Rigid Motion Correction for Positron Emission Tomography Brain Imaging

During a positron emission tomography (PET) scan the subject is required to be completely stationary for the duration of the scan. Any motion of the subject will translate into motion blur in the final reconstructed image and may reduce the diagnostic value of the image. Motion during a PET scan can be classified into two main categories: non-rigid and rigid motion. Non-rigid motion deforms the subject and is primarily caused by motion in the abdomen and thorax due to the cardiac and respiration cycles. Rigid motion is where the subject moves without any deformation, and occurs primarily during brain imaging when there is motion of the head. Non-rigid motion is largely periodic and there exist a number of techniques which attempt to handle it. Rigid motion of the head, however, is not periodic and cannot be compensated for with the same techniques. It is usually mitigated by using a head restraint, although with the high resolution of modern scanners this sometimes proves insufficient, or a sedative or anaesthesia, but these may carry significant risks to the patient.

In preclinical PET studies, an anaesthetic is usually used to ensure that the animal remains motionless for the duration of the scan (this of course does not solve the problem of motion due to the respiratory and cardiac cycles). However, the anaesthetic used may have a confounding effect on the PET tracer under study, and these effects are usually not well understood. This is especially problematic for translational studies since anaesthetics are usually avoided in the clinic and thus the findings of the preclinical studies may not be directly applicable.

The problem of head motion in the clinical and preclinical setting can be solved by tracking the motion of the head and correcting the PET data after acquisition for the recorded motion. The work presented in this thesis focusses entirely on developing and implementing such a rigid motion correction technique for brain imaging. The technique we implemented corrects each individual recorded event to where it would have been detected if the subject had not moved.

For preclinical studies, a motion correction technique was implemented which utilises a stereo-optical system to track a marker attached to the head of a rat. The experimental protocol and parameters, as well as the data processing, were optimised. The technique was applied in a proof-of-principle investigation into the effect of the anaesthetic isoflurane on the uptake of 18F-FDG (a very common tracer used for PET imaging) in the rat brain. Scanning of fully conscious and unrestrained rats was performed for more than an hour from the moment of injection, which had not been reported on before. A significant effect of the isoflurane was observed, which confirmed other studies investigating isoflurane using different approaches.

A motion correction protocol was also implemented in the clinical setting for four different PET scanners. In addition to proof-of-principle studies, the technique was also used in a clinical study involving long brain scans of dementia patients, who were prone to move.

In addition, an algorithmic technique of the form known as spatially variant resolution modelling was developed to improve the reconstruction, which could be used in conjunction with motion correction. Resolution modelling aims to improve the reconstruction by incorporating a model of the scanner uncertainties. To be accurate the model must account for variation in the uncertainties within the scanner field-of-view. Common techniques apply the model in image- or sinogram-space, and thus do not allow for the use of motion correction. The technique developed applies the model to the list-mode data directly, before motion correction, thereby correctly handling the spatial variance.