Collection of Work Projects

Transcription

1 EUTEMPE-RX: European Training and Education for Medical Physics Experts in Radiology MPE05: Anthropomorphic phantoms Collection of Work Projects Varna, September

2 2

3 Introduction: The present booklet is created from the report of the trainees in the MPE05 module of the pilot running of the EUTEMPE-RX course. The content of the reports is not changed. The reports have been reformatted to meet a uniform style for the whole collection. One or two slides used by the trainees during their oral presentations are also added to each work. 3

4 4

5 Table of Contents PROJECT ASSIGNMENT No PROJECT ASSIGNMENT No PROJECT ASSIGNMENT No PROJECT ASSIGNMENT No PROJECT ASSIGNMENT No PROJECT ASSIGNMENT No PROJECT ASSIGNMENT No PROJECT ASSIGNMENT No

6 6

7 EUTEMPE-RX: European Training and Education for Medical Physics Experts in Radiology MPE05: Anthropomorphic phantoms Applications of anthropomorphic phantoms for design and evaluation of advanced x-ray imaging techniques PROJECT ASSIGNMENT No. 1 Using computational breast phantom, perform a virtual study to determine the potential of breast tomosynthesis for detectability of breast abnormalities, compared to conventional mammography To be considered: Involved Phantom(s): Breast Phantom Small to Medium size, Glandular to Dense background tissue, Tree min 15 lactiferous branches in total, include skin and Cooper, Output volume matrix size 400 x 400 x 400 voxels (or alternatively 500 x 500 x 500 voxels), voxel size = 0.25 mm. Abnormalities: One spherical mass abnormality up to 10 mm in diameter and one cluster of microcalcifications with a microcalcification of up to 1 mm in diameter. Imaging techniques: Conventional mammography and Breast tomosynthesis. Assumed Incident Dose: 5-12 mgy. Any other parameters are by your discretion and preference. To be completed: Create a phantom, according to the provided specification and add the abnormalities. Obtain synthetic conventional mammographic image of the phantom at an energy in the range kev, selected by your discretion. Obtain 10 synthetic mammography images over an isocentric arc (-9 to 9) degree (every 2 degrees) and 26 synthetic mammography images over an isocentric arc (-25 to 25) degree (every 2 degree) at the same energy as above, keeping the total dose the same. Add a realistic level of noise, based on the required incident dose. Perform image reconstruction at different planes of interest. Compare subjectively (visually) the images. 7

8 Compute objective figure of merits (FOMs) CNR in the region of the abnormalities and compare. To be prepared: Report: a short written summary of the work and the obtained results; containing for example: o appropriate outcome images and profiles o tables with computed FOMs o discussion of the results and conclusions PowerPoint Presentation: to be used for a short (5-10 ) oral presentation. Useful hints: The tool for the phantom creation is the BreastSimulator tool. For introducing abnormalities use the BreastSimulator tool. Projection images are generated using the BreastSimulator tool. Image reconstruction is performed using FDKR tool. References: [1]. BreastSimulator User Guide. [2]. FDKR User Guide. 8

9 PROJECT IMPLEMENTATION REPORT A virtual study to determine the potential of breast tomosynthesis for detectability of breast abnormalities, compared to conventional mammography Şölen Çubukçu(Turkey) ; Agnieszka Kuchcinska (Poland) ; Nicola Pace (Italy) Introduction Virtual anthropomorphic phantoms can be used in order to assess the diagnostic possibilities of different imaging techniques. In particular, breast virtual phantoms containing abnormalities may be employed in a comparative evaluation of mammography and tomosynthesis. Materials and Methods We developed an anthropomorphic phantom to simulate a small dense breast using BreastSimulator software. Two abnormalities were inserted within the simulated breast: 1 spherical abnormal mass simulating breast cancer. Radius = 5mm 1 cluster (of radius = 10 mm) containing 5 microcalcifications each max microcalcification radius = 0.5 mm. Data used for the development of the model are displayed in table 1 and table 2. Simulation was performed considering incident photon energy of 20 kev. Two breast phantoms were simulated with same parameters except attenuation coefficient factor. Breast size Nipple size Total Ducts Cooper ligaments (radius) Matrix size Voxel size Nwalks Nsteps 40x40x40 mm 4x4x5mm 16 (4x4) (2mm) 400x400x Table 1: input parameters for the used breast model Element Attenuation 20keV (mm -1 ) Semiellypsoid, hyperboloid, nipple, ducts, cooper elements, skin microcalcifications Abnormal mass (phantom 1 )/ 0.90 (Phantom 2 ) Table 2: attenuation coefficients assumed 9

10 Correct localization of both abnormalities and microcalcification was based on generated model. Figures below show phantom visualization. Table 3 show example of localization for cluster and abnormality for phantom 2. µcaco3 1 µcaco3 2 µcaco3 3 µcaco3 4 µcaco3 5 mass X [mm] 11,9 14,42 12,62 13,63 12,15-3 Y [mm] 12,59 14,22 10,83 15,03 12,11-11 Z [mm] 9,26 3,64 3,93-2,97 6,21-12 Table 3: localization of abnormalities in phantom 2 Fıg 1: skin and duct system phantom 2 Fıg 2: Phantom 2 yellow color represent abnormality, green color Cooper ligaments Fıg 3: Phantom 2 basic visualisation After the generation of the simulated breast, XrayImagingSimulator was used in order to generate a synthetic X-ray image obtained through conventional mammography (single CC projection), a tomosynthesis reconstructed synthetic image generated from 10 planar acquisition ranging in angles between -9 and 9 degrees, and a tomosynthesis reconstructed synthetic image generated from 26 planar acquisitions ranging in angles between -25 and 25 degrees with step 2 degree. Reconstruction was made using FDKR Software. Images obtained were analyzed with ImageJ software in order to investigate differences in diagnostic capabilities for abnormal masses of the different radiographic techniques considered. Major outcome for comparison was the figure of merit (FOM) CNR in the region of the abnormalities. Other simulations accounting for different abnormality attenuation coefficient, different photon energy and different breast density were made in order to explore the variability of results. Results We obtained 3 images for phantom 1 simulating planar mammography (fig 4) and tomosynthesis acquisition (fig 5 and 6). Abnormal mass was well displayed in all the three images (fig 4 to 6). 10

