Digital breast tomosynthesis is a new technology that provides three-dimensional information of the breast and makes it possible to distinguish the cancer from overlying breast tissues. We are dedicated to optimizing image reconstruction and imaging configuration for a new multi-beam parallel digital breast tomosynthesis prototype system. Several commonly used algorithms from the typical image reconstruction models which were used for iso-centric tomosynthesis systems were investigated for our multi-beam parallel tomosynthesis imaging system. The representative algorithms, including back-projection (BP), filtered back-projection (FBP), matrix inversion tomosynthesis reconstruction (MITS), maximum likelihood expectation maximization (MLEM), ordered-subset maximum likelihood expectation maximization (OS-MLEM), simultaneous algebraic reconstruction technique (SART), were implemented to fit our system design. An accelerated MLEM algorithm was proposed, which significantly reduced the running time but had the same image quality. Furthermore, two statistical variants of BP reconstruction were validated for our tomosynthesis prototype system. Experiments based on phantoms and computer simulations show that the prototype system combined with our algorithms is capable of providing three-dimensional information of the objects with good image quality and has great potentials to improve digital breast tomosynthesis technology. Four methodologies were employed to optimize the reconstruction algorithms and different imaging configurations for the prototype system. A linear tomosynthesis imaging analysis tool was used to investigate blurring-out reconstruction algorithms. Computer simulations of sphere and wire objects aimed at the performance of out-of-plane artifact removal. A frequency-domain-based methodology, relative NEQ(f) analysis, was investigated to evaluate the overall system performance based on the propagation of signal and noise. Conclusions were made to determine the optimal image reconstruction algorithm and imaging configuration of this new multi-beam parallel digital breast tomosynthesis prototype system for better image quality and system performance.

The best way to conquer breast cancer is early detection of the disease. Research studies show that earlier detection results in the increase of life span of the affected person. Traditional two-dimensional mammography is the most prevalent method used in detecting breast cancer. Recently, a three-dimensional digital breast tomosynthesis (DBT) system has been created, which is hopeful to surpass the technology of traditional mammography systems. The DBT system can provide three-dimensional information, allowing physicians to reduce the amount of false negative screening in addition to better monitoring of breast cancer and to catch lesions that may be otherwise cancerous. In this research, the View Angle (VA) and number of projection images (N) were investigated and compared with parallel imaging configurations using two reconstruction algorithms, including Back Projection (BP) and Shift-And-Add (SAA). Modulation Transfer Function (MTF) analyses were conducted with both algorithms, in order to determine which method displayed better image qualities to ultimately improve the detection of breast cancer.

A tomossíntese digital mamária (Digital Breast Tomosynthesis - DBT) é um exame radiográfico utilizado para o rastreamento do câncer de mama, que busca superar a limitação da sobreposição de tecidos existente na mamografia digital 2D. Nessa técnica são adquiridas projeções radiográficas em diferentes ângulos, que são processadas para a reconstrução do volume da mama. Um grande desafio é a elaboração dos algoritmos para a reconstrução tomográfica, visto que há um número limitado de projeções adquiridas com baixas doses de radiação, abrangendo uma estreita faixa de ângulo. Outro fator importante é o ruído presente nas imagens, que pode impactar o diagnóstico do câncer pelos radiologistas. Esse trabalho tem como objetivo apresentar uma ferramenta de reconstrução de imagens para DBT e fazer um estudo do comportamento do sinal e do ruído nas imagens reconstruídas. Os métodos analíticos de retroprojeção simples e filtrada, bem como os interativos de máxima verossimilhança e algébricos foram avaliados. A validação dos algoritmos de reconstrução foi feita por meio de métricas objetivas e as imagens reconstruídas foram comparadas com imagens obtidas a partir de um software de reconstrução para DBT desenvolvido pelo Food and Drug Administration (FDA). A partir das análises objetivas e visuais, demonstrou-se o potencial da ferramenta desenvolvida nesse trabalho. O ruído pós-reconstrução foi investigado através da aquisição de imagens de phantoms homogêneos, utilizando dois sistemas comerciais de DBT. As curvas de valor médio de pixel, a variância do ruído e a relação sinal-ruído seguiram o mesmo padrão já demonstrado para as projeções. A análise do espectro de potência do ruído demonstrou que o processo de reconstrução correlaciona o ruído para ambos os equipamentos. / Digital Breast Tomosynthesis (DBT) is a radiographic examination used for breast cancer screening, which seeks to overcome the tissue superposition in 2D digital mammography. In this technique, radiographic projections are acquired at different angles, which are processed for the reconstruction of the breast volume. A major challenge is the elaboration of algorithms