, M. J. Value of one-view breast tomosynthesis versus two-view mammography in diagnostic workup of women with clinical signs and symptoms and in women recalled from screening. AJR 200(1), 226–231 (2013). 20. Gennaro, G. et al. Performance comparison of singleview digital breast tomosynthesis plus single-view digital mammography with two-view digital mammography. Eur. Radiol. 23(3), 664–672 (2013). 21. Thibault, F., Dromain, C., Breucq, C., Balleyguier, C. S., Malhaire, C., Steyaert, L., Tardivon, A., Baldan, E. and Drevon, H. Digital breast tomosynthesis versus mammography and breast ultrasound: a multireader performance study. Eur. Radiol. 23(9), 2441–2449 (2013). 22. Shin, S. U., Chang, J. M., Bae, M. S., Lee, S. H., Cho, N., Seo, M., Kim, W. H. and Moon, W. K. Comparative evaluation of average glandular dose and breast cancer detection between single-view digital breast tomosynthesis (DBT) plus single-view digital mammography (DM) and two-view DM: correlation with breast thickness and density. Eur. Radiol. 25(1), 1–8 (2015). 23. Smith, A. P., Rafferty, E. A. and Niklason, L. T. Clinical performance of breast tomosynthesis as a function of radiologist experience level. LNCS 5116, 61– 66 (2008). 24. Michell, M. J., Iqbal, A., Wasan, R. K., Evans, D. R., Peacock, C., Lawinski, C. P., Douiri, A., Wilson, R. and Whelehan, P. A comparison of the accuracy of film-screen mammography, full-field digital mammography, and digital breast tomosynthesis. Clin. Radiol. 67(10), 976–981 (2012). 25. Rafferty, E. A., Park, J. M., Philpotts, L. E., Poplack, S. P., Sumkin, J. H., Halpern, E. F. and Niklason, L. T. Assessing radiologist performance using combined digital
DIGITAL BREAST TOMOSYNTHESIS IN ONE OR TWO VIEWS
26.
27.
28.
30.
31.
32.
33.
34.
35.
36.
37. 38. 39.
40. 41.
42. 43.
44.
45.
46. 47.
48.
microcalcification clusters? Results of a multicentre study. Eur. Radiol. (2014). Spangler, M. L., Zuley, M. L., Sumkin, J. H., Abrams, G., Ganott, M. A., Hakim, C., Perrin, R., Chough, D. M., Shah, R. and Gur, D. Detection and classification of calcifications on digital breast tomosynthesis and 2D digital mammography: a comparison. AJR 196(2), 320–324 (2011). Marshall, N. W. and Bosmans, H. Measurements of system sharpness for two digital breast tomosynthesis systems. Phys. Med. Biol. 57(22), 7629– 7650 (2012). Qian, X. et al. High resolution stationary digital breast tomosynthesis using distributed carbon nanotube x-ray source array. Med. Phys. 39(4), 2090– 2099 (2012). Tani, H., Uchiyama, N., Machida, M., Kikuchi, M., Arai, Y., Otsuka, K., Jerebko, A., Fieselmann, A. and Mertelmeier, T. Assessing radiologist performance and microcalcifications visualization using combined 3D rotating mammogram (RM) and digital breast tomosynthesis (DBT). Breast Imaging Lect. Notes Comput. Sci. Vol. 8539, 142–149 (2014). Houssami, N. and Skaane, P. Overview of the evidence on digital breast tomosynthesis in breast cancer detection. Breast 22(2), 101–108 (2013). Svahn, T. M. Letter to the editor re: diagnostic accuracy of digital breast tomosynthesis versus digital mammography for benign and malignant lesions in breasts: a metaanalysis. Eur. Radiol. 24(4), 927 (2014). Zhou, X.-H., Obuchowski, N. A. and McClish, D. K. Statistical methods in diagnostic medicine. John Wiley & Sons, Inc. (2002). ISBN 0-471-34772-8. Skaane, P. et al. Prospective trial comparing full-field digital mammography (FFDM) versus combined FFDM and tomosynthesis in a population-based screening programme using independent double reading with arbitration. Eur. Radiol. 23(8), 2061–2071 (2013). Philpotts, L. Breastimaging: screening/emergingtechnologies (initial experience with digital breast tomosynthesis in screening mammography). Am. J. Roentgenol. 198(Supplement), (2012). Svahn, T. M., Houssami, N., Sechopoulos, I. and Mattsson, S. Review of radiation dose estimates in digital breast tomosynthesis relative to those in two-view fullfield digital mammography. Breast (2014). doi: 10.1016/ j.breast.2014.12.002. Gur, D. et al. Prevalence effect in a laboratory environment. Radiology 228(1), 10– 14 (2003). Reed, W. M., Ryan, J. T., McEntee, M. F., Evanoff, M. G. and Brennan, P. C. The effect of abnormality-prevalence expectation on expert observer performance and visual search. Radiology 258(3), 938–943 (2011). Obuchowski, N. A. One less bias to worry about. Radiology 232(1), 302 (2004).
