Abstract
Nuclear Medicine (NM) images quantify the spatial variations in concentration of an injected or inhaled radiopharmaceutical. Such variations can indicate the presence of disease well before it is discernible in other imaging modalities such as X-ray Computed Tomography (CT). For that reason NM is widely used despite the drawbacks of slow acquisition and faint, noisy and blurred images. The blurring in NM images is in part due to the inevitable breathing of the patient that takes place during the 3 to 15 minute acquisition periods. Breathing motion can lead to substantial errors in quantitation and hinder the accuracy with which lesions may be located. Blurring due to breathing motion becomes increasingly significant as instrument resolution improves. The non-invasive but robust approach to reducing breathing motion artefact presented in this thesis is to monitor patient breathing and relate the configuration of the external abdominal-thoracic surface to parameters that describe the configuration of the internal organs. These parameters are then used to reconstruct, register and then combine a set of respiratory-gated PET images in order to improve signal to noise ratio whilst compensating for breathing motion. The novel procedure presented here, termed Virtual Dissection (VD), requires the acquisition of a breathing cycle of 4D CT images which provides a 4D map of the internal organ and external surface configuration. For each frame in the cycle, organs are segmented and then individually registered to frame 1 in the cycle. Per-organ deformations are then merged to create a global deformation field for each frame. The motion fields are then used to realign the gated PET images to give the motion-corrected NM image. VD uses two fundamental image processing procedures: segmentation and registration. A data-specific segmentation algorithm was developed to perform a cycle-wise organ classification. Three independent registration methods, representing a range of registration characteristics were used, one of which was developed specifically for this work. The registration methods were chosen to demonstrate that VD enables the integration of different registration techniques and parameterisations to permit a choice of registration methods to be made on a per-organ basis in order to reach the optimal choice for the completed images. These developments provide a path, for a subset of organs, from the breathing cycle of CT images through VD to corrected CT images. These techniques have been validated using both simulated and patient data. Visual inspection of lesions in simulated PET images show a reduction in blurring and ten frame animations of CT and PET images show significant breathing motion compensation is achieved by the VD pipeline. Objective measures confirm these findings.