banner image
No data available.
Please log in to see this content.
You have no subscription access to this content.
No metrics data to plot.
The attempt to load metrics for this article has failed.
The attempt to plot a graph for these metrics has failed.
A low dose simulation tool for CT systems with energy integrating detectors
Rent this article for
Access full text Article
1. A. J. Britten, M. Crotty, H. Kiremidjian, A. Grundy, and E. J. Adam, “The addition of computer simulated noise to investigate radiation dose and image quality in images with spatial correlation of statistical noise: An example application to x-ray CT of the brain,” Br. J. Radiol. 77, 323328 (2004).
2. X. Li, E. Samei, D. M. DeLong, R. P. Jones, A. M. Gaca, C. L. Hollingsworth, C. M. Maxfield, J. G. Colsher, and D. P. Frush, “Pediatric MDCT: Towards assessing the diagnostic influence of dose reduction on the detection of small lung nodules,” Acad. Radiol. 16, 872880 (2009).
3. D. Frush, C. Slack, C. Hollingsworth, G. Bisset, L. Donnelly, J. Hsieh, T. Lavin-Wensell, and J. Mayo, “Computer-simulated radiation dose reduction for abdominal multidetector CT of pediatric patients,” AJR, Am. J. Roentgenol. 179, 11071113 (2002).
4. J. R. Mayo, K. P. Whittall, A. N. Leung, T. E. Hartman, C. S. Park, S. L. Primack, G. K. Chambers, M. K. Limkeman, T. L. Toth, and S. H. Fox, “Simulated dose reduction in conventional chest CT: Validation study,” Radiology 202, 453457 (1997).
5. O. Amir, D. Braunstein, and A. Altman, “Dose optimization tool,” Proc. SPIE 5029, 815821 (2003).
6. T. M. Benson and B. K. B. De Man, “Synthetic CT noise emulation in the raw data domain,” in IEEE Medical Imaging Conference Proceedings, 2010.
7. B. R. Whiting, “Signal statistics of x-ray computed tomography,” Proc. SPIE 4682, 5360 (2002).
8. B. R. Whiting, P. Massoumzadeh, O. A. Earl, J. A. O'Sullivan, D. L. Snyder, and J. F. Williamson, “Properties of preprocessed sinogram data in x-ray computed tomography,” Med. Phys. 33, 3290303 (2006).
9. P. Massoumzadeh, S. Don, C. Hildebolt, K. Bae, and B. Whiting, “Validation of CT dose-reduction simulation,” Med. Phys. 36, 17489 (2009).
10. P. Massoumzadeh, O. Earl, and B. Whiting, “Noise simulation in x-ray CT,” Proc. SPIE 5745, 898905 (2005).
11. A. Wang and N. Pelc, “Synthetic CT: Simulating low dose single and dual energy protocols from a dual energy scan,” Med. Phys. 38, 55515562 (2011).
12. A. Wang and N. Pelc, “Synthetic CT: Simulating arbitrary low dose single and dual energy protocols,” Proc. SPIE 7961 (2011).
13. D. Sennst, M. Kachelriess, C. Leidecker, B. Schmidt, O. Watzke, and W. Kalender, “An extensible software-based platform for reconstruction and evaluation of CT images,” Radiographics 24, 60113 (2004).
14. S. Nakanishi, M. D. Silver, and A. A. Zamyatin, “Low dose CT simulation using experimental noise model,” in f3d (Fully 3D, Potsdam, Germany, 2011), pp. 203205.
15. R. Luhta, M. Chappo, B. Harwood, R. Mattson, D. Salk, and C. Vrettos, “A new 2D-tiled detector for multislice CT,” Proc SPIE 6142, 61420U61420U (2006).
16. G. Casella and R. Berger, Statistical Inference (Duxbury Resource Center, Pacific Grove, CA, 2001).
17. W. H. Press, S. A. Teukolsky, W. T. Vetterling, and B. P. Flannery, Numerical Recipes in C: The Art of Scientific Computing, 2nd ed. (Cambridge University Press, New York, NY, 1992).
18. S. Žabić, Q. Wang, and K. M. Brown, “A low-dose simulation tool,” in Proceedings of the 2nd International Meeting on Image Formation in X-ray CT (University of Utah, Salt Lake City, 2012), pp. 6165.
19. J. M. Boone, SPEKTR v2.1, 1999, http://www.jhu.edu/istar/downloads.html (accessed July 31, 2012).
20. J. M. Boone and J. A. Seibert, “An accurate method for computer-generating tungsten anode x-ray spectra from 30 to 140 kV,” Med. Phys. 24, 16611670 (1997).
21. J. Fessler, “Statistical image reconstruction methods,” in SPIE Handbook on Medical Imaging—Volume II: Medical Image Processing and Analysis, edited by J. M. Fitzpatrick and M. Sonka (SPIE, Bellingham, WA, 2000).
22. G. B. Folland, Real Analysis: Modern Techniques and Their Applications (Wiley, Hoboken, NJ, 1984).
23. J. A. Fessler, “Hybrid Poisson/polynomial objective functions for tomographic image reconstruction from transmission scans,” IEEE Trans. Image Process. 4, 14391450 (1995).
24. D. S. Lalush and B. M. W. Tsui, “MAP-EM and WLS-AP-CG reconstruction methods for transmission imaging in cardiac spect,” in Proceedings of IEEE Nuclear Science Symposium and Medical Imaging Conference (IEEE, Norfolk, VA, 1993), Vol. 2, pp. 11171178.
25. J.-B. Thibault, C. A. Bouman, and K. D. Sauer, “A recursive filter for noise reduction in statistical iterative tomographic imaging,” Proc. SPIE 6065 (2006).
26. L. A. Feldkamp, L. C. Davis, and J. W. Kress, “Practical cone-beam algorithm,” J. Opt. Soc. Am. A 1, 612619 (1984).
27. H. Turbell, Cone-Beam Reconstruction Using Filtered Backprojection (Department of Electrical Engineering Linköpings Universitet, Linköping, 2001).
28. H. Schöndube, Helical Cone-Beam Computed Tomography Using the Differentiated Backprojection (Shaker Verlag, Erlagen, 2010).
29. G. T. Herman, Image Reconstruction from Projections: The Fundamentals of Computerized Tomography (Academic, New York, 1980).
30. K. Sauer and C. Bouman, “A local update strategy for iterative reconstruction from projections,” IEEE Trans. Signal. Process. 41, 534548 (1993).
31. J. Hsieh, “Adaptive streak artifact reduction in computed tomography resulting from excessive x-ray photon noise,” Med. Phys. 25, 21392147 (1998).
32. M. Kachelriess, O. Watzke, and W. Kalender, “Generalized multi-dimensional adaptive filtering for conventional and spiral single-slice, multi-slice, and cone-beam CT,” Med. Phys. 28, 475490 (2001).
33. P. F. Judy, M. F. Kijewski, and R. G. Swensson, “Observer detection performance loss: Target-size uncertainty,” Proc. SPIE 3036, 3947 (1997).