Page 7 of 7
Downloaded from http://rpd.oxfordjournals.org/ at University of Winnipeg on September 1, 2015
29.
mammography and breast tomosynthesis compared with digital mammography alone: results of a multicenter, multireader trial. Radiology 266(1), 104– 113 (2013). Skaane, P. et al. Comparison of digital mammography alone and digital mammography plus tomosynthesis in a population-based screening program. Radiology 267(1), 47– 56 (2013). Haas, B. M., Kalra, V., Geisel, J., Raghu, M., Durand, M. and Philpotts, L. E. Comparison of tomosynthesis plus digital mammography and digital mammography alone for breast cancer screening. Radiology 269(3), 694 –700 (2013). Rose, S. L., Tidwell, A. L., Bujnoch, L. J., Kushwaha, A. C., Nordmann, A. S. and Sexton, R. Jr. Implementation of breast tomosynthesis in a routine screening practice: an observational study. AJR 200(6), 1401–1408 (2013). Friedewald, S. M. et al. Breast cancer screening using tomosynthesis in combination with digital mammography. JAMA 311(24), 2499– 2507 (2014). Zackrisson, S., La˚ng, K., Timberg, P. and Andersson, I. Performance of one-view breast tomosynthesis versus twoview mammography in breast cancer screening: first results from the Malmo? breast tomosynthesis screening trial. ECR, Vienna (2014). Skaane, P., Bandos, A. I., Eben, E. B., Jebsen, I. N., Krager, M., Haakenaasen, U., Ekseth, U., Izadi, M., Hofvind, S. and Gullien, R. Two-view digital breast tomosynthesis screening with synthetically reconstructed projection images: comparison with digital breast tomosynthesis with full-field digital mammographic images. Radiology 271(3), 655– 663 (2014). Gilbert, F. J. et al. The TOMMY trial: a comparison of TOMosynthesis with digital MammographY in the UK NHS Breast Screening Programme—a multicentre retrospective reading study comparing the diagnostic performance of digital breast tomosynthesis and digital mammography with digital mammography alone. Health Technol. Assess. 19(4), 1–136 (2015). Evans, W. P., Warren Burhenne, L. J., Laurie, L., O’Shaughnessy, K. F. and Castellino, R. A. Invasive lobular carcinoma of the breast: mammographic characteristics and computer-aided detection. Radiology 225(1), 182– 189 (2002). Rafferty, E. A., Niklason, L. and Jameson-Meehan, L. Breast tomosynthesis: one view or two? In: Presented at the Radiological Society of North America (RSNA), Session SSG01-04 Breast Imaging: digital tomosynthesis (2006). Tagliafico, A., Mariscotti, G., Durando, M., Stevanin, C., Tagliafico, G., Martino, L., Bignotti, B., Calabrese, M. and Houssami, N. Characterisation of microcalcification clusters on 2D digital mammography (FFDM) and digital breast tomosynthesis (DBT): does DBT underestimate
doi:10.1093/rpd/ncv078
DIGITAL BREAST TOMOSYNTHESIS IN ONE OR TWO VIEWS AS A REPLACEMENT OR ADJUNCT TECHNIQUE TO FULL-FIELD DIGITAL MAMMOGRAPHY T. M. Svahn* and N. Houssami School of Public Health, Sydney Medical School, University of Sydney, Sydney 2006, NSW, Australia *Corresponding author: [email protected]
INTRODUCTION Breast cancer is one of the most common cancer types among women in western countries. Risk factors include reproductive factors (e.g. non-parity, late first birth, early menarche and late menopause), hormone replacement therapy, genetic factors, ionising radiation and high breast density on mammography. There is also evidence that life style, such as high alcohol consumption and low physical activity, may be associated with increased risk. Mammography has gained wide acceptance because of its ability to visualise non-palpable early-stage changes in the breast, which at screening has shown to reduce mortality by 20–30 %(1). The evolution from 2D mammography to 3D (digital breast tomosynthesis, DBT) reduces the essential problems of over-projected anatomy, for example, effect of tissue overlap and the masking effect of tissue density. At present, DBT is increasingly being used as a diagnostic tool, though, not yet considered the standard of care for breast cancer screening. Recent clinical trials embedded in population screening programmes have shown their immense potential in screening when interpreted in two views together with full-field digital mammography (FFDM), but promising results have also been reported when used as a stand-alone technique. Reported breast cancer detection rates have shown 30– 50 % increases in relation to that of FFDM, usually at reduced recall rates(2, 3). In a randomised 2D screen-film mammography trial, a two-view protocol was found to detect 24 % more women with breast cancer at a reduced recall rate by 15 % when compared with a one-view
protocol(4). Significant technological advancements have been made since then, but the problem of superimposed tissues owing to the 2D nature of the technique (FFDM) remains. Because DBT is a 3D imaging device, it has been hypothesised that additional views to the mediolateral oblique (MLO) projection would not be needed, but as depth resolution is limited due to constraints present in the image sampling(5), the use of two views to acquire DBT might be advantageous. In clinical studies using mammogram sets with ascertained outcomes (often referred to as observer performance studies, or reader studies), a variety of imaging protocols have been examined. In this paper, a review of such studies was performed, in addition to preliminary data assessed from screening trials with respect to imaging protocol to determine whether performance-related trends or patterns are apparent. MATERIALS AND METHODS Review and applied accuracy measures A literature search was performed on clinical studies reporting on breast cancer detection that compared the diagnostic tests using DBT and FFDM (PubMed search: July 2008 to September 2014, in addition to two proceedings papers; IWDM, SPIE and two ECRpresentations; literature search was performed by T.S.). The literature was restricted to studies where interpretation was done blinded to if the patient had cancer or not, with case mixtures consisting of abnormal and normal/benign findings in a retrospective design. A
# The Author 2015. Published by Oxford University Press. All rights reserved. For Permissions, please email: [email protected]
Downloaded from http://rpd.oxfordjournals.org/ at University of Winnipeg on September 1, 2015
Clinical studies using different imaging protocols to perform digital breast tomosynthesis (DBT) were reviewed (2008–14) to assess interpretive accuracy. Descriptive pooled statistics were used to estimate and summarise accuracy measures for each type of imaging protocol in relation to that of two-view full-field digital mammography (FFDM). In studies comparing multiple DBT imaging protocols, a trend of increased performance was often seen when including both the mediolateral oblique and craniocaudal views for DBT alone and even more so for DBT adjunct to FFDM. Overall, the average DAUC (%; sd) across studies for stand-alone DBT (relative to FFDM), in one and in two views, were 2.2 (+ + 3.7) and 5.9 (+ + 4.6), and when used together with FFDM, 3.9 (+ + 2.0) and 6.7 (+ + 0.9). With respect to individual studies, improvements in accuracy using DBT were present for different types of imaging protocols although the magnitude of the impact varied between studies, and some studies did not show significant improvements in comparison with FFDM. The most consistent effect of improvement in breast cancer detection was seen across studies for two-view DBT with FFDM. These summary findings may depend on the sampling constraints present in tomosynthesis imaging and on other factors discussed in this paper. In order to investigate these effects more thoroughly and how they might impact outcomes, comparative or randomized-controlled trials are warranted.
