PEPconnect

The Impact of the Angular Range in DBT (15° vs 50°)

This webinar by Prof. Dr. Wei Zhao (Stony Brook, New York), introduced by Prof. Hilde Bosmans and Dr. Axel Hebecker, gives an overview about the effect of angular range in digital breast tomosynthesis from both physics study and clinical experience.

Due to high data volume of clinical image files, restrictions in the bandwidth may compromise image quality depicted in this webinar.

Thank you for the organizers and the Siemens for inviting me to give this presentation. Today I'm going to do first your review of the physics of Thomas instances that's related to lesion conspicuity, so this review we include other people's work as well, and then I'll present a little bit about about our own clinical experience when comparing systems with different. Angular range. So this is my disclosure. As we've heard throughout the entire Aidablu DPI conference, the digital breast remove instances as main to sway to clinical practice, and it has shown improved sensitivity and specificity over 2D mammography, an also reduced callback rage. However, there's a large variety of differences between the systems in terms of acquisition that ranges from angular range of 15 degrees. Up to 50 degrees and also the number of projections also vary greatly. So what is the impact of these different in differences in design in lesion conspicuity? So I review the physics first, showing on the left is a digital mammogram, and the main thing that's obscuring the detection of lesion is the breast structural noise, so the breast structural noise can be quantitatively analyzed using power law computation by using a regional interest and repeat the noise power spectrum calculation over the entire image. And then the power law fitting gives us a exponent which is called beta and is well known that this data is 3 for mammograms and with any forms of tomography and extreme edit will be bressi T. So I shown by several papers by John Booms Group the beta is 2 for brushing T. An for Thomas instances that lies somewhere in between and so our group has done some analysis. I will show later, but when we look at mammogram side by side with tomosynthesis size, so in the middle is the slice from a narrow angle DBT and on the right is from a wide angle DBT. You you see that the background tissue structure is looks quite different. So in the wide angle it has a more of a tomography look. Where the there the tissue overlap is somewhat minimized and you see a mass here, which appears very solid. And when you look at the narrow angle, they also has a tissue separation, but not as much as a wide angle. And then somewhere in between a mammogram and a wide angle Tomo. And it detects see mass but the mass does not look as solid because it still has some of the overlapping breast tissue structure that has not been removed. So now if we do the same structural noise analysis by going through different regions of the image slides. And that's shown here. So this analysis was done on the collection of 13 patients that participated in our clinical study, which I'll describe later. So here you see that the beta is close to three for the narrow angle DBT, whereas for the wide angle DVT the beta is 2.339, so it's less. And if you look at on the right, so this is the average value that we have changed from all 13 patients. So you have the beta equals three that we analyze it from projection from the central projection of DVT. So which agrees with the mammogram an for narrow it's larger than the beta that we could achieve for the wide angle DVT. So in terms of signal or signal profile or slice sensitivity here I was site study that was also done by Doctor Boone's Group, where they imaged a object that sandwich between two lucite slaps. This is a disc shaped object where they acquired the entire breadth Siti projections, but they use a subset of it to mimic the slice sensitivity profile for limited angle tomography. So as you go in the top row. From full sampling of brushite down to limited angle to 15 degree, you see that the bleeding to the adjacent slice becomes more severe and also when you go from top to bottom where the diameter of the disk increases with from 2.5 or up to 15 millimeter you see that the contrast of the center of the lesion also is degraded, and this is usually affect the contrast. Off the lesion in the detection, whereas if you increase the angular range than this contrast can be improved an the bleeding to addition slides can also be decreased. Several group. Phantom studies an we've heard some studies here and also they're published. Study used one example here by the University University of Michigan Group, where they have a special system that was made by GE is in the staff and shoot mode with no focal spot motion and also the angular range can be up to 6 / 60 degrees. So they did Phantom Imaging with a modified. This is the swirl Phantom that's made by. See IRS and then they have to slap where they have mass masses of different size and contrast. So when they do imaging of this Phantom with different angular range, they found that the contrast noise ratio is the highest with the widest angle acquisition. So as you see in this image, so this is 16th degree angular range 24 and then 60 degrees. So you could see the mass insert lesion much more clearly in the wide angle acquisition. So what is? How does this translate to clinical practice? Then? The story becomes much more complicated, so in our place we have a. I'll be approved study where allow us to compare the lesion conspicuity in two different tomosynthesis systems. The narrow angle Hologic Selenia and also the wide angle which is Simmons inspiration. So our clinical protocol is any women that has a buyer at zero. That's coming back for Diagnostic Gnostic follow-up. We invite them to image on the wide angle tomosynthesis, and if they're callback is based on 2D mammogram, which is only four out of the 60 patients that we have, then we will invite them to image on both systems. So just the we have a total of 60 patients an between the patients. We have 39 masses. 