11 Inter-reader variability was assessed, with an agreement on detectability of abnormalities of 100% across the three Readers on all the three images. For phantom 2 detail tomosynthesis analysis was performed and very accurate localization of micro calcification was obtained. For 2D mammography FOM was also calculated for other suspicious part of the image in order to simulate false positive investigation for dense breast. For phantom 2 other region has higher FOM than real abnormal mass. In order to check possible improvement of system performance for dense breast and low difference of attenuation factor between abnormality and background Dual energy image simulation was done. Agreement between simulated Phantoms background was find, for both phantoms mean pixel value was 1, 46. ROI analysis yielded the results listed in tables 3, and figure D mammography -9 to 9 tomosynthesis -25 to 25 tomosynthesis ROI mass mean ROI mass stdev ROI bkgd mean ROI bkgd stdev CNR Table 3: ROI Analysis for Phantom 1 Fıg 4: synthetic X-ray image of phantom 1 obtained through conventional mammography Fıg 5: tomosynthesis reconstructed synthetic image of phantom 1 generated from 10 planar acquisition ranging in angles between -9 and 9 degrees Fıg 6: tomosynthesis reconstructed synthetic image of phantom 1 generated from 26 planar acquisitions ranging in angles between -25 and 25 degrees 11

12 Fıg 7: detaıl of the abnormal mass dısplayed by 2D mammogram (Phantom 1 ) Fıg8: detaıl of the abnormal mass dısplayed by -9 to 9 tomosynthesis (Phantom 1 ) Fıg 9: detaıl of the abnormal mass dısplayed by -25 to 25 tomosynthesis (Phantom 1 ) ABN FP ROImass mean ROImass stdev ROIbkgd mean ROIbkgdstdev CNR Fıg 10: Phantom 1 ; with abnormal mass attenuation coefficient 0,200 [1/mm] ABN FP ROImass mean ROImass stdev ROIbkgd mean ROIbkgdstdev CNR Fıg 11: Phantom 2 ; with abnormal mass attenuation coefficient 0,090 [1/mm] 12

13 Fig 12: Tomo angle 9 reconstuctions planes of phantom 2 contained microcalfification (automatic contrast of image adjustement). Fig 13: Tomo angle 9 reconstuctions planes of phantom 2 contained microcalfification (contrast of image adjusted by operator). Fig 14: Tomo angle 9 reconstuctions planes of phantom 2 contained microcalfification (automatic contrast of image adjustement). 13

14 Fıg 15: Low energy image phantom 1 (Image J) Fıg 16: High energy image phantom 1 (Image J) Fıg 17: Combined image phantom 1 (Image J) Fıg 18: Low energy, high energy, combined images (XraySimulator Conclusions Simualtion of 2D irradiation and obtained images showed that for high density breast 2D mammography can provide false positive results. Tomosynthesis was found better for microcalcification detectability, althought in case of very low difference in attenuation coefficient between abnormal mass and bacground it was found that any method that was under investigation can contribute to correct diagnosis Special care should be taken when choosing between phantom models (solid state vs. Voxels) and Software used for obtaining image simulation Using mathematical antropomorhic phantom can contribute to virtual clinical studies. From the presentation: 14

15 EUTEMPE-RX: European Training and Education for Medical Physics Experts in Radiology MPE05: Anthropomorphic phantoms Applications of anthropomorphic phantoms for design and evaluation of advanced x-ray imaging techniques PROJECT ASSIGNMENT No. 2 Using computational breast phantom, perform a virtual study to determine the potential of the cone-beam CT for detectability of breast abnormalities, compared to conventional mammography To be considered: Involved Phantom(s): Breast Phantom Small to Medium size, Glandular to Dense background tissue, Tree min 15 lactiferous branches in total, Cooper and skin included, Output volume matrix size 400 x 400 x 400 voxels, (or alternatively 500 x 500 x 500 voxels), voxel size = 0.25 mm. Abnormalities: Five single microcalcifications up to 1 mm in diameter, randomly distributed. Imaging techniques: Conventional mammography and Cone-beam CT. Assumed Incident Dose: 5-12 mgy. Any other parameters are by your discretion and preference. To be completed: Create a phantom, according to the provided specification and add the abnormalities. Obtain synthetic conventional mammographic image of the phantom at an energy in the range kev, selected by your discretion. Obtain 180 synthetic projection images over a complete arc (every 2 degrees) at the same energy as above. Add a realistic level of noise, based on the required incident dose. Perform image reconstruction at different planes of interest. Compare subjectively (visually) the images. Compute objective figure of merits (FOMs) CNR in the region of the abnormalities and compare. 15

16 To be prepared: Report: a short written summary of the work and the obtained results; containing for example: o appropriate outcome images and profiles o tables with computed FOMs o discussion of the results and conclusions PowerPoint Presentation: to be used for a short (5-10 ) oral presentation. Useful hints: The tool for the phantom creation is the BreastSimulator tool. For introducing abnormalities use the BreastSimulator tool. Projection images are generated using the BreastSimulator tool. Image reconstruction is performed using FDKR tool. References: [1]. BreastSimulator User Guide. [2]. FDKR User Guide. 16

17 PROJECT IMPLEMENTATION REPORT Using computational breast phantom, perform a virtual study to determine the potential of the cone-beam CT for detectability of breast abnormalities, compared to conventional mammography Yanka Baneva and Anna Radice Our work was to create phantom of breast with Breast Simulator tool: small to medium size glandular to dense background tissue tree - min 15 lactiferous branches in total cooper and skin included five single microcalcifications up to 1 mm in diameter, randomly distributed. For the image techniques we used conventional mammography and Cone-beam CT. We created the phantom with the provided specifications: Energy of the range of 19 kev Semi-ellipsoid: (x0,y0,z0)=(24,0,0); (xr,yr,zr)=(40,40,40) Semi-hyperbolid: (x0,y0,z0,free_offset)=(25,0,55,11), (xr,yr,zr,c_param)=(5,5,5,6) Additional: Cooper ligaments: number =50000; radius=3mm;att_coeff =0.079mm -1 Pectoralis: yes, att_coeff =0.097mm -1 Skin: yes, att_coeff=0.08mm -1 Nipple: (x0,y0,z0)=(-15,0,-2); (xr,yr,zr)=(4,4,5) Ducts: (x0,y0,z0)=(14.5,0,-2);(ϕ,θ)=(-10, -10 ), att_coeff=0.079mm -1 Abnormalities: 5 single microcalcifications up to 1 mm in diameter, randomly distributed. Major ducts Major ducts height(mm) Major duct radius(mm) Lact.ducts Lact.Ducts in.height (mm) Lact.Ducts, lim.height (mm) Lact.Ducts In.radius (mm) Texture: att_coeff= mm-1 Size Voxel(mm) Nwalk N step LPF Dilation Filt.size Increment Hurst. Coeff No No Lobule radius (mm) St_dev. 17

