No description available.

This thesis presents an investigation into Gyrodactylus species recognition, making use of machine learning classification and feature selection techniques, and explores image feature extraction to demonstrate proof of concept for an envisaged rapid, consistent and secure initial identification of pathogens by field workers and non-expert users. The design of the proposed cognitively inspired framework is able to provide confident discrimination recognition from its non-pathogenic congeners, which is sought in order to assist diagnostics during periods of a suspected outbreak. Accurate identification of pathogens is a key to their control in an aquaculture context and the monogenean worm genus Gyrodactylus provides an ideal test-bed for the selected techniques. In the proposed algorithm, the concept of classification using a single model is extended to include more than one model. In classifying multiple species of Gyrodactylus, experiments using 557 specimens of nine different species, two classifiers and three feature sets were performed. To combine these models, an ensemble based majority voting approach has been adopted. Experimental results with a database of Gyrodactylus species show the superior performance of the ensemble system. Comparison with single classification approaches indicates that the proposed framework produces a marked improvement in classification performance. The second contribution of this thesis is the exploration of image processing techniques. Active Shape Model (ASM) and Complex Network methods are applied to images of the attachment hooks of several species of Gyrodactylus to classify each species according to their true species type. ASM is used to provide landmark points to segment the contour of the image, while the Complex Network model is used to extract the information from the contour of an image. The current system aims to confidently classify species, which is notifiable pathogen of Atlantic salmon, to their true class with high degree of accuracy. Finally, some concluding remarks are made along with proposal for future work.

Dynamic contrast enhanced magnetic resonance imaging (DCE-MRI) of the breast is a widely used non-invasive approach to gather information about the underlying physiology of breast tumors. Recent studies indicate that breast tumor heterogeneity may reflect the presence of different levels of cellular aggressiveness or habitats within the tumor. This heterogeneity has been correlated to the variations in the contrast enhancement patterns within the tumor apparent on gadolinium-enhanced DCE-MRI. Although pathological and qualitative (based on contrast enhancement patterns) studies suggest the presence of clini- cal and molecular predictive tumor sub-regions, this has not been fully investigated in the quantitative domain. The new era of cancer imaging emphasizes the use of Radiomics to provide in vivo quan- titative prognostic and predictive imaging biomarkers. Thus Radiomics focuses on apply- ing image analysis techniques to quantify tumor radiographic properties to create mineable databases from radiological images. In this research work, the Radiomics approach was ap- plied to develop a novel computer aided diagnosis (CAD) model for quantifying intratumor heterogeneity not only within the tumor as a whole, but also within tumor habitats with an intent to build predictive models in breast cancer. The process of building these predictive models started with 2-D tumor segmentation followed by habitat extraction (based on vari- ations in contrast patterns and geometry) and textural kinetic feature extraction to quantify habitat heterogeneity. A new correlation based random subspace ensemble framework was developed to evaluate the textural kinetics from the individual tumor habitats. This new CAD framework was applied to predict two clinical and prognostic factors: Axillary lymph node (ALN) metastases and Estrogen receptor (ER) status. An AUC of more than 0.8 was achieved for classifying breast tumors based on number of ALN involvement. The highest AUC of 0.91 was achieved for classifying tumors with no ALN metastases from tumors with 4 or more ALN metastases. For classifying tumors based on ER status the highest AUC of 0.87 was achieved. These results were acquired by utilizing the textural kinetic features from the tumor habitat with rapid delayed washout. The results presented in this work showed that the heterogeneity within the tumor habitats which showed rapid contrast washout in the delayed phase, correlated with aggressive cellular phenotypes. This work hypothesizes that successfully quantifying these prognostic factors will prove to be clinically significant as it can improve the diagnostic accuracy. This, in turn, will im- prove the breast cancer treatment paradigm by providing more tailored treatment regimens for aggressive tumors.

Random Forest (RF) is an ensemble classification technique that was developed by Leo Breiman over a decade ago. Compared with other ensemble techniques, it has proved its accuracy and superiority. Many researchers, however, believe that there is still room for optimizing RF further by enhancing and improving its performance accuracy. This explains why there have been many extensions of RF where each extension employed a variety of techniques and strategies to improve certain aspect(s) of RF. The main focus of this dissertation is to develop new extensions of RF using new optimization techniques that, to the best of our knowledge, have never been used before to optimize RF. These techniques are clustering, the local outlier factor, diversified weighted subspaces, and replicator dynamics. Applying these techniques on RF produced four extensions which we have termed CLUB-DRF, LOFB-DRF, DSB-RF, and RDB-DR respectively. Experimental studies on 15 real datasets showed favorable results, demonstrating the potential of the proposed methods. Performance-wise, CLUB-DRF is ranked first in terms of accuracy and classifcation speed making it ideal for real-time applications, and for machines/devices with limited memory and processing power.

Orientador: Prof. Dr. Alexandre Noma / Dissertação (mestrado) - Universidade Federal do ABC, Programa de Pós-Graduação em Ciência da Computação, 2016. / A classificação supervisionada consiste em utilizar uma base de dados rotulada para avaliar o desempenho de um determinado classifcador. Mensurando tal desempenho, podemos inferir se, para o problema abordado, tal classifcador poderá ser empregado ou não. Métodos classicos de classificação utilizam um unico classifcador para a analise de um problema. Uma forma de melhorar o desempenho da classificação é empregar técnicas que misturam classifcadores, sejam com base em seus resultados ou nas caracteristicas intrinsecas que cada classicador possui. Neste trabalho, foram empregados os métodos Votação e Adaboost para combinar classifcadores e utilizando base de dados rotuladas provenientes de imagens satelitais extraídas da regi~ao da Amazonia Legal para classificar a cobertura do solo. Resultados obtidos mostraram que o algoritmo SVM por si so consegue resultados de classificação em torno dos 90% em casos gerais. Para casos especifios, a empregabilidade do Adaboost resultou em um acrescimo de, aproximadamente, 10% na taxa de acurácia para um tipo de classe em comparação o com o melhor resultado dos métodos tradicionais. / Supervised classification is based on using a labeled database to evaluate a given classifer's performance. Measuring such performance, it is possible to infer if, for the problem addressed, such a classifer can be employed or not. Classical classification methods use a single classier to analyze a problem. One way to improve classifcation's performance is to employ techniques that mix classifers, based on their results or by each classifer's intrinsic characteristics. In this paper, the methods Voting and Adaboost were used to combine classifers and using labeled data bases from satellite's images extracted from the Legal Amazon region to classify the soil cover. Results obtained showed that the SVM algorithm alone achieves classifcation results around 90 % in general cases. For specific cases, the employability of Adaboost resulted in an increase of approximately 10 % in the accuracy rate for a class type compared to the best result of the traditional methods.

No description available.