The Future Role of Synthetic Mammograms – Is it more than a 2D Image?

So if we tell talk about digital mammography, an teacher, breast Thomas synthesis, we will also next step is talk about synthetic mammogram an if this is a business solution between screen with 2D DBT Duty fulfilled, digital mammography, an this the best of synthesis alone. So this is my conflict of interest. Synthetic mammogram is not designed as a copy of a 2D full field digital mammography, but as a useful guide to evaluate all the information coming from all the slices of. The DBT the processing algorithm used to create synthetic 2D weights line structures as well as high contrast features more than background, fits an fiber Andra tissue, so this resulting in increasing conspicuity of architectural distortions. Speculations an micro classifications. So if you add. Full field digital mammography, 2D2 DBT. You certainly had an increased acquisition time and increase radiation dose. And of course an extra image is also an increase reading time. Do we need these two DI? Think yes, we are sure. She surely need this 2D image not only for comparison with previous to the images. Comparison with DBT images from different vendors can help us for the evaluation of all the DBT information. And as the end doesn't overview of the mammogram, it helps if you see the relation between the Masters, the microcalcifications position of the nipple. So this is a combination comparison of inside 2D of 2018 with the previous to the full field mammography. In 2016, if we replace the 2D full free digital mammography by a synthetic mammogram, and we combine synthetic mammography and tube T, we finally get, and it's very obvious, reduction in radiation to San. So these confirm that it's around 40% and also with your reduction in. Acquisition time, it's not clear that we will receive if we will get. Clearly a reduction in reading time. So if we get in this look at the overview, I mainly will talk about the detection performance in the combination of synthetic DBT and radiation. Those is proven, yet not, but I have no information on the workflow reading and reporting. I will also talk about alone only about screening because diagnostic imaging and synthetic mammography is and all different story. So what we can evaluate is the evaluation of synthetic and EBT compared to versus to the full fee. Digital mammography and EBT and also compare synthetic mammogram alone versus full field digital mammography. We know from the study of Oslo trial and bar scanner that improving the quality of the synthetic mammogram also had an improved cancer detection rate, so we have not. Sofia already showed you about these recent trials and large prospective trials. I will inform you also about some other retrospective trials, like in 2016 there was in the one ARM, DBT +2 D 15,000 patient and they had recall rate of eight point 8%, which is quite high. If you compared to the European study an the synthetic mammogram had recall rate of. 7% so it was a slightly decrease in the recall rate and bitter at all was also retrospective study. With 14,000 in the synthetic mom arm and again very slow, low reduction in recall rate. The ensure it all, it was a respective study, but he adjusted for age, race and density an between the two D and the synthetic to DBT arm there was 50% reduction of the recall rate, so now it appears to be like in the European level of 4% recall rate and the paper of free rent all adjusted. Also for priors and radiologists and also. Was able to find reduction in recall rate and here the information coming from the synthetic mammogram was more clear as he looked for all the adjusted for all confounders, so it was more the effect of the synthetic DBT compared as an effect compared to the to the DBT for the cancer detection rate. In all of these studies there was not more, there was no difference between the synthetic. Arm and to be the DBT arm. So it was 8000 in the Storm 2 trial and around 5 to 6000 in the three other studies. We can think about what should the synthetic mammogram alone perform, or what is the performance synthetic mammogram alone versus the full field. Digital mammography, and we've certainly look at the micro classifications in this