Osteoporosis is a bone disease associated with fracture risk. Accurate assessments of fracture risk, guidelines to initiate preventive intervention, and monitoring treatment response are of paramount importance in public health. Clinically, osteoporosis is defined by low bone mineral density, which explains 65-75% of the variance in bone stiffness. The remaining variability is due to the cumulative and synergistic effects of various factors, including trabecular bone micro-architecture. Osteoporostic imaging is critically important in identifying fracture risks for planning of therapeutic intervention and monitoring response to treatments. In this work, quantitative analysis of trabecular bone micro-architecture using volumetric imaging techniques and computational biomechanical simulation through finite element modeling (FEM) are applied on in vivo imaging for various human studies. The ability of imaging methods in characterizing trabecular bone micro-architecture was experimentally examined using MRI and multi-row detector CT. They were found suitable for cross-sectional and longitudinal studies in monitoring changes of trabecular micro-architectural quality in clinical research. A framework which consists of robust segmentation of in vivo images and quality mesh generator, was constructed for FEM analysis. The framework was experimentally demonstrated effcient and effective to predict bone strength under limited spatial resolution. The ability of distinguishing bone strengths of different groups were evaluated on various human studies. And the relation between FEM and image-based micro-architectural measures was explored. Quantitative analysis supports the hypothesis that trabecular bone have distinct structural properties in different anatomic sites and the osteoporosis related change of the micro-architecture also varies. It highlight the importance of standardizing the definition of bone scan locations and the segmentation of such well-defined regions. A shape modeling method was proposed to solve the problem and its application in human proximal femur using MRI were presented. The method was compared with manual segmentation and found highly accurate. Together with tools developed for quantitative analysis, this work facilitates future researches of trabecular bone micro-architecture in different anatomic sites.

In this work several algorithms for diaphragm detection in 2D views of cone-beam computed tomography (CBCT) raw data are developed. These algorithms are tested on 21 Siemens megavoltage CBCT scans of lungs and the result is compared against the diaphragm apex identified by human experts. Among these algorithms dynamic Hough transform is sufficiently quick and accurate for motion determination prior to radiation therapy. The diaphragm was successfully detected in all 21 data sets, even for views with poor image quality and confounding objects. Each CBCT scan analysis (200 frames) took about 38 seconds on a 2.66 GHz Intel quad-core 2 CPU. The average cranio-caudal position error was 1.707 ± 1.117 mm. Other directions were not assessed due to uncertainties in expert identification.

