Overcoming detector limitations of x-ray photon counting for preclinical microcomputed tomography

Abstract

Spectral computed tomography (CT) using photon counting detectors (PCDs) can provide accurate tissue composition measurements by utilizing the energy dependence of x-ray attenuation in different materials. PCDs are especially suited for K-edge imaging, revealing the spatial distribution of select imaging probes through quantitative material decomposition. We report on a prototype spectral micro-CT system with a CZT-based PCD (DxRay, Inc.) that has $16\times 16\text{pixels}$ of $0.5\times 0.5{\mathrm{mm}}^2$ , a thickness of 3 mm, and four energy thresholds. Due to the PCD's limited size ( $8\times 8{\mathrm{mm}}^2$ ), our system uses a translate-rotate projection acquisition strategy to cover a field of view relevant for preclinical imaging ( $\sim 4.5\mathrm{cm}$ ). Projection corrections were implemented to minimize artifacts associated with dead pixels and projection stitching. A sophisticated iterative algorithm was used to reconstruct both phantom and ex vivo mouse data. To achieve preclinically relevant spatial resolution, we trained a convolutional neural network to perform pan-sharpening between low-resolution PCD data ( $247\text{-}\mu m$ voxels) and high-resolution energy-integrating detector data ( $82\text{-}\mu m$ voxels), recovering a high-resolution estimate of the spectral contrast suitable for material decomposition. Long-term, preclinical spectral CT systems such as ours could serve in the developing field of theranostics (therapy and diagnostics) for cancer research.

Keywords: deep learning; image reconstruction; microcomputed tomography; photon counting; spectral computed tomography.