18 Skin Ducts Cooper ligaments Microcalcifications Figure 1. That is the image of the phantom with all parts on it Then we obtained synthetic conventional planar mammographic image at the same energy level, with Breast simulator, with: SID = 600 mm, SDD =900 mm, ray tracing step = 0.1 mm, image size 500x500 pixels; Resolution: 5 pixels/mm. Figure 2. The microcalcifications are visible on the picture. Next step was to obtain 180 synthetic projection images over a complete arc with step 2 degrees, at same energy as above and to reconstruct the images by using FDKR tool at different planes of interest. Figure 3. 18

19 The first image in fig.3 represents the axial plane at z= -9 mm, the second one is lateral, at the same z. On both pictures the calcifications are still visible. Figure 4. At fig.4 we see the axial, lateral and coronal reconstructions at z=30mm. The calcifications in the first two images are still visible while for the third one not. That means that it is very important to put the correct number for z in order to receive good image. The final task was to compute CNR in the region of the abnormalities and to compare. This time we decided to reconstruct coronal images at the same z-level where the 2 of 5 microcalcifications are located in the breast phantom: Image comparison (z1 = - 4 mm), mammographic image versus coronal reconstruction CBCT image. A) Mammo Figure 5. Area Mean StdDev Min Max CNR 1 signal background B) Tomo image Area Mean StdDev Min Max CNR 1 signal background

20 CNR percentage difference = Image comparison (z1 = - 10 mm), mammographic image versus coronal reconstruction CBCT image. z = 10 A) Mammo Figure 6. Area Mean StdDev Min Max CNR 1signal background B)Tomo image ROI Area Mean StdDev Min Max CNR 1signal background CNR percentage difference = As a conclusion we may say that: CBCT is visually better. CBCT allows reconstruction images at different planes. CBCT reconstruction images have higher CNR. From the presentation: 20

21 21

22 22

23 EUTEMPE-RX: European Training and Education for Medical Physics Experts in Radiology MPE05: Anthropomorphic phantoms Applications of anthropomorphic phantoms for design and evaluation of advanced x-ray imaging techniques PROJECT ASSIGNMENT No. 3 Using computational breast phantom, perform a virtual study to determine the potential of cone-beam CT for detectability of breast abnormalities, compared to conventional mammography To be considered: Involved Phantom(s): Voxel-based phantom composed from segmented CT slices: \Phantoms\CT Slices\VoxelBreast2, Voxel size = 0.3 mm. Abnormalities: One Irregular mass up to 10 mm in size and one cluster of microcalcifications with a microcalcification of up to 1 mm in diameter. Imaging techniques: Conventional mammography and Cone-beam CT. Assumed Incident Dose: 5-12 mgy. Any other parameters are by your discretion and preference. To be completed: Create a phantom, according to the provided specification and add the abnormalities. Assign relevant attenuation coefficients to the different phantom regions. Obtain synthetic conventional mammographic image of the phantom at an energy value, selected by your discretion. Obtain 180 synthetic projection images over a complete arc (every 2 degree) at the same energy as above. Add a realistic level of noise, based on the required incident dose. Perform image reconstruction at different planes of interest. Compare subjectively (visually) the images. Compute objective figure of merits (FOMs) CNR in the region of the abnormalities and compare. 23

24 To be prepared: Report: a short written summary of the work and the obtained results; containing for example: o appropriate outcome images and profiles o tables with computed FOMs o discussion of the results and conclusions PowerPoint Presentation: to be used for a short (5-10 ) oral presentation. Useful hints: The tool for the phantom creation is the XRayImageSimulator tool. Microcalcifications may be modelled as ellipsoids or spheres. Exploit the option Voxel-based phantom Monte Carlo to convert the object to voxel based model. For the Irregular abnormality, use the tool AbnormalityMaker in the Tools directory. Use the script file abnormality_maker.m and modify accordingly. For introducing abnormalities use the XRayImageSimulator -> Tools -> Add an abnormality. For assigning linear attenuation coefficients, use the Matlab/Octave script: assign_linear_attenuation_coefficient.m. For inspection of the created volume, ImageJ is used. Tissue Value Skin 255 Air 0 Adipose 85 Gland 170 Projection images are generated using the XRayImageSimulator tool. Image reconstruction is performed using FDKR tool. References: [1]. XRayImageSimulator User Guide. [2]. FDKR User Guide. 24

25 PROJECT IMPLEMENTATION REPORT Using computational breast phantom, perform a virtual study to determine the potential of cone-beam CT for detectability of breast abnormalities, compared to conventional mammography Eric Pace and Lucie Sukupova Three microcalcifications were added into a basic CT breast image. A conventional mammography image and corresponding CT plane were reconstructed, and this is shown in Fig. 1. Fig. 1: The conventional mammography image and the reconstructed CT image Images were compared not only qualitatively but also quantitatively by the use of contrast to noise ratio (CNR). The values are included in Table 1. Mammography Image CT image Mean Background CNR Table 1: Values for an objective assessment The CNR values above show that the CT reconstructed image provides a much higher value of CNR, therefore the microcalcifications ares more visible. While the microcalcifications are visible in the mammography image and may be considered of sufficient quality for the radiologists, the 25

26 aim of this study was to compare the CNR of both imaging methods. In the simulation the microcalcifications were quite large, whereas this would not be so in the clinical situation. Thus it is possible to have microcalcifications which are not visible at all in the mammographic image but would be visible in the CT reconstruction. From the presentation: 26

27 EUTEMPE-RX: European Training and Education for Medical Physics Experts in Radiology MPE05: Anthropomorphic phantoms Applications of anthropomorphic phantoms for design and evaluation of advanced x-ray imaging techniques PROJECT ASSIGNMENT No. 4 Using computational breast phantom, perform a virtual study to determine the potential of breast tomosynthesis for detectability of breast abnormalities, compared to conventional mammography To be considered: Involved Phantom(s): Voxel-based phantom composed from segmented CT slices: \Phantoms\CT Slices\CompressedBreastSlices, Voxel size 0.3 mm. Abnormalities: One Irregular mass up to 10 mm in size and one cluster of microcalcifications with a microcalcification of up to 1 mm in diameter. Imaging techniques: Conventional mammography and Breast tomosynthesis mammography. Assumed Incident Dose: 5-12 mgy. Any other parameters are by your discretion and preference. To be completed: Create a phantom, according to the provided specification and add the abnormalities. Assign relevant attenuation coefficients to the different phantom regions. Obtain synthetic conventional mammographic image of the phantom at an energy in the range kev, selected by your discretion. Obtain 10 synthetic mammography images over an arc (-9 to 9) degree (every 2 degrees) and 26 synthetic mammography images over an arc (-25 to 25) degree (every 2 degrees) at the same energy as above. Add a realistic level of noise, based on the required incident dose. Perform image reconstruction at different planes of interest. Compare subjectively (visually) the images. 27

