Thesis (doctoral)--University of Copenhagen, 2005.

published_or_final_version / Microbiology / Master / Master of Philosophy

Thesis (M. Phil.)--University of Hong Kong, 2002. / Includes bibliographical references (leaves 158-171).

El cáncer es una de las principales causas de muerte en el mundo. Los tumores cerebrales tienen una incidencia relativamente baja en comparación con otras patologías cancerígenas más generalizadas, pero la prognosis de algunos es muy pobre, contribuyendo significativamente a su morbilidad. La gestión clínica de una masa anormal en el cerebro es materia delicada y difícil, por lo que los expertos han de basarse en mediciones indirectas no invasivas de las características del tumor y de su crecimiento. En la práctica radiológica actual, estas mediciones se realizan a menudo mediante técnicas de resonancia magnética (MR), como la imagen (MRI) y la espectroscopia (MRS). La vasta información contenida en las señales de MR les hace ideales para la aplicación de técnicas de reconocimiento de patrones (PR). Durante las dos últimas décadas, estas técnicas se han aplicado con éxito al problema de la extracción de conocimiento a partir de datos de tumores cerebrales humanos, para su diagnóstico y pronóstico. No obstante, la discriminación de algunos tipos y subtipos de tumores, así como la delimitación precisa del área tumoral, continúan siendo un reto para los investigadores. En esta tesis, abordamos tales retos mediante la aplicación de un conjunto de técnicas avanzadas de PR. En primera instancia, se implementaron una variedad de técnicas comunes y bien conocidas en una herramienta integrada de software. Esta fue utilizada en tareas de reducción de dimensionalidad (DR), clasificación y evaluación de modelos, para el desarrollo de clasificadores apropiados para analizar datos de MRS. Posteriormente, se profundizó en el desarrollo de métodos de extracción de características (FE), para proponer un método que proporciona prototipos de señal interpretables a partir de los datos de MRS. En una siguiente fase, dos técnicas de descomposición espectral fueron utilizadas para extraer fuentes de MRS e identificar la que proporciona mejores resultados en el contexto de problemas de neuro-oncología, utilizando datos de MRS de un solo vóxel (SV). El mejor y más adecuado método de extracción de fuentes resultante se utilizó posteriormente para derivar fuentes correlacionadas con los espectros promedio de los tipos de tejidos estudiados. El primer enfoque se aplicó también en el contexto de datos de múltiples vóxeles (MV), donde se propone un mecanismo para delimitar la zona patológica del tumor. Las contribuciones se pueden resumir de la siguiente manera. En primer lugar, el desarrollo de una herramienta de software que ha permitido reproducir clasificadores previamente publicados, así como probar nuevas hipótesis. También se ha contribuido con un método de FE, cuyo rendimiento es comparable a su contraparte más comúnmente utilizada en el análisis de datos de MRS, al tiempo que mejora su interpretación. Por otra parte, hemos identificado la variante de descomposición espectral que mejor se adapta al análisis de datos de SV MRS, llamada Convex Non-negative Matrix Factorisation (NMF), mostrando su capacidad para discriminar entre tejido sano, necrosis y tumor en proliferación activa, con resultados que son comparables a los obtenidos en modo supervisado. Con los datos de MV, los resultados obtenidos evidencian que es posible conseguir una delimitación precisa de la zona patológica mediante la aplicación de Convex-NMF. Con esta tesis, ofrecemos una herramienta a radiólogos espectroscopistas para facilitar el desarrollo de clasificadores para el análisis de datos de MRS, en un amplio grupo de tipos tumorales. También ofrecemos una alternativa no supervisada para mejorar la discriminación entre tipos y subtipos tumorales, colocando este enfoque un paso por delante de los métodos supervisados clásicos. Esto nos ha permitido hacer frente a una de las principales fuentes de incertidumbre en el manejo clínico de tumores cerebrales: la dificultad de delimitar adecuadamente el área patológica. / Cancer is a leading cause of death worldwide. Tumours of the Central Nervous System and, among them, brain tumours have a relatively low incidence as compared to other more widespread cancer pathologies, but the prognosis of some of them is very poor, contributing significantly to morbidity. The clinical management of an abnormal mass in the brain is sensitive and difficult, making experts to rely on non-invasive indirect measurements of the tumour characteristics and growth. In current radiological practice, these data measurements are often provided by magnetic resonance (MR) techniques, such as imaging (MRI) and spectroscopy (MRS). The rich information contained in MR signals makes them ideally suited to the application of pattern recognition (PR) techniques. Over the last two decades, these techniques have been successfully used to address the problem of knowledge extraction from human brain tumour data, for their diagnosis and prognosis. Nevertheless, the discrimination of some tumour types and subtypes, along with the accurate delimitation of the tumour area, remained challenging. In this thesis, we approach these challenges using a set of advanced PR techniques. A variety of common and well-known dimensionality reduction (DR), classification, and evaluation methods are first gathered in a software tool, used for the development of classifiers that are suitable for the analysis of MRS data. We then delve into the feature extraction (FE) family of DR methods to propose a method that is robust in the presence of noise, not prone to overfitting, and which also provides interpretation of the extracted MRS signal prototypes. Two spectral decomposition techniques, in different algorithmic variants, are subsequently used to extract the sources of the MRS signals and identify the one that provides better results in the context of neuro-oncology, using single-voxel (SV) MRS data. The best and most adequate source extraction method is then used to derive sources correlated with the mean spectra of known tissue types. The former, an unsupervised approach, is also applied in this thesis in the multi-voxel (MV) context, where we propose a mechanism for delimiting the pathological area of the tumour. The contributions of this thesis can be summarised as follows. First, the development of a software tool allowed us to reproduce previously published MRS-based classifiers, and test new hypotheses that led to new publications. We also contributed a FE method, whose performance is comparable to its most commonly used counterpart in MRS data analysis, while improving on the interpretability. Moreover, we identified the spectral decomposition variant that best suits the analysis of SV MRS data, namely Convex Non-negative Matrix Factorisation (NMF), and showed its ability to discriminate between healthy tissue, necrosis, and actively proliferating tumour, with results that are comparable to those obtained in fully supervised mode. For MV data, we successfully benchmarked alternative spectral decomposition methods, and provided evidence supporting that very accurate delimitation can be achieved through the application of Convex-NMF. With this thesis, we provide spectroscopists with a tool that facilitates the development of classifiers for the analysis of MRS data, for a large group of tumour types; allowing them to concentrate on the interpretation of the results, without requiring a specialised mathematical expertise for testing their hypotheses. We also provide an unsupervised alternative to improve the discrimination between tumour types and subtypes, placing this approach one step ahead of classical label-requiring supervised methods for detection of the increasingly recognised molecular subtype heterogeneity within human brain tumours. This also allowed us to accurately tackle one of the main sources of uncertainty in the clinical management of brain tumours, which is the difficulty of appropriately delimiting the pathological area.

