Accuracy of respiratory motion compensated image reconstruction using 4DPET-derived deformation fields

PET quantitation in the thorax and upper abdomen is confounded by artifacts introduced by breathing motion. This has led to the emergence of a variety of techniques for motion compensated image reconstruction, many of which rely on motion information computed from a series of respiratory-gated anatomical images synchronized with the PET gates. A simpler alternative to this approach is to derive deformation fields directly from non-attenuation-corrected gated 4DPET images. The goal of this paper is to assess the accuracy of motion compensated image reconstruction based on PET-derived motion information using the ground truth and anatomically derived motion information as references. We used the Monte Carlo simulation software GATE to generate realistic PET images of the XCAT phantom with pulmonary lesions. Deformation fields were derived from 4DPET images in two passes. In the first pass, the fields were estimated from 4D non-attenuation-corrected PET images. In the second pass, 4D attenuation maps were generated using the first-pass motion estimates and then used to reconstruct 4D attenuation-corrected PET images. A second set of deformation fields were then computed from the 4D attenuation-corrected PET images. The PET-derived deformation fields were compared with the true XCAT deformations. For more realistic validation, gated pseudo-MR images were generated from the XCAT phantom. We then compared MR-derived deformation fields with the true deformations. Finally, motion compensated PET images were reconstructed using the different motion estimates, and error estimates were computed for individual organs and lesions.