28 Compute objective figure of merits (FOMs) CNR in the region of the abnormalities and compare. To be prepared: Report: a short written summary of the work and the obtained results; containing for example: o appropriate outcome images and profiles o tables with computed FOMs o discussion of the results and conclusions PowerPoint Presentation: to be used for a short (5-10 ) oral presentation. Useful hints: The tool for the phantom creation is the XRayImageSimulator tool. Microcalcifications may be modelled as ellipsoids or spheres. Exploit the option Voxel-based phantom Monte Carlo to convert the object to voxel based model. For the Irregular abnormality, use the tool AbnormalityMaker in the Tools directory. Use the script file abnormality_maker.m and modify accordingly. For introducing abnormalities use the XRayImageSimulator -> Tools -> Add an abnormality. For assigning linear attenuation coefficients, use the Matlab/Octave script: assign_linear_attenuation_coefficient.m. For visualization and inspection of the created volume, ImageJ is used. The initial voxel values in the compressed breast matrix are: Tissue Value Skin 7 Air 119 Adipose 255 Gland 151 Photon noise is added by using add_noise.m script in the mcode directory. Projection images are generated using the XRayImageSimulator tool. Image reconstruction is performed using FDKR tool. References: [1]. XRayImageSimulator User Guide. [2]. FDKR User Guide. 28

29 PROJECT IMPLEMENTATION REPORT Using computational breast phantom, perform a virtual study to determine the potential of breast tomosynthesis for detectability of breast abnormalities, compared to conventional mammography Marius Laurikaitis 1 and Anastasios Konstantinidis 2 1 Oncology Hospital of Kaunas Clinics, Kaunas, Lithuania, marius@medicinosfizika.lt 2 Christie Medical Physics and Engineering, The Christie NHS Foundation Trust, Manchester, UK, taskon25@yahoo.gr Project tasks Create a Voxel-based phantom composed from segmented CT slices. Assign relevant attenuation coefficients to different phantom regions. Obtain conventional Mammographic image of the phantom (energy 20 kev). Obtain Digital Breast Tomosynthesis (DBT) images (energy 20 kev, arc -9 o to 9 o and -25 o to 25 o ). Add quantum noise considering 5 mgy incident dose. Perform image reconstruction at different planes (z = -20, -5, 0, 5, 20 mm). Compare subjectively the images. Compute objective figure of merit, i.e. Contrast-to-Noise Ratio (CNR) in the region of abnormalities and compare. Materials and methods For Voxel-based phantom creation we used XRayImageSimulator [1]. The image size was 641 x 357 x 175 pixels. For adding abnormalities we used XRayImageSimulator [1]. The abnormality objects were masses with size of 200 x 200 x 200 pixels. For assign attenuation coefficients we used the Octave [2] script assign_linnear_attenuation_coefficient.m. For irradiating Voxel-based phantom and getting conventional Mammographic and Digital Breast Tomosynthesis images we used BreastSimulator [3]. For adding quantum noise (incident dose - 5 mgy) to the images we used the Octave [2] script add_noise.m. For reconstruction of Digital Breast Tomosynthesis images we used FDKR [4]. For visualization of images and extraction of objective parameters we used ImageJ [5]. Results and Discussion Figure 1 shows synthetic Mammographic (a) and DBT images using 10 projections (-9 o to 9 o arc) (b) and (c) 26 projections (-25 o to 25 o arc) (c) both for z=-5 mm plane. DBT resulted in better visual detection of abnormality compared to Mammography. In particular, the reconstructed image over 26 projections was better compared to 10 projections. 29

30 (a) (b) (c) Figure 1. Visualization using (a) Mammography and Digital Breast Tomosynthesis modalities with (b) -9 o to 9 o arc and (c) -25 o to 25 o arc (both for z=-5 mm) Figure 2 shows the regions of interest (ROIs) used for the calculation of CNR parameter. The used background ROI was larger in order to take into account background non-uniformities. Figure 2. Computation of CNR: CNR=(mean_object mean_back)/std_back Figure 3 shows projection DBT images for different arcs (-9 o to 9 o (a) and -25 o to 25 o (b)) and planes (z = -20, -5, 0, 5, 20 mm). The abnormality can be visually detected in the range z = -5, 0 and 5 mm but it is hard to define which plane is best. (a) z = -20 mm z = -5 mm z = 0 mm z = 5 mm z = 20 mm (b) Figure 3. Visualization using Digital Breast Tomosynthesis with (a) -9 o to 9 o arc and (b) -25 o to 25 o arc in different planes (z = -20, -5, 0, 5, 20 mm) Table 1 shows the CNR values for different conditions, i.e. using Mammography and DBT for 10 and 26 projection images and 5 planes of interest. Mammography results in the best CNR value. Regarding DBT application, the combination of 26 projections with z = -5 mm resulted in high CNR value. 30

31 Table 1. Comparison of CNR values for different irradiation parameters. Technique CNR Mammography Digital Breast Tomosynthesis z = -20 mm z = -5 mm z = 0 mm z = 5 mm z = 20 mm DBT (-9 to 9 ) DBT (-25 to 25 ) Conclusions 1. DBT resulted in better visual detection of abnormality compared to Mammography. In particular, the reconstructed image over 26 projections was better compared to 10 projections. 2. We were able to visually detect the abnormality in the range z=-5, 0 and 5 mm but we could not say which one was best. 3. The CNR value using Mammography was higher compared to DBT. 4. DBT for 26 projections resulted in better CNR values compared to 10 projections. 5. Reconstructed images for z=-5 mm had the best CNR values for both sets of projections. References 1. XRayImageSimulator User Guide 2. Octave (available online: 3. BreastSimulator User Guide 4. FDKR User Guide 5. ImageJ (available online: From the presentation: 31