paper on detection of microcalcification 71 cases. The detection of the classification was fair to moderate agreement, and you see the difference between the three readers. For the bar, its classification of the classifications, it wasn't moderate agreement. He's at all data. Prospective evaluation of 1000 screening mammograms. An concluded that the synthetic mammogram safely had significantly better conspicuity for classifications and distortions. While EFM was better for Masters and are symmetries, so again we see some difference is there between the synthetic mammogram and full field digital mammography. If we look at them alone without DBT. This is another case. Another study of 79 screen detected cancers and also there there were differences in Masters and a symmetries where more less more, often less conspicuous at synthetic mammogram compared to full field digital mammography. What about the density? It's very important that the synthetic mammogram at least repeat or shows the same density as a full field digital mammography as a lot of our risk assessments protocols include breast density. And in this paper in 2017, the agreement between consensus Barrett sensitive categories assigned 4 full Fiji mammography and synthetic to the was 80%. So a very good. Cap agreement and also the in the reader agreement various between 0.5 to 7.6 for the digital and the same for the synthetic mammogram order. Studies also repeated this sometimes synthetic a little bit less density reported compared to the full field. Digital mammography. We don't have only 2D synthetic mammogram. There's also the rotating 3D synthetic mammogram. Autorotating mammogram alone. In this paper of Rodriguez settled, they evaluated retrospectively 190 cases, 27 Millennium cases, and they compared synthetic mammogram and rotating mammogram in the rotating mammogram and synthetic. There was in the area under the curve was quite similar. They found also similar sense and sensitivity, but to specificity for the rotating mammogram was higher. They this set of images was also set to the transparent an here you see the results of the DM read by the computer and you see that indeed the DM also bothered computer had a higher area under the curve compared to the synthetic mammogram. You can also use cats on the synthetic mammogram, not only on the DBT slices but also on the synthetic mammogram, and this is a study by done by GE on the on 80 DBT cases. Look at the difference here. For this one reader who the performance disincentive T 100% for the EFM but turned out to below 50% for the synthetic mammogram but adding cats again increased to 94% sensitivity. The difference is is more regarding the mouse is an. You see if you add cat then area under the curve increased for the synthetic fan 0.5 to 7.8 an it was more pronounced. Impact was more pronounced for smaller lesions, so if you'd get again this differences in readers and the differences in performance of synthetic mammogram can be increased. In this study we also were curious if you have a better. Quality of synthetic mammogram. In this case, a new prototype of the inside of the synthetic mammogram by Siemens. We wanted to look if the quality regarding related to the older version and to the efmb was equal. And what about? Also the diagnostic accuracy. In this study we evaluated the synthetic mammogram alone versus the other tree two versions. So there were 184 cases which were acquired in the combo to mode. So in the same position, 130 normal cases and 71 abnormal cases. The study was only supported for by the synthetic Mamas provided data collection and statistical analysis was done by the University hospitals in love. First, we did a visual to look at the quality of the images we did. Visual grading study. You see two images here. From this is a full field digital mammography, and this is the version inside Version 2. The new prototype you see here. The classifications, the artifacts are low. You see in the previous you had a lot of artifacts, but they were kept a very low. You see tiny calcifications here. There a little bit more dense and irregular, but this is a postoperative area and also the area. The subcutaneous area was quietly preserved. Also the densities. So we