No description available.

No description available.

No description available.

No description available.

Object recognition systems require baseline information upon which to compare sensed data to enable a recognition task. The ability to integrate a diverse set of object recognition data for different components in a Building Information Model (BIM) will enable many autonomous systems to access and use these data in an on-demand learning capacity, and will accelerate the integration of object recognition systems in the construction environment. This research presents a new framework for linking feature descriptors to a BIM to support construction object recognition. The proposed framework is based upon the Property and External Reference Resource schemas within the IFC 2x3 TC1 architecture. Within this framework a new Property Set (Pset_ObjectRecognition) is suggested which provides an on-demand capability to access available feature descriptor information either embedded in the IFC model or referenced in an external model database. The Property Set is extensible, and can be modified and adjusted as required for future research and field implementation. With this framework multiple sets of feature descriptors associated with different sensing modalities and different algorithms can all be aggregated into one Property Set and assigned to either object types or object instances. / Ph. D.

Deep learning methods have achieved state-of-the-art performance in various tasks of medical image analysis. However, the success relies heavily on the expensive and time-consuming collection of large quantities of labeled data, which is not always available. This dissertation investigates the use of self-supervised and generative methods to enhance the label efficiency of deep learning models for 3D medical image analysis. Unlike natural images, medical images contain consistent anatomical contexts specific to the domain, which can be exploited as self-supervision signals to pre-train the model. Furthermore, generative methods can be utilized to synthesize additional samples, thereby increasing sample diversity. In the first part of the dissertation, we introduce self-supervised learning frameworks that learn anatomy-aware and disease-related representation. In order to learn disease-related representation, we propose two domain-specific contrasting strategies that leverage anatomical similarity across patients to create hard negative samples that incentivize learning fine-grained pathological features. In order to learn anatomy-sensitive representation, we develop a novel 3D convolutional layer with kernels that are conditionally parameterized based on the anatomical locations. We perform extensive experiments on large-scale datasets of CT scans, which show that our method improves the performance of many downstream tasks. In the second part of the dissertation, we introduce generative models capable of synthesizing high-resolution, anatomy-guided 3D medical images. Current generative models are typically limited to low-resolution outputs due to memory constraints, despite clinicians' need for high-resolution details in diagnoses. To overcome this, we present a hierarchical architecture that efficiently manages memory demands, enabling the generation of high-resolution images. In addition, diffusion-based generative models are becoming more prevalent in medical imaging. However, existing state-of-the-art methods often under-utilize the extensive information found in radiology reports and anatomical structures. To address these limitations, we propose a text-guided 3D image diffusion model that preserves anatomical details. We conduct experiments on downstream tasks and blind evaluation by radiologists, which demonstrate the clinical value of our proposed methodologies.

Diabetes is a rapidly growing disease with 415 million affected adults worldwide. The pancreatic endocrine cells, most importantly the insulin producing β-cells, play an important role in regulating blood glucose homeostasis. Type 1 diabetes (T1D) is characterized by the inability of the pancreas to secrete sufficient amounts of insulin due to autoimmune destruction of insulin producing β-cells. Type 2 diabetes (T2D) on the other hand is characterized by defects in insulin secretion and insulin sensitivity. Alterations in the β-cell mass (BCM) and/or function play a major role in the development and progression of the disease. Understanding BCM dynamics in disease models is therefore a key aspect for better interpretation of research results. In this thesis, we have used optical projection tomography (OPT) as a tool to evaluate a non-invasive imaging modality for β-cell scoring and to study disease dynamics in frequently used animal models for T1D and T2D. The possibility to monitor BCM in vivo would radically improve our competence in studying the pathogenesis of diabetes and in therapeutic interventions. Single photon emission computed tomography (SPECT) is a widely used technique that has become a promising approach to monitor changes in BCM in vivo. A key issue for using this approach is to evaluate the β-cell specificity and read out of the utilized radiotracers. This is most commonly performed by conventional stereological approaches, which rely on the extrapolation of 2D data. We developed a protocol for SPECT-OPT multimodal imaging that enables rapid and accurate cross evaluation of SPECT based assessments of BCM. While histological determination of islet spatial distribution was challenging, SPECT and OPT revealed similar distribution patterns of the radiotracer 111In-exendin-3 and insulin positive β-cell volumes respectively between different pancreatic lobes, both visually and quantitatively. We propose SPECT-OPT multimodal imaging as an accurate and better approach for validating the performance of β-cell radiotracers. The leptin deficient ob/ob mouse is a widely used model for studies of metabolic disturbances leading to T2D, including obesity and insulin resistance. By OPT imaging we created the first 3D-spatial and quantitative account of BCM distribution in this model. We observed a previously unreported degree of cystic lesions in hypertrophic islets, that were occupied by red blood cells (RBCs) and/or fibrin mesh. We propose that these lesions are formed by a mechanism involving the extravasation of RBCs/plasma due to increased blood flow and islet vessel instability. Further, our data indicate that the primary lobular compartments of the ob/ob pancreas have different potentials for expanding their β-cell population. Unawareness of these characteristics of β-cell expansion in ob/ob mice presented in this study may significantly influence ex vivo and in vivo assessments of this model in studies of β-cell adaptation and function. The tomographic data, on which this study was based, will be made publically available as a resource to the research community for the planning and interpretation of research involving this model. There are limited studies on early metabolic and functional changes of BCM in the settings of T1D. In order to assess initial metabolic alterations in BCM before the onset of diabetes, we characterized congenic diabetes prone Bio-breeding (BB) DR.lyp/lyp rats, a widely used model for T1D diabetes. We observed lower acute insulin response, reduced islet blood flow and a significant reduction in the BCM of small and medium sized islets at a very early stage (40 days), i.e. before insulitis and development of diabetes. Underlying changes in islet function may be a previously unrecognized factor of importance in the development of T1D.

