Search

Malignant pleural mesothelioma (MPM) has a rind-like growth pattern and may have both solid disease and pleural fluid, so it is very challenging to accurately categorize change in tumour burden using size-based tumour response criteria, eg. modified Response Evaluation Criteria in Solid Tumours (RECIST). Diffusion Weighted Magnetic Resonance Imaging (DWI) provides functional information of tissue microstructure and is a promising technique to enable segmentation of tumours due to its excellent contrast between tumour and suppressed background. The aim of this thesis is to evaluate the changes in tumour volume and functions during treatment thus assess disease treatment response using DWI. A novel method, T2-adjusted computed diffusion weighted magnetic resonance imaging (T2-cDWI), is proposed to improve the image contrast between solid tumour and pleural effusion, and may eliminate T2 shine-through effect. In T2-cDWI, T2-weighted echo-planar images at multiple (>2) echo times were acquired in addition to the standard DWI acquisition protocol. To acquire the solid tumour volume, a workflow including image segmentation and classification has been investigated on DWI images and further applied in a clinical mesothelioma study for treatment response evaluation. DWI imaging protocols were developed to dovetail with a routine clinically chest scan for treatment evaluation. Using previously developed workflow, volumetric tumour burden and DWI parameter values are obtained and shown to be correlated with disease progression after treatment. These results clearly show the potential for the developed tools as a novel and quantifiable biomarker for assessing the volumetric response of MPm to therapy compared with the conventional modified RECIST. In summary, this thesis has investigated and provided a new volumetric response assessment strategy for MPM patients using DWI images.

In recent years, topic group mining and massive crowd opinion analysis from on-line social network platforms have become some of the most important tasks not only in research area but also in industry. Systems of this sort can identify similar topics from a very large dataset, group them together based on the topic, and analyse the inclination of the content's owner. To solve this problem, which involves research from a number of different areas, an integrated platform needs to be proposed. Most community mining techniques treat the network as a graph where nodes represent users and edges reflect user relationship between two users. One obvious drawback of these approaches is that it can only utilise the explicit user relationships provided by on-line social network platforms. All other possible relationships will be ignored. Some on-line social network platforms restrict the length of content a user can publish. This causes traditional document clustering methods to perform poorly. Meanwhile, the restriction of content length also affects opinion mining performance since most content lacks contextual features. Hence, other context features that are not immediately or obviously related need to be investigated to improve performance in user inclination classification. This research proposes a novel three layered platform. Two core technologies of the platform are topic group mining and user inclination analysis. The integrated approach was evaluated by a series of experiments to examine each core technology. The results indicate that the proposed integrated platform is able to produce the following results. 1) Scores up to 0.82 by V-measure evaluation function in topic group mining. 2) High accuracy rate in inclination mining. 3) A flexible and adaptable platform design which can accommodate different on-line social networks easily.