see a much better quality here in this image and how we did the visual grading study. We did a side-by-side comparison we used for we evaluated six image quality criteria on the left on the right image and comparing to the left. So we evaluated opacities, classification, noise, contrast, sharpness and overall diagnostic quality. In this case, there was one experience. Reader for the valuation. Secondly, there was also a detection study using the following score as normal be 9 together as probably be 9 as corresponding to by at three and then possibly malignant, probably malignant and definitely millenniums. We also scored the breast density and the FM delight likeliness of the three. Which of the three groups there were two experience readers? One was experienced with synthetic mammogram and DBT. One was only experienced with DBT. This is statistically analysis and results as you already ordered could see on the images, there were visual grading for the study, showing for the opacities a significant, better representation of opacities inside version 2 compared to the inside version one, and I will see was seven points 0.7 also for the contrast much better contrast. I show you here the three images this. Breast contains suspicious mouse in the upper part of the breast. You have the EFM, the older for the current version version one, and then the prototype version 2 and you can appreciate that contrast is much better correlating to the EFM. Also the inhomogeneous area of the lesion is better correlating to this structure you see there is less noise an this is also for the further. Evaluation like for the microcalcifications significant better representation for microcalcifications, but a little bit less than for the opacities and the sharpness was the same between all the images. This is the case also postoperative case with a lot of different types of classifications. In it, you see the classifications look brighter. Some of the classifications come together and make a little bit larger. Calcifications some calcifications here where better seen on the. The synthetic mammogram compared to fulfill digit mammography and some law smaller classification. We're not visualized here. So overall the score was quite good noise. There was less noise in the newer version and overall diagnostic quality was significantly better for the version to compare to the version one. So as a conclusion for the visual grading study, inside version two is rated overall very similar to ever DM and Superior Tour inside version one. This of course reflects in the of them, likely the score where the median scored for the version two was eight on 10, an only for the version one was three on 10. Surprisingly, was the rock results in New Roc. We see that for one reader, the area under the curve for Seven Point 8 and for the new version Zero Point 77 here is the black line here, and this is the older version, so we based on the quality. We thought that rock or the accuracy which much higher for the version 2 and we see some difference also with the second reader where the. Version two scored lower. We need to know to go into detail and see if it's related to speculations, masses or microcalcifications. Or is it really related to the experience of the reader with a synthetic mammogram? If you look at the data and we D optimize the data in four score one we combine it to no follow up as we do in screening and scored 3, four and five as a recall, like in screening. Then reader one. Indeed, the cancer detection rate. Increased from version one to version two and also FTM has a higher cancer detection rate with also the highest sensitivity. An accuracy was also higher forever DM, but version two was certainly in the Magnum artist similar to ever DM for the reader. 