Improved PET/CT Respiratory Motion Compensation by Incorporating Changes in Lung Density

Elise C Emond,Brian F Hutton,Ludovica Brusaferri,Kris Thielemans,Alexandre Bousse

doi:10.1109/trpms.2020.3001094

Abstract

Positron emission tomography/computed tomography (PET/CT) lung imaging is highly sensitive to motion. Although several techniques exist to diminish motion artifacts, a few accounts for both tissue displacement and changes in density due to the compression and dilation of the lungs, which cause quantification errors. This article presents an experimental framework for joint activity image reconstruction and motion estimation in PET/CT, where the PET image and the motion are directly estimated from the raw data. Direct motion estimation methods for motion-compensated PET/CT are preferable as they require a single attenuation map only and result in optimal signal-to-noise ratio (SNR). Previous implementations, however, failed to address changes in density during respiration. We propose to account for such changes using the Jacobian determinant of the deformation fields. In a feasibility study, we demonstrate on a modified extended cardiac-torso (XCAT) phantom with breathing motion—where the lung density and activity vary—that our approach achieved better quantification in the lungs than conventional PET/CT joint activity image reconstruction and motion estimation that does not account for density changes. The proposed method resulted in lower bias and variance in the activity images, reduced mean relative activity error in the lung at the reference gate (−4.84% to −3.22%) and more realistic Jacobian determinant values.

Highlights

P ET lung imaging suffers from patient respiratory motion, affecting the image resolution, localization, and quantification due to the mismatch of the acquired PET data with the attenuation map, typically computed from a single snapshot CT acquisition
The proposed method proceeds by alternating between λ and θ updates, using block sequential regularized expectationmaximization (BSREM) algorithm proposed by Ahn et al [24] for λ—which will not be detailed here—and limitedmemory Broyden-Fletcher-Goldfarb-Shanno (L-BFGS) [25] for θ—for which we propose a summary in Appendix B
The results show that using a mass-preserving model in Joint reconstruction and motion estimation (JRM) reduces the image squared biases, especially in the lung region of the activity image

Summary

Introduction

P ET lung imaging suffers from patient respiratory motion, affecting the image resolution, localization, and quantification due to the mismatch of the acquired PET data with the attenuation map, typically computed from a single snapshot CT acquisition. A common approach to compensate for respiratory motion, known as motion-compensated image reconstruction (MCIR), is to use image registration to compute deformation fields between different respiratory states (determined using a respiratory signal [1]), in order to warp the attenuation map and match the PET data, while performing.

Objectives

Methods

Results

Conclusion