Le travail de thèse porte sur l’étude des phénomènes de l'absorption d'une onde de choc par un matériau granulaire, de l’impact balistique dans un matériau granulaire, et de l’impact balistique sur un tissu. Afin d'évaluer l'absorption d'une onde de choc par un matériau granulaire, un dispositif expérimental a été conçu, qui permet d'évaluer à la fois la pression incidente et réfléchie, et la déflexion et l'accélération d'une plaque de référence, en fonction du temps. Le comportement du système étudié est décrit par un modèle masse-ressort-amortisseur à un degré de liberté.La performance balistique d'un granulat, soumis à l'impact d'un projectile sphérique, est étudiée sur la base de l'évolution de la vitesse de la bille dans le matériau. L'impact dans le matériau granulaire peut être considéré comme un impact hydrodynamique. Au niveau de l'étude numérique, deux approches sont utilisées pour l'application de la sollicitation sur le matériau granulaire : application directe de la pression de l'onde de choc, et propagation de l'onde dans un domaine Eulérien. De même, pour le granulat, plusieurs modèles de comportement sont étudiés. Dans le modèle numérique, le matériau granulaire est modélisé par un matériau équivalent homogénéisé. Un protocole expérimental spécifique a été développé en vue de la mesure continue de la vitesse d'un projectile lors de l'impact sur un tissu. La confrontation entre les résultats expérimentaux et un modèle analytique sur la base du principe de la conservation de quantité de mouvement, démontre que le modèle analytique simple est capable de décrire l'évolution de la vitesse pendant la première phase après l'impact. / This thesis deals with the absorption of shock wave energy by a granular material, and with the study of ballistic impact in fabrics and granular materials. For the assessment of shock wave absorption by a granular material, a test set-up was developed. This set-up allows the evaluation of incident and reflected pressure, and deflection and acceleration of a reference plate as a function of time. An analytical mass-spring-dashpot model with one degree of freedom can predict the deflection correctly. The ballistic performance of a granular material, impacted by a spherical projectile, is evaluated in terms of velocity decay through the material. The experimental results indicate that the impact can be considered as a hydrodynamic impact.In the numerical study, two approaches are used to apply the pressure on the granular material: direct application of the pressure on the sample, and propagation of the shock wave through an Eulerian domain. For the granular material, considered as homogeneous and isotropic, different models have been analysed. A new test set-up was developed for the continuous measurement of the projectile velocity during the penetration in ballistic fabric. The experimental results are compared to an analytical model, based on the conservation of momentum. It appears that the very simple analytical model describes the velocity evolution very well during the first phase after impact.