2 The cancer detection rate of the two synthetic mammograms was lower compared to the EFM, so there was no difference there. Regarding the synthetic mammograms, but a large difference for the EFM. For the density, we saw a very good agreement cap agreement between inside version two and EFM for both of the readers to conclude the accuracy of the synthetic mammogram alone is lower than. I've of them and it's influenced by the lesion type and also by the experience of the lesion, so I think it's fair conclusion that we cannot use synthetic mammogram in screening alone. We also we always need to use it together with DBT an I think I prefer to first look activity and then at synthetic mammogram. The quality of the reconstruction algorithm has an impact on the performance on synthetic mammogram, but so far it's not clear in which impact is largest for the speculated mouse for the classifications. Offer all type of lesions of just related to the experience of delicious, but I should advise that the physical technical evaluation of these synthetic mammogram and images is a mandatory field in the physical technical evaluation of DBT and I also will my college acknowledgement for my Hilda Bosmans and Leslie got Martin who provide given me all the results for this study.

I@ UZ LJZ LEUVEN Conclusion Conclusion CAD and SM SM versus FFDM - SM versus FFDM Rotating Mx Results SM+DBT versus FFDM + DBT Visual grading study Visual grading study (1) LEUVEN Synthetic mammogram (SM) LEUVEN LEFUVEN 2DFFDM 2616 grading "sight 2D 2018 20 2018 2D 2018 2DFFDM 2016 2 (l -V InsightV1 AUc FFDM_ FFDM Insight V2 vs Insight V2 The future role of Synthetic What should the future of breast cancer screening with Performance of synthetic mammography (SM): The accuracy of synthetic mammography alone Visual grading study (1) Visual grading study (3) Visual grading study (2) Visual grading study (4) FFDM LEUVEN Cancer detection rate ACR breast density Study set-up UZ Leuven Statistical analysis Synthetic mammography Synthetic mammography in Conflict of Interest FFDM likeness score ROC results Detection study 2D FFPM + RECALL RATE Evaluation of quality of SM • Not designed as copy of 2D FFDM but to off set the AJR 2017: 209 AJR 2017; 209 lhsiOhtW Comparison Between Digital and InsightV2 InsightV? lhsi6htVf The future role of synthetic Synthetic mamm gr Euro—n Journal of Radiology Euroran Journal of Radiology The future role cf Synthetic mamm g digital breast tomosynthesis look like? is lower than FFDM. is lower than F F DM. 71 SM and DM cases, Hologic Mammography mammography Synthetic 2D Mammograms in Synthetic 2D Mammography in Synthetic 2D Mammogramsy in limitation of FFDM - a useful 2D guide. CAD 2D+DBT SM+DBT 2D*OBT 2D+DBT SM+DBT Microcalc Detection ( Agreement K; for RI Detection ( Agreement K; 0.4 for RI Breast Density Interpretation is influenced by the lesion type and the experience of the reader. Agreement between Overall diagnostic Sharpness Opacities Contrast Subjective image quality study with side-bv-side evaluation ROC Evaluation Of a computer-aided detection • Median score for InsightV1 : 3 • Data dichotomized • Detection Study (Receiver Operating Characteristic study) using following Bridging solution between 2D FFDM and DBT? Speaker's fees and travel support from Siemens Healthineers (with 184 cases acquired in combined FFDM-DBT mode 11000 1'1000 11000 Purpose: and for R2 and and for and for R3) and for R2 and for R3) and for R2 and 0,58 for R3) • all six IQ criteria The processing algorithm used to create s2D weights line Noise consensus BI-RADS Mammography mammography 3.7% 4.45% 5.45 o' and and Of and 01 and o' and o' depiction and (CAD)-enhanced 20 synthetic mammogram: Reduction in (reference Visual Grading Characteristic study) VGC analysis6 Comparedto FFDM + D T Compared to FFDM + DBT increased acquisition time FFDM + DBT In screening, SM should only be read additional to DBT. In screening, SM should only be read additional to DBT quality STORM-2 0.03 BIRADS (moderate agreement) BRADS (moderate agreement) 0.5 8.8% 8.8 • Introduction Attendance rate comparison with standard synthetic 20 comparison with standard synthetic 21) comparison with standard synthetic 20) comparison with standard synthetic 21:) comparison with standard synthetic MDEON approval) (combotomo) on a Inspiration Siemens system (Forcheim, density categories 2013-2015 2015-2016 9,672 exams - Score 1 and 2 (BRADS 1, 2, 3) - Score 1 and 2 (BRADS 1, 2, 3) -5 0 — R2: 689% agreement, kappa = 0.55 = 0.53 To compare the clinical image quality and diagnostic accuracy of a score structures ( architectural distortions and spiculations), as RI = 0.74 = 0.71 7.7% mortality 2013-2015 • radiati n dae• 'Q%reductiqn