32 32

33 EUTEMPE-RX: European Training and Education for Medical Physics Experts in Radiology MPE05: Anthropomorphic phantoms Applications of anthropomorphic phantoms for design and evaluation of advanced x-ray imaging techniques PROJECT ASSIGNMENT No. 5 Using computational breast phantom, perform a virtual study to determine the potential of the dual-energy imaging for detectability of breast abnormalities, compared to conventional mammography To be considered: Involved Phantom(s): Voxel-based phantom composed from segmented CT slices: \Phantoms\CT Slices\CompressedBreastSlices, Voxel size 0.3 mm. Abnormalities: One irregular mass and five individual microcalcifications of up to 1 mm in diameter. Imaging techniques: Conventional mammography and dual-energy mammography. Assumed Incident Dose: 5-12 mgy. Any other parameters are by your discretion and preference. To be completed: Create a phantom, according to the provided specification and add the abnormalities. Assign relevant attenuation coefficients to the different phantom regions. Obtain synthetic conventional mammographic images of the phantom at an energy in the range kev. Obtain synthetic mammography images at three high x-ray energies, for instance: 50 kev, 65 kev and 80 kev or use other ones of your preference. Add a realistic level of noise, based on the required incident dose. Compute dual-energy difference images. Compare subjectively (visually) the images. Compute objective figure of merits (FOMs) CNR in the region of the abnormalities and compare. 33

34 To be prepared: Report: a short written summary of the work and the obtained results; containing for example: o appropriate outcome images and profiles o tables with computed FOMs o discussion of the results and conclusions PowerPoint Presentation: to be used for a short (5-10 ) oral presentation. Useful hints: The tool for the phantom creation is the XRayImageSimulator tool. Microcalcifications may be modelled as ellipsoids or spheres. Exploit the option Voxel based phantom Monte Carlo to convert the object to voxel based model. For the Irregular abnormality, use the tool AbnormalityMaker in the Tools directory. Use the script file abnormality_maker.m and modify accordingly. For introducing abnormalities use the XRayImageSimulator -> Tools -> Add an abnormality. For assigning linear attenuation coefficients, use the Matlab/Octave script: assign_linear_attenuation_coefficient.m. For inspection of the created volume, ImageJ is used. The initial voxel values in the compressed breast matrix are: Tissue Value Skin 7 Air 119 Adipose 255 Gland 151 Projection images are generated using the XRayImageSimulator tool. Photon noise is added by using add_noise.m script in the mcode directory. Dual-energy images are computed using the script calculate_dual_energy_image.m in Matlab/Octave or it may be also performed using ImageJ. References: [1]. XRayImageSimulator User Guide. [2]. FDKR User Guide. 34

35 PROJECT IMPLEMENTATION REPORT Using computational breast phantom, perform a virtual study to determine the potential of the dual-energy imaging for detectability of breast abnormalities, compared to conventional mammography. Rita Demou and Simona Avramova-Cholakova The aim of this report was to investigate whether dual-energy imaging improves detectability of breast abnormalities compared to conventional mammography. Both a visual assessment and a comparison of the CNR were employed in order to evaluate the two techniques. METHODOLOGY Using XRaySimlator a 3D matrix was constructed from slices recorded in a CT modality. A 641x357x175 matrix of a 0.3mm voxel size was created. The model represented a compressed breast phantom. Then, a solid geometry phantom was constructed to represent five microcalcifications and an irregular mass. Table 1 records the spatial and content information of the abnormalities. The phantom was converted to a vocalized matrix of 84x84x84 pixels of a size of 0.3mm. Table 1. Specifications of abnormalities Type Center (mm) Dimensions (mm) Compound φ, θ, ψ degrees MCal Ellipsoid (0, 0, 0) (0.7, 1, 0.6) CaCO3 (0, 0, 0) MCal Ellipsoid (2, -2, 3) (0.9, 1.2, 1.0) CaCO3 (0, 0, 0) MCal Ellipsoid (0, 3, -3) (0.9, 0.3, 0.5) CaCO3 (0, 0, 0) MCal Ellipsoid (-2, 4, 5) (0.5, 0.3, 0.4) CaCO3 (0, 0, 0) MCal Ellipsoid (2, 2, 2) (0.9, 1.2, 0.8) CaCO3 (0, 0, 0) IrMass Ellipsoid (8, 8, 8) (10, 5, 7) Water (45, -75, 0) The two matrices were combined utilizing XRaySimulator. The matrix representing the abnormalities was positioned at position (180, 200, 70) inside the CT breast model. Using Image J, the combined model was visualized both as a stack of slices and as a 3D volume. Figure 1 represents the combined phantom. 35

36 Octave was used to transform modeled coefficients of the materials included in the combined phantom to linear attenuation coefficients (LAC). Phantoms were created taking into account the liner attenuation coefficients at 20, 50, 65 and 80 kev. Table 2 records modeled and linear attenuation coefficients. Figure 1. Slices and 3D Volume of the combined phantom Table 2. Modeled and Linear attenuation coefficients Object Modeled Coefficient LAC 20keV LAC 50keV LAC 65keV LAC 80keV Skin Air Adipose Gland Water CaCO Planar projections were simulated for the four phantoms using analytical image simulator in XRaySimulator. Set irradiation properties were identical for the four phantoms and are recorded in table 3. Table 3. Irratiation properties of simulated projections Source to Isocenter Distance Source to Detector Distance Ray Step Resolution 600 mm 650 mm 0.1 mm 5 pixels/mm Image Size 1024 x 1024 Geometry Output Fan Beam Line Integral 36

37 Using Octave the 20keV projection was subtracted from 50, 65and 80keV image, consecutively to obtain the dual energy images. Finally, using ImageJ, the Contrast-to-Noise Ratio (CNR) was determined for 20keV planar projection and for the three dual energy images, applying the formulae:, where MPVobj is the mean pixel value of the regions of interest (ROI) in the object, and MPVbckgr and STDEVbckgr are the mean pixel value and the standard deviation of the ROI in the background, respectively. RESULTS Figure 2 shows the visual assessment of the two different techniques. Figure 2. The 20keV planar projection and the dual energy images. The ROI, chosen for the CNR calculation, are depicted in Figures 3-6 for both the biggest microcalcification and the irregular mass. 37

38 Figure 3. ROI in the calcification. Figure 4. ROI in the background of calcification. Figure 5. ROI in the mass. Figure 6. ROI in the background of mass. The results from the CNR calculations are presented in table 4. Mean is for mean pixel value and StdDev is the standard deviation. Planar Mammo_20 is for the planar mammography simulated image and DE describes the dual energy images for the different energy combinations. Table 4. Input data and results from calculation of CNR. Planar Mammo_20 Planar Mammo_20 DE_20_80 DE_20_65 DE_20_50 DE_20_80 DE_20_65 DE_20_50 Area Mean StdDev CNR Mass Background Calcification Background Mass Background Mass Background Mass Background Calcification Background Calcification Background Calcification Background

