Accuracy analysis of image-based respiratory motion estimation and compensation in respiratory-gated PET reconstruction

Reconstructed images from respiratory-gated frames of positron emission tomography (PET) data can be used to estimate the patient’s respiratory motion estimation. The estimated motion can then be incorporated into iterative image reconstruction methods for respiratory motion compensation. The objective of this study is to investigate two factors in the PET images that can affect the accuracy of image-based motion estimation: image noise and artifacts due to attenuation map misregistration. 4D XCAT (formerly known as NCAT) digital phantom and Monte-Carlo methods were employed to simulate realistic F¹⁸-FDG oncological PET data acquisition of two different noise levels. Hot spherical lesions of different sizes, contrasts and locations in the XCAT phantom were used to simulate tumors in PET imaging. Respiratory-gated PET data were then reconstructed using OS-EM methods with attenuation, uniformity and random coincidence compensation. Two kinds of attenuation maps were used: the attenuation map registered with every respiratory-gated PET frame and that averaged over the respiratory cycle. The image registration method with B-Spline non-rigid transform was applied to the above reconstructed images for motion estimation. The estimated motion was then used to warp the reconstructed images to a reference frame for summation. The recovered contrasts of the lesions were compared for the final reconstructed images from two noise levels using different methods. The results show that the artifacts in PET images due to attenuation map misregistration is another significant factor besides image noise which can affect the motion estimation accuracy.