for tomographic reconstruction since there are a limited number of projections acquired with low doses of radiation, covering a narrow-angle range. Another important factor is the noise present in this modality that can impact the diagnosis of tumors by radiologists. This work aims to present an image reconstruction toolbox for DBT and study the signal and noise behavior in the reconstructed slices. The analytical methods of simple and filtered back-projection, as well as the maximum likelihood and algebraic iterative methods were evaluated. The validation of the reconstruction algorithms was done by objective metrics and the reconstructed images were compared with the images obtained from a reconstruction software for DBT developed by the Food and Drug Administration (FDA). Through the objective and visual analysis, the potential of the toolbox developed in this work was demonstrated. The noise after reconstruction was investigated by means of the acquisition of homogeneous phantom images, using two commercial DBT systems. The mean pixel value, the noise variance and the signal-to-noise ratio follow the same curve shape already shown for the projection domain. The analysis of noise power spectrum demonstrated that the process of reconstruction correlates the noise for both systems used.

Conventional 2D mammography was the most effective approach to detecting early stage breast cancer in the past decades of years. Tomosynthetic breast imaging is a potentially more valuable 3D technique for breast cancer detection. The limitations of current tomosynthesis systems include a longer scanning time than a conventional digital X-ray modality and a low spatial resolution due to the movement of the single X-ray source. Dr.Otto Zhou's group proposed the concept of stationary digital breast tomosynthesis (s-DBT) using a Carbon Nano-Tube (CNT) based X-ray source array. Instead of mechanically moving a single X-ray tube, s-DBT applies a stationary X-ray source array, which generates X-ray beams from different view angles by electronically activating the individual source prepositioned at the corresponding view angle, therefore eliminating the focal spot motion blurring from sources. The scanning speed is determined only by the detector readout time and the number of sources regardless of the angular coverage spans, such that the blur from patient's motion can be reduced due to the quick scan. S-DBT is potentially a promising modality to improve the early breast cancer detection by providing decent image quality with fast scan and low radiation dose. DBT system acquires a limited number of noisy 2D projections over a limited angular range and then mathematically reconstructs a 3D breast. 3D reconstruction is faced with the challenges of cone-beam and flat-panel geometry, highly incomplete sampling and huge reconstructed volume. In this research, we investigated several representative reconstruction methods such as Filtered backprojection method (FBP), Simultaneous algebraic reconstruction technique (SART) and Maximum likelihood (ML). We also compared our proposed statistical iterative reconstruction (IR) with particular prior and computational technique to these representative methods. Of all available reconstruction methods in this research, our proposed statistical IR appears particularly promising since it provides the flexibility of accurate physical noise modeling and geometric system description. In the following chapters, we present multiple key techniques of statistical IR to tomosynthesis imaging data to demonstrate significant image quality improvement over conventional techniques. These techniques include the physical modeling with a local voxel-pair based prior with the flexibility in its parameters to fine-tune image quality, the pre-computed parameter κ incorporated with the prior to remove the data dependence and to achieve a predictable resolution property, an effective ray-driven technique to compute the forward and backprojection and an over-sampled ray-driven method to perform high resolution reconstruction with a practical region of interest (ROI) technique. In addition, to solve the estimation problem with a fast computation, we also present a semi-quantitative method to optimize the relaxation parameter in a relaxed order-subsets framework and an optimization transfer based algorithm framework which potentially allows less iterations to achieve an acceptable convergence. The phantom data is acquired with the s-DBT prototype system to assess the performance of these particular techniques and compare our proposed method to those representatives. The value of IR is demonstrated in improving the detectability of low contrast and tiny micro-calcification, in reducing cross plane artifacts, in improving resolution and lowering noise in reconstructed images. In particular, noise power spectrum analysis (NPS) indicates a superior noise spectral property of our proposed statistical IR, especially in the high frequency range. With the decent noise property, statistical IR also provides a remarkable reconstruction MTF in general and in different areas within a focus plane. Although computational load remains a significant challenge for practical development, combined with the advancing computational techniques such as graphic computing, the superior image quality provided by statistical IR will be realized to benefit the diagnostics in real clinical applications.