39 Figures 7 and 8 present a summary of the results in table 4. CONCLUSIONS Figure 7. Figure 8. Both visual assessment and comparison of the CNR clearly shows that dual energy technique significantly improves the detectability of breast abnormalities compared to conventional mammography. The improvement of the contrast of the micro-calcifications is more significant while increasing energy difference of the dual energy images. In the case of the low contrast mass, the CNR becomes worst when a low energy difference is employed. However, when energies above 65keV are used to get the high energy projections, the CNR improves in comparison with conventional mammography. From the presentation: 39

40 40

41 EUTEMPE-RX: European Training and Education for Medical Physics Experts in Radiology MPE05: Anthropomorphic phantoms Applications of anthropomorphic phantoms for design and evaluation of advanced x-ray imaging techniques PROJECT ASSIGNMENT No. 6 Using computational breast phantom, perform a virtual study to determine the potential of dual-energy imaging for detectability of breast abnormalities, compared to conventional mammography To be considered: Involved Phantom(s): LUCMR Small to Medium size, Average compressed breast thickness, Spheres sized mm in diameter, bulk material. Abnormalities: Two spherical mass abnormalities of different materials and five individual microcalcifications of up to 1 mm in diameter. Imaging techniques: Conventional mammography and dual-energy mammography. Assumed Incident Dose: 5-12 mgy. Any other parameters are by your discretion and preference. To be completed: Create a phantom, according to the provided specification and add the abnormalities. Choose appropriate materials from table 1. Obtain synthetic conventional mammographic images of the phantom at two energies in the range kev. Obtain synthetic mammography images at three higher x-ray energies, for instance: 45 kev, 60 kev and 75 kev. Add a realistic level of noise, based on the required incident dose. Compute dual-energy difference images. Compare subjectively (visually) the images. Compute objective figure of merits (FOMs) CNR in the region of the abnormalities and compare. To be prepared: Report: a short written summary of the work and the obtained results; containing for example: 41

42 o appropriate outcome images and profiles o tables with computed FOMs o discussion of the results and conclusions PowerPoint Presentation: to be used for a short (5-10 ) oral presentation. Useful hints: The tool for the phantom creation is the LUCMRGen tool. Take into account table 1 that contains list of modelled in XRayImageSimulator materials. For introducing abnormalities use the XRayImageSimulator tool. Take into account table 1. Projection images are generated using the XRayImageSimulator tool. Photon noise is added by using add_noise.m script file in the mcode directory. Modify the script file accordingly. The image subtraction may be performed in ImageJ but more flexibility is provided by Matlab/Octave. Dual-energy images are computed using the script calculate_dual_energy_image.m in Matlab/Octave or it may be also performed using ImageJ. Table 1: Modelled Materials C Al Si Cu Pb water bone compact ICRU-44 lung tissue - inflated brain, grey/white matter breast mass 75% gland 25% fat plexiglas (lucite, PMMA) ideal detector adipose ICRU-44 muscle ICRU-44 Gd2O2S (phosphor) air CsI (phosphor) bone cortical ICRU-44 polystyrene BaFBr gland CaCO3 soft tissue ICRU- 44 Breast 50% gland 50% fat ideal air SrFBr paraffin wax, C25H52 skin gray matter white matter blood SiO2 silicon gel epoxy resin nylon-6,6 water vapour ethanol ethanol vapour polyethylene povidoneiodine References: [1]. Bliznakova, K., I. Buliev, et al. (2013). Development and Evaluation of a Software Phantom Dedicated for Breast Imaging Experimentation: Preliminary (First) Results. 5th Pan-Hellenic Conference on Biomedical Technology (ELEBIT 2013). 4-6 April 2013, Athens, Greece. [2]. Cockmartin, L., N. Marshall, et al. (2012). Design and evaluation of a phantom with structured background for digital mammography and breast tomosynthesis. Philadelphia, PA LNCS:

43 PROJECT IMPLEMENTATION REPORT Potential of dual energy imaging for detectability of breast abnormalities Ciprian Cindea and Oliver Diaz Purpose : To explore the potential of computer simulation using phantom models and study the detection performance of several breast lesions in dual energy (DE) mammography and conventional mammography. Methods: L UCMR phantom was created using the LUCMRGen software. The phantom contains a PMMA case of semicircular shape (radius 40 and height 35mm). It was filled with 2246 water spheres of different radii to produce breast texture. Five spheres (1 mm 3 radius) of CaCO3 were randomly inserted within the LUCMR phantom to simulate microcalcifications. Moreover, two masses were modelled as spheres of radius 3 and 6 mm 3 containing breast mass tissue (75% glandular and 25% adipose tissue) and 100% glandular tissue respectively. XRAYImagingSimulator was used to generate the projection images for several monoenergetic beams: 17, 23, 45, 60, 75 kev. Poisson noise was added to the idealised projected images assuming 6 mgy entrance dose. DE images were generated from the subtraction of the noisy projections and contrast to noise ratio (CNR) was used to analyse the detection performance of the simulated lesions. Results: From all the 6 DE combinations of low high energy, the best configuration observed was 1760 kev. This setup illustrates an improvement of 191% and 196% in CNR for microcalcifications and masses respectively when compared with conventional energy projections (17 and 23 kev). Conclusions: Results have demonstrated that, according to CNR, DE is superior to mammography for lesion detection as background tissue could be eliminated from the image. All microcalcifications could be visually differentiated in both DE and conventional mammography, however, breast masses, which have similar attenuation coefficient to background tissue, were only visible in DE images. Computer simulations using phantom models have demonstrated great potential to optimize imaging parameters. LUCMR phantom model Conventional mammography Dual Energy projection 43

44 From the presentation: 44

45 EUTEMPE-RX: European Training and Education for Medical Physics Experts in Radiology MPE05: Anthropomorphic phantoms Applications of anthropomorphic phantoms for design and evaluation of advanced x-ray imaging techniques PROJECT ASSIGNMENT No. 7 Using computational breast phantom, perform a virtual study to determine the potential of breast tomosynthesis for detectability of breast abnormalities, compared to conventional mammography To be considered: Involved Phantom(s): LUCMR Small to Medium size, Average compressed breast thickness, Spheres sized mm in diameter, bulk material. Abnormalities: Two spherical mass abnormalities of different materials and five individual microcalcifications of up to 1 mm in diameter. Imaging techniques: Conventional mammography and breast tomosynthesis mammography. Assumed Incident Dose: 5-12 mgy. Any other parameters are by your discretion and preference. To be completed: Create a phantom, according to the provided specification and add the abnormalities. Choose appropriate materials from table 1. Obtain synthetic conventional mammographic images of the phantom at an energy selected by your discretion. Obtain 26 synthetic mammography images over an arc of your discretion. Add a realistic level of noise, based on the required incident dose. Perform image reconstruction at the planes of interest. Compare subjectively (visually) the images. Compute objective figure of merits (FOMs) CNR in the region of the abnormalities and compare. 45

