Convolution/deconvolution of generalized Gaussian kernels with applications to proton/photon physics and electron capture of charged particles

Abstract

Scatter processes of photons lead to blurring of images produced by CT (computed tomography) or CBCT (cone beam computed tomography) in the KV domain or portal imaging in the MV domain (KV: kilovolt age, MV: megavoltage). Multiple scatter is described by, at least, one Gaussian kernel. In various situations, this approximation is crude, and we need two/three Gaussian kernels to account for the long-range tails (Landau tails), which appear in the Molière scatter of protons, energy straggling and electron capture of charged particles passing through matter and Compton scatter of photons. The ideal image (source function) is subjected to Gaussian convolutions to yield a blurred image recorded by a detector array. The inverse problem is to obtain the ideal source image from measured image. Deconvolution methods of linear combinations of two/three Gaussian kernels with different parameters s0, s1, s2 can be derived via an inhomogeneous Fredholm integral equation of second kind (IFIE2) and Liouville - Neumann series (LNS) to provide the source function ρ. The determination of scatter parameter functions s0, s1, s2 can be best determined by Monte-Carlo simulations. A particular advantage of this procedure is given, if the scatter functions s0, s1, s2 are not constant and depend on coordinates. This fact implies that the scatter functions can be calibrated according to the electron density ρelectron provided by image reconstructions. The convergence criterion of LNS can always be satisfied with regard to the above mentioned cases. A generalization of the present theory is given by an analysis of convolution problems based on the Dirac equation and Fermi-Dirac statistics leading to Landau tails. This generalization is applied to Bethe-Bloch equation (BBE) of charged particles to analyze electron capture. The methodology can readily be extended to other disciplines of physics.