Assessment of the advantages of learning tomography over conventional linear optical tomography

Abstract

Optical diffraction tomography provides a powerful tool for imaging the 3-D refractive index (RI) distribution. However the method suffers from two major issues. The first one is the missing cone problem that comes from limited illumination angles. The missing cone problem can be dealt with sparsity based regularization [1]. The second problem is the nonlinear relationship between the refractive index distribution and the scattered field. The assumption of single scattering, i.e. linearization of the problem, is the most commonly used approximation (Wolf transform based on first Rytov approximation). Recently, we proposed a beam propagation method (BPM) based reconstruction algorithm that takes nonlinearity into account [2,3]. In the present contribution, linear and nonlinear forward models are systematically compared for a spherical object against the analytical solution provided by Mie theory [4], which serves as a ground truth. The sparsity based regularization step in the algorithm is kept the same for a fair comparison of the linear and nonlinear forward models.