Stabilizing Laplacian Inversion in Fokker–Planck Image Retrieval using the Transport-of-Intensity Equation

Samantha J Alloo, Kaye S Morgan · 2024

Read Stabilizing Laplacian Inversion in Fokker–Planck Image Retrieval using the Transport-of-Intensity Equation by Samantha J Alloo, Kaye S Morgan (2024) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2412-15513.txt.

Introduction

Samantha Alloo and Kaye Morgan present a 2024 paper on stabilizing X-ray image reconstruction. Speckle-based X-ray imaging can yield three kinds of pictures at once: absorption, phase, and dark-field. Their retrieval method rests on an equation whose inversion is numerically shaky. A standard fix adds regularization, but the result swings wildly with the chosen strength, and picking it is slow and subjective.

The pair automates that choice. A related, steadier equation gives a trustworthy phase picture, which serves as ground truth while their algorithm tunes the regularization step by step. On synchrotron data from a four-rod sample, the method recovered clean phase and dark-field images with no hand tuning. The authors expect the trick to help other retrieval schemes with the same instability. It makes their imaging pipeline friendlier for routine use.

Catalog record

CREATORSamantha J Alloo, Kaye S Morgan
YEAR2024
COLLECTIONPAPERS
FORMATTXT
FILE SIZE34 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2412-15513

