This dissertation addresses the difficulties of semantic segmentation when dealing with an extensive collection of images and 3D point clouds. Due to the ubiquity of digital cameras that help capture the world around us, as well as the advanced scanning techniques that are able to record 3D replicas of real cities, the sheer amount of visual data available presents many opportunities for both academic research and industrial applications. But the mere quantity of data also poses a tremendous challenge. In particular, the problem of distilling useful information from such a large repository of visual data has attracted ongoing interests in the fields of computer vision and data mining. Structural Semantics are fundamental to understanding both natural and man-made objects. Buildings, for example, are like languages in that they are made up of repeated structures or patterns that can be captured in images. In order to find these recurring patterns in images, I present an unsupervised frequent visual pattern mining approach that goes beyond co-location to identify spatially coherent visual patterns, regardless of their shape, size, locations and orientation. First, my approach categorizes visual items from scale-invariant image primitives with similar appearance using a suite of polynomial-time algorithms that have been designed to identify consistent structural associations among visual items, representing frequent visual patterns. After detecting repetitive image patterns, I use unsupervised and automatic segmentation of the identified patterns to generate more semantically meaningful representations. The underlying assumption is that pixels capturing the same portion of image patterns are visually consistent, while pixels that come from different backdrops are usually inconsistent. I further extend this approach to perform automatic segmentation of foreground objects from an Internet photo collection of landmark locations. New scanning technologies have successfully advanced the digital acquisition of large-scale urban landscapes. In addressing semantic segmentation and reconstruction of this data using LiDAR point clouds and geo-registered images of large-scale residential areas, I develop a complete system that simultaneously uses classification and segmentation methods to first identify different object categories and then apply category-specific reconstruction techniques to create visually pleasing and complete scene models.

The biological, technical and computational aspects of this project collectively focused on using electron tomography (ET) for the high-resolution (10-20 nm) 3D reconstruction of entire insulin-secreting beta cells within islets of Langerhans isolated from mouse pancreata. Islets were cultured overnight to represent either steady-state (non-stimulated) or elevated glucose (stimulated) conditions, prior to fast-freezing, freeze-substitution, plastic embedment and cutting into 250-400 nm thick sections for tomographic imaging using intermediate voltage electron microscopy (EM). 3D images (tomograms) of each section were used to evaluate the performance of the new technical and computational approaches developed, and make biological comparisons of intercellular structure-function. Analysis focused on key compartments/organelles of the insulin-secretory pathway - Golgi apparatus, mitochondria, insulin secretory granules and multi-granular bodies. To allow the application of ET to entire mammalian cells, several technical limitations were addressed. Since segmenting (delimiting compartments of interest) tomograms manually, represented the major ërate-limiting stepí of ET, an interactive approach for 3D segmentation using novel interpolation algorithms (crude smooth, pointwise smooth and spherical interpolation) to iteratively predict the shape of 3D surfaces between user-drawn contours was developed. The performance of these tools in segmenting a range of compartment types was examined, and found to significantly enhance the speed and accuracy of manual segmentation. To better compensate for the physical collapse of plastic sections in the EM, a novel method was developed for estimating section collapse by analyzing approximately spherical organelles. Using this method on mature insulin granules in high-resolution datasets, coupled with measurements from the whole cell reconstructions, section collapse was found to be substantially less (~25%) than the value (40%) previously used to re-scale 3D models. Other new approaches developed to further improve the accuracy and quality of tomograms, included interactive tools for fiducial tracking, and the use of larger gold particles, a ëreduced second axisí to account for the missing wedge problem, and deformation grids to account for anisotropic deformation. As well as affording more efficient and precise mapping of cell ultrastructure in 3D for subsequent quantitative analysis, these developments provided new insights for future automated (hybrid) segmentation pipelines and new computational approaches for improving quality and isotropic accuracy of volumetric image data. The Interpolator and DrawingTools for segmentation, AnalysisTools for estimating section collapse and BeadHelper for tracking fiducial particles, written as plug-ins for the IMOD software package distributed by the University of Colorado, are now being used by the wider ET community with significant positive feedback. Using the novel approaches developed, four insulin-secreting beta cells - two from the periphery of an islet frozen 1 hr after stimulation with 11 mM glucose, and two from the periphery of another islet under steady-state 5.6 mM glucose conditions - were reconstructed in their entirety in 3D. Quantitative data on the key compartments/organelles provided new information regarding global changes in cellular organization, and enabled robust comparisons of each pair of functionally equivalent cells at unprecedented spatial resolution. Relative differences in the number, dimensions, architecture and distribution of organelles per cubic micron of cellular volume (including mitochondrial branching) reflected differences in the cellsí individual capacity/readiness to respond to secretagogue stimulation. In the two stimulated cells this was reflected by inverse relationships between the number/size of mature granules versus immature granules, the number/size of mitochondria, and the volume of the trans-Golgi network relative to the entire Golgi ribbon. Complementary stereological analysis of whole islets indicated which cells were the most representative under stimulated versus non-stimulated conditions, and revealed a marked natural heterogeneity between cells both within and between individual islets. Overall, this project led to significant improvements in efficiency and accuracy for segmenting cellular compartments/organelles, and in image quality and accuracy for tomogram computation and reconstruction through use of the newly developed techniques. The improved 3D reconstruction and analysis of pancreatic beta cells in toto in native tissue provided a powerful approach for quantitatively mapping the organelles involved in insulin synthesis/secretion at unprecedented detail, and afforded a level of insight into the complex 3D organization of mammalian cells not previously achieved by any other analytical technique or imaging method.