23 is symmetry for architectural distortion and 919 pairs of microcalcification that we could compare between the two systems. So we have a double reader study and with a radiology resident, an experienced radiologist and the readers are asked to compare the conspicuity between the two Tomoe scans and give a score. Them minus two would mean that the lesion is seen much better on the narrow angle and then a + 2 means the lesion is seen much better on the wide angle. When they do the read, they actually blinded to which type of tomo? It is and the data was compiled afterwards. So here is the result, and so the bar indicate the mean score for each of the readers, and then the bar represents the 95% confidence interval. So for asymmetry and masses we see that the wide angle tomosynthesis provide significantly better. Can't lesion conspicuity and for the actual distortion, although one reader provide better score but the other one, the confidence interval crosses 0. So here we do not see significant difference and this is perhaps because we only have 4 cases with architectural distortion. An for microcalcification we see that the narrow angle performs better than the wide angle, and for one of the readers this confidence interval also crosses 0. So now I verbally summarizes results. So in terms of breast density, we have majority of them is scattered fibroglandular fiber, glandular, itean or heterogeneously dense breast, and this is consistent with the general population. And the dose between the two different systems with respect to the breast compressed breast thickness, we can see that is similar and then those on the narrow angle is slightly higher than the wide angle by about 15%. OK, so let's resume. I guess I was talking about we have to exclude 6 cases before we were cut off and so the reason for these is be cause our study we recruit from the diagnostic callbacks and 56 out of the 60 patients their screening was done actually on the narrow angle DBT. So that's why there would be certain lesions that was seen on the narrow angle that was actually not seen. On the wide and actually is because it was deemed to be normal tissue overlap in the narrow angle DVT, which was actually confirmed by spot compression study or all the combination of diagnostic follow-up imaging. And. So there are so as I said, the six excluded cases they are due to the fact that the abnormality was called on the narrow angle and then it was seen as normal tissue overlap in DBT and so these were potential reduction in the callback. So there is one particular case where the there angle DBT was has a diagnosis of asymmetry. But on the wide angle it was seen on both view, so it turned into a focal asymmetry. That example is shown here. So the narrow angle. See, See View shows a area. Of abnormality, but that's not shown on the narrow angle, so this is a symmetry, but this area is shown on the wide angle. See See view as well as the ML. Love you so that is turned into a focal asymmetry. So this is a comparison example of masses on the wide angle and. The narrow angle let me try to change my presentation to a pointer of laser. So this in this case the mass is seen as more solid with better lesion. Margin is seem much better on the wide compared to the narrow angle. Here you have the effect of overlapping tissue. This example was actually called back with the abnormality here. Based on the narrow angle DBT, but on the wide angle you can actually see that this is actually this journalist muscle and then so it seemed much more clearly here, so there is no abnormality, so the lesion was downgraded. Here shows an example of multiple masses. Shown on arrow at wide angle DVT so this is wide angle. So as you step through it you see several solid masses. That you could entered. You can see it much better on the narrow angle. Here's another one. So as the system step through it, you see that there is the the tissue changes is very slow and when you see a mass it does not appear as solid as the other one. So this is another comparison pair where the masses seem much better on the wide angle. Also, this is a case where the density where the mass is in is quite high, so this suggests that perhaps wide angle could improve the detection of abnormalities in dense tissue that requires a bit more study to investigate. So here we show a comparison of microcalcifications between the narrow angle and the wide angle. In this particular case, the lesion is seen much better on the narrow angle DVT compared to the wide angle. So this in the case where a magnification view was called for for the diagnostic work up. So the microcalcification in this case is seen equivalently between the narrow angle at the wide angle Thomas instances. So how does this kind of difference in wide angle narrow angle for microcalcification compared to other studies? So here I used to examples of other people study. The first one is done by. Also the University of Michigan Group on the prototype G system with Steven Shoot. So here on the right hand side you see that the top row is subtle microcalcification. This is medium and this is very obvious microcalcifications. While you see no differences in the microcalcification detection in the medium to obvious microcalcification groups. For the very subtle micro calx you can see them better with the narrowest angle compared to the increased angular range. So it was thought that the reduced detection in the wider angle could possibly due to the smaller dose that was used for each