in radiati n dae• 'Q%reductiqnin radiati n dae• conventional conventional digital Most sign in digital 6 image quality criteria using a 5-point rating scale (-2, -1, O, +1, +2) 6 image quality criteria using a 5-point rating scale (-2, -1, O, 0 p Bernardi et al All recalls incl All reca Is incl assigned for FFDM and The quality of the reconstruction algorithm has an impact on the Is it more than a 2D image? The future role • increased radiation dose increased radiation dose mammography Germany). - Score 3, 4, 5 (BIRADS 4a, 4b, 4c, 5) - Score 3, 4, 5 (BRADS 4a, - Score 3, 4, 5 (BRADS 4a, 4b, 4c, 5) — R2: 68% agreement t kappa 0.55 - Score 3, 4, 5 (BIRADS 4a, 4b, 4c, 5) 0.55 RI AUC R2 0.77 = 0.71 Bernardi et al • Performance of synthetic mammography (SM): The accuracy of synthetic mammography alone prototype version of synthetic mammogram (InsightV2) to the current - Definitely normal/benign (BRADS 1 and 2) Performance of synthetic mammography (SM): - InsightV1 vs FFDM -0.1 - InsightV2 vs FFDM -0.1 Insight V2 vs FFDM well as high contrast features ( calcifications), more than The future role of — The future role of synthetic mammograms — InsGhtV2 EUROPEAN SOCIETY OF RADIOLOGY • Median score for FFDM: 10 • Representation and characterization of the opacities • Representation and characterization of microcalcifications • Representation and characterization of the opalcificatiens synthetic 2D was 80.3% synthetic 2D was AUC • SIEMENS .. SIEMENS Auc No significant difference between the compared modalities il confidence interval ot AUC ot VGC curve contains 06 of synthetic Dense breasts Zuckerman et al performance of SM. et al 5,366 5.366 • 0 89 RI R2 RI 083 083 7.83% AX • - Probably benign (BRADS 3) - Probably benign (BIRADS 3) - Probably (BIRADS 3) v30%'réduction in -acqui tioq tirie in acqyi tior) tirie • InsightV2 vs FFDM: • InsightV2 • InsightV2 vs FFDM version (InsightV1) and to Full-Field Digital Mammography (FFDM) y-30% reduction in acqyi tion time specificity • Increased reading time (95% Cl, 75.4-84.5%) — InsightV2 vs InsightV1 0.6 background fat and fibroglandular tissue. - SM + DBT versus FFDM + DBT Personalized I' AUC 05: preference for if AUC preference InsightNew • RI RI Is it just a bridging solution? 32,076 2017 2016 32,016 8.7% 3.7% 840/0 8.8% 5.83% 8.83% 7.83% 8.7% 6.39% 7.13% 4.45% 7.63% 3.7% Strengths and Weaknesses of 5.035 Zuckerman et al mammography mammogramsy mammograms SM had significantly better 5.457 7.3% Radiation dose with 0.73 (95% Cl, with K 0.73 (95% Cl, Insight VI vs Insight VI vs FFDM peo.003 Physical-technical evaluation of these SM images is a mandatory vs — 113 abnormal cases — Possibly malignant (BIRADS 4a) — 113 abnyormal cases Overall presence of noise • R2 Ins•ghtVI R2 Insight V2 InstghtVI R2 Insight Vv Insight vs Ins ghtV I Ins ghtV1 RI Insight Vs Insight V 2 vs screening Synthetic Mammography in 2016 — RI: 70% agreement, kappa = 0.53 - SM versus FFDM conspicuity for calcifications and conspicuity. for calcifications and Reading time?? — Is it just a Eos Ambinder et al Aujero et al Retrospective 7,843 14,722 • Resulting in increasing conspicuity of these lesions Insight TABLE of TABLE o' 63% 18% 63% Insight VI Insight vs InsightVI R2 0.5 80 DB T cases (23 malignant) 80 DBT cases (23 malignant) 80, DBT cases (23 malignant) vs 0.66-0.79). = 0.71 0.69 Chantal Van Ongeval - • RI R2 Dngeval Trapezoidal curve of ROC and VGC data Screening' Radiographics 2017 Overall contrast architectural distortions (P .001), architectural distortions (P architectural distortions (P < .001), architectural distortions (P <001), field in the physical-technical evaluation of DB T. Ins ghtV I Insight vs Ins ghtV 1 InsightV I Rotating Insight, Siemens. 