Correction of motion Artefacts in HR-pQCT using Cycle-Consistent Adversarial Networks

Abstract

High-resolution peripheral quantitative computed tomography (HR-pQCT) can provide important information about age-related changes in bone microstructure and strength. However, in elderly patients, uncontrollable tremors often induce motion artefacts that affect the accuracy of HR-pQCT measurements. Repeat acquisition protocols are commonly used to address this issue; however, they are ineffective in these patients, resulting in motion-blurred and streaked images. Deblurring these scans computationally is a difficult inverse problem that is severely ill-posed. Therefore, we present a deep learning approach to suppress motion-induced artefacts in HR-pQCT scans.