| Summary: | <p>Cancer is one of the main causes of premature death worldwide, leading to almost one in three deaths in the UK. Medical imaging plays a fundamental role in the diagnosis and treatment of cancer patients. This thesis focuses on the assessment of early stage clinical trials of novel cancer treatment, which is performed based on dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) and perfusion computed tomography (pCT). From a clinical viewpoint, the objectives are tumour microenvironment characterisation, which involves the assessment of tumour perfusion and vascularity, and early prediction of response to treatment. From a methodological viewpoint, this thesis is at the interface between pharmacokinetic (PK) modelling and image registration for dynamic images. Major challenges addressed in this work are PK model selection and arterial input function derivation, as well as intrasequence motion correction for dynamic imaging data. Registration of dynamic imaging data is a particularly challenging problem due to the variety of motion types that appear in these data: bulk patient motion, periodic motion due to breathing, and random motion of small features due to peristalsis. Moreover, contrast-enhanced modalities such as DCE-MRI and pCT pose additional challenges as image intensities change due to contrast inflow. The key contributions of this thesis are the derivation of patient-specific arterial input functions from pCT data, which are employed for performing PK modelling of contemporaneous DCE-MRI scans of the same patients, and the development of an MRF-based discrete optimisation framework for the nonrigid registration of dynamic sequences. While the former was applied on DCE-MRI and pCT images from an early stage trial of new treatment for colorectal cancer, the latter was employed for the motion correction of DCE-MRI images from the same trial. Additionally, this new registration framework was applied to dynamic CT images of the lung to demonstrate the advantage of temporal regularisation in an application where there is periodic motion (i.e. breathing). Experimental validation shows that the proposed registration framework improves over state-of-the-art methods based on continuous optimisation in terms of registration accuracy and computational complexity.</p>
|
|---|