Computerized tomography and reproducing kernels (Q6972396)

This paper proposes a new method for computerized tomography (CT) based on reproducing kernel Hilbert spaces (RKHSs). The authors treat the X-ray transform as an operator between RKHSs instead of traditional \(\mathscr{L}^2\) spaces. This leads to a more stable and flexible way to deal with tomographic reconstruction, especially when the data are discrete and noisy.\N\NThe main idea of the paper is to replace the usual function space with an RKHS, so that kernel methods can be used to solve the reconstruction problem. The authors reformulate the reconstruction task as a ridge regression problem and propose a new method called kernel reconstruction (KR). This method does not depend on the Fourier transform and gives a clear way to control smoothness using the kernel function. They also show that this idea is related to the well-known representer theorem.\N\NThis method has several benefits. It can be used with many kinds of kernels, including rotation-invariant ones. It does not require changing the problem into a fixed dimension, and the Gram matrix can be computed ahead of time, which saves time in real applications. Also, the method can reduce artifacts in the reconstructed images, which is very helpful in practice.\N\NThis is an interesting and well-written paper. The authors present a new point of view for tomography that is both practical and mathematically solid. I think the ideas in this paper could be useful for both theory and applications in CT and related problems.