Les implants cochléaires (IC) sont utilisés pour traiter la surdité profonde en insérant chirurgicalement un réseau d'électrodes dans l'organe de l'audition, la cochlée. Les images tomodensitométriques (TDM) pré et post-opératoires sont utilisées couramment pour la planification chirurgicale et l'évaluation de l'implantation cochléaire. Cependant, en raison de la petite taille et de la topologie complexe de la cochlée, l'information anatomique qui peut être extraite des images est limitée. Le premier axe de ce travail vise à définir des méthodes automatiques de traitement d'images adaptées à la forme en spirale de la cochlée pour étudier en étudier la variabilité à partir d'images de micro-TDM (μTDM) haute résolution. Le deuxième axe vise à développer et à évaluer un nouveau modèle paramétrique de forme cochléaire. Le modèle est appliqué pour extraire des paramètres cliniquement pertinents spécifiques au patient, tels que la profondeur d'insertion maximale des portes électrode. Grâce à la quantification de l'incertitude, fournie par le modèle, la fiabilité des segmentations issues de TDM a pu être évaluée par rapport à la vérité terrain fournie par μTDM. Enfin, le dernier axe concerne un modèle de forme cochléaire (et de ses sous-structures) et d'apparence combiné dans un cadre bayésien probabiliste génératif. La méthode de segmentation proposée a été appliquée à une grande base de données de 987 images de TDM et a permis la caractérisation statistique de la variabilité anatomique cochléaire ainsi que la quantification de la symétrie bilatérale. Ce travail ouvre la voie à de nouvelles applications cliniques telles que l'amélioration du diagnostic en identifiant les formes cochléaires pathologiques ; la planification préopératoire du choix de l'électrode et de l'axe d'insertion ; l'estimation postopératoire de la position de l'électrode et évaluation de l'implantation ; et la simulation d'implantation cochléaire. / Cochlear implants (CI) are used to treat hearing loss by surgically inserting an electrode array into the organ of hearing, the cochlea. Pre- and post-operative CT images are used routinely for surgery planning and evaluation of cochlear implantation. However, due to the small size and the complex topology of the cochlea, the anatomical information that can be extracted from the images is limited. The first focus of this work aims at defining automatic image processing methods adapted to the spiral shape of the cochlea to study the cochlear shape variability from high-resolution μCT images. The second focus aims at developing and evaluating a new parametric cochlear shape model. The model is applied to extract patient-specific clinically relevant metrics such as the maximal insertion depth of CI electrode arrays. Thanks to the uncertainty quantification, provided by the model, we can assess the reliability of CT-based segmentation as compared to the ground truth segmentation provided by μCT scans. Finally, the last focus concerns a joint model of the cochlear shape (and its substructures) model and its appearance within a generative probabilistic Bayesian framework. The proposed segmentation method was applied to a large database of 987 CT images and allowed the statistical characterization of the cochlear anatomical variability along with the quantification of the bilateral symmetry. This work paves the way to novel clinical applications such as improved diagnosis by identifying pathological cochlear shapes; preoperative optimal electrode design and insertion axis planning; postoperative electrode position estimation and implantation evaluation; and cochlear implantation simulation.

[pt] As expressões faciais transmitem muita informação sobre um indivíduo, tornando a capacidade de interpretá-las uma tarefa muito importante, com aplicações em diversas áreas, tais como Interação Homem Máquina, Jogos Digitais, storytelling interativo e TV/Cinema digital. Esta dissertação discute o processo de reconhecimento de emoções em tempo real usando ASM (Active Shape Model) e SVM (Support Vector Machine) e apresenta uma comparação entre duas formas comumente utilizadas na etapa de extração de atributos: faces neutra e média. Como não existe tal comparação na literatura, os resultados apresentados são valiosos para o desenvolvimento de aplicações envolvendo expressões de emoção em tempo real. O presente trabalho considera seis tipos de emoções: felicidade, tristeza, raiva, medo, surpresa e desgosto. / [en] The facial expressions provide a high amount of information about a person, making the ability to interpret them a high valued task that can be used in several fields of Informatics such as Human Machine Interface, Digital Games, interactive storytelling and digital TV/Cinema. This dissertation discusses the process of recognizing emotions in real time using ASM (Active Shape Model) and SVM (Support Vector Machine) and presents a comparison between two commonly used ways when extracting the attributes: neutral face and average. As such comparison can not be found in the literature, the results presented are valuable to the development of applications that deal with emotion expression in real time. The current study considers six types of emotions: happiness, sadness, anger, fear, surprise and disgust.

