HD FOV Somaris X

White paper HD FOV Technical principles and phantom measurements evaluating HU accuracy and skin-line accuracy in the extended field of view (FOV) region for VA30, version 4.0 Matthias Baer, PhD, Siemens Healthcare GmbH Harald Schoendube, PhD, Siemens Healthcare GmbH SIEMENS Healthineers White paper · HD FOV Introduction Accuracy of dose calculation in radiotherapy often depends on precise assignment of the electron density to each voxel in an image and precise reconstruction of patient geometry. Imaging is typically carried out on a CT simulator (CT scanner that can be used for treatment simulation and has features that support radiotherapy) with a wide-bore size (typically at least 80 cm) to accommodate bariatric cases, immobilization devices, patient positioning systems, etc. The CT simulator typically has a scan field of view (sFOV) of 50 cm –60 cm, and in most cases the patient boundary is well within the scan field of view (sFOV). However, in certain cases, for example, in patients with higher BMI or when using a positioning device such as a breast board, part of the patient’s anatomy extends beyond the sFOV. The limited data in the regions outside the sFOV often leads to inaccurate patient geometry and often reduced accuracy on the HU in the extended FOV region. To overcome this, extended FOV algorithms1, 2 have been developed and constantly been improved over the years to increase the accuracy of reconstruction outside the sFOV. In this publication, we describe our latest implementation of extended FOV reconstruction. Furthermore, we demonstrate HU and geometric accuracy in the extended FOV region using phantom studies. 2 HD FOV · White paper Contents Introduction 4 Theory 5 Phantom studies 6 Results and discussion 9 Known limitations 13 Conclusion 15 References 16 3 White paper · HD FOV Theory The current implementation of the HD FOV1 algorithm a) Mass conserving boundary estimation, b) Sinogram consists of several steps (shown in figure 1) to reduce estimation, and c) Artifact reduction; as shown in figure 1. artifacts in the regions outside the sFOV. These include: a) Mass conserving boundary estimation b) Sinogram estimation c) Artifact reduction Measured Projections V V V V V V Measured Detruncation Initial Recon Contour Binary Image Forward Mixing Final Final HD FOV Projections Estimation Projection of Detruncation Recon (with Binary Image truncation) Figure 1: HD FOV algorithm flowchart that shows the 3 main subparts, i.e., Mass conserving boundary estimation, sinogram estimation, and final artifact reduction. Mass conserving boundary estimation To reconstruct information in the extended FOV region, projection angles that see the entire object/patient and one needs to start with an estimate of the patient lower otherwise. boundary from the limited data available. In our newest implementation, we start with the assumption that every To de-truncate the data, a mass consistency condition projection that covers the entire object/patient should is applied and an initial estimate of projections is created result in a constant mass. When the object or patient such that the mass is identical in all the projections (as body being imaged lies at least partly outside the sFOV, seen in figure 2). Extrapolation is carried out in projection this condition is violated and hence the data for those space for the truncated projections using a cosine shaped projections is truncated during measurements as seen on function such that mass is consistent across all the the left in figure 2. The projection mass that is defined as projections. To avoid any artifacts from an abrupt transition the normalized cumulative sum of the attenuation values from measured data to extrapolated data, the algorithm of a projection varies as a function of the angle (0–360°). ensures that the transition between measured and As can be seen, the normalized projection mass is 1 for extrapolated data is smooth. Projection mass Projection mass 1 1 0.9 0.9 0.8 0.8 0° 180° 360° 0° 180° 360° Mass consistency de-truncation Scan field of view Figure 2: Sinogram de-truncation to ensure mass consistency across all the projections. 1 The image quality for the area outside the 50 cm (or 60 cm for applicable scanners, see table 1) scan field of view does not meet the image quality of the area inside the 50 cm (or 60 cm, see table 1) scan field of view. Image artifacts may appear, depending on the patient setup and anatomy scanned. HD FOV cannot be used for scan FOV smaller than 50 cm (or 60 cm, see table 1). Current version of the HD FOV Algorithm is 4.0. 