La fibrose pulmonaire est une maladie pulmonaire interstitielle caractérisée par une production irréversible de tissus conjonctifs. Le pronostic de la maladie est plus faible que celui de plusieurs cancers. Dans les dernières années, cette pathologie a été identifiée comme un risque de complication suite à des traitements de radiothérapie. Développer une toxicité post-radique peut compromettre les bénéfices de la radiothérapie, ce qui fait de la fibrose pulmonaire une contre-indication relative. Localiser manuellement la présence de fibrose sur des images de tomodensitométrie (CT) est un problème difficile pouvant nécessiter l’intervention de plusieurs experts pour un seul patient. L’objectif de ce projet est de segmenter automatiquement la fibrose pulmonaire sur des images CT. Des réseaux de neurones complètement convolutifs ont été développés et implémentés pour effectuer une assignation automatique de tissus pulmonaires. Sur une coupe axiale donnée en entrée, l’assignation est réalisée pour l’ensemble des voxels pulmonaires en une seule inférence. L’optimisation des paramètres a été réalisée dans des contextes d’apprentissage supervisé et semi-supervisé en minimisant des variantes de l’entropie croisée entre les prédictions et des annotations manuelles d’experts. Les données utilisées consistent en des images CT haute résolution ainsi que des délinéations réalisées par des radiologistes et des radio-oncologues. Les cartes de segmentation prédites ont été comparées par rapport à des segmentations manuelles afin de valider les tissus assignés par les réseaux convolutifs. Les résultats obtenus suggèrent que des applications en radio-oncologie sont envisageables, telles que le dépistage de la fibrose avant la planification de traitements et l’évaluation de la progression de la fibrose pendant et suivant les traitements de radiothérapie. / Pulmonary fibrosis is an interstitial lung disease characterized by an irreversible production of scarring tissue. Pulmonary fibrosis has a particularly poor prognosis, with a mean survival after diagnosis lower than many cancers. This pathology was recently identified as a risk for complication following radiation therapy treatments. Pulmonary toxicity can lead to severe conditions that compromise the benefits provided by radiation therapy, making pulmonary fibrosis a relative contraindication to treatments. Manual segmentation of fibrosis on computed tomography (CT) images is a difficult task that can involve many experts for a single patient. The aim of this project is to perform automatic segmentation of pulmonary fibrosis on CT images. Fully convolutional neural networks were developed and implemented to automatically assign lung tissues. For an input CT slice, every lung voxel is assigned a tissue in a single inference. Parameters optimization was performed in a supervised and semi-supervised manner by minimizing variants of the cross-entropy between the prediction and manual annotations produced by experts. The dataset employed consists of high resolution CT scans and delineations made by radiologists and radiation oncologists. Predicted segmentation maps were compared with manual segmentations to validate the tissues assigned by the convolutional networks. Results suggest that radiation oncology applications could be developed. Possible applications include pulmonary fibrosis screening prior to treatment planning and assessment of fibrosis progression during and post-treatment.