Holography has the ability to render three-dimensional information of a recorded scene by capturing both the amplitude and phase of light incident on the recording medium. The application of digital camera technology and high-speed computing means digital holograms can be analysed numerically and novel applications can be found for this technology. This thesis explores the potential for both inline and off-axis digital holographic microscopy to study the three-dimensional swimming behaviour of bacteria. A high-magnification (225x) digital holographic microscope was designed and constructed with the ability to switch easily between inline and off-axis imaging modalities. Hardware aspects, in particular the illumination source, the choice of camera and data transfer rates, were considered. Novel strategies for off-axis holography combining dark field microscopy were designed and implemented. The localisation accuracy of the inline imaging modality was assessed by studying samples of polystyrene microspheres. The microscope is sensitive to stage drift on the order of angstroms per second and can successfully localise microspheres in dilute suspensions at least 100μm from the objective specimen plane. As a simple test of the capabilities of the microscope, the diffusion coefficient of a 0.5μm microsphere was found to be isotropic and consistent with the theoretical value. Amplitude and phase image reconstructions from the off-axis modality are demonstrated. High-magnification dark field off-axis holographic microscopy is shown to be superior to inline microscopy in localising 100nm gold nanoparticles. An artifact from our method of dark-field imaging, however, restricts the depth range to 15μm. A lower-magnification (45x) configuration of the microscope was used to study the 3D swimming behaviour of wild type Escherichia coli as a qualitative demonstration of the potential for this instrument in microbiological applications.

Nesta pesquisa, o metodo GPR foi empregado para localizar e mapear urnas funerarias enterradas, visando orientar as escavacoes arqueologicas e auxiliar nas medidas de protecao de sitios arqueologicos na regiao de influencia direta do aproveitamento hidreletrico de Dardanelos, proximo a Aripuana, MT. Um estudo arqueologico previo seria necessario para verificar a presenca de sitios arqueologicos, pois a regiao seria submersa, afetando todo e qualquer possivel artefato presente no sitio. Na area de influencia da usina de hidreletrica ja havia um sitio conhecido, o sitio de Dardanelos, sendo este o objeto da presente pesquisa. Dados GPR obtidos com a antena blindada de 200 MHz foram processados e analisados, e os resultados apresentados na forma de perfis 2D e em 3D na forma de depth-slices. Apos a aquisicao e processamento dos dados foram identificadas as anomalias GPR e interpretadas a fim de identificar os alvos de interesse arqueologico e raizes de arvores, evitando assim, que haja ambiguidade na caracterizacao dos alvos de interesse. A analise 3D gerada a partir dos perfis de reflexao 2D permitiu diferenciar com clareza os alvos de interesse das raizes de arvores, uma vez que nela podemos visualizar um padrao mais alongado ao inves de pontual, como e apresentado quando temos um artefato arqueologico. Ainda, atraves da conversao do tempo de percurso da onda eletromagnetica em profundidade, podemos identificar a profundidade dos alvos. Esta conversao tambem ajuda a esclarecer as ambiguidades, uma vez que as raizes sao mais rasas e os artefatos mais profundos. / In this research, GPR method was used to locate and map buried indigenous urns, aiming to guide and assist the archaeological excavations in order to guide protections acts of archaeological sites in the region directly affected by the hydroelectric of Dardanelos, near to Aripuana, MT. A preliminary archaeological study would be necessary to investigate the presence of archaeological sites, because the area would go underwater, affecting any possible artifact present on the site. In the area of influence of the hydroelectric plant there was already a known site, the site of the Dardanelos, which is the subject of this research. The GPR data obtained with shielded antenna 200 MHz were processed and analyzed, and the results presented as 2D and 3D profiles in the form of depth-slices. After processing the GPR data anomalies were identified and interpreted to identify the targets of archeological interest and roots of trees, thus avoiding ambiguity in the characterization of targets of interest. The 3D analysis generated from the 2D reflection profiles allowed to differentiate clearly the targets of interest from the roots of trees, since they can display a more elongated pattern rather than punctual, as shown when we have an archaeological artifact. Further, by converting the travel time of the electromagnetic wave in depth, we can identify the depth of targets. This conversion also helps to clarify the ambiguities, since the roots are shallower and the artifacts are deeper.

Nous présentons dans ce manuscrit une technique d’imagerie basée sur l’holographie numérique. Elle permet d’imager en 3D et dans le temps la circulation sanguine chez une larve de poisson-zèbre. L’information 3D est acquise en une seule image de la caméra, ce qui permet de suivre le mouvement des globules rouges dans le système vasculaire. Nous évoquerons dans un premier temps les techniques de bio-imagerie et d’imagerie du flux sanguin traditionnelles, puis nous rappellerons les principes de l’holographie. Ensuite, nous décrirons la méthode d’imagerie que nous avons développée et les résultats expérimentaux obtenus. Nous compléterons, en présentant les différentes améliorations que nous avons apportées à la technique. Enfin, nous discuterons brièvement de l’application du compressed sensing à l’imagerie de la circulation sanguine dans le poisson-zèbre. / In this manuscript, we present an imaging technique based on digital holography.It enables to image in 3D and in time the blood circulation in a zebrafish larva. The 3D information is acquired in a single frame of the camera, which makes possible to track the movement of red blood cells in the vascular system. We will first discuss the traditional techniques of bio and blood flow imaging, then we will remind the principles of holography. Afterwards, we will describe the imaging method we developed and the experimental results obtained. We will then present the improvements that have been made to the technique. Finally, we will briefly discuss the application of the compressed sensing to the blood flow imaging in zebrafish.