An Analytical Model for Fast Computation of Scatter Estimation in KV Cone-Beam CT Images
J Liu1,2*, J Bourland1,2,3, (1) Department of Physics, Wake Forest University, Winston-Salem, NC (2)Department of Radiation Oncology, Wake Forest School of Medicine, Winston-Salem, NC (3)Department of Biomedical Engineering, Wake Forest School of Medicine, Winston-Salem, NCSU-E-I-4 Sunday 3:00PM - 6:00PM Room: Exhibit Hall
Purpose: To develop an analytical model for fast and accurate scatter estimation in kV cone-beam CT images.
Methods: With a point source of 100 keV x-rays we used the Klein-Nishina (KN) formula for Compton scattering differential cross-section and integrated the point scatter along an interaction line of tissue-equivalent medium to obtain a beam-scatter-kernel (BSK). We limited our derivation to the case of small imaging field and small scattering angles. The Taylor series of the integrand was approximated by terms up to the 4th order. The integral was then simplified to an analytical form. To calculate attenuation in the medium for scattered photons, we assumed all the scattered photons from a BSK originate from the beam center and have the same energy as the primary photons. We then performed robust calculations using the point scatter theorem for a cubic water phantom (10x10x10 cm3) and treated the results as theoretical expectations for inter-comparison.
Results: Compared to the point-kernel method, BSK calculations show a +10% systematic difference for either a perpendicular or a tilted beam for the absolute value of scatter fluence. However, the largest difference on the relative scatter distribution is as low as 0.4% and 2.5% for perpendicular and tilted beams, respectively. The largest relative difference for the whole phantom is about -1.8%. In computation time, our integrated scatter model is faster by over 2 orders than the point kernel method.
Conclusion: Our kV scatter model shows a very high computational efficiency in scatter estimation. The expected behavior of scattering distribution is observed and well estimated. We are validating more thoroughly for small imaging fields (~10x10cm2) and the model is expected to have great potential for imaging small objects and with interior reconstruction algorithms. Future work will extend the frame for multiple photon energies and tissue non-homogeneity.