In the field of craniomaxillofacial (CMF) surgery, surgical planning is an important and necessary procedure due to the complex nature of the craniofacial skeleton. Computed tomography (CT) has brought about a revolution in virtual diagnosis, surgical planning and simulation, and evaluation of treatment outcomes. It provides high-quality 3D image and model of skull for Computer-aided surgical planning system (CSPS). During the planning process, one of the essential steps is to reestablish the dental occlusion. In the first project, a new approach is presented to automatically and efficiently reestablish dental occlusion. It includes two steps. The first step is to initially position the models based on dental curves and a point matching technique. The second step is to reposition the models to the final desired occlusion based on iterative surface-based minimum distance mapping with collision constraints. With linearization of rotation matrix, the alignment is modeled by solving quadratic programming. The simulation was completed on 12 sets of digital dental models. Two sets of dental models were partially edentulous, and another two sets have first premolar extractions for orthodontic treatment. Two validation methods were applied to the articulated models. The results show that using the proposed method, the dental models can be successfully articulated with a small degree of deviations from the occlusion achieved with the gold-standard method. Low contrast resolution in CBCT image has become its major limitation in building skull model. Intensive hand-segmentation is required to reconstruct the skull model. Thin bone images are particularly affected by this limitation. In the second project, a novel segmentation approach is presented based on wavelet active shape model (WASM) for a particular interest in the outer surface of the anterior wall of maxilla. 19 CBCT datasets are used to conduct two experiments. This model-based segmentation approach is validated and compared with three different segmentation approaches. The results show that the performance of this model-based segmentation approach is better than those of the other approaches. It can achieve 0.25 +/- 0.2mm of surface error distance from the ground truth of the bone surface. Field of view (FOV) can be reduced in order to reduce unnecessary radiation dose in CBCT. This ROI imaging is common in most of the dentomaxillofacial imaging and orthodontic practices. However, a truncation effect is created due to the truncation of projection images and becomes one of the limitation in CBCT. In the third project, a method for small region of interest (ROI) imaging and reconstruction of the image of ROI in CBCT and two experiments for measurement of dosage are presented. The first experiment shows at least 60% and 70% of radiation dose can be reduced. It also demonstrates that the image quality was still acceptable with little variation of gray by using the traditional truncation correction approach for ROI imaging. The second experiment demonstrates that the images reconstructed by CBCT reconstruction algorithms without truncation correction can be degraded to unacceptable image quality.

Optimal surface finding (OSF), a graph-based optimization approach to image segmentation, represents a powerful framework for medical image segmentation and analysis. In many applications, a pre-segmentation is required to enable OSF graph construction. Also, the cost function design is critical for the success of OSF. In this thesis, two issues in the context of OSF segmentation are addressed. First, a robust model-based segmentation method suitable for OSF initialization is introduced. Second, an OSF-based segmentation refinement approach is presented. For segmenting complex anatomical structures (e.g., lungs), a rough initial segmentation is required to apply an OSF-based approach. For this purpose, a novel robust active shape model (RASM) is presented. The RASM matching in combination with OSF is investigated in the context of segmenting lungs with large lung cancer masses in 3D CT scans. The robustness and effectiveness of this approach is demonstrated on 30 lung scans containing 20 normal lungs and 40 diseased lungs where conventional segmentation methods frequently fail to deliver usable results. The developed RASM approach is generally applicable and suitable for large organs/structures. While providing high levels of performance in most cases, OSF-based approaches may fail in a local region in the presence of pathology or other local challenges. A new (generic) interactive refinement approach for correcting local segmentation errors based on the OSF segmentation framework is proposed. Following the automated segmentation, the user can inspect the result and correct local or regional segmentation inaccuracies by (iteratively) providing clues regarding the location of the correct surface. This expert information is utilized to modify the previously calculated cost function, locally re-optimizing the underlying modified graph without a need to start the new optimization from scratch. For refinement, a hybrid desktop/virtual reality user interface based on stereoscopic visualization technology and advanced interaction techniques is utilized for efficient interaction with the segmentations (surfaces). The proposed generic interactive refinement method is adapted to three applications. First, two refinement tools for 3D lung segmentation are proposed, and the performance is assessed on 30 test cases from 18 CT lung scans. Second, in a feasibility study, the approach is expanded to 4D OSF-based lung segmentation refinement and an assessment of performance is provided. Finally, a dual-surface OSF-based intravascular ultrasound (IVUS) image segmentation framework is introduced, application specific segmentation refinement methods are developed, and an evaluation on 41 test cases is presented. As demonstrated by experiments, OSF-based segmentation refinement is a promising approach to address challenges in medical image segmentation.

