Medical Image Reconstruction Using Kernel Based Methods

TL;DR

This paper introduces a flexible kernel-based method for medical image reconstruction from Radon transform data, addressing limitations of classical techniques with a novel interpolation approach and regularization.

Contribution

It applies Hermite-Birkhoff interpolation with kernel functions to improve Radon transform inversion, offering a new, adaptable reconstruction method in medical imaging.

Findings

Kernel methods provide flexible reconstruction options.

Compared to classical algorithms, the new method shows promising behavior.

Regularization is essential due to potential infinities in kernel Radon transforms.

Abstract

The image reconstruction problem consists in finding an approximation of a function f starting from its Radon transform Rf. This problem arises in the ambit of medical imaging when one tries to reconstruct the internal structure of the body, starting from its X-ray tomography. The classical approach to this problem is based on the Back-Projection Formula. This formula gives an analytical inversion of the Radon transform, provided that all the values of Rf are known. In applications only a discrete set of values of Rf is given, thus, one can only obtain an approximation of f. Another class of methods, called ART, can be used to solve the reconstruction problem. Following the ideas contained in ART, we try to apply the Hermite-Birkhoff interpolation to the reconstruction problem. It turns out that, since the Radon transform of a kernel basis function can be infinity, a regularization technique is needed. The method we present here is then based on positive definite kernel functions and it is very flexible thanks to the possibility to choose different kernels and parameters. We study the behavior of the methods and compare them with classical algorithms.