T. M. SVAHN AND N. HOUSSAMI
false positives (Figure 1). All five radiologists performed best in the combined setting. Similarly, another study found that DBT MLO þ FFDM improved accuracy significantly (relative FFDM). However, this study also evaluated two-view DBT þ FFDM, which doubled the increase in sensitivity and in overall accuracy (relative FFDM), while further reducing the recall rate(14). The difference between the two DBT imaging protocols was statistically significant. The benefits were found ‘not’ to correlate with radiologists´ experience levels, hence suggested to apply to a vast majority of radiologists, which has been validated in screening trials. In the study by Thibault et al. (15) various imaging protocols, including conventional FFDM, resulted in a similar accuracy. Sensitivity markedly increased for protocols including ultrasonography but was counterbalanced by a loss in specificity. A high proportion of cancers were not detected by single imaging techniques(15). Pooled accuracy across studies by DBT imaging protocol For stand-alone DBT, the average DAUC (%; sd) across studies performed in one and in two views, respectively, were 2.2 (+3.7) and 5.9 (+4.6), and when used adjunct to FFDM in one and in two views, 3.9 (+2.0) and 6.7 (+0.9) (Table 1A,B). The average improvement in sensitivity and specificity increased for DBT of both views used adjunct to FFDM relative to
RESULTS Studies reported in 2008–14 Multiple imaging protocols in paired study designs In many of the studies, there is a gradual increase in performance present with additional imaging information. Yet, a number of studies only found significant improvements in breast cancer detection for the DBT imaging protocol associated with ‘most’ imaging information (relative FFDM), covering ‘both’ MLO and CC views, stand-alone or adjunct to FFDM(7, 12 – 14). In one of the studies, performance was highest for a combined setting of DBT MLO þ FFDM CC when compared with stand-alone one-view DBT, and FFDM(7), associated with more breast cancers detected at fewer
Figure 1. The accuracy (area under the curve) was highest for DBT MLO þ FFDM CC, 5 % higher than one-view DBT and 11 % higher than FFDM alone, with more breast cancers localised and identified (LLF) at a fewer false positives (FPF) as described by the endpoints of the solid lines of the AFROC plot (7).
Page 2 of 7
Downloaded from http://rpd.oxfordjournals.org/ at University of Winnipeg on September 1, 2015
common study type used is where several radiologists individually interpret all images for each modality under investigation, the multiple-reader multiple-case (MRMC) design, which is done to maximise the sensitivity of detecting differences between imaging modalities (statistical power). Unlike other studies included, the TOMMY trial is population based using consecutive sampling of FFDM-recalled cases at screening. To assess whether there are trends or consistent patterns apparent, information was extracted on sensitivity and specificity, diagnostic accuracy and recall rate for assessment. As sensitivity and specificity are anti-correlated and depend on the decision threshold of the radiologist, it is valuable to use measures that combine the effects seen on both, denoted diagnostic accuracy. Diagnostic accuracy was estimated by the area under the ROC curve (AUC) or by area under the alternative free-response ROC curve (JAFROC)(6). The main difference between the methods is that ROC considers the cases as a whole (e.g. if they are abnormal or normal/benign), whereas the free-response method also considers locations of individual breast cancers, thereby providing increased realism and power(6). The AUC was here used as a unified measure across studies (inferred AUC was estimated from freeresponse ROC data(6 – 10)). AUC can be defined as the average sensitivity value for all possible values of specificities(11). Note that, as a stand-alone technique, FFDM refers to interpretation of both MLO and CC views and is hence ‘always’ the baseline modality that the DBT modality is compared with. The estimated accuracy changes were obtained by subtracting results from the DBT modality with that of FFDM (for instance, DAUC ¼ AUCDBTmodality AUCFFDM ), which were pooled across studies by imaging protocol. Cancer detection rates and recall or false-positive rates were assessed from screening trials. At first, results obtained from studies where several different imaging protocols were investigated based on paired study designs, mostly the MRMC type, are summarised.