In recent years, image segmentation by using deep neural networks has made great progress. However, reaching a good result by training with a small amount of data remains to be a challenge. To find a good way to improve the accuracy of segmentation with limited datasets, we implemented a new automatic chest radiographs segmentation experiment based on preliminary works by Chunliang using deep learning neural network combined with shape context information. When the process was conducted, the datasets were put into origin U-net at first. After the preliminary process, the segmented images were then repaired through a new network with shape context information. In this experiment, we created a new network structure by rebuilding the U-net into a 2-input structure and refined the processing pipeline step. In this proposed pipeline, the datasets and shape context were trained together through the new network model by iteration. The proposed method was evaluated on 247 posterior-anterior chest radiographs of public datasets and n-folds cross-validation was also used. The outcome shows that compared to origin U-net, the proposed pipeline reaches higher accuracy when trained with limited datasets. Here the "limited" datasets refer to 1-20 images in the medical image field. A better outcome with higher accuracy can be reached if the second structure is further refined and shape context generator's parameter is fine-tuned in the future. / Under de senaste åren har bildsegmentering med hjälp av djupa neurala nätverk gjort stora framsteg. Att nå ett bra resultat med träning med en liten mängd data kvarstår emellertid som en utmaning. För att hitta ett bra sätt att förbättra noggrannheten i segmenteringen med begränsade datamängder så implementerade vi en ny segmentering för automatiska röntgenbilder av bröstkorgsdiagram baserat på tidigare forskning av Chunliang. Detta tillvägagångssätt använder djupt lärande neurala nätverk kombinerat med "shape context" information. I detta experiment skapade vi en ny nätverkstruktur genom omkonfiguration av U-nätverket till en 2-inputstruktur och förfinade pipeline processeringssteget där bilden och "shape contexten" var tränade tillsammans genom den nya nätverksmodellen genom iteration.Den föreslagna metoden utvärderades på dataset med 247 bröströntgenfotografier, och n-faldig korsvalidering användes för utvärdering. Resultatet visar att den föreslagna pipelinen jämfört med ursprungs U-nätverket når högre noggrannhet när de tränas med begränsade datamängder. De "begränsade" dataseten här hänvisar till 1-20 bilder inom det medicinska fältet. Ett bättre resultat med högre noggrannhet kan nås om den andra strukturen förfinas ytterligare och "shape context-generatorns" parameter finjusteras.

This dissertation thesis deals with the automatic system for basic emotional facial expressions recognition from static images. Generally the system is divided into the three independent parts, which are linked together in some way. The first part deals with automatic face detection from color images. In this part they were proposed the face detector based on skin color and the methods for eyes and lips position localization from detected faces using color maps. A part of this is modified Viola-Jones face detector, which was even experimentally used for eyes detection. The both face detectors were tested on the Georgia Tech Face Database. Another part of the automatic system is features extraction process, which consists of two statistical methods and of one method based on image filtering using set of Gabor’s filters. For purposes of this thesis they were experimentally used some combinations of features extracted using these methods. The last part of the automatic system is mathematical classifier, which is represented by feed-forward neural network. The automatic system is utilized by adding an accurate particular facial features localization using active shape model. The whole automatic system was benchmarked on recognizing of basic emotional facial expressions using the Japanese Female Facial Expression database.