Image of FIG. 1.

Click to view

FIG. 1.

This figure shows a difference between the real electronic detector noise with offsets subtracted and the random Gaussian noise with the same variance. The gain setting agrees with the Gaussian random only in the first moment.

Image of FIG. 2.

Click to view

FIG. 2.

In this figure we compare frequency spectra of the 6000 samples of a real detector noise signal associated to the 417 μs integration period (left) and we compare it to a signal of 6000 Gaussian noise realization (right) with the same noise variance as the detector noise. We can see that the lower frequencies of the real detector noise are suppressed compared to the uniform spectrum of the Gaussian signal.

Image of FIG. 3.

Click to view

FIG. 3.

In this figure, we see reconstructions from scans at 121 mA and integration period 138 μs of a 30 cm diameter water phantom at 100 kV. Images are 10 mm thick, they are zoomed-in close to the isocenter and visualized at a window of 350 HU, centered at 25 HU. Isocenter is denoted by the white lines in each image. The image (a) comes from the real scan, the image (b) comes from our simulation which uses a real sample of detector noise, and the image (c) is another simulation, but this time we used a zero-mean Gaussian signal for detector noise, with uniform variance channel to channel. Clearly, ring artifacts that are also in the real scan are correctly simulated in the center image, while the image on the right failed to reproduce the artifacts.

Image of FIG. 4.

Click to view

FIG. 4.

Figure shows (a) noiseless reconstruction of the phantom used in Subsection VIII A . The outer ellipse has the long axis of 350 mm and short axis of 210 mm and HU of 0. The circle at the top has diameter of 70 mm and 750 HU. Circle at the bottom has diameter of 50 mm and HU of 1000. (b) The average of 1000 reconstructions from direct noise simulations each equivalent to a 150 mA s scan. The rest of the figures represent the mean of 1000 reconstructions from Poisson simulations of 50 mA s using: (c) direct realizations and (d) method from Sec. IV . (e)–(g) Three different state of the art methods: Benson's (Ref. 6 ), Massoumzadeh's (Refs. 9,10 ), and Amir's (Ref. 5 ), respectively. We consider image (c) to be the ground truth in this experiment, in order to compare low dose simulation tools to a direct Poisson method. We observe that the mean levels are correctly estimated by our method, underestimated by the methods in (e) and (g), and overestimated by the method in (f). Images are in HU, they are centered at 25 HU, and represented in window of 150 HU. The dashed horizontal line represents the profile location used in Figs. 5 and 6 .

Image of FIG. 5.

Click to view

FIG. 5.

Figure represents the profile mean of the central row in reconstructed volumes for direct noise simulation (full line), method from Sec. IV (labeled as “Our method”), Benson's method (Ref. 6 ) (labeled as Method 1), Massoumzadeh's method (Refs. 9,10 ) (labeled as Method 2), and Amir's method (Ref. 5 ) (labeled as Method 3). Mean levels are most closely calculated by our method.

Image of FIG. 6.

Click to view

FIG. 6.