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Stabilizing Laplacian Inversion in Fokker–Planck Image Retrieval using the Transport-of-Intensity Equation Samantha J Alloo, Kaye S Morgan Abstract X-ray attenuation, phase, and dark-field (multimodal) images provide complementary information to one another. Different experimental techniques can capture these contrast mechanisms, and the corresponding images can be retrieved using various theoretical algorithms. Our previous works developed the ‘Multimodal Intrinsic Speckle-Tracking’ (MIST) algorithm, which is suitable for multimodal image retrieval from intensity data acquired using a speckle-based X-ray imaging (SBXI) technique. MIST is based on the X-ray Fokker–Planck equation, requiring the inversion of derivative operators that are often numerically unstable. These numerical instabilities can be addressed by employing appropriate regularization techniques, such as Tikhonov regularization. However, the output from such regularization is highly sensitive to the choice of the Tikhonov regularization parameter, making it crucial to select this value carefully and optimally. In this work, we present an automated iterative algorithm to optimize the regularization of the inverse Laplacian operator in our most recently published and most general MIST variant, addressing the operator’s instability near the Fourier-space origin. Our algorithm leverages the inherent stability of the phase solution obtained from the transport-of-intensity equation for SBXI, using it as a reliable ground truth for the more complex X-ray Fokker–Planck-based algorithms that incorporate the dark-field signal. We applied the algorithm to an experimental SBXI dataset collected using synchrotron light of a sample comprised of four different rods. The phase and dark-field images of the four-rod sample were optimally retrieved using our developed algorithm, eliminating the tedious and subjective task of selecting a suitable Tikhonov regularization parameter. The developed regularization-optimization algorithm makes MIST a more user-friendly multimodal-retrieval algorithm by eliminating the need for manual parameter selection. We anticipate that our optimization algorithm can also be applied to other image retrieval approaches derived from the Fokker–Planck equation, such as those designed for free-space propagation techniques, as they also involve the unstable inverse Laplacian operator. 20.12.2024 Keywords : X-ray dark-field, X-ray phase, Fokker–Planck equation, Transport-of-Intensity equation, Tikhonov regularization 1 Introduction Attenuation, phase, and dark-field are three distinct contrast channels in X-ray imaging, each providing unique insights into a sample’s properties. The attenuation image reveals structural density by quantifying how many X-ray photons are absorbed or scattered out of the primary beam. In contrast, the phase channel captures X-ray refraction within the sample, allowing materials with weak attenuation (i.e., those that are almost invisible in attenuation images) to be distinguished [ 1 ] . The dark-field channel visualizes structures that induce phase variations smaller than the system’s spatial resolution, thus offering information about features that are otherwise undetectable in attenuation or phase images. Numerous studies highlight the broad applicability of these three contrast channels across different fields [ 2 , 3 , 4 , 5 , 6 , 7 , 8 , 9 ] . Several multimodal experimental techniques have been developed to recover attenuation, phase, and dark-field images, each offering specific advantages and limitations. For example, using free-space propagation [ 10 , 11 ] , grating-interferometry [ 12 ] , edge-illumination [ 13 ] , two-dimensional (2D) single grid [ 14 , 15 ] , and speckle-based X-ray imaging (SBXI) [ 16 ] . Each technique employs a unique image acquisition procedure, exploiting slightly different physical changes in the imaging system to capture multimodal sample information. Then, an appropriate retrieval algorithm is required to extract multimodal signals from the experimental intensity data. These algorithms decode the contrast seen in raw images and how that contrast relates to the sample’s attenuation, phase, and dark-field characteristics based on the specific setup. Figure 1: Speckle-based X-ray imaging (SBXI) setup with a paraxial monochromatic X-ray wavefield. A spatially random membrane imprints speckles into the X-ray wavefield to generate a reference-speckle pattern I_{R}(\textit{{r}}) at the detector. When a sample is introduced, a sample-plus-speckle image I_{S}(\textit{{r}}) is formed. All images are captured at the detector, positioned at a distance z=\Delta downstream from the sample. Our research group has been developing the so-called Fokker–Planck equation (FPE) for paraxial X-ray imaging [ 17 , 18 ] , which can also be thought of as a diffusive extension to the transport-of-intensity equation (TIE) [ 19 ] . In this context, the X-ray imaging FPE models how the intensity of an X-ray wavefield, refracted and diffused by a sample, evolves as it propagates along the optical axis. The TIE can be viewed as a simplified version of the Fokker–Planck equation, neglecting X-ray diffusion caused by the sample. Specifically, the TIE is a mathematical framework that describes how an X-ray wavefield’s intensity changes with propagation due to a phase shift introduced by some refracting object. The equation has proven successful in solving the phase-retrieval problem (recovering just phase information), being the basis of the widely adopted Paganin method [ 20 ] , which is suitable for free-space propagation phase-contrast X-ray imaging data, and has been implemented in both light microscopy [ 21 , 22 , 23 ] and electron [ 24 , 25 , 26 , 27 ] microscopy. To also include sample-induced X-ray diffusion, described by an effective diffusion coefficient, the X-ray FPE [ 17 , 18 ] should be used instead of the TIE to model the forward problem. The X-ray FPE has been successfully applied to both free-space propagation [ 11 , 28 , 29 ] and SBXI [ 30 , 31 , 32 , 33 , 34 , 35 , 36 ] , modeling the forward problem and enabling the retrieval of multimodal images from experimental data by solving the corresponding inverse problem. In terms of SBXI, which is the focus of this paper, the FPE establishes a relationship between the reference-speckle image and sample-plus-speckle image. The reference-speckle image is captured when a spatially random membrane is placed in the beam path, while the sample-plus-speckle image is recorded when a sample is introduced into this reference-speckle field. The experimental set-up of SBXI using a monochromatic paraxial X-ray source, which can be achieved at synchrotron facilities, is shown in Fig. 1 . We have been developing an FPE-based multimodal retrieval algorithm for SBXI, known as Multimodal Intrinsic Speckle-Tracking (MIST) [ 30 ] . Since its inception in 2020, several MIST variants have been published, each aiming to optimize the solution to the multimodal inverse problem. The ultimate goal is to solve the SBXI FPE fully–or with minimal assumptions about sample properties or experimental conditions–in the most mathematically stable manner, thereby enabling the recovery of high-resolution multimodal images of a sample. In each published MIST variant [ 30 , 31 , 32 , 33 , 35 ] , the generality of the algorithm has been expanded by relaxing assumptions about the sample’s attenuation, phase, and diffusion characteristics. However, this comes with the trade-off of requiring more input parameters [ 35 ] , which are necessary to stabilize the algorithm and address particular ill-posed mathematical operations. The most recently published and most general MIST algorithm is presented in Alloo et al. [ 35 ] . Notably, when the required input parameters in this MIST algorithm are chosen correctly, the recovered multimodal images exhibit superior image quality compared to earlier MIST variants [ 30 , 31 , 33 ] . Similar to other FPE-derived multimodal retrieval algorithms [ 30

Read the complete text