Currently, mammography is the most common imaging technology used in breast screening. Low dose X-rays are passed through the breast to generate images called mammograms. One type of breast abnormality is a cluster of microcalcifications. Usually, in benign cases, microcalcifications result from the death of fat cells or are due to secretion by the lobules. However, in some cases, clusters of microcalcifications are indicative of early breast cancer, partly because of the secretions by cancer cells or the death of such cells. Due to the different attenuation characteristics of normal breast tissue and microcalcifications, the latter ideally appear as bright white spots and this allows detection and analysis for breast cancer classification. Microcalcification detection is one of the primary foci of screening and has led to the development of computer-aided detection (CAD) systems. However, a fundamental limitation of mammography is that it gives a 2D view of the tightly compressed 3D breast. The depths of entities within the breast are lost after this imaging process, even though the breast tissue is spread out as a result of the compression force applied to the breast. The superimposition of tissues can occlude cancers and this has led to the development of digital breast tomosynthesis (DBT). DBT is a three-dimensional imaging involving an X-ray tube moving in an arc around the breast, over a limited angular range, producing multiple images, which further undergo a reconstruction step to form a three-dimensional volume of breast. However, reconstruction remains the subject of research and small microcalcifications are "smeared" in depth by current algorithms, preventing detailed analysis of the geometry of a cluster. By using the geometry of the DBT acquisition system, we derive the "epipolar" trajectory of a microcalcification. As a first application of the epipolars, we develop a clustering algorithm after using the Hough transform to find corresponding points generated from a microcalcification. Noise points can also be isolated. In addition, we show how microcalcification projections can be detected adaptively. Epipolar analysis has also led to a novel detection algorithm for DBT using a Bayesian method, which estimates a maximum a posterior (MAP) labelling in each individual image and subsequently for all projections iteratively. Not only does this algorithm output the binary decision of whether a pixel is a microcalcification, it can predict the approximate depth of the microcalcification in the breast if it is. Based on the epipolar analysis, reconstruction of just a region of interest (ROI) e.g. microcalcification clusters is possible and it is more straightforward than any existing method using reconstruction slices. This potentially enables future classification of breast cancer when more clinical data becomes available.

Bröst består av mjölkkörtlar, subkutant fett och bindväv. Det finns också kärl och lymfa i brösten. Både män och kvinnor har bröst. Olika sjukdomar kan drabba brösten av benigna och maligna slag. Den mest använda undersökningsmetoden för att upptäcka bröstcancer är mammografi. Vid ytterligare undersökning av brösten kan digital bröst-tomosyntes (DBT) förekomma. DBT är en sorts begränsad vinkel-tomografi som producerar bilder på brösten i sektioner. Åsikter om DBT är motstridiga, en del studier säger att tomosyntes är bättre än mammografi gällande sensitivitet och andra säger att det är sämre eller ekvivalent. För att få kunskap om tomosyntes, mammografi och vad som skiljer i sensitivitet krävs det en sammanfattning av olika studier. Syftet med studien är att jämföra sensitivitet vid bröstundersökningar inom mammografi och tomosyntes. Via en systematisk litteraturstudie sammanfattas ett resultat utifrån kvantitativa artiklar som kvalitetsgranskas och analyseras. Arbetet har genomgått en etisk egengranskning. Resultatet skapades via hypotesprövning och SPSS och de påvisar att det finns en signifikant skillnad i sensitivitet mellan DBT och mammografi vilket innebär att DBT har högre sensitivitet sett till medelvärde och median.