4 HD FOV · White paper An initial estimate of the image is reconstructed (seen in frequency dependent threshold is applied to the initial figure 3a) using these extrapolated projections. Based on reconstruction. In contrast to the predecessor version this reconstruction a refined estimation of the parts of the the frequency dependency of the thresholding is now patient lying outside the sFOV is done. To do so, as a first extended to the z direction of the volume. With the prior step the initial reconstruction (after applying the mass version of HD FOV the thresholding was only frequency conserving data estimation) of the object is binarized to dependent in the axial plane. Extending the frequency get an approximation of the boundaries. If the binarization dependency to the z direction helps to avoid step artifacts would be done by a simple threshold this would often lead in the patient outline in z direction which where observable to a non-continuous result of the binarization as can be in some cases with the predecessor version. Overall the seen in figure 3b: Patient positioning system and patient frequency depended threshold helps in reducing the table are covered only partly in the binary image and we sensitivity of binarization with respect to artifacts and can also notice that the patient body shows certain finally leads to a more continuous estimate of the object/ non-anatomical discontinuities. To get a more reliable and patient boundaries (see figure 3c). more stable estimate of the object/patient boundary a Sinogram estimation and artifact reduction The measured data is now extrapolated and mixed with To remove these residual truncation artifacts a final the forward projections from the binary image (see figure de-truncation step is carried out in the projection space 1). This step can potentially result in a final image that is and the final image is reconstructed. This final a close representation of the object/patient being imaged. de-truncation is applied to all projection data after the This however is true only when the truncation in the mixing step (forward projections of binary image and measured data is fully captured in the binary image. As measured data) and ensures that all projections fade can be seen from figure 3, truncation can occur from the out smoothly to zero. CT table, immobilization devices, or positioning devices. a) b) c) Figure 3: a) Initial estimated image with mass consistency, b) Binary representation of a created by simple thresholding. c) Binary representation of a created with frequency dependent thresholding. 5 White paper · HD FOV Phantom studies To evaluate the performance of the new HD FOV algorithm For position two and three the height was adjusted such we set up a study using a 33 cm diameter Gammex that parts of the phantom were located outside the scan Electron Density phantom (Gammex, Middleton, WI, USA). field of view of the CT scanner. The absolute phantom The images were acquired on a diagnostic CT scanner positions were adjusted for the different geometries of with 70 cm bore (SOMATOM go.Up), on a diagnostic the investigated CT scanners. Images were reconstructed 82 cm bore CT scanner (SOMATOM X.cite) and on a 85 cm using a FOV of 700 mm on the 70 cm bore CT scanner. bore CT scanner (SOMATOM go.Open Pro) designed for The reconstructed FOV on the 82 cm scanner was set to radiotherapy.1 815 mm and to 850 mm on the 85 cm bore CT scanner. The phantom was imaged at three different positions An overview of the used phantom positions, reconstructed in the scanner as shown in figure 4. The three positions FOVs, and scan filed of views is given in table 1 (see also were chosen so that for the first position the phantom figure 4, figure 6, and figure 7). was completely contained within the scan field of view. CT Scanner Position 1 Position 2 Position 3 Reconstructed Scan field of FOV view (sFOV) SOMATOM go.Up Within sFOV Outer phantom Outer phantom edge at 580 mm edge at 680 mm 700 mm 500 mm SOMATOM Within sFOV Outer phantom Outer phantom X.cite edge at 630 mm edge at 750 mm 815 mm 500 mm SOMATOM go.Open Pro Within sFOV Outer phantom Outer phantom edge at 740 mm edge at 810 mm 850 mm 600 mm Table 1: Positions of the Gammex phantom that were used on the different scanners for the evaluation of HDFOV. HU accuracy HU accuracy in the extended field of view region was mean HU value within the ROI was compared to a evaluated by measuring the mean HU value in a region-of- reference mean HU value from the ROI in the case when interest (ROI). Therefore an ROI was placed at the 12 the phantom is contained within the sFOV, left image in o’clock position within the Gammex phantom as indicated figure 4, figure 6, and figure 7. These measurements were by the yellow circle in figure 4, figure 6, and figure 7 carried out using a water insert next to the ROI and then below. This measurement was repeated on 6 central slices repeated by placing a bone insert next to the ROI. and the average value over all slices was computed. The 1 Please note that SOMATOM go.Sim achieves comparable results for HU and skin-line accuracy (see chapter Conclusion). 