La sécurité routière est un enjeu majeur de santé publique et de protection des personnes. D'après l'organisation mondiale de la santé (OMS), près de 1,2 millions de personnes meurent chaque année dans le monde suite à des accidents de la route (2015). D’après des données accidentologiques, 36,7% des blessures graves ont pour origine des lésions au thorax (Page et collab., 2012). La biomécanique en sécurité passive a pour rôle d'améliorer notre compréhension du corps humain dans le but de construire de meilleurs outils pour évaluer le risque de blessure.Les modèles numériques d'être humain sont employés pour simuler virtuellement les conditions d'un accident. Aujourd'hui, ils sont de plus en plus utilisés par les constructeurs automobiles et équipementiers pour mieux comprendre les mécanismes lésionnels. Cependant, ils n’existent que dans certaines tailles et ne prennent alors pas en compte les variations morphologiques observées dans la population.L'imagerie médicale 3D donne accès aux géométries des différentes structures anatomiques composant le corps humain. Les hôpitaux regorgent aujourd'hui de quantités d'images 3D couvrant une très large partie de la population en termes d'âge, de corpulence et de sexe.L’objectif global de cette thèse est de modéliser statistiquement la géométrie 3D de la cage thoracique à partir d'images médicales afin de personnaliser un modèle numérique de corps humain pour simuler par éléments finis des conditions de choc automobile. Le premier objectif est d’élaborer un protocole de segmentation une base de CT-scans de manière à obtenir des données géométriques adaptées à la construction d’un modèle statistique de forme de la cage thoracique.Le deuxième objectif est de construire un modèle statistique de forme de la cage thoracique, en prenant en compte sa structure articulée.Le troisième objectif est d’utiliser le modèle statistique de la cage thoracique pour déformer un modèle numérique d’être humain, de manière à étudier l’influence de certains paramètres sur le risque de blessure. / Road safety is a major issue of public health and personal safety. According to the World Health Organization (WHO), nearly 1.2 million people die each year worldwide due to road accidents (2015). According to accident data, 36.7% of serious injuries are caused by thoracic injuries (Page et al., 2012). The aim of biomechanics in passive safety is to improve our understanding of the human body in order to build better tools for assessing the risk of injury.Numerical human body models are used to virtually simulate the conditions of an accident. Today, they are increasingly used by car manufacturers and equipment manufacturers to better understand injury mechanisms. However, they exist only in few sizes and do not take into account the morphological variations observed in the population.3D medical imaging gives access to the geometries of the different anatomical structures that make up the human body. Today, hospitals are full of 3D images covering a very large part of the population in terms of age, body size and sex.The overall objective of this thesis is to statistically model the 3D geometry of the rib cage from medical images in order to personalize a numerical human body model to simulate car crash conditions.The first objective is to develop a segmentation process based on CT-scans in order to obtain geometric data adapted to the construction of a statistical model of shape of the rib cage.The second objective is to build a statistical model of the shape of the rib cage, taking into account its articulated structure.The third objective is to use the statistical model of the rib cage to deform a numerical human body model, in order to study the influence of certain parameters on the risk of injury.

Medical imaging technologies such as MRI, CT, PET, etc. enable the use of higher resolution 3D digital image data for research and clinical treatment. The new technologies provide improved spatial resolution at the cost of increased data processing time. Manual identification of anatomical landmarks is still a common practice in many neuroimaging and other medical imaging applications but it is labor-intensive, subjective, and suffers from intra-/inter- rater inconsistency. This work explored one way of estimating a landmark constellation automatically, consistently, and efficiently. The proposed method demonstrated a successful application on how to effectively utilize image processing in tackling clinical challenges. It is shown that the cooperation of spatial localization using linear model prediction with evolutionary principal components and local search estimation using statistical shape models is capable of effectively extracting important landmark detection information from both morphometric relationships of landmarks and consistent intensity distribution of images. It is accurate (compared to 1.6 mm root mean squared errors of manual labeling of brain landmarks), consistent, reliable in predicting many salient midbrain point landmarks such as ac, pc, MPJ, etc. in a longitudinal, multisubject environment, and throughout large datasets with different modalities and image information such as orientation, spacing, and origin. The framework of linear model estimation method using evolutionary principal components and the idea of local search using statistical shape models are generalized to the detection task for arbitrary number of landmarks in other organs, creatures, or even any other physical objects in the world as long as the landmarks present intensity consistency and satisfy regularity in spatial organization.