projection view in that particular system, which was a see my direct detector with 100 Micron pixel size and also perhaps due to oblique entry that may have blurred the calcification more severely in the oblique angle. And so the second study I would like to talk about is actually was just presented by Linda on the first day of the IW by conference on the FDA Phantom study on five different clinical systems. So over there she saw no clear trend as a function of angular range, but there were some differences between systems, but due to the fact that the the design between different systems. For example, Steven Shoot versus. Continuous travel or? Different detector technology or even the dose is different, so it's different to draw any trend with the angular separation. So in terms of system design, then how do we? Improve one, we are given the same total dose and same system design as we see here. How would we possibly improve the microcalcification detection so the motivation is that we would like to maintain the wide angle because of its vantage in detection, detection of abnormalities. In by reducing breast tissue overlap and however we would like to improve the microcalcification detection by identifying the factors affecting its detection. For example, the focal spot motion an you could have improvement in systems using stationary source or Steven Shoot. But you could also reduce the X Ray pulses in continuous motion systems, and that could also reduce the effective focal spot. In terms of reconstruction, we actually heard several talks from EU Penn Group on the Superresolution reconstruction, so that's one example that you could improve the sharpness of microcalcification if the focal spot motion effect is minimized. And also there are other more advanced reconstructions, for example, mmore other reconstruction method that could smooth the appearance of quantum noise that could also improve the microcalcification detection. Furthermore, there have been a lot of publications and different groups doing denoising techniques in the projection domain. Try to make the load those images look similar to higher dose acquisitions. And also the angular dose or angular separation could also be made differently so that you could focus more dose to the central views. And that has shown to produce better results for microcalcification detection. So as I listed some of the references here, but I would like to go over in more detail of one of our own studies. So this is a couple of slides from a paper we presented at SPE 2020 by shall we Don? In which we used the FDA victory tool to simulate common instances with different. Acquisition strategy for a 4.9 CM digital Phantom with 1.6 milligram of total dose. So the Microcalcification is a very subtle one. And we simulated the normal and uniform those distribution by dividing the total dose uniformly between 25 projection views. Or we could have more those focused to the central views. For example, here we focus more dose more than twice of the dose to the central view compared to the normal. But then the periphery will only give 1/2 of the normal dose, so the total those remains the same. So the direction of the arrow we focus more and more those towards the central views. You can see that the microcalcification is not detectable in the uniform, but then it starts to become visible on the non uniform dose distribution. So observers study was conducted using a four AFC observer study tool that was developed by Doctor Bassmans Group and so here is result of percent correct detection of the microcalcification as a function of the different angular dose distribution schemes. You can see that overall conditions the non uniform scheme provided better detection compared to the uniform dose distribution. In terms of other factors, you see that if we assume a perfect system without any. So that is the yellow curve without any focal spot motion. An sorry the blue curve without any focal spot motion and no scatter radiation. That which causes increased quantum noise. So we see that the. Scatter radiation has very little effect on the total score, but then if you start to add focal spot motion then the percent correct start to decrease. So this is 1 factor that the system implementation could really use a lot of care in the design. So on the right is a visibility score for the five for the six microcalcification clusters you also see the same improvement with the non uniform dose distribution. In summary, I provided overview of the physics of digital breast tomo synthesis in terms of the bright structural noise and how it depends on the angular range and also how the signal profile changes as a function of angular range and why that leads to a better detectability of mass with wide angular range. And also I presented our clinical experience with a small. Number of patients and we show that we have better detection of mass and focalor symmetries in the wide angle versus a narrow angle, and there were a few cases which shows that the wide angle DPT can potentially further reduce callbacks. And this is becauses separate the overlapping tissues better in wide angle compared to the narrow angle. So in the future direction in the future work we would like to extend the clinical study to a larger patient sample size and also with some indication that the microcalcification in the wide angle is perhaps degraded due to multiple factors. Would like to improve the system acquisition as well as reconstruction algorithms. To improve their visualization of microcalcifications. Finally, I'd like to thank our colleagues, which are students in our labs. Current and former students, as well as our clinical collaborators, as well as very helpful discussions from our collaborators in Siemens. Healthineers. Thank you for your attention.