6 HD FOV · White paper Phantom diameter accuracy (skin-line accuracy) Skin-line accuracy was evaluated by measuring 75 line averaged over all 75 line profiles and all slices. As for profiles distributed in equally spaced angle increments the evaluation of the HU values the case where the entire within the yellow shaded area (some example lines are phantom is contained within the sFOV was used as plotted in figure 5). The angle range for the line profiles reference (left image of figure 5, figure 6, and figure 7). was chosen so that the whole range of truncation in the is covered for phantom position 2, as shown by the The study comparing FWHM was done using a water yellow shaded region in figure 5, figure 6, and figure 7. insert in the extended FOV and then repeated with a bone insert in the extended FOV. The datasheet value The full-width-at half-max (FWHM) of the 75 line profiles for the diameter of the Gammex phantom3 is 330 mm. was computed for the 6 central image slices of the reconstruction. The reported diameter is the FWHM O 580 mm 680 mm 700 mm (bore size) 500 mm (sFOV) Figure 4: Setup for imaging the Gammex phantom at 3 different heights (from left to right) on the 70 cm bore CT scanner SOMATOM go.Up: phantom completely in the scan field of view, outer phantom edge at 580 mm, and outer phantom edge at 680 mm. 580 mm 680 mm 700 mm (bore size) 500 mm (sFOV) Figure 5: Setup to measure skin-line accuracy on the 70 cm bore CT scanner SOMATOM go.Up. 75 line profiles in the 6 central slices were used to report the accuracy of the diameter of the phantom. 7 White paper · HD FOV O 630 mm 750 mm 815 mm (bore size) 500 mm (sFOV) Figure 6: Measurement setup on the 82 cm bore CT scanner SOMATOM X.cite. Phantom positions were adjusted to the bore size of the scanner. 75 line profiles in the 6 central slices were used to report the accuracy of the diameter of the phantom. ... 740 mm 810 mm 850 mm (bore size) 600 mm (sFOV) Figure 7: Measurement setup on the 85 cm bore CT scanner SOMATOM go.Open Pro. Phantom positions were adjusted to the bore size of the CT scanner. 75 line profiles in the 6 central slices were used to report the accuracy of the diameter of the phantom. 8 HD FOV · White paper Results and discussion HU accuracy Measurement of the mean HU values within the ROI was edge at position 3 (table 1). Using the image where the performed in the 6 central slices. The average of these entire phantom is contained within the scan field of view measurements is shown in table 2 for the different as reference, we see that the difference in the mean HU positions of the phantom: contained within scan field of values is within ± 20 HU. view, phantom edge at position 2 (table 1) and phantom 580 mm 680 mm 700 mm (bore size) 500 mm (sFOV) Figure 8: Results from SOMATOM go.Up. Gammex phantom with water insert next to the ROI reconstructed at different phantom positions (from left to right): phantom completely in the scan field of view, outer phantom edge at 580 mm, and outer phantom edge at 700 mm. 630 mm 750 mm 815 mm (bore size) 500 mm (sFOV) Figure 9: Results from SOMATOM X.cite. Gammex phantom with water insert next to the ROI reconstructed at different phantom positions (from left to right): phantom completely in the scan field of view, outer phantom edge at 630 mm, and outer phantom edge at 750 mm. 9 White paper · HD FOV With water insert Position SOMATOM go.Up SOMATOM X.cite SOMATOM go.Open Pro HUROI – HUROI_Reference HUROI – HUROI_Reference HUROI – HUROI_Reference Phantom at position 1 (within sFOV) 0.00 HU 0.00 HU 0.00 HU Phantom at position 2 9.75 HU 9.52 HU 10.04 HU Phantom at position 3 -0.97 HU 15.54 HU 12.13 HU Table 2: HU values measured in an ROI (averaged over 6 central slices) at different phantom positions: phantom completely in the scan field of view, small part of the phantom out of the sFOV, and large part of the phantom out of the sFOV. Reported values are in proximity to a water insert. The same measurements repeated in the proximity of a 82 cm bore CT scanner we see a larger HU deviation of bone insert (figure 11, figure 12, and figure 13) show more than ± 20 HU (see table 3). Good HU value stability consistent results with the difference within ± 20 HU for and consistency even in cases where large parts of the most of the cases (see table 3). Only for the case with phantom lie significantly outside the scan field of view the bone insert and phantom edge at 750 mm on the implies reliable dose calculations and dosimetry outside the scan field of view. 740 mm 810 mm 850 mm (bore size) 600 mm (sFOV) Figure 10: Results from SOMATOM go.Open Pro. Gammex phantom with water insert next to the ROI reconstructed at different phantom positions (from left to right): phantom completely in the scan field of view, outer phantom edge at 740 mm, and outer phantom edge at 810 mm. 10 HD FOV · White paper 700 mm (bore size) 580 mm 680 mm 500 mm (sFOV) Figure 11: Results from SOMATOM go.Up. Gammex phantom with bone insert next to the ROI reconstructed at different phantom positions (from left to right): phantom completely in the scan field of view, small part of the phantom out of the sFOV, and large part of the phantom out of the sFOV. 815 mm (bore size) 630 mm 750 mm 500 mm (sFOV) Figure 12: Results from SOMATOM X.cite. Gammex phantom with bone insert next to the ROI reconstructed at different phantom positions (from left to right): phantom completely in the scan field of view, small part of the phantom out of the sFOV, and large part of the phantom out of the sFOV. 850 mm (bore size) 740 mm 810 mm 600 mm (sFOV) Figure 13: Results from SOMATOM go.Open Pro. Gammex phantom with bone insert next to the ROI reconstructed at different phantom positions (from left to right): phantom completely in the scan field of view, small part of the phantom out of the sFOV, and large part of the phantom out of the sFOV. 11 White paper · HD FOV With bone insert Position SOMATOM go.Up SOMATOM X.cite SOMATOM go.Open Pro HUROI – HUROI_Reference HUROI – HUROI_Reference HUROI – HUROI_Reference Phantom at position 1 (within sFOV) 0.00 HU 0.00 HU 0.00 HU Phantom at position 2 9.83 HU -18.39 HU -5.66 HU Phantom at position 3 -14.47 HU 41.25 HU 16.87 HU Table 3: HU values measured in an ROI (averaged over 6 central slices) at different phantom positions: phantom completely in the scan field of view, small part of the phantom out of the sFOV, and large part of the phantom out of the sFOV. Reported values are in proximity to a bone insert. Phantom diameter accuracy (skin-line accuracy) The true diameter of the Gammex phantom used in this The results show that the accuracy of the diameter study is 330 mm (data sheet value)3. Mean phantom measurement is 2 mm (see table 3 and table 4). The diameter was measured as the average diameter from 75 results in this phantom setup imply that HD FOV produces line profiles for the 6 central image slices. The case where a consistent phantom geometry even in cases where large the entire phantom contained within the scan field of view parts of the phantom lie outside the scan field of view of was used as reference. the CT scanner. With water insert SOMATOM go.Up SOMATOM X.cite SOMATOM go.Open Pro Position Difference to Difference to Difference to reference diameter reference diameter reference diameter Phantom at position 1 (within sFOV) 0.0 mm 0.00 mm 0.00 mm Phantom at position 2 -0.77 mm -0.95 mm -1.27 mm Phantom at position 3 -0.88 mm -1.03 mm -2.24 mm Table 4: Mean diameter of the Gammex phantom measured in central 6 slices in presence of a bone insert at the top edge of the phantom. Difference to the reference (phantom completely within the scan field of view) is used as a metric to estimate skin-line accuracy. With bone insert SOMATOM go.Up SOMATOM X.cite SOMATOM go.Open Pro Position Difference to Difference to Difference to reference diameter reference diameter reference diameter Phantom at position 1 (within sFOV) 0.00 mm 0.00 mm 0.00 mm Phantom at position 2 -0.71 mm 0.38 mm -1.92 mm Phantom at position 3 -1.03 mm -2.05 mm -2.46 mm Table 5: Mean diameter of the Gammex phantom measured in central 6 slices in presence of a bone insert at the top edge of the phantom. Difference to the reference (phantom completely within the scan field of view) is used as a metric to estimate skin-line accuracy. 12 HD FOV · White paper Known limitations The HD FOV algorithm tries to estimate data in regions sFOV. Even in this case the extended projection data is that were not covered during the measurement. To do so attached smoothly to the measured projections. Typically the algorithm uses the principle of mass conservation in the high density region, e.g. a bone, is followed by a projection data. This principle holds true only in the case lower density tissue, e.g. soft tissue or fat, which causes of a 2D data acquisition in fan or parallel beam geometry. a discontinuity in the real projection data which may not Siemens Healthineers CT scanners have a cone beam be reassembled sufficiently by the extrapolation. As a geometry and typically use a 3D spiral scan mode for result the final reconstructed image may show artifacts data acquisition. This violates the principle of mass in these cases. conservation in projection data which is only approximately true for spiral cone beam CT. This is a known limitation of For correcting a given projection at a certain view angle the HD FOV algorithm and may cause artifacts in HD FOV needs the data from prior projections: If the reconstructed images. Further due to its design the scanned patient’s anatomy changes dramatically in algorithm is prone to artifacts in the case where low z direction this may lead to artifacts. This issue is even density regions are located at the