Search

Breast focal asymmetry is an important sign of early breast cancer and might be a developing masses or other underlying breast cancer. However, breast focal asymmetry is very difficult to recognise, even for the specialists of radiologists because focal asymmetries lack the boundary and consistent shape of masses and are similar to normal fibroglandular tissues. The computer-aided detection (CADe) for focal asymmetry is a novel research topic in the field of computer-aided detection of breast cancer. The detection of focal asymmetry has the significant value to save women’s live in the early stage of breast cancer. At present, there is only one existing dedicated detection solution for focal asymmetry, which was proposed in 2008. The aim of this research is to develop a novel dedicated computer-aided detection solution for focal asymmetry and overcome the weakness of the existing approach. The objectives of this research are to propose a novel solution for detecting focal asymmetry, a novel breast segmentation approach, a novel pectoral muscle detection approach and a dedicated focal asymmetry detection approach. The literature review of previous researches for focal asymmetry, asymmetric breast findings, breast segmentation and pectoral muscle detection have been extensively carried out. Based on the literature review, a range of hypotheses have been proposed for breast segmentation, pectoral muscle detection and focal asymmetry detection. Furthermore, a novel framework and three novel approaches have been proposed and developed in this research. Firstly, the automatic breast segmentation (ABS) approach was proposed to segment the skin-line of breast from mammographic images and preserve the nipple at the same time, a task which very few breast segmentation methods claimed. The experimental results showed that the ABS approach can adequately segment the breast skin-line and preserve the nipple if it is in profile. The proposed ABS approach is one of three existing breast segmentation approaches that can segment breast skin-line and preserve the nipple at the same time. Secondly, a novel approach for detecting pectoral muscle has been proposed to detect and remove pectoral muscle from the segmented breast areas. The experimental results showed that the proposed maximum intensity change (MIC) approach can detect pectoral muscle with high quality. Thirdly, the novel approach for detecting focal asymmetry has been proposed in this research and various features have been extracted to classify a suspicious region into the category of focal asymmetry or non-focal asymmetry. The experimental results indicated that the proposed detection approach can detect focal asymmetry with the 81.8% sensitivity and the 0.333 false positive regions per image. This research has proposed a novel framework and three novel approaches to build the computer-aided detection solution for focal asymmetry. This research overcomes the weakness of current approach for detecting focal asymmetry. The proposed detection approach is the second dedicated solution for detecting breast focal asymmetry in existence.

Abstract: In this thesis, we study the mathematical modelling of some problems that involve surface effects. These include an optical biosensor, which uses optical principles qualitatively to convert chemical and biochemical concentrations into electrical signals. A typical sensor of this type was constructed in Badley et al., [6], and Jones et al., [18],but diffusion was considered in only one direction in [18] to simulate the reaction between the antigen and the antibody. For realistic applications, we propose the biosensor model in R3. Our theoretical approach is explicitly presented since it is simple and directly applicable to the numerical part of the thesis. In particular, we present existence and uniqueness results based on Maximum Principle and weak solution arguments. These ideas are later applied to systems and to the numerical analysis of the approximate discretized problems.It should be noted that without one dimensional symmetry, the equations can not be decoupled in order to reduce the problem to a single equation. We also show the long time monotonic convergence to the steady state. Next, a finite volume method is applied to the equations, and we obtain existence and uniqueness for the approximate solution as well as the convergence of the the first order temporal norm and the L2 spatial norm. We illustrate the results via some numerical simulations. Finally we consider a mathematically related system motivated by lagoon ecology. We show that under suitable conditions on the coe±cients, the system has a periodic solution under harvesting conditions. The mathematical techniques now depend on estimates for periodic parabolic problems.

Recently, social multimedia-sharing websites, which allow users to upload, annotate, and share online photo or video collections, have become increasingly popular. The user tags or annotations constitute the new multimedia "meta-data". We present an image search system that exploits both image textual and visual information. First, we use focused crawling and DOM Tree based web data extraction methods to extract image textual features from social networking image collections. Second, we propose the concept of visual words to handle the image's visual content for fast indexing and searching. We also develop several user friendly search options to allow users to query the index using words and image feature descriptions (visual words). The developed image search system tries to bridge the gap between the scalable industrial image search engines, which are based on keyword search, and the slower content based image retrieval systems developed mostly in the academic field and designed to search based on image content only. We have implemented a working prototype by crawling and indexing over 16,056 images from flickr.com, one of the most popular image sharing websites. Our experimental results on a working prototype confirm the efficiency and effectiveness of the methods, that we proposed.