DIGITAL BREAST TOMOSYNTHESIS IN ONE OR TWO VIEWS Table 1. Clinical studies using digital breast tomosynthesis alone (DBT; A), in one and in two views, and as adjunct to two-view digital mammography (FFDM; B) in comparison with standard two-view FFDM. Study
Subjects Radiologists (abnormal)
DAUC (%, sd)
D (sensitivity/ specificity)
Benign/normals (false recalls)
Cancer cases
(211.2)
(210.5)
212.5 210.7 269.7
21.2 219.3 252.5
231
224
22.6 210
þ3.3 þ5.7
(211.2)
(0)
217.5 25 28.8
21.2 21.2 þ2.9
272.7 238.5*
268.3 23.9
255.6
234.2
242.6* 230*
þ5.7
269.7* 238.3* 245.7*
27.8 þ1.1 þ1.0
245.3
0
The results of the studies are presented in reader-averaged difference, increase (þ) or decrease (2), in tomosynthesis performance relative to two-view FFDM. The change in recall rate (%) is relative to two-view FFDM; (RRDBT modality 2 RRFFDM)/RRFFDM. Significance is indicated by an asterisk (*). a Pooled average estimates are applied (e.g. not adjusted/weighted for study size). b The studies by Wallis et al. and Zanca et al. are overlapping; hence, data points were only included in the pooled avg. estimates from the study by Zanca et al. c Screen-film mammography (SFM) was also included in this comparison. The results are thus presented from when the readers had read SFM and FFDM, in comparison with those from when they had read SFM, FFDM and DBT. d The studies by Svahn et al.(7), Gennaro et al.(20), Thibault et al.(21) and Shin et al.(22) evaluated DBT MLO view plus FFDM CC, whereas Rafferty et al.(14) and Waldherr et al.(19) evaluated DBT MLO plus two-view FFDM.
Page 3 of 7
Downloaded from http://rpd.oxfordjournals.org/ at University of Winnipeg on September 1, 2015
A. Studies comparing stand-alone DBT (as a replacement to FFDM) versus FFDM One-view DBT 376 (63) 6 þ1.5 24.5/þ4.1 Gennaro et al.(16) Svane et al.(17) 144 (76) 2 20.07 23.9/þ5.1 (8) 185 (89) 5 þ9.4* þ10.8*/20.8 Svahn et al. Wallis et al. (12) 130 (40) 10 (þ0.1)b Zanca et al.(18) Exp. readers 130 (40) 5 þ1 Inexp. readers 5 21 144 (86) 2 (in con.) þ14*/þ2.1 Waldherr et al. (19) 130 (55) 7 þ2.3 27/þ11 Thibault et al.(15) a Average 208 (68) 6 þ2.2 (+3.7) þ1.9/þ4.3 Two-view DBT 30 (25) 9 þ2 Good et al.(10) 125 (35) 6 þ5/þ4 Gur et al.(13) (9) 501 (111) 1 (21.5; 0/21.7 Teertstra et al. classical metric) 130 (40) 10 (þ7.9*)b Wallis et al.(12) (18) Zanca et al. Exp. readers 130 (40) 5 þ4.7 Inexp. readers 5 þ11.0* a Average 197 (53) 6 þ5.9 (+4.6) þ2.5/þ1.2 B. Studies comparing DBT adjunct with FFDM versus FFDM d One-view DBT Svahn et al.(7) 50 (25) 5 þ7.1 þ9.6/22 469 (68) 6 þ2.1 þ3.4/þ1.9 Gennaro et al. (20) (21) 130 (55) 7 þ2.4 25/þ11 Thibault et al. 144 (86) 2 (in con.) þ17.5*/21.4 Waldherr et al.(19) Rafferty et al.(14) 310 (51) 15 þ3.6* þ8.7*/20.2 149 (102) 3 þ4.1* þ7.1*/23.4 Shin et al.(22) a Average 209 (65) 6 þ3.9 (+2.0) þ6.9/þ0.4 Two-view DBT 316 (48) 12 þ7.1* Smith et al.(23) 125 (30) 8 þ5/þ12 Gur et al.(13) (24) c 738 (204) 6 þ7.2* Michell et al. (25) 312 (48) 12 þ7.2* þ10.7/þ5.1 Rafferty et al. Rafferty et al. (25) 312 (51) 15 þ7.1* þ16*/21.7 310 (51) 15 þ6.8* þ16*/21.7 Rafferty et al.(14) (3) 5224 (1087) 3 þ5* þ3*/þ12* Gilbert et al. TOMMY trial a Average 1048 (217) 10 þ6.7 (+0.9) þ10.1/þ6.5
Relative change in recall rate (%)
T. M. SVAHN AND N. HOUSSAMI
other imaging protocols (was rarely reported for twoview DBT; Table 1A). The reduction of false recall rates was highest when DBT was combined with FFDM. All screening trials evaluating two-view DBT þ FFDM were associated with improvements over FFDM both in cancer detection rates (10 –53 %) and in reduced recall rates or false-positive rates (15– 38 %). Stand-alone one-view DBT (MBTST) was associated with an increase in cancer detection rate by 43 % at an equally increased recall rate by 43 %. DISCUSSION
Page 4 of 7
Downloaded from http://rpd.oxfordjournals.org/ at University of Winnipeg on September 1, 2015
The clinical application of DBT is currently evolving. Therefore, it seems timely to present a review of the findings from clinical studies using DBT in various imaging protocols. As indicated, results vary across studies but some findings are evident; in particular, increased accuracy was generally seen in studies that added DBT to FFDM and more so when two-view DBT was used (as shown in the summary evidence tables). For instance, the improvement in avg. sensitivity was about five times higher for two-view DBT adjunct to FFDM in comparison with that of stand-alone one-view DBT at a slight increase in avg. specificity. In screening trials, two-view DBT þ FFDM was always associated with an increase in cancer detection rate at a reduced recall rate or FP rate, whereas at one-view DBT cancer detection rates increased as much as the recall rate, relative to that of FFDM. There are several possible explanations for the results. DBT images are collected within a limited angular range and reconstructed into pseudo-tomographic images with partial blurring outside a selected plane and, hence, effects from overlapping tissues (anatomical noise) are not completely eliminated(5). In addition, an additional 2D examination may yield a different compression clearing some possible tissue superimpositions, as well as another dose delivery, and hence