Machine learning is a growing topic with possibilities that seems endless with growing areas of applications. The field of metallography today is highly dependent on the operators’ knowledge and technical equipment to perform segmentation and analysis of the microstructure. Having expert dependents is both costly and very time-consuming. Some automatic segmentation is possible using SEM but not for all materials and only having to depend on one machine will create a bottleneck. In this thesis, a traditional supervised machine learning model has been built with a Random Forest (RF) classifier. The model performs automatic segmentation of complex microstructure features from images taken using light optical- and scanning electron microscopes. Two types of material, High-Strength-Low-Alloy (HSLA) steel with in-grain carbides and grain boundary carbides, and nitrocarburized steel with different amounts of porosity were analyzed in this work. Using a bank of feature extractors together with labeled ground truth data one model for each material was trained and used for the segmentation of new data. The model trained for the HSLA steel was able to effectively segment and analyze the carbides with a small amount of training. The model could separate the two types of carbides which is not possible with traditional thresholding. However, the model trained on nitrocarburized steel showcased difficulties in detecting the porosity. The result was however improved with a different approach to the labeling. The result implies that further development can be made to improve the model. / Maskininlärning är ett växande område där möjligheterna verkar oändliga med växande applikationsområden. Området för metallografi är idag till stor utsträckning beroende av operatörens kunskap och de tekniska instrumenten som finns tillgängliga för att genomföra segmentering och analys av mikrostrukturen. Viss automatisk segmentering är möjlig genom att använda SEM, men det är inte möjligt för alla material samt att behöva vara beroende av endast en maskin kommer skapa en flaskhals. I denna uppsats har en traditionell övervakad maskininlärnings modell skapats med en Random Forest klassificerare. Modellen genomför automatisk segmentering av komplexa mikrostrukturer på bilder från både ljusoptiskt- och svepelektron-mikroskop. Två olika typer av material, Hög-Styrka-Låg-Legerat (HSLA) stål med karbider och korngräns karbider, samt nitrokarburerat stål med varierande mängd porositet analyserades i detta arbete. Genom användningen av en särdragsextraktions bank tillsammans med annoterad grundsannings data tränades en modell för vartdera materialet och användes för segmentering av ny bild data. Modellen som tränades för HSLA stålet kunde effektivt segmentera och analysera karbiderna med en liten mängd träning. Modellen kunde separera de två typerna av karbider vilket inte varit möjligt med traditionellt tröskelvärde. Den modell som tränades för det nitrokarburerade stålet visade emellertid upp svårigheter i att detektera porositeten. Resultatet kunde dock förbättras genom ett annorlunda tillvägagångssätt för annoteringen. Resultatet vittnar om att vidareutveckling kan göras för att förbättra slutresultatet.

Multiple sclerosis is a neurological disease causing a degeneration of myelin around the axons in the central nervous system. This process leaves traces in the form of lesions, which can be distinguished in an MRI examination. It is important to detect these at an early stage to state diagnosis and initiate medication.  In this Master's Thesis, an automatic segmentation algorithm was developed, with the purpose of segmenting possible multiple sclerosis lesions. Secondly, a progression model was developed with the purpose of estimating the state of each individual lesion. The implementation was based on synthetic contrast weighted images, segmentation maps and quantitative relaxation maps produced by SyMRI (SyntheticMR, Linköping, Sweden). The automatic segmentation algorithm has a relatively high sensitivity but low precision, causing a large number of false positives. The algorithm performed better in the cerebrum compared to the cerebellum. The large number of false positives appeared mainly due to partial volume effects, creating hyperintense artifacts in synthetic T2W FLAIR images. A larger amount of data would have been desirable to create a more robust algorithm. The progression model showed promising results, with a clear correlation to the synthetic contrast-weighted images and segmentation maps available in SyMRI. The progression model could be useful in disease monitoring, medical decisions and diagnosis of Multiple Sclerosis.

No description available.

