Advancement of bright-field hyperspectral X-ray tomographic imaging

Abstract

X-ray computed tomography (CT) has become a powerful tool for 3D evaluation of samples, expanding from healthcare applications to security scanning, archaeology, and non-destructive testing across a range of industries. To date, the majority of X-ray systems implement 'black and white' imaging, providing insight into the relative electron density of structures, but no further material discrimination is achievable. With the creation of pixelated energy-sensitive X-ray detectors, previously lost energy-based information may now be captured, providing both spatial and spectral detail about the sample at every pixel. The development of hyperspectral sensors in recent years has enabled this spectral dimension to be recorded and visualised with excellent energy resolution. Characteristic features of elements within a sample, such as absorption edges and fluorescence peaks, may now be used as fingerprints for chemical mapping across a full reconstructed volume. Hyperspectral imaging offers a wide range of imaging modalities, enabling exploration in the direct X-ray beam (bright-field) and off-axis (dark-field), for full flexibility in the range of chemical and crystallographic properties that may be studied. Bright-field imaging provides an appealing alternative to conventional X-ray CT given the simple requirement to replace the old detector for a hyperspectral one. However, despite the significant potential of bright-field hyperspectral CT, the technique is still in its infancy, and requires further advancement in a number of areas. This thesis focuses on advancing the field of bright-field imaging in terms of the reconstruction, analysis, and wider applications of the technique moving forward. Limited studies to date have explored the reconstruction techniques required to handle hyperspectral datasets. With the additional spectral dimension, 4D datasets emerge. The benefit of high energy resolution, however, is counteracted by poor signal-to-noise ratio. A novel form of regularised iterative reconstruction algorithm is described, and evaluated in a research study on both a phantom sample and stained biological specimen. The extent of noise removal is assessed, with a scan time reduction of 36x possible without loss of image quality. Advancements in the range of bright-field spectral analyses are also explored. The ability to extract quantitative information on chemical concentration and distribution is evaluated for the field of bioimaging, through a research study on multiple simultaneous contrast agent staining. The extraction of absolute chemical concentration values in each voxel are shown for double- and triple-stained specimens, performed for the first time in hyperspectral CT via absorption edge fitting of a set of calibration phantoms. Finally, potential routes for the further development of hyperspectral imaging are discussed.

Date of Award	1 Aug 2023
Original language	English
Awarding Institution	The University of Manchester
Supervisor	Robert Cernik (Supervisor) & Philip Withers (Supervisor)