a different quantum noise pattern. However, this problem was usually not treated by the papers reviewed, but may have limited impact if acquisitions are performed under the same compression (i.e. 2D & 3D - same view)(31, 32). Owing to the incomplete elimination of anatomical noise, a cancer can be undetected or misinterpreted. Breast cancers such as small spiculated tumours might preferentially be oriented in one plane and difficult to identify in other planes. As known in 2D mammography, lobular carcinomas are usually best visualised in the CC projection(33), which seem to apply for DBT as well(34). In the same manner, the presence of anatomical noise might adversely affect the radiologists’ ability to confirm (and hence dismiss) normal/ benign cases. Another reason for the greater stability (sd) seen for DBT adjunct to FFDM might be in calcification detection. FFDM is the current reference standard for detection of calcifications, and studies have reported lower image quality or partially poorer
interpretation for them at DBT(8, 35, 36). This might depend on motion artefacts caused by the longer DBT acquisition time compared with FFDM; for example, due to patient movement between projections, blur associated with the focal spot movement (in continuous movement machines) or the reduced spatial resolution at acquisition (e.g. from binning of pixels)(37). Effects from these factors such as unsharpness of details may influence morphology-based decisions but can be solved by technical progress(38). In addition, tomosynthesis reduces tissue overlap, structural noise and thus helps to better show breast masses. Microcalcifications are, on the other hand, less affected by it and may in that way be more appropriately perceived by 2D views, which enable an overview of calcifications that are spread out in the breast at various depths. The use of synthetic 2D views, thicker slice thicknesses, for example, ‘slabbing’ and other developed methods,(39) can help the radiologist to maintain the accuracy of calcification detection and may ease the comparison with prior years’ screening images. Besides the influence of image protocol used, there are other possible explanations for different results in studies, as discussed by Houssami et al. (40) An important aspect is case difficulty. Cases that are more likely detectable by only one modality are essential to show differences between medical imaging devices, if any, but having higher occurrences of them in relation to the general population could result in enhanced effects(41, 42). Spectrum effects(42) may be pronounced when comparing a per-view more powerful technique (DBT) with another one (FFDM); and relevant to consider if DBT is acquired (for one-view DBT studies) only in the view where the abnormality was most visible at FFDM, or if case selection is mainly determined by difficulty level at FFDM. In order to obtain generalisable results, it is crucial that the case sampling mirror the usual spectrum of findings from the target population, hence allowing both comparative techniques to demonstrate their full detection potential. The pooled accuracy across studies can even out such effects (Table 1A,B), if present, and thus provide a more reliable average, though system-specific differences are not explicitly being addressed. In general, the fact that so many studies have found substantial benefits should be regarded as highly promising and would justify larger service-based trials. The publication of several large screening trials has in recent years provided insight of the clinical impact of DBT (Table 2). These have been conducted using different designs, with and without double reading and arbitration management. With regard to imaging protocols, only two different approaches have been investigated: two-view DBT plus FFDM(3, 26 – 28, 43, 44) and stand-alone one-view DBT (MBTST)(30). All trials show substantial improvements in breast cancer detection (Table 2). Stand-alone one-view DBT (MBTST) was, however, also associated with a 43 % significant
DIGITAL BREAST TOMOSYNTHESIS IN ONE OR TWO VIEWS Table 2. The results of screening trials in breast cancer detection rate, recall rate or false-positive rate. Studya
12501 7292 13174 (7058/6116) 23355 (13856/9499) (281187/173663) 7500
6.1 5.3 5.2 4 4.2 6.3
Recall rate
4.4 11.7 8.7 10.7 2.6
b Falsepositive rate per 1000 women
Detection per 1000 women
61
8.0 (þ27*) 8.1 (þ53*) 5.7 (þ9.6) 5.4 (þ35) 5.4 (þ29) 8.9 (þ43*)
Recall rate
b Falsepositive rate per 1000 women
53 (215*) 3.5 (220*) 8.4 (228.2*) 5.5 (237.5*) 9.1 (215) 3.8 (þ43*)
The Malmo¨ (MBTST) trial compares stand-alone one-view DBT relative FFDM, whereas the other trials compare two-view DBT adjunct to FFDM relative FFDM. The numbers in the parentheses show the percentage increase or decrease in performance measure relative to that of FFDM. Statistical significance is indicated with an asterisk. a The trials performed at Yale University and at Houston Breast Center are based on a retrospective study design using nonpaired data with single independent reading. The OTST, STORM and MBTST trials are prospective using paired data. b False-positive rate is presented before arbitration management.