O câncer de mama é a doença que mais acomete as mulheres em todo o mundo, sendo o tratamento mais eficaz se for diagnosticada em estágio inicial. A partir de 2011, nos programas de rastreamento de países desenvolvidos, vem sendo empregada uma nova modalidade de exame, a tomossíntese digital mamária (Digital Breast Tomosynthesis - DBT), que possui diversas vantagens se comparada à mamografia digital. No exame, o médico radiologista busca por sinais suspeitos na imagem, como: nódulos, microcalcificações e distorção arquitetural mamária (DAM). Sendo que, este último pode representar o estágio mais inicial de um câncer em formação, podendo se manifestar antes da formação de qualquer outra lesão. No entanto, a DAM é difícil de ser detectada pois modifica o tecido mamário de forma sutil, não havendo qualquer formação de massa ou a borda definida. Os sistemas computacionais de auxílio ao diagnóstico (Computer-Aided Detection - CAD) vêm apresentando alto desempenho na detecção de nódulos e microcalcificações mamárias, mas para o caso da DAM, o desempenho ainda é insatisfatório. Algumas limitações são normalmente reportadas nos algoritmos adotados para detectar automaticamente a DAM. O presente trabalho tem por objetivo propor novas abordagens para aumentar a precisão dos métodos computacionais de detecção: o uso de descritores de micro-padrões local para discriminação de áreas suspeitas; redução de falsos-positivos; uso do volume 3D fornecido pelo exame de DBT e; uso de arquitetura de aprendizagem profunda para discriminação e classificação de regiões suspeitas. Os diversos testes efetuados em cada proposta mostraram que é possível melhorar as taxas de detecção da DAM, mesmo para imagens de DBT onde ainda não há um esquema computacional de detecção bem estabelecido. / Breast cancer is the disease that most affects women worldwide and is the most effective treatment if it is diagnosed at early stages. Since 2011, in developed countries screening programs has been employed a new exam, the digital breast tomosynthesis (DBT), which has several advantages compared to the digital mammography. In the exam, the radiologist looks for suspicious signs in the image such as masses, microcalcifications and architectural distortion of breast (ADB). Since the later may represent the earliest sign of a cancer in formation, it may manifests before the formation of any other lesion. However, ADB is difficult to be detected due to its subtly changes the breast tissue, with no mass or defined shape. Computer-aided detection (CAD) systems have shown good results in the detection of masses and microcalcifications, however, for ADB the performance is still poor. Several weakness are reported in the pipeline adopted to automatic detection of ADB. The present work aims to propose new approaches to increase the accuracy of the current CAD pipeline: the use of local micro-pattern descriptors to discriminate suspicious areas; false-positives reduction; automatic detection of ADB in DBT images using and tridimensionality of the exam and; use of deep learning architecture to discriminate and classify suspicious regions. The several tests performed on each proposal showed that it is possible to improve the detection rates even for DBT images where there is no established CAD pipeline.