border of the sFOV and stronger for phantom scans when phantoms which are if those regions are followed by regions with a higher quite short (relative to the detector width) in z direction density further out e.g. if the border of the sFOV lies in are scanned in spiral mode. Then the transition from air the patient’s lung. As described in section Mass to phantom and the short phantom length causes Conserving Boundary Estimation the algorithm attaches artifacts in HD FOV reconstructions. the extrapolated data smoothly to the measured data. When using HD FOV one should always be aware of the In the case where the border of the sFOV is in the lung fact that the algorithm estimates data were no data was region of the patient, true projection data outside the measured. Therefore the regions beyond the border of sFOV would increase again once the thorax wall and the the sFOV are only an educated guess of the real patient ribs are reached. This shape of the projection data cannot shape and HU values. Inside the sFOV all data needed for always be reassembled accurately by the algorithm and the reconstruction were measured during the scan and the current HD FOV algorithm is known to be prone to images resemble the patient anatomy correctly – as in artifacts in these situations. A similar effect can occur standard non-HD FOV reconstructions. Outside the sFOV when high density objects are located at the edge of the data is at least partly estimated by HD FOV. 13 White paper · HD FOV Scan & recon guide To receive comparable measurements, please use Phantom positioning: following scan parameter: Try to find a low-density phantom holder. Polystyrene were • Protocol: Default AbdomenSeq; used at the whitepaper. A low-density holder is quite • CareDose: off important as high density holder will cause artifacts and • 200 mAs will influence the results with the phantom (HU and • 120 kV diameter accuracy). The phantom should be shifted • Kernel Qr40 vertically out of the scan field of view. • ADMIRE/SAFIRE = off • 1 mm slices Phantom scanning and reconstruction: • IBHC: Bone. 1. 3 scans at three different positions. Phantom completely Only one sequence at the phantom region inside sFOV, but at the upper part (see figures 5,6,7) at one z-position without table feed. 2. Height is adjusted that outer phantom edge is contained within a circle with a diameter of appr. 580 mm, 630 mm, Phantom configuration: respectively 740 mm depending on the used CT scanner Only use 33 cm diameter Gammex Electron Density (see figures 5, 6, 7) phantom (Gammex, Middleton, WI, USA). Make sure 3. Height is adjusted that outer phantom edge is contained that no very high or very low density inserts are located within a circle with a diameter of appr. 680 mm, 750 mm, at the border of the scan field of view as this will cause respectively 810 mm depending on the used CT scanner artifacts. Try to move all high/low density inserts into (see figures 5, 6, 7) the region which is in the scan field of view and put mainly water-like inserts in the region outside the scan For scan 2+3 the reconstructed FOV is set to highest value. field of view. Whenever the white paper refers to a bone insert the insert with a density of 1.28 g/mm³ is meant. When measuring the accuracy HU and diameter please consider the way these values were measured and evaluated stated above. Please note, that we report a mean deviation for both HU value accuracy and skin- line accuracy. 14 HD FOV · White paper Conclusion The new HD FOV continues to improve upon the earlier generations of extended FOV algorithms in Siemens Healthineers CT scanners. It enables the reconstruction of images while significantly improving the visualization of anatomy in the regions outside the scan field of view of 50/60 cm. Based on the findings in a phantom study, an HU value accuracy of ± 20 HU was achieved with skin-line accuracy of 2 mm for three different CT scanners (SOMATOM go.Up, SOMATOM X.cite, SOMATOM go.Open Pro). SOMATOM go.Sim achieves the same HU and skin line accuracy as SOMATOM go.Open Pro. Only for the extreme case where large parts of the phantom lie outside the sFOV we see a larger deviation of about 40 HU on the 82 cm bore CT scanner. 15 References [1] Gysbrechts, S; Scheelen, I; and Bruder, H. HD Field of View in Computed Tomography for RT Planning: Whitepaper. Siemens Healthcare; 2012. [2] Gersh, J; and Mistry, N. Evaluating the Evolution of Extended Field of View Algorithms in CT Simulators from Siemens Healthineers: Whitepaper, Siemens Medical Solutions USA, Inc., 2017. [3] https://www.sunnuclear.com/documents/datasheets/ gammex/ct_electron_density_phantom.pdf Siemens Healthineers Headquarters Siemens Healthcare GmbH Henkestr. 127 91052 Erlangen, Germany Phone: +49 9131 84-0 siemens-healthineers.com Published by Siemens Healthcare GmbH · 8742 0620 online · ©Siemens Healthcare GmbH, 2020