Comparison of Two-dimensional and Three-dimensional Mammography Performance in Breast Cancer Screening

Ladies and gentlemen, dear colleagues, clip the first. Say that I'm very pleased and thankful that four I was invited to present our results here at Vienna had breast cancer at Breast Care Day. This study was sponsored by Siemens, including my travel expenses. We start with the question that we were asked to show at the beginning of our presentations here, and that is what should the future of breast cancer screening with tomosynthesis look like. We can talk about different attributes of screening, which can affect its future, but screening remains centered on breast cancer detection. Our study compared. Cancer detection performance of two dimensional and three dimensional 3 dimensional mammography in control settings only reader study. Indigital mammography era and before screen exam consisted of four views resulting in four to the images with introduction of tomosynthesis for additional views are obtained. So the radiation those received that screening exam is effectively doubled. With more images to look at, reading a screening exam takes more time anywhere between two and three times longer. Increased reading time and increased radiation are the price we pay for increased better performance of tomography. One way to reduce the radiation dose and possibly the reading time might be to abandon to the acquisition completely. In that way we would eliminate its radiation dose and decrease the overall number of images within the single exam. We can go even further, and amid three DCC projection thus further decreasing the radiation dose and cutting back the reading time to explore. Feasibility of such an approach, we compared these three modalities as standalone exams to the exam. One view 3D with emolo projection only and full 3D exam with both projections in 3D acquisition, we retrospectively. Identified asymptomatic women age 40 and older who under its annual screening exam in the period of 14 months and headed them on Siemens Wideangle Tomography unit. Women with history of surgery and women with implants were excluded. There were over 3100 exams, of which over 300 were called back for callback rate of 10%. Unlike in Europe, this percentage is fairly standard in United States. 22 cancers were found in 20 women or in 6.3 women per thousand screen. We selected the images for the study from three groups within the study population. First, we including screening exams of all cancer patients. Then we matched cancer patients by age and density and then selected matching exams from two groups. First, those patients who are called back from screening and had benign results on biopsy. And finally, from those patients who were not called back from screening. One screening exam consists of 2D and 3D images acquired in the same compression for each view. So each breast has 2D and 3D images. In both projections, I'll final image set consisted of 132 breasts, 22 with cancer, 22 with benign biopsy proven lesion, and 88 negative or benign breasts. Six readers participated in the study. There were four fellowship trained breast imagers with six to 20 years of mammography experience and two years of tomography experience each, and we had two breast imaging fellows in the middle of their training. The readers were blinded for results, and they were informed that the image set was enriched. Readers evaluated. The cases and two reading sessions separated by at least one month washout. In the first reading session. They first evaluated standalone One View 3D images only. This read was followed by evaluation of full 3D exam including both 3D projections, envelope and CC. Then in the second read, readers evaluated standalone to the exam first and then combine to the end full 3D exam. Each breast was evaluated separately for each read. Readers identified the lesions they would recall and assign probability of malignancy to each lesion, so called forced by rats. Please note if this was real screening, all the regions recalled weather Red. Here is 3. Four or five would be bar at 0 needs additional imaging since in real life you're not supposed to assign the final Baratz category prior to conclusion of diagnostic process. Hence the name forced Pirates we used by rats as a measurable probability of malignancy due to its high reproducibility to account for broad range of by rats for category. Anywhere chance before being cancer from 2% to 95% we used proposed direct