Digital pathology, mathematical modelling and algorithm development to enhance tissue research in cancer diagnostics

• Main Author: McCavigan, A.
• Published: Queen's University Belfast 2012
• Subjects: 616.994075
• Online Access: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.557956

This thesis combines the disciplines of mathematical modelling, image analysis and digital pathology to develop new •• 1proaches which can be used to enhance tissue-based research and pathological diagnosis. The development of Tissue Microarray (TMA) technology has paved the way for simultaneous analysis of hundreds of tissue samples in a single study which have all been subject to the same experimental conditions. Advances in virtual microscopy mean that it is now possible to access quantities of highly informative image data and analyse these using quantitative digital imaging tools. This thesis proposes a novel algorithmic technique to identify, locate and associate tissue cores on virtual TMAs with their corresponding patient metadata. This uses a combination of image search strategies together with mapping logic. Results show 100% accuracy of this technique in defining the positional information of 2860 tissue cores. Identification of tumour areas automatically would underpin the automated analysis of TMAs and whole slide scans in pathology. A Markov Random Field (MRF) model was adopted as a possible candidate for image pattern classification .id applied to a set of non-small cell lung cancer (NSCLC) images to automatically identify tumour regions. Results show a mean sensitivity for classification of tumour pixels ranging from 85.03% to 91.09%. Pathological processes and the development of cancer is a dynamic process that is expressed in the form of images which are routinely interpreted by pathologists. As part of this work, discrete and continuous models were developed that examine the dynamics of nuclear distribution in squamous epithelial layers at various Cervical Intraepithelial Neoplasia (ClN) grades. The computational models presented in this thesis provide a set of tools that can be used to assist and eliminate the inherent subjectivity in diagnostic interpretation of NSCLC and cervical cancer. This work is also significant for the future of biomarker discovery, given that it presents a methodology for the identification of hundreds of tissues cores in a single study, and associates these with their corresponding patient information, combined with a technique to automatically segment tumour regions within which biomarker expression levels can be quantified.