The exceptional properties of nanocrystals (NCs) are strongly influenced by many different characteristics, such as their size and shape, but also by characteristics on the atomic scale, such as their crystal structure, their surface structure, as well as by potential microstructure defects. While the size and shape of NCs are frequently determined in a statistical manner, atomic-scale characteristics are usually quantified only for a small number of individual NCs and thus with limited statistical relevance. Within this work, a characterization workflow was established that is capable of determining relevant NC characteristics simultaneously in a sufficiently detailed and statistically relevant manner. The workflow is based on transmission electron microscopy, networked by a correlative multiscale approach that combines atomic-scale information on NCs obtained from high-resolution imaging with statistical information on NCs obtained from low-resolution imaging, assisted by a semi-automatic segmentation routine. The approach is complemented by other characterization techniques, such as X-ray diffraction, UV-vis spectroscopy, dynamic light scattering, or alternating gradient magnetometry. The general applicability of the developed workflow is illustrated on several examples, i.e., on the classification of Au NCs with different structures, on the statistical determination of the facet configurations of Au nanorods, on the study of the hierarchical structure of multi-core iron oxide nanoflowers and its influence on their magnetic properties, and on the evaluation of the interplay between size, morphology, microstructure defects, and optoelectronic properties of CdSe NCs.:List of abbreviations and symbols 1 Introduction 1.1 Types of nanocrystals 1.2 Characterization of nanocrystals 1.3 Motivation and outline of this thesis 2 Materials and methods 2.1 Nanocrystal synthesis 2.1.1 Au nanocrystals 2.1.2 Au nanorods 2.1.3 Multi-core iron oxide nanoparticles 2.1.4 CdSe nanocrystals 2.2 Nanocrystal characterization 2.2.1 Transmission electron microscopy 2.2.2 X-ray diffraction 2.2.3 UV-vis spectroscopy 2.2.3.1 Au nanocrystals 2.2.3.2 Au nanorods 2.2.3.3 CdSe nanocrystals 2.2.4 Dynamic light scattering 2.2.5 Alternating gradient magnetometry 2.3 Methodical development 2.3.1 Correlative multiscale approach – Statistical information beyond size and shape 2.3.2 Semi-automatic segmentation routine 3 Classification of Au nanocrystals with comparable size but different morphology and defect structure 3.1 Introduction 3.1.1 Morphologies and structures of Au nanocrystals 3.1.2 Localized surface plasmon resonance of Au nanocrystals 3.1.3 Motivation and outline 3.2 Results 3.2.1 Microstructural characteristics of the Au nanocrystals 3.2.2 Insufficiency of two-dimensional size and shape for an unambiguous classification of the Au nanocrystals 3.2.3 Statistical classification of the Au nanocrystals 3.2.4 Advantage of a multidimensional characterization of the Au nanocrystals 3.2.5 Estimation of the density of planar defects in the Au nanoplates 3.3 Discussion 3.4 Conclusions 4 Statistical determination of the facet configurations of Au nanorods 4.1 Introduction 4.1.1 Growth mechanism and facet formation of Au nanorods 4.1.2 Localized surface plasmon resonance of Au nanorods 4.1.3 Catalytic activity of Au nanorods 4.1.4 Motivation and outline 4.2 Results 4.2.1 Statistical determination of the size and shape of the Au nanorods 4.2.2 Microstructural characteristics and facet configurations of the Au nanorods 4.2.3 Statistical determination of the facet configurations of the Au nanorods 4.3 Discussion 4.4 Conclusions 5 Influence of the hierarchical architecture of multi-core iron oxide nanoflowers on their magnetic properties 5.1 Introduction 5.1.1 Phase composition and phase distribution in iron oxide nanoparticles 5.1.2 Magnetic properties of iron oxide nanoparticles 5.1.3 Mono-core vs. multi-core iron oxide nanoparticles 5.1.4 Motivation and outline 5.2 Results 5.2.1 Phase composition, vacancy ordering, and antiphase boundaries 5.2.2 Arrangement and coherence of individual cores within the iron oxide nanoflowers 5.2.3 Statistical determination of particle, core, and shell size 5.2.4 Influence of the coherence of the cores on the magnetic properties 5.3 Discussion 5.4 Conclusions 6 Interplay between size, morphology, microstructure defects, and optoelectronic properties of CdSe nanocrystals 6.1 Introduction 6.1.1 Polymorphism in CdSe nanocrystals 6.1.2 Optoelectronic properties of CdSe nanocrystals 6.1.3 Nucleation, growth, and coarsening of CdSe nanocrystals 6.1.4 Motivation and outline 6.2 Results 6.2.1 Influence of the synthesis temperature on the optoelectronic properties of the CdSe nanocrystals 6.2.2 Microstructural characteristics of the CdSe nanocrystals 6.2.3 Statistical determination of size, shape, and amount of oriented attachment of the CdSe nanocrystals 6.3 Discussion 6.4 Conclusions 7 Summary and outlook References Publications

