Search

Medical imaging is a cornerstone of modern healthcare. The ability to acquire images from inside a patient has revolutionised the way doctors diagnose and treat diseases, with almost all clinical pipelines now involving imaging to some degree. The development of these imaging methods has led to the field of medical image computing, where a multitude of tools and techniques have been proposed to aid clinicians and researchers in interpreting and analysing these images. One such family of techniques involves generating synthetic medical images. Image synthesis techniques are wide and varied, ranging from basic phantoms, to disease atlases, to high resolution photo-realistic subject-specific images. Their applications are similarly diverse: for developing novel acquisition protocols, training and testing algorithms, visualising changes in disease, predicting particular image types from others, and improving image quality and resolution. This thesis examines the use of medical image synthesis with a particular focus on applications in neurodegenerative diseases. A method for synthesising subject-specific non-pathological images from pathological images is first proposed and used for the unsupervised brain lesion segmentation. We next show how generative adversarial networks (GANs) can be used to both analyse the structural changes seen in patients with Alzheimer's disease, and to add or remove these changes from patient images to produce a subject-specific prediction of disease progression. Finally we investigate how GAN-derived synthetic data can be used to increase the size of training datasets, and under what conditions this additional data can lead to an improvement across a variety of segmentation tasks. Within this context we explore two situations: where only a small amount of labelled data is available, and where a large amount of unlabelled data is also available. We show that the proposed methods can lead to significant improvements in segmentation results, especially when a small amount of labelled data is available.

Open channel intake structures occur in many different applications such as navigable water courses, irrigation, power station facilities for cooling or steam production and run-of-the-river hydropower facilities. All these applications suffer from problems such as energy losses due to asymmetric flow patterns and the commonly associated problem of sedimentation. The flow structure and the associated energy losses at a generalised open channel intake structure are investigated using three-dimensional numerical and physical modelling techniques. Intake angles of 15, 30 and 45 degrees with flow split ratios (intake channel flow / main channel flow) ranging from 0.5 to 3.0 and Reynolds numbers ranging from 1x105 to 3x105 are investigated. A predictive hypothesis for the position and shape of the dividing streamplane is presented and tested with reasonable correlation. The hypothesis can be applied to channel width ratios of 0.5 (intake channel / main channel) with a plane, smooth, rectangular cross sectional area. The flow structure is also described using flow visualisation techniques in the physical model and streamline plotting, velocity vector and velocity profile graphics. A study is also undertaken to describe the turbulent kinetic energy in the flow domain. Predictions near the bed of the flow domain are poor in all analyses with improved results with depth towards the surface. Areas of secondary circulation cells, flow reversal and stagnation zones are identified by numerical and physical models. Validation of the numerical code is undertaken using three dimensional velocity and root mean square velocity data collected using the Acoustic Doppler Velocimeter (ADV). Calibration of the ADV is undertaken and the results closely match the manufacturers stated calibration data. The numerical code used is a finite volume method solving the Reynolds Averaged Navier Stokes equations and the k-s turbulence model. Validation of the code indicated that it was suitable for the investigation of the flow structure in this instance but failed to predict completely areas of high velocity gradients or flow reversal.

The work in this thesis was undertaken to examine clinically significant questions at a time of evolving Left Ventricular Assist Device (LVAD) technology. The main theme of the research was to increase our understanding of the interaction between left ventricle and newer continuous flow LVADs through both in-vitro models and clinical studies with the ultimate aim of improving patient care. The hydrodynamic performance of the two continuous flow LVADs (Thoratec HeartMate II and HeartWare HVAD) was compared in a normothermic, human blood-filled pulsatile mock circulation model under conditions of steady flow and under simulated physiological conditions. These experiments were repeated using dextrose in order to determine its suitability as a mock circulation blood analogue. This study found that clinically representative pulsatile relationships are better represented as H/Q loops instead of linear plots. This allows for greater understanding of the instantaneous H/Q relationship of LVADs with the native LV. The relationship between LVAD pump speed and exercise capacity was studied in patients receiving support from a continuous flow HM II and examined the influence of residual LV function on this relationship. This study found that exercise time and pkVO2 are sensitive to pump speed in patients with poor LV function receiving HM II support. By contrast, insensitivity of exercise parameters such as pkVO2 and exercise time to changes in pump speed may be a non-invasive indicator for LV functional recovery in patients receiving support with rotary LVADs. Effect of temporarily increasing pump speed on exercise capacity was then studied in a group of patients supported with continuous flow LVADs. The study showed that it was safe and feasible to temporarily increase LVAD speed during exercise and this resulted in a significant increase in peak oxygen consumption. Furthermore, improved exercise capacity was observed in both axial and centrifugal pump devices. Radial strain and radial strain rate values were measured by Speckle Tracking Echocardiography to assess LV function and mechanics. The effect of reducing pump speed on these measurements was also examined. My result showed that the radial strain values are significantly lower than normal range and in keeping with advanced heart failure patients. I found the measurements to be relatively independent of degree of LVAD speed and may potentially reflect underlying LV function. Lastly, I investigated the feasibility of recording acoustic waveforms from a LVAD using an electronic stethoscope and analysing the frequency components of the LVAD acoustic spectrum. After recording the findings in patients with normal pump function, we compared it to patients with pump thrombosis. The acoustic profile is primarily determined by rotation speed and device design. In pump thrombosis cases, there were atypical spectral peaks during pump thrombus episodes. Acoustic monitoring is a simple non-invasive method which may have clinical utility in the diagnosis of pump thrombus episodes in patients supported by a rotary LVAD.