20 20 20 "d 20 "d 20 — Probably malignant. (BIRADS 4b and 4c) bridging Insight VI vs Insight V2 p-o.137 p:o.137 p=o.137 p-O.137 3 modalities: FFDM, current Insight (insightVI)and new 2015-2016 FFbM 7.13% 7.83% 7,060/0 — R2: 79% agreement, kappa = 0.70 — R2: 79% agreement, kappa 7.1% = 0.70 82% Senoclaire, GE Senåclaire, GE • Evaluation of the quality of a new prototype of • - Retrospective, monocentric study RI — Comparison with previous 2DFFDM = 0.69 while FFDM had significantly better Obuchowski-Rockette method with (co-)variance analysis based on jackknife method Overall sharpness Overall diagnostic Retrospective Workflow, reading Ambinder et al Am binder et al 8.8% 5.03 5.52% 8.25 5.8% 5.57 5.457 5.82% Giess et al., Freer et al. Aueer et al. Van Ongeval Chantal, Lesley Cockmartin solution? — Definitely malignant. (BIRADS 5) conspicuibj for masses and "Uses the CAD detections to Detection • In screening? protoype Insight (InsightV2) Zuckerman et al Aujero et al petrospective Retrospective Petrospective retrospective 32,016 32,076 15,571 30,561 iconspicuity for masses and conspicuity for masses and 16,173 10,571 15,571 Healthineers LeuvenIBelgium R2 0.77 R2 = 0.77 • Overall diagnostic quality Overall diagnostic quality 190 cases Insight (Synthetic mammogram Siemens) & reporting — Comparison of the DBT images of different systems InsightV2 is rated overall very similar prosp 1206 screeningmx 0.5 083 0.83 0.54 0.3 000 020 040 080 000 020 04.0 080 0M 020 040 060 080 0" 020 040 060 080 0.00 0,20 0.40 080 0.00 0.20 040 080 performance create a CAD-enhanced D...t.170 20 70 70 Postema Sandra, Keupers Machteld, Renate Prevos, Hilde Bosmans McNemar test asymmetries (P < .001) 2017 Adj for age, Adj age, 3.7% 4.45% 4.45% ACR breast density (5th edition) (27 malignant, 31 benign, 132 normal) (27 mallgnant, 31 benign, 132 normal) (27 mallgnant, 31 benign, (27 malignant, 31 benign, 1 32 normal) Insight VIvs InsightVI Insight V2 Insight VI InsightVI R 82% 72% Aujero et al Zuckerman et al Aujero et al. Freer et al. Petrospective petrospective Retrospective 5.3 5.03 6.03 6.4 6.3 6.1 1 experienced reader Insight vs Ins•ghtVI R2 pool • In diagnostic imaging? S' synthetic image by blending the 02 — Overview of the mammogram (relation mass, microcal race, density 0 00 020 080 080 020 SM alone RM alone Support for this study was provided by Siemens (Forcheim, Germany) who Significant better representation of opacities in InsightV2 I V2 V2 39% 74% 39% Insight V2 vs FFDM Insight Insight V2vs Insight V 2 vs More invasive p 20.000 diagnostics to FFDM and superior to InsightV1 Significant better representation of microcalcifications in Sigoificant better representation of microcalcifications in five detections from the planes and position Of nipple) FFDM likeness ( IQ-point rating scale) were scored FFDM likeness (10-point rating scale) were scored Masses and asymmetries were more RM alone created synthetic 2D images for the reading study. Freer et al Aujero et al Freer et al. Aujero et al. 21 ,435 1,019 9,525 Chikarmane et al's Zuckerman et al compared to InsightV1 9th Doy 9th Dog 9th Day into the corresponding standard 0.0 0.3 0.2 02 0.2 often less conspicuous at SM than at often less conspicuous at than at often less copspicuous at SM than at @ Siemens Healthcare GmbH, 2019 InsightV2 lower than for opacities 7.83% 7.63% 6.39% 4.39% 4.45% 4.35% 5.52% 4.45% 8.7% 7.83% 9.525 Data collection and statistical analysis was done by the University Hospitals Adj priors, age, Adj pciors, age, Adj age, Adj priors. age, Freer et al. Synthesized Mammography: Clinical Evidence, • Two experienced readers • Two experiencéd readers Synthetic mamm gr synthetic image." FFDM when compared with Significant better noise level and overall diagnostic review of SM and FFDM 79 Significant better image contrast in InsightV2 compared density, Leuven. • SQ A ROC • A ROC No significant difference in image sharpness Appearance, and Implementation Acknowledgment = invasive Breast Care Day 0 0.1 0.2 0 a 0.7 08 0.9 0 0.1 0.2 0.7 08 0.9 0 0.1 0.2 05 0.7 08 0.9 calcifications. UNIVERSITY HOSPITALS LEUVEN CE n The statements by Siemens Healthineers• customers described herein are based on results that were screening-detected cancers radiologist quality for InsightV2 compared to InsightVI in situ = in situ Collection of anonymized DB T data was approved by the ethical committee of the Collection of anonymized DBT data was approved by the ethical committee of the to InsightVI InsightVI InstghtV I Prof Hilde Bosmans, Lesley Cockmartin ECR 2019 -RAD S c d -RADS c d -RADS c d c positive Fraction in •G NGC ) to 2 to hospital and informed consent was waived. 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 calcifications. SIEMENS .. SIEMENS ... Healthineers Healthineers •