increase in recall rate, which results in a slightly better positive predictive value for FFDM (PPV; cancers detected/100 recalls), and a comparable TP:FP ratio between the techniques as often seen for that imaging protocol (Table 1A). Other screening trials have often been associated with substantial PPV increases of 50 to 110 % for two-view DBT plus FFDM. It should be noted that the presented findings (Table 2) partially comprise interim analyses, or preliminary results presented at international conferences. Hence, the final results of some of the studies have not yet been published. The authors’ pooled analysis was not adjusted/ weighted for population size; hence, it is not a studyadjusted meta-analysis. A meta-analysis would generally use sensitivity and specificity pairs without including the rating data. By using the DAUC, the impact of rating data is being acknowledged. Variable thresholds were partially used for reporting sensitivity and specificity, which could even out between the different groups but is avoided in the DAUC. In studies, where several imaging protocols were investigated based on the same populations of cases and readers (MRMC), the comparisons were consistent. Radiation dose typically increased with the number of views and is described more elsewhere(45), but was not addressed in relation to performance. MRMC studies are sensitive at detecting differences between the performances of modalities. In general, they subject only a limited number of patients to radiation dose. However, they do not necessarily declare where the clinical operating point is, but (AUCs) have however shown to be robust to effects of cancer prevalence(46 – 48). Another potential limitation is that the studies are performed under experimental settings with no direct link between radiologist decision and patients health or outcome. Interestingly, the operating points of the TOMMY
trial were within the range of those from other studies of the same imaging protocol, with the same ‘total’ accuracy increments in sensitivity and specificity (Sn þ Sp 14–17 %), indicating that valuable estimates have been obtained from reader studies. Moreover, differences in radiologist detection threshold should not substantially affect the relative differences found between imaging protocols. Considering that these have been relatively large (Table 1A,B) and present in DBT systems of various designs(7, 12, 14), a more comprehensive evaluation of the effects and their impact on outcomes would be desirable. Preferably performed in comparative or randomized-controlled trials and perhaps for a couple of different types of DBT units, as system-specific factors vary. Although variable factors are present, and there may be a large proportion of experience involved, the 3D representation of the breast in DBT is incomplete, which is true for any DBT system. Therefore, analogous to 2D mammography(4), a relevant difference could be present between a oneand a two-view DBT setting even at higher performance levels (alone and/or adjunct to FFDM), as shown in the authors’ findings of individual studies and as an overall pattern of studies performed over 6 y (2008–14). In future practice, the increasing use of tomosynthesis with and without additional views needs to be considered in relation to the potential increase in radiation dose, and the additional resources needed such time to review the cases and additional costs that may have logistical implications for breast screening programmes.
ACKNOWLEDGEMENTS The authors thank So¨ren Mattsson, Dev Chakraborty, Juan Min Chang, Nancy Obuchowski, Jennifer Bullen,
Page 5 of 7
Downloaded from http://rpd.oxfordjournals.org/ at University of Winnipeg on September 1, 2015
Oslo (OTST)(26) Trento/Verona (STORM)(3) Yale(27) Houston, Breast Center(28) Friedewald et al.(29) Malmo¨ (MBTST)(30)
Detection per 1000 women
T. M. SVAHN AND N. HOUSSAMI
Fiona Gilbert and Federica Zanca for their valuable contributions. FUNDING N.H. receives research support via a National Breast Cancer Foundation (NBCF Australia) Practitioner Fellowship. T.M.S. was partially supported by the Franke & Margareta Bergqvist foundation. REFERENCES
Page 6 of 7
Downloaded from http://rpd.oxfordjournals.org/ at University of Winnipeg on September 1, 2015
1. Glasziou, P. and Houssami, N. The evidence base for breast cancer screening. Prevent. Med. 53(3), 100–102 (2011). 2. Skaane, P. et al. Prospective trial comparing full-field digital mammography (FFDM) versus combined FFDM and tomosynthesis in a population-based screening programme using independent double reading with arbitration. Eur.Radiol. 23(8), 2061–2071 (2013). 3. Ciatto, S. et al. Integration of 3D digital mammography with tomosynthesis for population breast-cancer screening (STORM): a prospective comparison study. Lancet Oncol. 14(7), 583–589 (2013). 4. Wald, N. J., Murphy, P., Major, P., Parkes, C., Townsend, J. and Frost, C. UKCCCR multicentre randomised controlled trial of one and two view mammography in breast cancer screening. BMJ 311(7014), 1189–1193 (1995). 5. Reiser, I. and Sechopoulos, I. A review of digital breast tomosynthesis. Med. Phys. Int. J., in press (2014). 6. Chakraborty, D. P. New developments in observer performance methodology in medical imaging. Semin. Nucl. Med. 41(6), 401– 418 (2011). 7. Svahn, T., Andersson, I., Chakraborty, D., Svensson, S., Ikeda, D., Fornvik, D., Mattsson, S., Tingberg, A. and Zackrisson, S. The diagnostic accuracy of dual-view digital mammography, single-view breast tomosynthesis and a dual-view combination of breast tomosynthesis and digital mammography in a free-response observer performance study. Radiat. Prot. Dosim. 139(1– 3), 113– 117 (2010). 8. Svahn, T. M., Chakraborty, D. P., Ikeda, D., Zackrisson, S., Do, Y., Mattsson, S. and Andersson, I. Breast tomosynthesis and digital mammography: a comparison of diagnostic accuracy. Br. J. Radiol. 85(1019), e1074–e1082 (2012). 9. Teertstra, H. J., Loo, C. E., van den Bosch, M. A., van Tinteren, H., Rutgers, E. J., Muller, S. H. and Gilhuijs, K. G. Breast tomosynthesis in clinical practice: initial results. Eur. Radiol. 20(1), 16– 24 (2010). 10. Good, W. F., Abrams, G. S., Catullo, V. J., Chough, D. M., Ganott, M. A., Hakim, C. M. and Gur, D. Digital breast tomosynthesis: a pilot observer study. AJR 190(4), 865– 869 (2008). 11. Hanley, J. A. and McNeil, B. J. The meaning and use of the area under a receiver operating characteristic (ROC) curve. Radiology 143, 29– 36 (1982). 12. Wallis, M. G., Moa, E., Zanca, F., Leifland, K. and Danielsson, M. Two-view and single-view tomosynthesis versus full-field digital mammography: high-resolution Xray imaging observer study. Radiology 262(3), 788– 796 (2012).