Le but de cette étude est la prédiction de la durée de vie à l'initiation de la fissure d'un contact acier/aluminium sous la sollicitation de fretting fatigue. La prédiction des durées de vie a été déterminée à l'amorçage des fissures en fonction des différents chargements appliqués. Un dispositif expérimental adapté sur un banc d'essai multiaxial de fatigue a été mis au point. Des essais ont été réalisés sur un alliage d’aluminium de la série 6082T6. Les résultats obtenus montrent un abattement de la durée de vie par rapport à la fatigue classique. La modélisation par éléments finis a été menée en utilisant un code de calcul Abaqus. Les champs de contraintes et de déformations identifiés ont été utilisés sur différents critères de fatigue multiaxiale basés sur les plans critiques. Les résultats de la prédiction donne une sous estimation de la durée de vie en fretting fatigue. Ceci est dû aux forts gradients présents dans la zone de contact. La prédiction a été améliorée par couplage d’une méthode de champs moyens avec les critères de fatigue multiaxiale. Les estimations obtenues présentent une bonne corrélation avec les résultats expérimentaux. Néanmoins, la taille de la zone d'élaboration optimale semble dépendre du critère de fatigue multiaxiale utilisé et des conditions de chargement appliquées. Cette zone ne semble pas être liée directement avec la microstructure pour le matériau étudié.Dans une structure les sollicitations sont variables. La démarche ci-dessus a été couplée avec deux lois d'endommagement: loi de Miner et loi dite de la contrainte endommagée pour la prédiction avec des chargements par blocs. Les prédictions obtenues sont satisfaisantes. / The purpose of this study is to predict fretting fatigue crack nucleation of a mono contact steel/aluminum at different applied loads.An experimental device was first designed and adapted on a multiaxial fatigue apparatus. Tests were conducted aluminum alloy at constant amplitude loading under both different load levels and load ratios. As expected, the fretting fatigue life was found lower compared to uniaxial fatigue under the same loading conditions. Finite element analysis was conducted using \textit{Abaqus} software. The computed stress and strain fields were used to estimate the parameters of different multiaxial fatigue criteria based on the critical plane approach. When using stress and strain values corresponding to the material point exhibiting the maximum value of the considered parameter, we have found that, whatever the multiaxial fatigue parameter, conservative fatigue life estimates due to the severe gradients in the contact zone. An averaging method of the mechanical quantities over a given reference volume was then used to attenuate these gradients effects. The estimates show a good correlation with experimental results. However, the size of the reference volume depends on both the multiaxial fatigue criteria and on loading conditions applied. So that it could not be directly linked to the grain size for the material studied. Finally, we made an attempt to extend these criteria and the developed method to variable loadings. Fretting fatigue tests using two and four blocks loadings were performed and the previous criteria were coupled with two damage laws. The estimates we have obtained seems very promising.