sub categories AB&C to get more precise results. For each modality we looked at cancer detection rate and callback rate. Statistical significance was tested with Friedman test with Posthoc Bonferroni adjustment for multiple comparisons. We also evaluated effect size of each modality. This graph. Summarizes the results for cancer detection rate to the exam is Blue 3D one view 3D is orange and full. 3D example is Gray. When reading one view 3D compared to 2D exam, five of the six reader increased their cancer detection rate. While there was no change for one reader for overall increase of 15% in cancer detection rate between one view 3D and to the exam with addition of CC projection for full 3D exam, there was additional 6% increase in cancer detection rate. 2521% increased in cancer detection rate with full 3D and 2D exam was statistically significant. The results are best viewed. Presented in Box Plus chart and using cancer detection for breast. There is zero. Percent indicates that all of the readers detected cancer in that breast, and 100% indicates that sorry for zero percent indicates that none of the readers detected the cancer in that breast, and 100% indicates that all of the readers detected the cancer in that breast. Whiskers here represent lowest date lowest data point or minimum. So for example, into the exam this whisker at zero percent indicates that there was at least one cancer that none of the readers detected on 2D. Colored boxes represent where the 50% of data of results are, so you can see that the boxes are smaller as we move towards the 3D towards full 3D, indicating better agreement among the readers. Readers confidence that recalled lesion represents cancer expressed through the Pirates category assigned for each lesion increased with one view 3D compared to 2D readers confidence further increased with. Full 3D exam but remain within the same sub category. X marks represent mean cancer detection for breast, but the results are not in normal distribution and the better represented with median value. Indicated by splitline. Medium cancer detection rate progressed was 67% for two D, 92% for one view 3D and 100 for full 3D exam. Please note that 100% here means that cancer was detected by all readers in more than half of cancer breasts. One should we should be aware that practical benefit can be present without statistical significance. We evaluated effect size, effect size shows how large the effect of modality was on cancer detection rate progressed. It is complementary to statistical significant test. And does not have to give the same results. It is an indicator of practical magnitude of effect. It ranges from zero to one and these are the cutoff points. Effect size also showed significant advantage of full 3D over 2D while there was medium to large practical effect of 1 view. A 3D to versus 2D and but also of full 3D compared to one view 3D. Call back rates varied widely among the readers and were dependent on readers experience. However, improvement in cancer detection rate with 3D imaging came with no significant difference in callback rates among modalities. The comparison between one view 3D and two do exam may not be statistically significant due to small sample size. The head in the study with bigger sample, more cancers we may have seen that this difference was statistically significant with no statistical significantly changes in callback rate. One view 3D trended to have better cancer detection rate. Then through the exam, while full 3D, was statistically significant, increased cancer detection rate compared to 2D. Our results suggest that 3D exam is all we need for screening and that in regard to cancer detection, we could omit to the acquisition at all or meeting to the acquisition would have beneficial effect on radiation dose. And it would also reduce the number of images in the exam with unknown effect on screen of reading time. Thank you for your attention.