Orientador: Denis Henrique Pinheiro Salvadeo / Resumo: Tomossíntese Mamária Digital (DBT) é uma técnica de imageamento radiográfico, com aquisição de projeções em ângulos limitados utilizando dose reduzida de radiação. Ela tem por objetivo reconstruir fatias tomográficas do interior da mama, possibilitando o diagnóstico precoce de possíveis lesões e aumentando, consequentemente, a probabilidade de cura do paciente. Contudo, devido ao fato de que DBT utiliza doses baixas de radiação, a imagem gerada contém mais ruído que a mamografia digital. Embora a qualidade do exame esteja diretamente relacionada com a dose utilizada, espera-se que a dose de radiação empregada no exame seja a mais baixa possível, mas ainda com qualidade suficiente para que o diagnóstico possa ser realizado, conforme o princípio As Low As Reasonably Achievable (ALARA). Uma das etapas importantes para se buscar o princípio ALARA é a reconstrução tomográfica, que consiste em um software que gera as fatias do interior da mama a partir de um conjunto de projeções 2D de DBT adquiridas. Por outro lado, técnicas de Aprendizado de Máquina, especialmente redes neurais com aprendizado profundo, que recentemente tem evoluído consideravelmente o estado da arte em diversos problemas de Visão Computacional e Processamento de Imagens, tem características adequadas para serem aplicadas também na etapa de reconstrução. Deste modo, este trabalho investigou uma arquitetura básica de rede neural artificial com aprendizado profundo que seja capaz de reconstruir imagens de DBT, espe... (Resumo completo, clicar acesso eletrônico abaixo) / Abstract: Digital Breast Tomosynthesis (DBT) is a technique of radiographic imaging, with acquisition of projections at limited angles using reduced dose of radiation. It aims to reconstruct tomographic slices inside the breast, making possible the early diagnosis of possible lesions and, consequently, increasing the probability of cure of the patient. However, due to the fact that DBT uses low doses of radiation, the generated image contains more noise than digital mammography. Although the quality of the exam is directly related to the dose applied, the radiation dose used in the examination is expected to be as low as possible, but still keeping enough quality for the diagnosis to be made, as determined by the As Low As Reasonably Achievable (ALARA) principle. One of the important steps to achieve the ALARA principle is the tomographic reconstruction, which consists of a software that generates slices inside the breast from an acquired set of 2D DBT projections. On the other hand, Machine Learning techniques, especially neural networks with deep learning, that have recently evolved considerably the state-of-the-art in several problems in Computer Vision and Image Processing areas, it has suitable characteristics to be applied also in the reconstruction step. Thus, this work investigated a basic architecture of artificial neural network with deep learning that is capable to reconstruct DBT images, especially focused on noise reduction. Furthermore, considering an additional filtering... (Complete abstract click electronic access below) / Mestre

Esse trabalho apresenta uma nova metodologia para a correção do espectro de potência do ruído no processo de simulação de aquisições de imagens de tomossíntese digital mamária (Digital Breast Tomosynthesis - DBT) com doses reduzidas de radiação. A simulação é realizada por meio da inserção de ruído quântico dependente do sinal em imagens previamente adquiridas com a dose padrão de radiação. A DBT utiliza a mesma tecnologia de raios X que a mamografia digital, porém com a capacidade de prover ao médico exames do volume tridimensional da mama, minimizando o problema de superposição de tecidos. Apesar de ser o sucessor da mamografia, estudos têm mostrado que a otimização da relação entre a dose de radiação e a qualidade da imagem adquirida ainda não está bem estabelecida na DBT. Devido à impossibilidade de realizar diversas exposições de radiação a uma mesma paciente para os estudos de otimização da dose de radiação, é desejável que exista um método capaz de simular com exatidão diversas exposições tendo como base uma imagem clínica de referência. Embora existam diversos métodos para a simulação da redução de dose em exames mamográficos, o mesmo não pode ser dito quanto a imagens de DBT. O método desenvolvido para simulação da redução da dose de radiação em imagens de DBT se baseia em uma abordagem de inserção de ruído por meio de uma transformada de estabilização de variância, que já foi utilizada para simulação da redução de dose em exames de mamografia digital. Porém, esse trabalho propõe a inclusão da correção do espectro de potência do ruído para otimizar o desempenho do método de inserção de ruído para exames de DBT. Os resultados obtidos mostraram que, quando comparando a imagens de referência, a as imagens simuladas apresentaram erro menores que 1% para a análise do valor médio e desvio padrão e erro próximo de 5% para a análise do espectro de potência, apresentado resultados até 64% melhores que métodos não otimizados para DBT. / This work presents a new methodology for noise power spectrum correction in the simulation of digital breast tomosynthesis (DBT) images with reduced dose of radiation. The simulation is performed by inserting a signal-dependent quantum noise into previously acquired images with the standard dose of radiation. Using the same X-ray technology as a standard mammography, the DBT is capable of reconstructing the inner tissues of the patients\' breasts as a three-dimensional volume, providing more resources for cancer detection than its bi-dimensional counterpart and minimizing tissue overlapping. Despite being the successor to mammography, studies have shown that the optimization of the relationship between radiation dose and image quality is not well established in DBT yet. Due to the impossibility of exposing the same patient to multiple exams with different doses each, a simulation method able to mimic clinical images with high reliability is desirable. Despite the number of methods proposed for dose reduction simulation in mammography, scarcely any may be used in DBT. The method developed for simulation of radiation dose reduction in DBT images is based on a noise insertion approach using a variance-stabilizing transformation, which has already been used to simulate dose reduction in digital mammography exams. However, this work proposes the inclusion of the noise power spectrum correction to optimize the performance of the noise insertion method for DBT exams. The results showed that, when compared with reference images, the simulated images achieved less than 1% error for mean and standard deviation values and close to 5% error for power spectrum analysis, improving in up to 64% when compared with non-optimized for DBT simulation methods.

