No description available.

This thesis explores possible improvements, using parallel computing, to the PSF-alignment and image subtraction algorithm found in HOTPANTS. In time-domain astronomy the PSF-alignment and image subtraction algorithm OIS is used to identify transient events. hotpants is a software package based on OIS, the software package ISIS, and other subsequent research done to improve OIS. A parallel GPU implementation of the algorithm from HOTPANTS – henceforth known as BACH –was created for this thesis. The goal of this thesis is to answer the questions: “what parts of HOTPANTS are most suited for parallelisation?” and “how does bach perform compared to HOTPANTS and SFFT?”, another PSF-alignment and image subtraction tool. The authors found that the parts most susceptible to parallelisation were the convolution and subtraction steps. However, the subtraction did not display a significant improvement to its sequential counterpart. The other parts of HOTPANTS were deemed too complex to implement in parallel on the GPU. However, some parts could probably either be partly parallelised on the GPU or parallelised usingthe CPU. BACH was always as fast as or faster than HOTPANTS; it was generally 2 times faster, but was up to 4.5 times faster in some test cases. It was also faster than SFFT, but this result was not equivalent to the result presented in [15], which is why the authors of this thesis believe something was wrong with either the installation of SFFT or the hardware used to test it.

No description available.

O presente trabalho tem por objetivo propor um modelo envolvendo subtração digital de imagens obtidas a diferentes níveis de energia do feixe de raios X, para possibilitar a detecção de lesões malignas da mama que, no modo tradicional de realização do exame, seriam totalmente camufladas quando superpostas por tecido de absorção semelhante. A pesquisa tem aplicação mais direcionada às avaliações de imagens referentes aos casos de mamas densas, que apresentam tradicionalmente baixo contraste em função da presença maior de tecido fibroglandular de alta densidade. Para possibilitar essa investigação, a pesquisa trabalha com a geração de imagens através de simulação computacional dos principais tecidos mamários envolvidos adiposo, fibroglandular e o próprio carcinoma. Por esse procedimento, é possível observar o comportamento das variações de níveis de cinza nas imagens mamográficas a partir dos coeficientes de absorção daqueles tecidos, considerados com diferentes espessuras e submetidos a diferentes valores de energia, dentro da faixa típica utilizada no exame mamográfico. Foi considerada para referência do procedimento uma mama comprimida totalizando 4,5 cm de espessura total. Os resultados apontaram basicamente que: (a) se o carcinoma tiver espessura menor que 0,8 cm, aparentemente, com exposição na faixa de 14 a 17 keV e com pequena variação de energia na aquisição da segunda imagem sua visualização é totalmente comprometida quando camuflado por tecido fibroso; (b) se o carcinoma tiver espessura maior que 0,4 cm, possivelmente será detectado, mesmo que camuflado por tecido fibroso, com exposição na faixa de 19 a 25 keV; (c) para carcinomas camuflados, de espessura entre 0,4 e 2,0 cm, considerando diferença maior de energias na aquisição das imagens, a realização do procedimento proposto permitirá destacá-los na imagem resultante da subtração digital entre imagens produzidas por exposições de 14 a 22 keV, representando, portanto, uma nova ferramenta metodológica para possibilitar e identificar lesões malignas que não seriam detectadas no exame típico, sobretudo em casos de mamas densas. / This work intends to propose a model involving subtraction of digital images obtained at different levels of energy in the X-ray beam, to permit the detection of malign lesions of the breast that in the traditional way of performing the examination, would be completely hidden when overlapped by tissue of similar absorption. The research has more directed application to the evaluations of referring images to the cases of dense breasts that traditionally present low contrast in function of the tissue presence biggest to fibrousglandular of high density. In order to make possible this investigation, the research works with the generation of images by computational simulation of main involved mammary tissues - adipose, fibrousglandular and the proper carcinoma. For this procedure, it is possible to observe the behavior of the variations of gray levels in the mammographic images from the coefficients of absorption of those tissues, considered with different thicknesses and submitted to different values of energy, inside of the used typical band in the mammographic examination. A compressed breast was considered for reference of the procedure totalizing 4,5 cm of total thickness. The results had pointed basically that: (a) if the carcinoma will have lesser thickness that 0,8 cm, apparently, with exposition in the band of 14 to 17 keV and with small variation of the energy in the second image acquisition, its visualization is quite damaged when masked for fibrous tissue; (b) if the carcinoma will have bigger thickness that 0,4 cm, will possibly be detected, even masked for fibrous tissue, with exposition in the band of 19 to 25 keV; (c) for masked carcinomas, with thickness between 0,4 and 2,0 cm, considering larger difference of energies in the acquisition of the images, the accomplishment of the proposed procedure will allow to highlight them in the resultant image of the digital subtraction among images produced by expositions of 14 to 22 keV, representing, therefore, a new methodological tool to make possible and identify malign lesions that would not be detected in the typical examination, especially in cases of dense breasts.