Les propriétés mécaniques (dureté et module d’élasticité) peuvent s’étudier à différentes échelles en fonction de la composition et de la microstructure des matériaux (nombre de phases, répartition des constituants, diamètre des grains…). Dans ce travail nous étudions ces comportements aux échelles nano, micro et macroscopiques, grâce à une gamme d’appareils expérimentaux en indentation instrumentée qui permettent d’appliquer des forces variant de 1 mN à 2,5 kN et de provoquer des enfoncements mesurables de 10 nm à 3 mm. Une première partie consiste à optimiser les conditions expérimentales, dépouiller et interpréter les données en suivant une démarche métrologique rigoureuse, dans le but d’effectuer le raccordement entre les résultats obtenus aux trois échelles de mesure étudiées. Ces essais sont effectués sur des matériaux métalliques industriels mais suffisamment homogènes (aciers et alliage d’aluminium). Nous proposons par exemple une méthode alternative à celle d’Oliver et Pharr pour estimer le module d’élasticité à partir de la courbe de décharge en indentation. Cette méthodologie est ensuite appliquée à l’étude du comportement sous indentation d’un matériau composite, hautement hétérogène, utilisé pour la fabrication de semelles de freinage dans le domaine ferroviaire. Les données obtenues (répartition spatiale en surface et en volume de la dureté et des modules d’élasticité) sont destinées à être utilisées dans les modèles simulant le bruit de crissement au freinage, afin d’en atténuer les effets. / The mechanical properties (hardness and elastic modulus) can be studied at different scales as a function of the composition and microstructure of materials (number of phases, constituent distribution, grain size…). In this work, we studied this mechanical response at the nano, micro and macroscopic scales, through different instruments that allow to apply loads ranging from 20 mN up to 2 kN generating penetrations depths between 100 nm up to 2.5 mm. The first stage consisted in the improvement of the experimental conditions, developing and interpreting the experimental data, following a rigorous metrological methodology, in order to achieve the connection between the results obtained at the three studied scales. These tests were performed in metallic industrial materials, which are sufficiently homogeneous at the tested penetration depths (steels and aluminum alloy). For example, we had proposed an alternative approach to the Oliver and Pharr method to fit the unloading curve to estimate the elastic modulus. Afterwards, this methodology is applied to the study of the response obtained by indentation in a composite material, which is highly heterogeneous, used in the fabrication of brake pads in the railway industry. The obtained results (spatial distribution on the surface and through the volume of hardness and elastic modulus) are going to be valuables in the applications of models to study the squeal noise during braking, looking to reduce its effects.

Studien omfattar 19 elever i år 6. Syftet var att undersöka om det går att se och utvärdera några förändringar i elevernas resultat i läsförståelse efter att de systematiskt tränats i denna färdighet under sex veckor. Eleverna fördelades på en test- och kontrollgrupp. Elevernas prestationer mättes före och efter insatserna med material från nationella prov. Därefter gjordes en intervention av övningar för att stärka läsförståelsen hos eleverna i testgruppen. Eleverna i testgruppen följs upp och arbetet beskrivs mer ingående av mig som deltagande observatör. Övningarna genomfördes 3-4 gånger i veckan med 20min vid varje tillfälle och under sex veckor. Testresultaten redovisas i tabellform och arbetsprocessen följs upp och beskrivs. I resultaten kan man utläsa en förändring i positivriktning hos testgruppen och i klassens resultat.

Since the successful implementations in some prestigious companies such as Motorola, GE & Allied Signal the ¡§Six Sigma¡¨ quality management system has become more and more companies¡¦ priority approach in upgrading their overall competitiveness. With its actual progress made in competitiveness and the shining financial results in the past the Six Sigma has attracted numerous followers and it is now indeed the main quality management stream after the TQM era. No matter the Six Sigma implementation is still under evaluation or has been accepted and ready to kick off, it is no other than one of the main strategic plans of almost all the contemporary companies. However, in spite that Six Sigma has been maturely implemented for some time and there have been some global companies and local companies here in Taiwan that have phased it in already, there are still a lot of local companies that rather wait-and-see and decide not to be the front runners. In Taiwan the number of Six-Sigma companies that have truly phased in and paid off are still quite limited so far. Because of that this research summarized the existing literature based on the key success factors and then validated this hypothesis through expert validation. After the hypothesis was validated, there followed investigations upon the middle and upper level managers who were mostly also students of the NSYSU¡¦s EMBA, MBA (on-job), MBA (night-session), English MBA, MBA (HR, on-job) and ISU¡¦s EMBA and PMBA programs, to have an in-depth understanding on what were theirs and the temporary companies¡¦ wants and anticipated results, what were the differences between their perceptions on the Six Sigma¡¦s key implementation factors, what were the correlations between different independent and dependent variables, and whether there still embedded any tacit factors that could otherwise result in failures. From the result of this research we concluded that no matter what industries and what scales most of the companies have faced with pressure of having to implement Six Sigma. In addition, they believed that Six Sigma truly would be able to help them upgrading their overall competitiveness and corporate images. Also, we learned that there existed differences in independent variables vs. different dependent variables. The perceptions of the middle and upper managements were found not in line with the key success factors in certain degrees in some ways that could otherwise contribute to failures if we do not pay attention to prevent them from happening. Generally speaking, no matter how many employees the companies may have, what capital and revenue sizes they are, all the companies should know of their industries¡¦ own features well for deciding their own directions. They should have appropriate benchmarking and consultants, and need to seek for helps from outside if necessary so that they could take advantage of this new Six Sigma era.