4AFC Observer Study Result Z J 24-27 U" 2020 ß vs. Angle: From Clinical Data Angular dose distribution scheme Angular Dose Distribution Clinical Investigation Asymmetry on Narrow Angle to Focal Asymmetry on vs. Angle: From Clinical Data Tomosynthesis Acquisition Geometrie Strategies for Improving MC Detectio Strategies for Improving MC Detectio1 Strategies for Improving MC Detect101 Signal Profile / Slice Sensitivity Published Phantom Study: MC Detecti( Published Phantom Study: Phantom Study: Mass Detection Breast Structural Noise Double Reader Study Acknowledgments Summary Results MC Mc Detecti Score of 2 WIDE Stony Brook Medicine Wide-angle DBT DEPA NARROW WIDE Patient breast density: Sternalis Muscle Reader Readers: Chan et al, Radiology 2014 Physics of DBT Maintain wide angle advantage: mass, asymmetry, architect IRB approved prospective study Used VICRE, 4.9 cm digital phantom, 1.6 mGy total dose Goodsitt al, PMB 2014 ß decreases with wider angle - better detectability Nosratieh et al, Med Phys 2012, use subset of bCT PI Nosratieh et al, Med Phys 2012, use subset of bCT p Nosratieh et al, Med Phys 2012, use subset of bCT pi al, PMB 2014 NARROW better visualization on 240 40 400 300 Architectural Reader Hailiang Huang, MS X-ray Source Cluster of microcalcifications: 120 microns • Fatty: 4, Scattered Fibroglandular: 23, Heterogeneously Dense: 30, Extremely • 2nd year Radiology Resident : Reader 1 • In-plane breast structural noise decreases with wider angular range 5.0 1.0 Step-and-shoot Focal spot motion (FSM) Wide angle Wide Angle DBT • 16 - 64 degrees 17 - 60 degrees 16 - 64 degrees 16 - 64 qegrees Impact of Tomosynthesis Angular Range on Lesion Xiaoyu Duan, MS distortion 1.0 150 1.00 3600 1200 1200 3600 1.00 1800 3600 3200 3.08 1200 Patients undergoing diagnostic work up (BIRADS 0) were Uniform: Normal dose / 25 views DBT Central DBT Central Narrow-angl Wide-angle arrow-angl Wide-angle Dense: 3 4.5 3.5 • Breast Radiologist with 30+ year experience: Reader 2 David A. Scaduto, PhD • Mass detectability increases with angular range 17 - 60 degrees • 17 - 60 degrees 16 - 64 degrees R 0.9 • Stationary source or Step-and-shoot Manufacturer Angular Range Projections Projection DBT DBT • CNR highest with wide angle acquisition Impact of Non-uniform: central: N *Normal, peripheral: 0.5*normal Yue-houng Hu, PhD Conspicuity: Physics Study and Clinical Experience imaged on both DBT systems: Narrow angle Narrow angle DB T Narrow-angle DB T Wide angle Wide-angle DBT Wide angle DBT Dose and Compression Thickness: —3.5 3.5 Sensitivity for subtle MC Breast structural noise Breast Structural Noise Clinical potentials for wide angle DBT Lesion conspicuity compared on 5-point scale • Shorter x-ray pulses Wide-angle DBT Wide angle ß = 2.87 Average 3.08 2.81 2.39 2.81. 