X-ray Scatter In 3D Microscopes

Abstract

X-ray computed tomography (XCT) is a non-destructive imaging technique that enables 3D internal structure examination of objects. This is achieved by employing X-rays to penetrate the object of interest, and these X-rays are measured by a detector to form images. XCT is very useful in medicine and industrial inspection \cite{ou2021}, it can be used to examine an aircraft's part to ensure that there is no internal fracture. However, such an object would introduce \textbf{object scatter}, reducing image contrast, resulting blurry images that prevent us from seeing the small fractures. This is because some X-rays straight from the X-ray source are deflected or scattered by the object to become secondary X-rays that reach detector regions not initially aligned with these primary X-rays. In this work, we demonstrate a method to measure the spatial distribution of object scatter and subsequently remove it. We used an array of small tungsten spheres to block X-rays from reaching the detector, creating shadow regions on the detector. An array of these spheres is called a Beam-Stop-Array (BSA). By design, only X-rays scattered from the object can reach these shadow regions, thereby contributing extra counts in the shadow regions. Assuming object scatter distribution is spatially slowly variant, we can obtain object scatter distribution with extra counts in shadow regions by interpolation. During the project, we found that X-rays can also scatter: (i) within the source and from the cabinet wall of the XCT system; (ii) within the detector. We called these confounding factors as they can also contribute extra counts in the shadow regions, thereby reducing the accuracy of the object scatter measurement. After addressing these confounding factors, the extra counts in the shadow regions are solely due to object scatter. In our experiment, after measuring and correcting for object scatter, the root-mean-square-error (RMSE) of the attenuation profile of the image of an acrylic cylinder and the theoretical profile drops from $2.84\times10^{-2}$ to $2.07\times10^{-2}$. This drop in RMSE indicates that our method successfully increased the accuracy of images.