Like in Fig. ( 5 ) we see five profile lines, all through the middle section of the phantom, but this time we represent the noise variance from 1000 images. Noise levels of images simulated by our method are the closest to the direct simulations when compared to the other methods.

Image of FIG. 7.

Click to view

FIG. 7.

These graphs represent four statistical moments of a signal of 6000 x rays through 15 cm of polyoxymethylene across one row of the detector system. The real signal is acquired at 120 kv, 431 μs, and 50 mA. The simulations are at the same current, where the input data were 100 mA. There are 672 channels. Graph on the far left shows that the mean of the simulation and real data match almost exactly. Center left graph shows that also the variance of the two signals matches quite well. Two graphs compare the skew and kurtosis of the two signals. Skew matches well, and there is a mismatch in kurtosis.

Image of FIG. 8.

Click to view

FIG. 8.

In this figure we illustrate that noise power spectrum in images reconstructed from our simulation matches the real scans. Graph on the left is obtained for the tube current of 250 mA, and graph on the right is obtained for the tube current of 30 mA. Integration period was 413 μs and tube voltage was 120 kV, scanned object was a circular water phantom with 35 cm diameter. Tube current for the input to the simulator was 500 mA.

Image of FIG. 9.

Click to view

FIG. 9.

In this figure we show reconstructed images at (a) 250 mA real data, (b) 30 mA real data, (c) 250 mA simulated data, and (d) 30 mA simulated data, so that the reader can appreciate closeness of our simulations. The same images are used in Fig. 8 for the calculation of noise power spectra and in Table I for noise variance comparison.

Image of FIG. 10.

Click to view

FIG. 10.

This figure represents the sinogram of the central slice in the 10 mA anthropomorphic pelvis scan used in the Sec. VIII B . The white dots represent readings in which no photons were detected. The views with such readings are considered photon starved projections.

Image of FIG. 11.

Click to view

FIG. 11.

This figure represents the comparison between the real scans (a) and different noise simulation algorithms: ours (b), Benson's (c), and Massoumzadeh's (d). Images are 10 mm slabs, centered at 75 HU with a window of 700 HU. Black arrows in (a) and (b) indicate detector noise manifestation in the image space, which occur when the data are photon starved.

Image of FIG. 12.

Click to view

FIG. 12.

This figure represents an image from a high dose (200 mA) anthropomorphic pelvis scan used in Sec. VIII B . The image is a 10 mm slab, centered at 75 HU with a window of 700 HU.

Image of FIG. 13.

Click to view

FIG. 13.

This figure illustrates that it the physical effects such as scatter and beam hardening can have a subtle impact on our noise simulation tool. Image (a) is from the real scan, (b) is from the noise simulation that starts from the corrected data, and (c) is from the simulation that was corrected for the physical effects after the simulation. Noise measured in the circular ROI is slightly higher in image (c) (46 HU standard deviation) than in images (a) and (b) (36 and 37 HU standard deviation).

Image of FIG. 14.

Click to view

FIG. 14.

This figure contains reconstructions from (a) the real data set at 644 mA, (b) our simulation data at the level 80 mA, and (c) real data set at 80 mA. Tube voltage was 120 kV, integration period was 137 μs, and data which were used in reconstruction of (a) were also input for simulation tool. Images are visualized at 450 HU window and centered at 75 HU.


Generic image for table

Click to view


Image noise in real scan and simulations.


Article metrics loading...




This paper introduces a new strategy for simulating low-dose computed tomography (CT) scans using real scans of a higher dose as an input. The tool is verified against simulations and real scans and compared to other approaches found in the literature.


The conditional variance identity is used to properly account for the variance of the input high-dose data, and a formula is derived for generating a new Poisson noise realization which has the same mean and variance as the true low-dose data. The authors also derive a formula for the inclusion of real samples of detector noise, properly scaled according to the level of the simulated x-ray signals.


The proposed method is shown to match real scans in number of experiments. Noise standard deviation measurements in simulated low-dose reconstructions of a 35 cm water phantom match real scans in a range from 500 to 10 mA with less than 5% error. Mean and variance of individual detector channels are shown to match closely across the detector array. Finally, the visual appearance of noise and streak artifacts is shown to match in real scans even under conditions of photon-starvation (with tube currents as low as 10 and 80 mA). Additionally, the proposed method is shown to be more accurate than previous approaches (1) in achieving the correct mean and variance in reconstructed images from pure-Poisson noise simulations (with no detector noise) under photon-starvation conditions, and (2) in simulating the correct noise level and detector noise artifacts in real low-dose scans.


The proposed method can accurately simulate low-dose CT data starting from high-dose data, including effects from photon starvation and detector noise. This is potentially a very useful tool in helping to determine minimum dose requirements for a wide range of clinical protocols and advanced reconstruction algorithms.


Full text loading...

This is a required field
Please enter a valid email address
752b84549af89a08dbdd7fdb8b9568b5 journal.articlezxybnytfddd
Scitation: A low dose simulation tool for CT systems with energy integrating detectors