2.39 central 7 = 2.33 X central 5 = 3 X central 3 = 4.67 X SIEMENS uniform Calcifications • Observer preference increases with wider angle Observer preference increases with wider angle 00.9 3.0 5.08 0.08 5.0 3.01 1.0 1.01 Radiologists and clinical staff 2.5 mm 15 mm higher for narrow angle .61 • Narrow angle: 2.439nGy, 64.2 mm 150 Hologic • Narrow angle: 2.43 mGy, 64.2 mm • Wideow angle: 2.03 mGy, 64.4 mm 3.08 0.001 0.01 Reconstruction: enhance MC and reduce noise • Better detection for masses and focal asymmetries • Narrow angle: Hologic Selenia, AR = 150 • Narrow angle: Hologic Selenia, AR = 1500 • Narrow angle: Hologic Se!enia, AR = 150 2.5 00.5 3.0 25 00.9 3.5 4.5 Standard ADS-2 ADS-I ADS-3 ADS-4 ADS-I • Power law Power law 0.25 0.22 0.24 0.22 0.24 0.25 Anastasia Plaunova, MD -2 -1 Lesion seen mucghtly better on Narrow than Wide 00.7 Depends on recon, detector Deviation = 2.0 •e Average FSM, with scatter) Average FSM, with scatter) Average ('"th FSM, witscatter) Average (with FSM, scatter) • Downgrade lesions and reduce callbacks • Downgrade lesions and reduceeallbacks 16d17p 2.39 24d9p 39 o- Average (with FSM, noth scatter) Average ('With FSM, scatter) • Narrow angle: 2.43 mGy, 64.2 mm • Wideow angle: 2.03 mGy, 64.4 mm • Wideow angle: 2.03 mGy, 64.4 mm'i Paul R. Fisher, MD Image noise 0.001 Healthineers 25 0.5 —o- Average (with FSM, with scatter) —o- Average (noth FSM, with scatter) —o- Average (no FSM, VGth scatter) o- Average (with FSM, noth scatter) —k Average ('"th FSM, with scatter) —ky Average (noith FSM, with scatter) •e- Average (noith FSM, with scatter) —o-y Average (with FSM, with scatter) •ey Average (with FSM, with scatter) •e- Average (wWith FSM, with scatter) — 1.5 • Wide angle: Siemens MAMMOMAT Inspiration, AR = 500 Wide angle: Siemens MAMMOMAT Inspiration, AR = 500 oblique blur —o- Average (no FSM, scatter) o-y Average (with FSM, with scatter) Average (no FSM, scatter) • Increase with breast density (sample size N = 13) • Separate overlapping tissue in areas of high density -2 -1 Lesion seen slightly better on Narrow than Wide Lesion seen mucghtly bette& on Narrow than Wide Chunling Liu, MD, PhD Asymmetry Wei Zhao, PhD o- Average FSM, no scatter) Average (no FSM no scatter) b 0.6 5 mm 15 mm 0.96 0.01 0.22 0.95 6 Excluded cases: • Denoising through deep learning Kim Rinaldi, RT •e Average (with FSM, with scatter) o- Average (wWith FSM, no scatter) •e Average (with FSM, no scatter) —cy Average (no FSM, VGth scatter) —o-k Average (noth FSM, no scatter) —o- Average (no FSM, VGth scatter) •e Average (with FSM, noth scatter) o- Average (w'ith FSM, no scatter) —k- Average (noith FSM, no scatter) o- Average (no FSM, no scatter) L. Ikejimba, IWDM 2020 L. Ikejimba, bWDM 2020 —cy Average (noith FSM, with scatter) —k Average (no FSM, noth scatter) —o- Average (with FSM, with scatter) —k- Average (with FSM, noth scatter) —o- Average (no FSM, noth scatter) —o- Average (noith FSM, with scatter) —X- Average (no FSrJ no scatter) Average (no FSM," scatter) Average (no FSM, noth scatter) —o- Average (no FSM, with scatter) •e Average (with FSM, no scatter) Average (no ESM, no scatter) 00.8 1.5 0.9 00.9 0.25 00.25 00.5 Planmed 3.0 1.0 30 308 • =3 for mammograms Lesion seen equally well on Narrow and Wide 60 patients: total of 39 mass, 23 asymmetry, 4 architectural 60 patients: total of 390mass, 23 asymmetry, 4 architectural Future directions: • Angular