46 To be prepared: Report: a short written summary of the work and the obtained results; containing for example: o appropriate outcome images and profiles o tables with computed FOMs o discussion of the results and conclusions PowerPoint Presentation: to be used for a short (5-10 ) oral presentation. Useful hints: The tool for the phantom creation is the LUCMRGen tool. Take into account table 1 that contains list of modelled in XRayImageSimulator materials. For introducing abnormalities use the XRayImageSimulator tool. Take into account table 1. Projection images are generated using the XRayImageSimulator tool. Photon noise is added by using add_noise.m script file in the mcode directory. Modify the script file accordingly. Image reconstruction is performed using FDKR tool. Table 1: Modelled Materials C Al Si Cu Pb water bone compact ICRU-44 lung tissue - inflated brain, grey/white matter breast mass 75% gland 25% fat plexiglas (lucite, PMMA) ideal detector adipose ICRU-44 muscle ICRU-44 Gd2O2S (phosphor) air CsI (phosphor) bone cortical ICRU-44 polystyrene BaFBr gland CaCO3 soft tissue ICRU-44 Breast 50% gland 50% fat ideal air SrFBr paraffin wax, C25H52 skin gray matter white matter blood SiO2 silicon gel epoxy resin nylon-6,6 water vapour ethanol ethanol vapour polyethylene povidoneiodine References: [1]. Bliznakova, K., I. Buliev, et al. (2013). Development and Evaluation of a Software Phantom Dedicated for Breast Imaging Experimentation: Preliminary (First) Results. 5th Pan-Hellenic Conference on Biomedical Technology (ELEBIT 2013). 4-6 April 2013, Athens, Greece. [2]. Cockmartin, L., N. Marshall, et al. (2012). Design and evaluation of a phantom with structured background for digital mammography and breast tomosynthesis. Philadelphia, PA LNCS:

47 PROJECT IMPLEMENTATION REPORT A virtual phantom study to evaluate the detectability of breast abnormalities in tomosynthesis compared to conventional mammography Lesley COCKMARTIN and Loredana BOGDAN PURPOSE To perform a virtual study to determine the potential of breast tomosynthesis compared to conventional mammography for detectability of breast abnormalities using a computational structured phantom. MATERIALS AND METHODS A virtual structured phantom was designed using the LUCMRGen tool that constructs solid geometrical objects. The compressed breast-shaped container was created based on semicylindrical objects and randomly filled with 1430 differently sized spheres (radii of 7.94 mm; 6.35 mm; 4.76 mm; 3.18 mm and 1.50 mm). The container dimensions are 120 x 60 x 35 mm and the wall thickness was chosen at 3 mm. Afterwards different attenuation coefficients representing different materials were assigned to the objects: skin tissue for the container, glandular tissue as bulk material and adipose tissue for the background spheres. Breast abnormalities i.e. masses and microcalcifications were afterwards added as spherical objects. For the representation of two masses, two spheres with diameter of 8 mm were created and consisted of two different materials (75% glandular/25% adipose and 50% glandular/50% adipose tissue). The masses were positioned at two different heights in the phantom. Four individual microcalcifications were also inserted as small spheres of radii 125, 150, 175 and 200 µm made of Calcium Carbonate, at the same height in the phantom as the masses. Next, projection images were generated using the XRayImageSimulator. An iso-centric geometry with source-to-isocenter distance of 600 mm and source-to-detector distance of 650 mm was selected together with a beam energy of 20 kev. The detector image size was assigned as 150 x 150 mm with a pixel size of 0.1 mm. For 2D mammography a cranio-caudal projection with the x-ray tube positioned at 0 was acquired, while for tomosynthesis 26 projections at different angles ranging from -25 to +25 with 2 increment were acquired. A Poison noise distribution was simulated for 8 mgy entrance dose and added to the projection images. Finally, the 26 tomosynthesis projections were reconstructed into a 3D volume using the FDKR tool and applying the filtered back-projection algorithm. Images of both modalities were visually compared and contrast-to-noise ratio (CNR) was calculated in the images with pixel values consisting of line integrals. RESULTS AND DISCUSSION Microcalcifications were equally visible in 2D mammography and tomosynthesis whereas masses were better visualized and easier detected in tomosynthesis compared to 2D mammography (Figure 1). CNR values for microcalcifications ranged from 2.15 to 5.25 in 2D mammography and from to in tomosynthesis (Table 1). While CNR values were higher in tomosynthesis images compared to mammograms, their visibility and detectability was rated equally. For masses, CNR values were equal to 3.19 and 0.75 for 2D mammography and 1.51 and 0.56 in tomosynthesis for the 50% glandular and 75% glandular mass respectively (Table 1). Higher CNR 47

48 values were found in 2D mammography although mass visibility was better in tomosynthesis (Figure 1). These lower CNR values may be due to the limited dynamic range of reconstructed tomosynthesis images. The better visibility of masses in tomosynthesis can be assigned to the reduction of structure overlap and therefore increasing the visibility of the mass in the reconstructed in-focus plane. Note that higher CNR values were found for the 50% glandular sphere compared to the 75% glandular sphere. We assume this is due to the random generation of the phantom with unknown percent division of background materials: skin (wall), 100% glandular (bulk) and adipose tissue (spheres). CONCLUSIONS Microcalcifications were equally visible in both modalities while masses were better represented in tomosynthesis images. In the future, CNR may be replaced by other quantitative metrics for the assessment of the detectability of breast abnormalities. Table 1. Contrast-to-noise ratio values for masses and microcalcifications embedded in a structured phantom measured in 2D mammographic projection and tomosynthesis reconstructed images. Material Dimension (radius in mm) CNR 2D mammography CNR Tomosynthesis Mass 1 50% gland Mass 2 75% gland µcalc 1 CaCO µcalc 2 CaCO µcalc 3 CaCO µcalc 4 CaCO

49 Figure 1. 2D projection (A and B) and tomosynthesis reconstructed in-focus plane (C and D) images of two masses and four individual microcalcifications. 49