13. Gur, D., Abrams, G. S., Chough, D. M., Ganott, M. A., Hakim, C. M., Perrin, R. L., Rathfon, G. Y., Sumkin, J. H., Zuley, M. L. and Bandos, A. I. Digital breast tomosynthesis: observer performance study. AJR 193(2), 586– 591 (2009). 14. Rafferty, E. A., Park, J. M., Philpotts, L. E., Poplack, S. P., Sumkin, J. H., Halpern, E. F. and Niklason, L. T. Diagnostic accuracy and recall rates for digital mammography and digital mammography combined with one-view and two-view tomosynthesis: results of an enriched reader study. AJR 202(2), 273–281 (2014). 15. Thibault, F., Dromain, C., Breucq, C., Balleyguier, C. S., Malhaire, C., Steyaert, L., Tardivon, A., Baldan, E. and Drevon, H. Digital breast tomosynthesis versus mammography and breast ultrasound: a multireader performance study. Eur. Radiol. 23(9), 2441–2449 (2013). 16. Gennaro, G. et al. Digital breast tomosynthesis versus digital mammography: a clinical performance study. Eur. Radiol. 20(7), 1545– 1553 (2010). 17. Svane, G., Azavedo, E., Lindman, K., Urech, M., Nilsson, J., Weber, N., Lindqvist, L. and Ullberg, C. Clinical experience of photon counting breast tomosynthesis: comparison with traditional mammography. Acta Radiol. 52(2), 134–142 (2011). 18. Zanca, F., Wallis, M., Moa, E., Leifland, K., Danielsson, M., Oyen, R. and Bosmans, H. Diagnostic accuracy of digital mammography versus tomosynthesis: effect of radiologists’ experience. In: Proc. SPIE 8318, Medical Imaging 2012: Image Perception, Observer Performance, and Technology Assessment, 83180W (February 23, 2012). doi:10.1117/12905276 (2012). 19. Waldherr, C., Cerny, P., Altermatt, H. J., Berclaz, G., Ciriolo, M., Buser, K. and Sonnenschein, M. J. Value of one-view breast tomosynthesis versus two-view mammography in diagnostic workup of women with clinical signs and symptoms and in women recalled from screening. AJR 200(1), 226–231 (2013). 20. Gennaro, G. et al. Performance comparison of singleview digital breast tomosynthesis plus single-view digital mammography with two-view digital mammography. Eur. Radiol. 23(3), 664–672 (2013). 21. Thibault, F., Dromain, C., Breucq, C., Balleyguier, C. S., Malhaire, C., Steyaert, L., Tardivon, A., Baldan, E. and Drevon, H. Digital breast tomosynthesis versus mammography and breast ultrasound: a multireader performance study. Eur. Radiol. 23(9), 2441–2449 (2013). 22. Shin, S. U., Chang, J. M., Bae, M. S., Lee, S. H., Cho, N., Seo, M., Kim, W. H. and Moon, W. K. Comparative evaluation of average glandular dose and breast cancer detection between single-view digital breast tomosynthesis (DBT) plus single-view digital mammography (DM) and two-view DM: correlation with breast thickness and density. Eur. Radiol. 25(1), 1–8 (2015). 23. Smith, A. P., Rafferty, E. A. and Niklason, L. T. Clinical performance of breast tomosynthesis as a function of radiologist experience level. LNCS 5116, 61– 66 (2008). 24. Michell, M. J., Iqbal, A., Wasan, R. K., Evans, D. R., Peacock, C., Lawinski, C. P., Douiri, A., Wilson, R. and Whelehan, P. A comparison of the accuracy of film-screen mammography, full-field digital mammography, and digital breast tomosynthesis. Clin. Radiol. 67(10), 976–981 (2012). 25. Rafferty, E. A., Park, J. M., Philpotts, L. E., Poplack, S. P., Sumkin, J. H., Halpern, E. F. and Niklason, L. T. Assessing radiologist performance using combined digital
DIGITAL BREAST TOMOSYNTHESIS IN ONE OR TWO VIEWS
26.
27.
28.
30.
31.
32.
33.
34.
35.
36.
37. 38. 39.
40. 41.
42. 43.
44.
45.
46. 47.