dose distribution: Focus more dose to central views • Angular dosedistribution: Focus more dose to central views 0.9 2.25 1.0 0.76 00.9 H Huang al, RSNA 2018 H Huang al, H Huang al, RSNA • Mass or focal asymmetry only seen on narrow angle DBT Discussion with scientists from Siemens Healthineers 5 clinical systems ADS-I ADS-2 ADS-3 ADS4 ADS-2 ADS-I ADS-3 distortion: 19 calcification comparison pairs Physics Study and Clinical Experience Lesion seen slightly better on Wide than Narrow@ Lesion seen slightly better on Wide than Narrow distortion, 19 calcification comparison pairs distortion, 15 calcification comparison pairs • = 2 for breast@CT • 2 for CT • = 2 for breast CT • = 2 for 2 for CT Extend clinical study to a larger sample size Angular dose distribution scheme Angular Dose Distribution Angular dose distribution scheme Angular Dose Distribution • = IMS 3.0 240 40 300 30 Mass 15 mm 10 mm Axel R. Hebecker, PhD No clear trend with angle • Same areas of interest only showed overlapping tissue on wide angle DBT Same areas of interest only showed overlapping tissue on wide angle DB T o. • Improve visualization of calcifications through enhanced reconstruction algorithms and • Improve visualization of cabifications through enhanced reconstruction algorithms and Thomas Mertelmeier, PhD Lesion seen much better on Wide than Narrow Lesion seen slightly better on Wide than Narrow Lesion seen much on Wide than Narrow Lesion seen slightly better on Wide than Narrow@ Final diagnostic work-up showed superimposed tissue • Increased detectability with lower ß Increased detectability with lower ß (Qiang et al. Med Phys 2012) (Zheng et al, Med Phys 2019) Impacted by dose and other acquisition protocols 500 Siemens Sebastian Vogt, PhD 25 2.5 (Acciavatti & Maidment, Med Phys 2012) IE-5 IE-6 More dose on central fewer projections -2 Readers were blinded to the type of DBT system Readers were blinded tb the type of DBT system factors Steffen Kappler, PhD **One case showed asymmetry on narrow angle DBT which was focal (Sahu et al. ISBI 2019) (Sahu al. ISBI 2019) al. ISBI 2019) ISBI 2019) Narrow angle Narrow-angle DBT Stony Brook Chen et. al. Med Phys 2012 Chen et. al. PMB 2013 Narrow angle Narrow-angle DB T Narrow-angle DBT NARROW Narrow angle Narrow angle DB T Equally Narrow-angle DBT Wide-angle DBT Narrow-angle DB T Wide angle Wide angle DBT (Nishikawa et al 2007) asymmetry on wide angle WIDE 15 mm 10 mm Mammogram Wide angle Wide-angle DBT Frequency (mm- Frequency (mm (Das, Gifford et al. 2009) (Nosratieh et al, 2012) (Nishikawa et al 2007) NARROW WIDE DBT slice DBT slice Medicine much better slightly better well slightly better much better NARROW WIDE X. Duan et al, SPIE 2020 X. al, SPIE 2020 X. Duan et al, 2020 Narrow angle Narrow angle (MLO) Narrow angle DB T Narrow-angle DB T Wide angle (MLO) Wide angle Wide-angle DBT Wide angle (CC) Narrow angle (MLO) (Hu and Zhao 2011) (Zhao al, 2009) (Zhao et al, 2009) (Nishikawa et al 2007) X. SPIE 2020 Digital Radiological Imaging Laboratory

  • breast tomosynthesis
  • DBT
  • wide-angle tomosynthesis
  • dose
  • image quality