La detección de peatones continua siendo un problema muy difícil en escenarios reales, donde diferentes situaciones como cambios en iluminación, imágenes ruidosas, objetos inesperados, escenarios sin control y la variabilidad en la apariencia de los objetos ocurren constantemente. Todos estos problemas fuerzan el desarrollo de detectores más robustos para aplicaciones relevantes como lo son los vehículos autónomos basados en visión, vigilancia inteligente y el seguimiento de peatones para el análisis del comportamiento. Los detectores de peatones basados en visión más confiables deciden basándose en descriptores extraídos usando un único sensor y capturando características complementarias, e.g., apariencia y textura. Estas características son extraídas de una única imagen, ignorando la información temporal, o incluyendo esta información en un paso de post procesamiento e.g., seguimiento o coherencia temporal. Teniendo en cuenta estos hechos, nos formulamos la siguiente pregunta: ¿Podemos generar detectores de peatones más robustos mediante la introducción de nuevas fuentes de información en el paso de extracción de características? Para responder a esta pregunta desarrollamos diferentes propuestas para introducir nuevas fuentes de información a detectores de peatones bien conocidos. Empezamos por la inclusión de información temporal siguiendo el paradigma del aprendizaje secuencial apilado (SSL siglas en inglés), el cual sugiere que la información extraída de las muestras vecinas en una secuencia pueden mejorar la exactitud de un clasificador base. Después nos enfocamos en la inclusión de información complementaria proveniente de sensores diferentes como nubes de puntos 3D (LIDAR - profundidad), imágenes infrarrojas (FIR) o mapas de disparidad (par estéreo de cámaras). Para tal fin desarrollamos un marco multimodal en el cual información proveniente de diferentes sensores es usada para incrementar la exactitud en la detección (aumentando la redundancia de la información). Finalmente proponemos un detector multi-vista, esta propuesta multi-vista divide el problema de detección en n sub-problemas. Cada uno de estos sub-problemas detectara objetos en una vista específica dada, reduciendo así el problema de la variabilidad que se tiene cuando un único detector es usado para todo el problema. Demostramos que estas propuestas obtienen resultados competitivos con otros métodos en el estado del arte, pero envés de diseñar nuevas características, reutilizamos las existentes para mejorar el desempeño. / Pedestrian detection continues to be an extremely challenging problem in real scenarios, in which situations like illumination changes, noisy images, unexpected objects, uncontrolled scenarios and variant appearance of objects occur constantly. All these problems force the development of more robust detectors for relevant applications like vision-based autonomous vehicles, intelligent surveillance, and pedestrian tracking for behavior analysis. Most reliable vision-based pedestrian detectors base their decision on features extracted using a single sensor capturing complementary features, e.g., appearance, and texture. These features usually are extracted from the current frame, ignoring temporal information, or including it in a post process step e.g., tracking or temporal coherence. Taking into account these issues we formulate the following question: can we generate more robust pedestrian detectors by introducing new information sources in the feature extraction step? In order to answer this question we develop different approaches for introducing new information sources to well-known pedestrian detectors. We start by the inclusion of temporal information following the Stacked Sequential Learning (SSL) paradigm which suggests that information extracted from the neighboring samples in a sequence can improve the accuracy of a base classifier. We then focus on the inclusion of complementary information from different sensors like 3D point clouds (LIDAR - depth), far infrared images (FIR), or disparity maps (stereo pair cameras). For this end we develop a multi-modal framework in which information from different sensors is used for increasing detection accuracy (by increasing information redundancy). Finally we propose a multi-view pedestrian detector, this multi-view approach splits the detection problem in n sub-problems. Each sub-problem will detect objects in a given specific view reducing in that way the variability problem faced when a single detectors is used for the whole problem. We show that these approaches obtain competitive results with other state-of-the-art methods but instead of design new features, we reuse existing ones boosting their performance.