Esse trabalho apresenta uma nova metodologia para a correção do espectro de potência do ruído no processo de simulação de aquisições de imagens de tomossíntese digital mamária (Digital Breast Tomosynthesis - DBT) com doses reduzidas de radiação. A simulação é realizada por meio da inserção de ruído quântico dependente do sinal em imagens previamente adquiridas com a dose padrão de radiação. A DBT utiliza a mesma tecnologia de raios X que a mamografia digital, porém com a capacidade de prover ao médico exames do volume tridimensional da mama, minimizando o problema de superposição de tecidos. Apesar de ser o sucessor da mamografia, estudos têm mostrado que a otimização da relação entre a dose de radiação e a qualidade da imagem adquirida ainda não está bem estabelecida na DBT. Devido à impossibilidade de realizar diversas exposições de radiação a uma mesma paciente para os estudos de otimização da dose de radiação, é desejável que exista um método capaz de simular com exatidão diversas exposições tendo como base uma imagem clínica de referência. Embora existam diversos métodos para a simulação da redução de dose em exames mamográficos, o mesmo não pode ser dito quanto a imagens de DBT. O método desenvolvido para simulação da redução da dose de radiação em imagens de DBT se baseia em uma abordagem de inserção de ruído por meio de uma transformada de estabilização de variância, que já foi utilizada para simulação da redução de dose em exames de mamografia digital. Porém, esse trabalho propõe a inclusão da correção do espectro de potência do ruído para otimizar o desempenho do método de inserção de ruído para exames de DBT. Os resultados obtidos mostraram que, quando comparando a imagens de referência, a as imagens simuladas apresentaram erro menores que 1% para a análise do valor médio e desvio padrão e erro próximo de 5% para a análise do espectro de potência, apresentado resultados até 64% melhores que métodos não otimizados para DBT. / This work presents a new methodology for noise power spectrum correction in the simulation of digital breast tomosynthesis (DBT) images with reduced dose of radiation. The simulation is performed by inserting a signal-dependent quantum noise into previously acquired images with the standard dose of radiation. Using the same X-ray technology as a standard mammography, the DBT is capable of reconstructing the inner tissues of the patients\' breasts as a three-dimensional volume, providing more resources for cancer detection than its bi-dimensional counterpart and minimizing tissue overlapping. Despite being the successor to mammography, studies have shown that the optimization of the relationship between radiation dose and image quality is not well established in DBT yet. Due to the impossibility of exposing the same patient to multiple exams with different doses each, a simulation method able to mimic clinical images with high reliability is desirable. Despite the number of methods proposed for dose reduction simulation in mammography, scarcely any may be used in DBT. The method developed for simulation of radiation dose reduction in DBT images is based on a noise insertion approach using a variance-stabilizing transformation, which has already been used to simulate dose reduction in digital mammography exams. However, this work proposes the inclusion of the noise power spectrum correction to optimize the performance of the noise insertion method for DBT exams. The results showed that, when compared with reference images, the simulated images achieved less than 1% error for mean and standard deviation values and close to 5% error for power spectrum analysis, improving in up to 64% when compared with non-optimized for DBT simulation methods.