48.
microcalcification clusters? Results of a multicentre study. Eur. Radiol. (2014). Spangler, M. L., Zuley, M. L., Sumkin, J. H., Abrams, G., Ganott, M. A., Hakim, C., Perrin, R., Chough, D. M., Shah, R. and Gur, D. Detection and classification of calcifications on digital breast tomosynthesis and 2D digital mammography: a comparison. AJR 196(2), 320–324 (2011). Marshall, N. W. and Bosmans, H. Measurements of system sharpness for two digital breast tomosynthesis systems. Phys. Med. Biol. 57(22), 7629– 7650 (2012). Qian, X. et al. High resolution stationary digital breast tomosynthesis using distributed carbon nanotube x-ray source array. Med. Phys. 39(4), 2090– 2099 (2012). Tani, H., Uchiyama, N., Machida, M., Kikuchi, M., Arai, Y., Otsuka, K., Jerebko, A., Fieselmann, A. and Mertelmeier, T. Assessing radiologist performance and microcalcifications visualization using combined 3D rotating mammogram (RM) and digital breast tomosynthesis (DBT). Breast Imaging Lect. Notes Comput. Sci. Vol. 8539, 142–149 (2014). Houssami, N. and Skaane, P. Overview of the evidence on digital breast tomosynthesis in breast cancer detection. Breast 22(2), 101–108 (2013). Svahn, T. M. Letter to the editor re: diagnostic accuracy of digital breast tomosynthesis versus digital mammography for benign and malignant lesions in breasts: a metaanalysis. Eur. Radiol. 24(4), 927 (2014). Zhou, X.-H., Obuchowski, N. A. and McClish, D. K. Statistical methods in diagnostic medicine. John Wiley & Sons, Inc. (2002). ISBN 0-471-34772-8. Skaane, P. et al. Prospective trial comparing full-field digital mammography (FFDM) versus combined FFDM and tomosynthesis in a population-based screening programme using independent double reading with arbitration. Eur. Radiol. 23(8), 2061–2071 (2013). Philpotts, L. Breastimaging: screening/emergingtechnologies (initial experience with digital breast tomosynthesis in screening mammography). Am. J. Roentgenol. 198(Supplement), (2012). Svahn, T. M., Houssami, N., Sechopoulos, I. and Mattsson, S. Review of radiation dose estimates in digital breast tomosynthesis relative to those in two-view fullfield digital mammography. Breast (2014). doi: 10.1016/ j.breast.2014.12.002. Gur, D. et al. Prevalence effect in a laboratory environment. Radiology 228(1), 10– 14 (2003). Reed, W. M., Ryan, J. T., McEntee, M. F., Evanoff, M. G. and Brennan, P. C. The effect of abnormality-prevalence expectation on expert observer performance and visual search. Radiology 258(3), 938–943 (2011). Obuchowski, N. A. One less bias to worry about. Radiology 232(1), 302 (2004).
Page 7 of 7
Downloaded from http://rpd.oxfordjournals.org/ at University of Winnipeg on September 1, 2015
29.
mammography and breast tomosynthesis compared with digital mammography alone: results of a multicenter, multireader trial. Radiology 266(1), 104– 113 (2013). Skaane, P. et al. Comparison of digital mammography alone and digital mammography plus tomosynthesis in a population-based screening program. Radiology 267(1), 47– 56 (2013). Haas, B. M., Kalra, V., Geisel, J., Raghu, M., Durand, M. and Philpotts, L. E. Comparison of tomosynthesis plus digital mammography and digital mammography alone for breast cancer screening. Radiology 269(3), 694 –700 (2013). Rose, S. L., Tidwell, A. L., Bujnoch, L. J., Kushwaha, A. C., Nordmann, A. S. and Sexton, R. Jr. Implementation of breast tomosynthesis in a routine screening practice: an observational study. AJR 200(6), 1401–1408 (2013). Friedewald, S. M. et al. Breast cancer screening using tomosynthesis in combination with digital mammography. JAMA 311(24), 2499– 2507 (2014). Zackrisson, S., La˚ng, K., Timberg, P. and Andersson, I. Performance of one-view breast tomosynthesis versus twoview mammography in breast cancer screening: first results from the Malmo? breast tomosynthesis screening trial. ECR, Vienna (2014). Skaane, P., Bandos, A. I., Eben, E. B., Jebsen, I. N., Krager, M., Haakenaasen, U., Ekseth, U., Izadi, M., Hofvind, S. and Gullien, R. Two-view digital breast tomosynthesis screening with synthetically reconstructed projection images: comparison with digital breast tomosynthesis with full-field digital mammographic images. Radiology 271(3), 655– 663 (2014). Gilbert, F. J. et al. The TOMMY trial: a comparison of TOMosynthesis with digital MammographY in the UK NHS Breast Screening Programme—a multicentre retrospective reading study comparing the diagnostic performance of digital breast tomosynthesis and digital mammography with digital mammography alone. Health Technol. Assess. 19(4), 1–136 (2015). Evans, W. P., Warren Burhenne, L. J., Laurie, L., O’Shaughnessy, K. F. and Castellino, R. A. Invasive lobular carcinoma of the breast: mammographic characteristics and computer-aided detection. Radiology 225(1), 182– 189 (2002). Rafferty, E. A., Niklason, L. and Jameson-Meehan, L. Breast tomosynthesis: one view or two? In: Presented at the Radiological Society of North America (RSNA), Session SSG01-04 Breast Imaging: digital tomosynthesis (2006). Tagliafico, A., Mariscotti, G., Durando, M., Stevanin, C., Tagliafico, G., Martino, L., Bignotti, B., Calabrese, M. and Houssami, N. Characterisation of microcalcification clusters on 2D digital mammography (FFDM) and digital breast tomosynthesis (DBT): does DBT underestimate
