Abstract
Positron emission tomography (PET) is a modality with high temporal resolution but long acquisition times. This can result in blurred images due to subject motion. Respiratory motion in particular is an unavoidable source of degradation, which can cause issues with quantification and clinical interpretation. An important characteristic of respiratory motion is its pseudo-cyclic nature, which has previously been exploited to form mathematical models which describe the motion, driven by a small number of parameters. The aim of this project is to use this form of motion modelling to estimate motion using information acquired from both dynamic magnetic resonance (MR) scans and from the acquired PET data itself to correct for the effects of motion. The use of motion models in this way can overcome the high levels of noise which otherwise characterise the estimation problem.First the feasibility of using motion models is investigated using synthetic data consisting of individual PET gates simulated using real motion information. The PET gates are registered using constraints provided by a motion model derived from MR images. A novelty of this approach is that this is the first time PET data have been used to indirectly drive a parameterised motion model.
The next part of the project attempts to formalise the motion estimation process by incorporating the reduced-parameter motion model into the PET image reconstruction. An analytical gradient for a single motion parameter that drives the model is derived from the same objective function used to estimate the image. This results in significant noise averaging, providing robustness to the high level of noise typically found in PET data acquired over short time frames. This is shown to improve robustness to noise well enough that the number of gates can be increased and overall motion correction performance improved.
Finally, the formulation is extended further to also model photon attenuation effects. This is shown to improve the performance of the algorithm when dealing with synthetic data that includes attenuation.
Date of Award | 2017 |
---|---|
Original language | English |
Awarding Institution |
|
Supervisor | Andrew King (Supervisor) & Paul Marsden (Supervisor) |