Spectral X-ray imaging using deep learning

Abstract

X-ray Computed Tomography (CT) is a versatile imaging technique widely used in medicine, industry, and beyond. It operates by measuring the linear attenuation coefficient, which depends on photon energy. Traditional X-ray sources used in CT systems are polyenergetic, generating photons across a spectrum of energies. However, conventional CT systems rely on energy-integrating detectors that cannot resolve individual photon energies, limiting their ability to leverage the energy dependency of the attenuation coefficient for precise material characterisation.

Spectral X-ray imaging addresses this limitation by capturing energy-sensitive information using photon-counting detector technology. This approach enables improved material decomposition and characterisation by exploiting the spectral properties of the X-ray beam. By combining spatial and spectral data, spectral imaging provides insights into material composition and density that are unattainable with conventional CT. Applications range from non-destructive testing and material analysis to medical diagnostics and security screening, where such precise characterisation is critical.

Despite its significant potential, spectral imaging faces challenges such as noise from photon division across multiple energy channels, which lowers per-channel counts, increases statistical noise -often requiring longer scan times to collect sufficient data for accurate reconstruction- along with the need for optimised computational methods to balance speed and image quality. Moreover, scalable algorithms and cost-effective detectors are needed to make spectral imaging more accessible for broader applications.

This thesis investigates advanced computational methods and deep learning techniques to enhance Spectral X-ray imaging. Key contributions include novel image enhancement techniques, optimised scanning strategies, precise modelling of energy-dependent absorption spectra, and improved material recovery methods. By addressing these challenges, this work aims to expand the capabilities of Spectral X-ray imaging and establish its practical applications across diverse fields.

More information

Identifiers

ORCID for Raziye Kubra Kumrular: orcid.org/0000-0002-0976-3683

ORCID for Thomas Blumensath: orcid.org/0000-0002-7489-265X

Catalogue record

Export record