[ES] El síndrome metabólico se define como un conjunto de trastornos (e.g., niveles elevados de presión arterial, niveles elevados de glucosa en sangre, exceso de grasa abdominal o niveles elevados de colesterol o triglicéridos) que afectan a un individuo al mismo tiempo. La presencia de uno de estos factores no implica un riesgo elevado para la salud, sin embargo, presentar varios de ellos aumenta la probabilidad de sufrir enfermedades secundarias como la enfermedad cardiovascular o la diabetes tipo II. Las enfermedades difusas hepáticas son todas aquellas enfermedades que afectan a las células funcionales del hígado, los hepatocitos, alterando, de este modo, la función hepática. En estos procesos, los hepatocitos se ven sustituidos por adipocitos y tejido fibroso. La enfermedad de hígado graso no alcohólico es una afección reversible originada por la acumulación de triglicéridos en los hepatocitos. El alcoholismo, la obesidad, y la diabetes son las causas más comunes de esta enfermedad. Este estado del hígado es reversible si se cambia la dieta del paciente, sin embargo, si este no se cuida, la enfermedad puede ir avanzando hacia estadios más severos, desencadenando fibrosis, cirrosis e incluso carcinoma hepatocelular (CHC). La temprana detección de todos estos procesos es de gran importancia en la mejora del pronóstico de los pacientes. Así, las técnicas de imagen en combinación con modelos computacionales permiten caracterizar el tejido mediante la extracción de parámetros objetivos, conocidos como biomarcadores de imagen, relacionados con estos procesos fisiológicos y patológicos, permitiendo una estadificación más precisa de las enfermedades. Además, gracias a las técnicas de inteligencia artificial, se pueden desarrollar algoritmos de segmentación automática que permitan realizar dicha caracterización de manera completamente automática y acelerar, de este modo, el flujo radiológico. Por todo esto, en la presente tesis doctoral, se presenta una metodología para el desarrollo de modelos de segmentación y cuantificación automática, siendo aplicada a tres casos de uso. Para el estudio del síndrome metabólico se propone un método de segmentación automática de la grasa visceral y subcutánea en imágenes de tomografía computarizada (TC), para el estudio de la enfermedad hepática difusa se propone un método de segmentación hepática y cuantificación de la grasa y hierro hepáticos en imágenes de resonancia magnética (RM), y, finalmente, para el estudio del CHC, se propone un método de segmentación hepática y cuantificación de los descriptores de la curva de perfusión en imágenes de RM. Todo esto se ha integrado en una plataforma que permite su integración en la práctica clínica. Así, se han adaptado los algoritmos desarrollados para ser ejecutados en contenedores Docker de forma que, dada una imagen de entrada, generen los parámetros cuantitativos de salida junto con un informe que resuma dichos resultados; se han implementado herramientas para que los usuarios puedan interactuar con las segmentaciones generadas por los algoritmos de segmentación automática desarrollados; finalmente, éstos se han implementado de forma que generen dichas segmentaciones en formatos estándar como DICOM RT Struct o DICOM Seg, para garantizar la interoperabilidad con el resto de sistemas sanitarios. / [CA] La síndrome metabòlica es defineix com un conjunt de trastorns (e.g., nivells elevats de pressió arterial, nivells elevats de glucosa en sang, excés de greix abdominal o nivells elevats de colesterol o triglicèrids) que afecten un individu al mateix temps. La presència d'un d'aquests factors no implica un risc elevat per a la salut, no obstant això, presentar diversos d'ells augmenta la probabilitat de patir malalties secundàries com la malaltia cardiovascular o la diabetis tipus II. Les malalties difuses hepàtiques són totes aquelles malalties que afecten les cèl·lules funcionals del fetge, els hepatòcits, alterant, d'aquesta manera, la funció hepàtica. En aquests processos, els hepatòcits es veuen substituïts per adipòcits i teixit fibrós. La malaltia de fetge gras no alcohòlic és una afecció reversible originada per l'acumulació de triglicèrids en els hepatòcits. L'alcoholisme, l'obesitat, i la diabetis són les causes més comunes d'aquesta malaltia. Aquest estat del fetge és reversible si es canvia la dieta del pacient, no obstant això, si aquest no es cuida, la malaltia pot anar avançant cap a estadis més severs, desencadenant fibrosis, cirrosis i fins i tot carcinoma hepatocel·lular (CHC). La primerenca detecció de tots aquests processos és de gran importància en la millora del pronòstic dels pacients. Així, les tècniques d'imatge en combinació amb models computacionals permeten caracteritzar el teixit mitjançant l'extracció paràmetres objectius, coneguts com biomarcadores d'imatge, relacionats amb