br distinguishing the extent of invasive tumour from DCIS compo


distinguishing the extent of invasive tumour from DCIS compo-nents using mammography and also using tomosynthesis.
Variability in measurement was comparable for both tomo-synthesis and mammography, but was greater in dense breasts for both tests. For women with high breast density, 95% LOA indicated that the “true” (pathologic) tumour size may be up to 166% above or 60% below the imaging measurement. Sensitivity analysis again suggested that this relatively large range of measurement error was mostly attributable to a small number of cases with extensive DCIS components reported on histology. When two cases with extensive DCIS at pathology were excluded (one low and one high density), LOA for both tomosynthesis and mammography decreased mark-edly in women with dense breasts, and were comparable to LOA in the low density subgroup (which were also slightly lower compared with the main analysis). However, even within this more limited range, measurement errors may potentially impact man-agement when treatment decisions are informed by these tests. For example, underestimation may result in inadequate resection requiring reintervention for positive margins; overestimation may lead to upstaging and/or a change in treatment approach (e.g. mastectomy instead of breast conserving surgery).
A proposed advantage of tomosynthesis is the Rapamycin of tis-sue superimposition apparent on mammographic images, thereby potentially improving imaging of malignancy in women with dense breasts. Studies in the screening setting provide evidence for this advantage, with tomosynthesis demonstrated to detect additional cancers (not detected at mammography) particularly in high-density subgroups [6]. However, our study suggests that the enhanced detection capability of tomosynthesis does not
necessarily translate into enhanced measurement of tumour size. For cancers detected only by tomosynthesis, frequently in women with dense breasts, tumour size measurement had substantial measurement variability, with LOA that were larger than corre-sponding estimates for cancers measured by both tests. This is likely to reflect the fact that mammographically-occult cancers represent “challenging” cases, with tumour and breast density characteristics that make detection and imaging size measurement inherently difficult. Based on these results, tomosynthesis does not appear to improve size measurement when standard mammo-graphic images also visualise the cancer; for cancers detected by tomosynthesis but not mammography, there is potential that those cancers may be mis-sized.
Previous studies have used statistical methods that may lead to misleading conclusions about the accuracy of tomosynthesis for measuring tumour size, or have not used commercially available tomosynthesis technology, with unknown applicability to current practice. Our study addresses all those limitations, and further ex-tends the evidence base by reporting relative mean differences, which provide better estimates of measurement error with increasing tumour size; stratifying analyses by density; exploring possible causes of measurement inaccuracy through sensitivity analysis; comparing tomosynthesis and mammography using paired measurements; and assessing measurement accuracy for cancers detected by tomosynthesis alone. However, a limitation of our study was that radiologists read mammography and tomo-synthesis images sequentially, meaning that interpretation of tomosynthesis was not blinded to mammography. This may have resulted in tomosynthesis size measurements being more similar to mammography than if readers were blinded to mammographic measurement. There is therefore the potential for our study to have underestimated the differences between tests.
The clinical application and evaluation of tomosynthesis in breast cancer imaging and diagnosis has emerged rapidly in recent years, and includes its application in screening, assessment, and staging [17e19]. To date, there are no guidelines recommending its use in place of mammography in breast screening or diagnosis, however there is strong evidence that using tomosynthesis in population breast screening improves detection metrics relative to mammography screening. A comprehensive meta-analysis [17] has estimated the incremental cancer detection rate from tomosyn-thesis to be 1.6 cancers per 1000 screens, with an absolute reduc-tion in recall of 2.2%, compared to 2D mammography. In addition, there are several RCTs of tomosynthesis screening in progress [20]. There is relatively less evidence on use of tomosynthesis in diag-nosis, as highlighted in a recent review which reported that tomosynthesis for assessment or investigation of screen-detected abnormalities and lesions had been evaluated in few studies (with methodological limitations) and these studies generally show that tomosynthesis can potentially improve specificity over mammography [21].