50 From the presentation: 50

51 EUTEMPE-RX: European Training and Education for Medical Physics Experts in Radiology MPE05: Anthropomorphic phantoms Applications of anthropomorphic phantoms for design and evaluation of advanced x-ray imaging techniques PROJECT ASSIGNMENT No. 8 Using computational breast phantom, perform a virtual study to determine the influence of breast thickness on the detectability of breast abnormalities in conventional mammography and breast tomosynthesis To be considered: Involved Phantom(s): LUCMR Small to Medium size, Two different heights 20 and 60 mm Spheres sized mm in diameter, bulk material. Abnormalities: One spherical mass abnormality and one individual microcalcification of up to 1 mm in diameter, placed at different z-level. Imaging techniques: Conventional mammography and breast tomosynthesis mammography. Assumed Incident Dose: 5-12 mgy. Any other parameters are by your discretion and preference. To be completed: Create the phantoms, according to the provided specification and add the abnormalities. Choose appropriate materials from table 1. Obtain synthetic conventional mammographic images of the two phantoms at an energy selected by your discretion. Obtain 21 synthetic mammography images of the three phantoms over an arc -20 to 20 degrees (every 2 degrees) using the same energy. Add a realistic level of noise, based on the required incident dose. Perform image reconstruction at the plane of interest, where abnormalities are located. Compare subjectively (visually) the images. Compute objective figure of merits (FOMs) CNR in the region of the abnormalities and compare. To be prepared: Report: a short written summary of the work and the obtained results; containing for example: 51

52 o appropriate outcome images and profiles o tables with computed FOMs o discussion of the results and conclusions PowerPoint Presentation: to be used for a short (5-10 ) oral presentation. Useful hints: The tool for the phantom creation is the LUCMRGen tool. Take into account table 1 that contains list of modelled in XRayImageSimulator materials. For introducing abnormalities use the XRayImageSimulator tool. Take into account table 1. Projection images are generated using the XRayImageSimulator tool. Photon noise is added by using add_noise.m script file in the mcode directory. Modify the script file accordingly. Image reconstruction is performed using FDKR tool. Table 1: Modelled Materials C Al Si Cu Pb water bone compact ICRU-44 lung tissue - inflated brain, grey/white matter breast mass 75% gland 25% fat plexiglas (lucite, PMMA) ideal detector adipose ICRU-44 muscle ICRU-44 Gd2O2S (phosphor) air CsI (phosphor) bone cortical ICRU-44 polystyrene BaFBr gland CaCO3 soft tissue ICRU-44 Breast 50% gland 50% fat ideal air SrFBr paraffin wax, C25H52 skin gray matter white matter blood SiO2 silicon gel epoxy resin nylon-6,6 water vapour ethanol ethanol vapour polyethylene povidoneiodine References: [1]. Bliznakova, K., I. Buliev, et al. (2013). Development and Evaluation of a Software Phantom Dedicated for Breast Imaging Experimentation: Preliminary (First) Results. 5th Pan-Hellenic Conference on Biomedical Technology (ELEBIT 2013). 4-6 April 2013, Athens, Greece. [2]. Cockmartin, L., N. Marshall, et al. (2012). Design and evaluation of a phantom with structured background for digital mammography and breast tomosynthesis. Philadelphia, PA LNCS:

53 PROJECT IMPLEMENTATION REPORT Influence of breast thickness on the delectability of breast abnormalities on conventional mammography and breast tomosynthesis D.Petrov 1, A.Hustuc 2 1 dimitarbp@gmail.com 2 ahustuc@gmail.com Abstract In order to study the difference in detectability for different phantom thicknesses, for two imaging modalities Digital breast tomosynthesis (DBT) and digital mammography. Two different type lesions were placed inside the phantom case. Images were modified for specific dose level, adding blurriness and noise according to the chosen dose level. Detectability measurements has been done in CNR figure of merit. The results show better detectability in digital mammography for thinner phantom, and not significant difference for DBT modality. Method Using LUCMRGen home-made program, two anthropomorphic sphere phantoms has created with the following parameters: Container radius, mm Height, mm Spheres radius [mm] 1,59; 3,18; 5,0; 6,0; 7,0 1,59; 3,18; 5,0; 6,0; 7,0 No. of ellipsoids Material container Container filling material 453 objects skin gland 1383 objects skin gland Sphere material Adipose IRCU- 44 Adipose IRCU- 44 In order to study detectability we simulated two lesion types in each phantom. We chose breast calcification and non-spiculated mass lesions, with the following parameters: Lesions Abnormality material Object type Dimensions Calcification CaCO 3 Ellipsoid 1x1x1 Non-spiculated mass Breast mass Ellipsoid 6x6x6 We simulated a projection images using XrayImagingSimulator software to obtain a single digital mammography image and a set of DBT images with initial beam energy 27keV. System SID [mm] SDD [mm] Angles [deg] Image resolution Digital mammography px/mm Image sizes, [px] 512x x

54 DBT From -20 to +20, with step 1 2px/mm 512x x1024 To acquire real images, we used a home-made tool for adding noise and blurring in order to simulate 10mGy dose. For the DBT projection images a FBP reconstruction has been made, to acquire a stack of coronar slice images. Results For measurement we used ImageJ, with circular ROIs matching the size of the lesions. We calculated Contrast to Noise Ratio(CNR): System Phantom height CNR masses CNR Calcifications Digital Mammography Digital Breast Tomosynthesis 20mm 2,06 3,78 60mm 1,66 2,67 20mm 0,73 13,88 60mm 0,97 12,30 Discussion and conclusions In general using CNR as a figure of merit, we cannot compare the detectability of lesions between DM and DBT, because of the difference of the dynamic range. This causes a lower CNR for DBT, but lesions are easily visible. Human observer reading in 4AFC paradigm should be made and an evaluation of the detectability in percent correct figure of merit should be taken into account. Compare the imaging method In our particular case the CNR of microcalcification in DBT is greater for both phantom heights compared to digital mammography, and opposite for CNR comparison for masses. Compare the lesions For both imaging methods the CNR for masses is lower than the CNR for calcifications, which follows the human detectability of the images. Although the ratio for DBT is larger than the same ratio for digital mammography(15,82 and 1,73 respectively), this could be caused by the dynamic range difference between the two modalities. Compare the phantom 54

55 The phantoms are similar and the only difference is the height. This difference permitted to fit more objects (ellipsoids) in 60 mm phantom. For mammography this causes lower detectability for the thicker phantom, for both of the abnormalities, because of the more overlapping objects. For DBT since the CNR was evaluated in the central plane of the lesions and no overlapping was observed. In this case CNR20mm is lower, than CNR60mm for the mass lesions, and opposite for the microcalcifications, so we cannot conclude that there is a specific detectability difference depending of the phantom thickness. Two phantoms were simulated with different lesion types and different height. Projection images were acquired, then noise was added and comparison of the detectability in the two modalities has been made. From the presentation: 55

56 56