aquests processos fisiològics i patològics, permetent una estratificació més precisa de les malalties. A més, gràcies a les tècniques d'intel·ligència artificial, es poden desenvolupar algorismes de segmentació automàtica que permeten realitzar aquesta caracterització de manera completament automàtica i accelerar, d'aquesta manera, el flux radiològic. Per tot això, en la present tesi doctoral, es presenta una metodologia per al desenvolupament de models de segmentació i quantificació automàtica, sent aplicada a tres casos d'ús. Per a l'estudi de la síndrome metabòlica es proposa un mètode de segmentació automàtica del greix visceral i subcutani en imatges de tomografia computada (TC), per a l'estudi de la malaltia hepàtica difusa es proposa un mètode segmentació hepàtica i quantificació del greix i ferro hepàtics en imatges de ressonància magnètica (RM), i, finalment, per a l'estudi del CHC, es proposa un mètode de segmentació hepàtica i quantificació dels descriptors de la corba de perfusió en imatges de RM. Tot això s'ha integrat en una plataforma que permet la seua integració en la pràctica clínica. Així, s'han adaptat els algorismes desenvolupats per a ser executats en contenidors Docker de manera que, donada una imatge d'entrada, generen els paràmetres quantitatius d'eixida juntament amb un informe que resumisca aquests resultats; s'han implementat eines perquè els usuaris puguen interactuar amb les segmentacions generades pels algorismes de segmentació automàtica desenvolupats; finalment, aquests s'han implementat de manera que generen aquestes segmentacions en formats estàndard com DICOM RT Struct o DICOM Seg, per a garantir la interoperabilitat amb la resta de sistemes sanitaris. / [EN] Metabolic syndrome is defined as a group of disorders (e.g., high blood pressure, high blood glucose levels, excess abdominal fat, or high cholesterol or triglyceride levels) that affect an individual at the same time. The presence of one of these factors does not imply an elevated health risk; however, having several of them increases the probability of secondary diseases such as cardiovascular disease or type II diabetes. Diffuse liver diseases are all those diseases that affect the functional cells of the liver, the hepatocytes, thus altering liver function. In these processes, the hepatocytes are replaced by adipocytes and fibrous tissue. Non-alcoholic fatty liver disease is a reversible condition caused by the accumulation of triglycerides in hepatocytes. Alcoholism, obesity, and diabetes are the most common causes of this disease. This liver condition is reversible if the patient's diet is changed; however, if the patient is not cared for, the disease can progress to more severe stages, triggering fibrosis, cirrhosis and even hepatocellular carcinoma (HCC). Early detection of all these processes is of great importance in improving patient prognosis. Thus, imaging techniques in combination with computational models allow tissue characterization by extracting objective parameters, known as imaging biomarkers, related to these physiological and pathological processes, allowing a more accurate statification of diseases. Moreover, thanks to artificial intelligence techniques, it is possible to develop automatic segmentation algorithms that allow to perform such characterization in a fully automatic way and thus accelerate the radiological workflow. Therefore, in this PhD, a methodology for the development of automatic segmentation and quantification models is presented and applied to three use cases. For the study of metabolic syndrome, a method of automatic segmentation of visceral and subcutaneous fat in computed tomography (CT) images is proposed; for the study of diffuse liver disease, a method of liver segmentation and quantification of hepatic fat and iron in magnetic resonance imaging (MRI) is proposed; and, finally, for the study of HCC, a method of liver segmentation and quantification of perfusion curve descriptors in MRI is proposed. All this has been integrated into a platform that allows its integration into clinical practice. Thus, the developed algorithms have been adapted to be executed in Docker containers so that, given an input image, they generate the quantitative output parameters together with a report summarizing these results; tools have been implemented so that users can interact with the segmentations generated by the automatic segmentation algorithms developed; finally, these have been implemented so that they generate these segmentations in standard formats such as DICOM RT Struct or DICOM Seg, to ensure interoperability with other health systems. / Jimenez Pastor, AM. (2023). Aprendizaje profundo y biomarcadores de imagen en el estudio de enfermedades metabólicas y hepáticas a partir de resonancia magnética y tomografía computarizada [Tesis doctoral]. Universitat Politècnica de València. https://doi.org/10.4995/Thesis/10251/202602