O presente trabalho tem por objetivo propor um modelo envolvendo subtração digital de imagens obtidas a diferentes níveis de energia do feixe de raios X, para possibilitar a detecção de lesões malignas da mama que, no modo tradicional de realização do exame, seriam totalmente camufladas quando superpostas por tecido de absorção semelhante. A pesquisa tem aplicação mais direcionada às avaliações de imagens referentes aos casos de mamas densas, que apresentam tradicionalmente baixo contraste em função da presença maior de tecido fibroglandular de alta densidade. Para possibilitar essa investigação, a pesquisa trabalha com a geração de imagens através de simulação computacional dos principais tecidos mamários envolvidos adiposo, fibroglandular e o próprio carcinoma. Por esse procedimento, é possível observar o comportamento das variações de níveis de cinza nas imagens mamográficas a partir dos coeficientes de absorção daqueles tecidos, considerados com diferentes espessuras e submetidos a diferentes valores de energia, dentro da faixa típica utilizada no exame mamográfico. Foi considerada para referência do procedimento uma mama comprimida totalizando 4,5 cm de espessura total. Os resultados apontaram basicamente que: (a) se o carcinoma tiver espessura menor que 0,8 cm, aparentemente, com exposição na faixa de 14 a 17 keV e com pequena variação de energia na aquisição da segunda imagem sua visualização é totalmente comprometida quando camuflado por tecido fibroso; (b) se o carcinoma tiver espessura maior que 0,4 cm, possivelmente será detectado, mesmo que camuflado por tecido fibroso, com exposição na faixa de 19 a 25 keV; (c) para carcinomas camuflados, de espessura entre 0,4 e 2,0 cm, considerando diferença maior de energias na aquisição das imagens, a realização do procedimento proposto permitirá destacá-los na imagem resultante da subtração digital entre imagens produzidas por exposições de 14 a 22 keV, representando, portanto, uma nova ferramenta metodológica para possibilitar e identificar lesões malignas que não seriam detectadas no exame típico, sobretudo em casos de mamas densas. / This work intends to propose a model involving subtraction of digital images obtained at different levels of energy in the X-ray beam, to permit the detection of malign lesions of the breast that in the traditional way of performing the examination, would be completely hidden when overlapped by tissue of similar absorption. The research has more directed application to the evaluations of referring images to the cases of dense breasts that traditionally present low contrast in function of the tissue presence biggest to fibrousglandular of high density. In order to make possible this investigation, the research works with the generation of images by computational simulation of main involved mammary tissues - adipose, fibrousglandular and the proper carcinoma. For this procedure, it is possible to observe the behavior of the variations of gray levels in the mammographic images from the coefficients of absorption of those tissues, considered with different thicknesses and submitted to different values of energy, inside of the used typical band in the mammographic examination. A compressed breast was considered for reference of the procedure totalizing 4,5 cm of total thickness. The results had pointed basically that: (a) if the carcinoma will have lesser thickness that 0,8 cm, apparently, with exposition in the band of 14 to 17 keV and with small variation of the energy in the second image acquisition, its visualization is quite damaged when masked for fibrous tissue; (b) if the carcinoma will have bigger thickness that 0,4 cm, will possibly be detected, even masked for fibrous tissue, with exposition in the band of 19 to 25 keV; (c) for masked carcinomas, with thickness between 0,4 and 2,0 cm, considering larger difference of energies in the acquisition of the images, the accomplishment of the proposed procedure will allow to highlight them in the resultant image of the digital subtraction among images produced by expositions of 14 to 22 keV, representing, therefore, a new methodological tool to make possible and identify malign lesions that would not be detected in the typical examination, especially in cases of dense breasts.

Cette thèse présente plusieurs contributions au software developé pour le traitement d’images dans le cadre du LSST. Notre objectif est d'utiliser le code et les algorithmes LSST existants, afin de créer un pipeline dédié à la détection des supernovae de type Ia. Pour la détection des supernovae nous utilisons une technique appelée soustraction optimale d'images qui implique la construction de coadditions. Nous étudions aussi le comportement des différents objets dans le temps et construisons des courbes de lumière qui représentent leur cycle de vie en fonction de l'intensité lumineuse de chaque détection sur plusieurs nuits. Enfin, pour analyser un nombre excessif de candidats, nous utilisons des algorithmes d'apprentissage machine.Notre première contribution concerne le développement des taches de coaddition automatisée adaptées pour construire des images de référence et de science avec un haut rapport signal-sur-bruit. La contribution suivante est lié à l’addition de mesures et l’étude de résidus des images d’analyse de différence, y-compris la sélection des seuils adaptés et l'étiquetage basée sur les valeurs quantitativess des résidus pour identifier les mauvaises détections, les artéfacts et les flux réellement significatifs. Notre suivante contribution est un algorithme pour sélectionner et générer les courbes de lumière candidates. Finalement, on applique une classification machine learning pour trouver des type Ia supernovae en utilisant la méthode random forest. Ces résultats ont permis l’identification des supernovae de type Ia simulées et réelles parmis les candidats avec une haute précision. / This thesis will present several contributions to the software developed for the LSST telescope with the purpose of contributing to the detection of type Ia supernovae. Our objective is to use the existing LSST code and algorithms, in order to create a type Ia supernovae detection dedicated pipeline.Since detecting supernovae requires a special type of processing, we use a technique known as the Optimal Image Subtraction which implies the construction of coadditions. Afterwards, we study the behavior of the different objects through time and build light curves that represent their life cycle in terms of the light intensity of each detection on several nights. Lastly, in order to analyze an excessive number of candidates, we employ machine learning algorithms to identify what curves are more probable to be type Ia supernovae. Our first contribution concerns the development of adapted and automatized coaddition tasks for building high signal-to-noise reference and science images. The next contribution is related to the addition of measurements and study of the residuals on difference image analysis, including the selection with adapted thresholding and the assignation of labels. We also propose, as contributions, an algorithm to select and generate the different candidate light curves through the selection of objects with recurrent detections through time and in the different bandpasses. Finally, we apply the machine learning classification approach to find type Ia supernovae by means of using a random forest classifier and based strictly on geometrical features that are present in the light curves.

No description available.