br Corresponding author br E
* Corresponding author.
E-mail address: [email protected] (M.L. Marinovich).
tomosynthesis technology allows for software reconstruction of synthetic 2D-mammography images from 3D acquisitions, thereby roughly halving the radiation dose necessary to obtain separate 2D and 3D scans .
Tomosynthesis in the screening setting has been evaluated in prospective studies that show detection of additional cancers when the test is added to mammography [3e5]. Given such improve-ments in BC detection, the application of tomosynthesis in pre-treatment assessment of cancer extent is an important area for further study, particularly given the increasing adoption of tomo-synthesis into Sorafenib screening practice. Current guidelines for the diagnosis and evaluation of BC note that tomosynthesis may improve accuracy in this setting, particularly for women with dense
breasts , but do not yet recommend its routine use . More accurate measurement of tumour size at initial staging has the potential to better inform surgical management.
A small number of studies have evaluated the accuracy of tomosynthesis in measuring primary BC size in the staging setting, with pathologic tumour size as the reference standard. Compared with 2D mammography, tomosynthesis has been found to show slightly better agreement with pathologic tumour size, but studies are inconsistent in showing a tendency for underestimation  or overestimation  of tumour size by both tests. In addition, those studies used prototype tomosynthesis technologies, and thus have uncertain applicability to commercially-available units used in current practice. In a study using recent technology, an identical proportion (91%) of 2D-mammography and tomosynthesis mea-surements agreed with pathologic measurements within ±10 mm ; however, measurement errors within that relatively large range may have implications for treatment decisions, and differ-ences between tests may be evident with a more precise margin of error . Studies have reported higher Pearson correlations with pathology for tomosynthesis compared with 2D-mammography , but the limitations of such correlations are well-documented: Pearson's correlation does not assess agreement between mea-surements and may result in misleading conclusions .
Given the paucity of previous studies and their above-stated limitations , we undertook a retrospective study of tumour size estimation by tomosynthesis, using prospectively collected data from the Screening with Tomosynthesis Or standard Mammography-2 (STORM-2) population-based BC screening trial , and applying the recommended analytic methods to assess agreement between imaging and pathologic measurements . Agreement for tomosynthesis is compared with 2D-mammog-raphy, both when tests are interpreted by the same reader, or where different readers perform tumour size measurements.
All pathologically confirmed cancers detected in STORM-2 , a prospective population-based screening study of BC detection and false positive recall, were included in Schwann cells analysis. STORM-2 screened 9672 women using two parallel arms whereby double-readings of the same screening examinations were conducted: 1) sequential standard 2D followed by integrated 2D/3D-mammog-raphy versus 2) sequential synthetic 2D followed by integrated synthetic 2D/3D-mammography. An outline of the STORM-2 trial pathway is presented in Fig. S1 (online Appendix). Participants were asymptomatic women undergoing biennial screening in Trento, Italy. Women were recalled for further assessment based on recall by any reader in either arm. A total of 90 cancers in 85 women were detected and are the subject of this retrospective sub-study of STORM-2 that focuses on tumour size.
2.2. Imaging details
Participants in STORM-2 had digital mammography with 2D and 3D (tomosynthesis) mammography (acquired with Selenia® Di-mensions Unit operated in COMBO© mode; Hologic, Bedford MA, USA). Both 2D and 3D images were acquired at the same screening examination with a single breast positioning per view. Mediolateral oblique and cranio-caudal views were obtained for 2D and 3D ac-quisitions. Additional trial details have been described by Bernardi et al. .
Breast density based on 2D-mammographic findings was assessed as the majority score from screen-readers of STORM-2 (or
by arbitration, as required)  according to the American College of Radiology Breast Imaging Reporting and Data System (BI-RADS) classification: BI-RADS 1 (almost entirely fatty); BI-RADS 2 (scat-tered areas of fibroglandular density); BI-RADS 3 (heterogeneously dense); and BI-RADS 4 (extremely dense). For analysis, those cat-egories were collapsed into “low density” (BI-RADS 1 or 2) and “high density” (BI-RADS 3 or 4) groupings.