Feasibility of respiratory motion compensated reconstruction using 4D PET-derived deformation fields

Objectives With the advent of simultaneous PET/MR technology, it is now possible to generate high-resolution 4D anatomical images from which accurate deformation fields and hence 4D attenuation maps can be derived and utilized for motion compensated image reconstruction (MCIR) for PET. In principle, it is feasible to estimate deformation fields from gated, non-attenuation corrected (NAC) PET images. However, so far no systematic study has been conducted to quantitatively evaluate PET-based motion estimation approaches and determine their feasibility. Our objective therefore is to compare two PET-based motion estimation techniques with MCIR based on accurate 4D attenuation maps.

Methods We performed realistic simulations using GATE to compare the following 3 methods on the XCAT torso phantom with 4 pulmonary lesions for 6 respiratory gates. Method 0: We performed MCIR with 4D attenuation maps based on the true deformation fields. As this is the best case scenario, we treated this result as our reference image for quantitative comparison. Method 1: We first derived deformation fields from gated NAC PET images and then performed MCIR with 4D attenuation maps based on the NAC PET derived deformation fields. Method 2: We first used the NAC PET derived deformation fields to warp the attenuation map to generate 4D gated and attenuation corrected (AC) PET images. We then used the AC PET images to re-compute the deformation fields. Finally, we performed MCIR with 4D attenuation maps based on the AC PET derived deformation fields.

Results The motion compensated image from Method 1 exhibited significant bias and artifacts, particularly near the heart and the diaphragm. Method 2 suppressed some of the artifacts. However, compared to Method 0, a large absolute bias ranging from 50 % to 400% was observed in the lesions.

Conclusions This study demonstrates that, although MCIR based on 4DPET-derived deformation fields is attractive due to its inherent simplicity, there are significant tradeoffs in terms of image bias and artifacts.