| The overall objective of this dissertation study is to develop a four-dimensional (4D) positron emission tomography (PET) image reconstruction method with respiratory motion compensation to improve the detectability of small cancer lesions (<10mm) in clinical oncological 18F-fluorodeoxyglucose (FDG) PET images.;The first step of our method is to estimate the respiratory motion based on respiratory-gated PET sinograms. The following iterative approach was developed and applied. At each iteration, a CT image-based attenuation map at a reference respiratory phase is first transformed by the respiratory motion estimate of last iteration into attenuation maps whose respiratory phases match with those of the respiratory-gated PET sinograms. These attenuation maps are then used for attenuation compensation in the reconstruction of a new set of respiratory-gated PET images, which are subsequently input into a group-wise B-spline non-rigid image registration method for the update of the respiratory motion estimation. The above iterative process is terminated if the respiratory motion estimate of the current iteration satisfies a certain criteria. The second step of our method is to model the respiratory motion estimate in the system matrix of a 4D iterative maximum likelihood (ML) expecation maximization (EM) image reconstruction algorithm which employs all frames of the 4D respiratory-gated PET sinograms to reconstruct a three-dimensional (3D) PET image with respiratory motion compensation. We developed an acceleration scheme for the 4D ML-EM algorithm by employing the ordered subsets (OS) of the PET data to update the image estimation, which we call the 4D OS-EM image reconstruction algorithm.;We evaluated the above 4D PET image reconstruction method with a computer simulated respiratory-gated PET dataset obtained using the realistic 4D eXtended NURBS-based CArdiac-Torso (XCAT) phantom and the Monte-Carlo simulation methodology. Results showed that our method achieved an average of 2--3mm accuracy for estimating the respiratory motion of the lung and liver regions. The quality of the PET images reconstructed by our method, in terms of small lesion detectability, was significantly better than that of the images reconstructed by the conventional 3D ML-EM image reconstruction method without respiratory motion compensation. We also evaluated our respiratory motion estimation method in a preliminary study using patient data and observed similar improvements for the PET image quality.;In conclusion, we developed an accurate and practical 4D PET image reconstruction method with respiratory motion compensation for oncological 18F-FDG PET imaging. |