Cancer detection rate Cancer detection per breast Cancer detection rate per modality Cancer detection per breaste Cancer detection per modalte What should the future of breast cancer screening with Conclusion: What should the future of breast cancer Probability of malignancy Detection performance Screening performance Statistical methods 2 reading sessions 2nd reading sessions Detection performance of screening exam Readers Reading sessions Rnd reading sessions Results Results — Effect size Material and Methods Results — Callback Rate 2nd reading session: 2st reading sessions Radiation dose of screening exam Screening exam screening. A Image set - patients Image set Limitations BIRADS 3 Cancer detection rate per modality Cancer detection per ate Cancer detection per breast Cancer detection rate digital breast tomosynthesis look like? screening with digital breast tomosynthesis look like? Digital Breast Tomosynthesis digital breast tomosynthesis Comparison of BIRADS5 IRADS 3 BIRADS5 3 BIRADS5 4 BIRADS BIRADS5 - BRADS - BIRADS 4 BRADS BIRADS BRADS BIRADS 4 BIRADS - 3D 2D • Asymptomatic women age 40 and older Three groups of screening exams based on results of Study was sponsored by Siemens Screening exam consists of FFDM (2D) and DBT (3D) images, Cancer detection rate • Cancer Detection Rate (CDR) Cancer detection per modalt • How large the effect of modality was on CDR Separated by at least 1 month wash-out period Six readers: 2D and 3D mammography performance 2D and 3D Call-back rate per modality Cancer detection rate per modality 5% +6% full field digital mammography Full Field Digital Mammography digital breast tomosynthesis look like? digital breast tomosynthesis Cancer detection per ate Cancer detection per breaste Cancer detection per breast comparisons adjusted for multiple comparisons, were used to 41 4B 4B mammography performance in 82% / • Screening exam (Aug 1, 2016 - Sep 30, 2017) • Indicator of the practical magnitude of an effect (Cohen, 1988). • Indicator of the practical magnitude of an effect (Cohen, 2D and 3D acquisition in the same compression, for each view. • Call-back Rate (cbR) 1988). 82% 1st reading session: one-view 3D (MLO) 1st reading session: 3D (MLO & CC) 3D (MLO + CC) full 3D (MLO + CC) 3D MLO in screening. A reader study. +215% +15% Comparison of • Siemens wide-angle DBT unit, Mammomat Inspirion 3D 6 — 20 yrs. mammography experience, 2 yrs. of DBT [Small: r = 0.1 Medium: r = 0.3 Large: r = 0.5] [Small: r = 0.1 Medium: r = 0.3 Large: r = 0.51 1st reading session: 2nd reading session: FFDM DBT • 2D vs. Full 3D 2D + full 3D 3D 20 2D 2D + 3D 2D and 3D EUROPEAN socle OF RADIOLOGY EUROPEAN socle RADIOLOGY EUROPEAN socle r OF RADIOLOGY EUROPEAN RADIOLOGY EUROPEAN socle or RADIOLOGY EUROPEAN or RADIOLOGY OF RADIOLOGY RADIOLOGY or RADIOLOGY EUROPEAN EUROPEAN OF EUROPEAN socle OF EUROPEAN RADIOLOGY EUROPEAN OF RADIOLOGY EUROPEAN OFcle RADIOLOGY EUROPEAN OF RADIOLXGY EUROPEAN socle I r OF RADIOLOGY EUROPEAN soot OF RADIOLOGY EUROPEAN socle I r RADIOLOGY EUR-OPEAN socle OF EUROPEAN soote OF EUROPpEAN OF EUROPEAN soot OF SIEMENS .. • Statistical significance was considered at the a = .05 threshold 2D and 3D 2D < 3D Exclusion criteria: Hx of surgery, implants Two breast imaging fellows in the middle of their training Full Field Digital Mammography Digital Breast Tomosynthesis Full Field Digital Mammography Digital Breast Tomosynthesis digital breast tomosynthesis look like? digital breast tomosynthesis mammography Radiation dose 2. 22 patients with biopsy-proven benign breast lesion • 2D vs. Full 3D r = .70 This is a large effect. performance in 3D full 3D exam only Medical University of South Carolina BIRADS 4 BIRADS Each breast was evaluated separately for each read 2D < 3D 2D and 3D 2D 30% one-view 3D w only Screening exam screening. A Imaging set consisted of 132 breasts: r = .49 This is a medium to large effect. Charleston, SC • 2D vs. 3D MLO • 3164 exams in 3070 patients Healthineers reader study. reader study Readers identified the lesions they would recall and assigned 20 Medical University of South Carolina, Workflow, reading 2D vs. one-view 3D (MLO) vs. full 3D (MLO + CC) 22 with cancer, 22 with biopsy-proven benign lesion (matched by age and qualitative BIRADS density assessment) r = .48 This is a medium to large effect. Detection performance & reporting Average BIRADS for recalled lesions probability of malignancy (BIRADS categories) to each lesion, so and 88 negative or benign. Reader6 number of images Readers Readerl Readers Readerl fu113D vs. fu113D vs. 2D • 22 cancers were found in 20 women (6.3 0/00 ) called forced BIRADS one-view 3D vs. 20 one-view 3D w only 20 Breast Imaging Reporting and Data System BIRADS5 3 BRADS - BIRADS5 - BRADS BIRADS 4 3D MLO & CC D MLO & CC 2D 3D MLO D MLO 3D MLO 3D MLO & CC achieved in the customer's unique setting. Because there is no "typical" hospital or laboratory and many variables exist (e.g., hospital size, samples mix, case mix, level of IT and/or automation adoption) there can SIEMENS .. Healthineers Healthineers •