Conselho Nacional de Desenvolvimento CientÃfico e TecnolÃgico / A anÃlise de imagens mÃdicas por meio de tÃcnicas de visÃo computacional tornou-se bastante promissora, principalmente pelo fato de aperfeiÃoar a acurÃcia diagnÃstica de diversas patologias. Por essas razÃo, a Pneumologia à considerada atualmente uma Ãrea de concentraÃÃo de projetos que envolvem mÃtodos de Processamento Digital de Imagens. A segmentaÃÃo de vasos sanguÃneos pulmonares à de bastante auxÃlio na detecÃÃo de cardiopatias pulmonares. Esse processo à realizado atravÃs da anÃlise dos resultados obtidos por exame de diagnÃstico por imagem, os quais se destacam as radiografias torÃcicas, tomografia computadorizada (TC) do tÃrax, ressonÃncia magnÃtica, cintilografia pulmonar e angiografia. A hipertensÃo pulmonar e o cÃncer sÃo exemplos de doenÃas que podem ser diagnosticadas com menor subjetividade ao realizar a segmentaÃÃo de vasos, visualizaÃÃo em trÃs dimensÃes e extraÃÃo de seus atributos. Devido a essa importÃncia, diversos algoritmos sÃo desenvolvidos com intuito de obter uma segmentaÃÃo Ãtima destas estruturas. Dentre estes, encontram-se os mÃtodos por contornos ativos, LÃgica Fuzzy, Crescimento de RegiÃes, Filtragem Multi-escalar 3D e algoritmo Expectation Maximization (EM). Nesta dissertaÃÃo, sÃo segmentados os vasos sanguÃneos pulmonares de imagens de tomografia computadorizada do tÃrax utilizando-se trÃs mÃtodos: uma combinaÃÃo de Crescimento de RegiÃes 3D controlado por uma funÃÃo de pertinÃncia gaussiana e limiarizaÃÃo; um mÃtodo hÃbrido de segmentaÃÃo por Conectividade Fuzzy e limiarizaÃÃo; por m, a segmentaÃÃo utilizando o classicador K-mÃdias. Os resultados obtidos pelas segmentaÃÃes sÃo analisados e comparados por meio de uma anÃlise dos coecientes de similaridade e sensibilidade. Os resultados da aplicaÃÃo dos trÃs mÃtodos sÃo caracterizados aceitÃveis e compatÃveis com os observados na literatura. / Medical image analysis using computer vision techniques has become quite promising because of its improvement on the diagnostic accuracy of various pathologies. For this reason, pulmonology became an area of high concentration of projects involving methods of Digital Image Processing. The blood vessels segmentation in the lung is an important aid in the detection of pulmonary heart diseases. This process is performed by analyzing the results obtained with known diagnostic imaging exams, like chest Xrays, computed tomography (CT) scan, magnetic resonance imaging, scintigraphy and angiography. Pulmonary hypertension and cancer are examples of diseases that can be diagnosed with less subjectivity if performing vessels segmentation, three-dimensional visualization and attribute extraction of these images. Thus, several algorithms are developed with the objective of obtaining an optimal segmentation of these structures. Among those algorithms are active contours, fuzzy logic, 3D Region Growing, 3D multi-scale ltering algorithm and Expectation Maximization (EM). In this study, the blood vessels were extracted from lung CT scans of the chest using three methods. The rst is a combination of 3D Region Growing controlled by a Gaussian membership function and thresholding, the second is a hybrid segmentation by thresholding and Fuzzy Connectedness. Finally,the third refers to segmentation using the K-means classier. The results and evaluation of applying these algorithms are presented.

Medical imaging is an important tool for diagnosis and treatment planning today. However as the demand for efficiency increases at the same time as the data volumes grow immensely, the need for computer assisted analysis, such as image segmentation, to help and guide the practitioner increases. Medical image segmentation could be used for various different tasks, the localization and delineation of pathologies such as cancer tumors is just one example. Numerous problems with noise and image artifacts in the generated images make the segmentation a difficult task, and the developer is forced to choose between speed and performance. In clinical practise, however, this is impossible as both speed and performance are crucial. One solution to this problem might be to involve the user more in the segmentation, using interactivite algorithms where the user might influence the segmentation for an improved result. This thesis has concentrated on finding a fast and interactive segmentation method for liver tumor segmentation. Various different methods were explored, and a few were chosen for implementation and further development. Two methods appeared to be the most promising, Bayesian Region Growing (BRG) and Level Set. An interactive Level Set algorithm emerged as the best alternative for the interactivity of the algorithm, and could be used in combination with both BRG and Level Set. A new data term based on a probability model instead of image edges was also explored for the Level Set-method, and proved to be more promising than the original one. The probability based Level Set and the BRG method both provided good quality results, but the fastest of the two was the BRG-method, which could segment a tumor present in 25 CT image slices in less than 10 seconds when implemented in Matlab and mex-C++ code on an ACPI x64-based PC with two 2.4 GHz Intel(R) Core(TM) 2CPU and 8 GB RAM memory. The interactive Level Set could be succesfully used as an interactive addition to the automatic method, but its usefulness was somewhat reduced by its slow processing time ( 1.5 s/slice) and the relative complexity of the needed user interactions.

Laser scanners in combination with devices for accurate orientation like Coordinate Measuring Machines (CMM) are often used in Geometric Reverse Engineering (GRE) to measure point data. The industrial robot as a device for orientation has relatively low accuracy but the advantage of being numerically controlled, fast, flexible, rather cheap and compatible with industrial environments. It is therefore of interest to investigate if it can be used in this application. This thesis will describe a measuring system consisting of a laser profile scanner mounted on an industrial robot with a turntable. It will also give an introduction to Geometric Reverse Engineering (GRE) and describe an automatic GRE process using this measuring system. The thesis also presents a detailed accuracy analysis supported by experiments that show how 2D profile data can be used to achieve a higher accuracy than the basic accuracy of the robot. The core topic of the thesis is the investigation of a new technique for planar segmentation. The new method is implemented in the GRE system and compared with an implementation of a more traditional method. Results from practical experiments show that the new method is much faster while equally accurate or better.

Background One of the most used treatments for hyperthyroidism, is therapy with radioactive iodine (131I), which is accumulated in the thyroid gland. To determine the activity of 131I to be administered for a certain absorbed dose, the volume of the gland is of great importance but the historically used methods for estimating the functional volume of the gland are based on large approximations. The use of SPECT images enables increased accuracy of functional volume determination. However, there is a need for more realistic phantom studies and improved image segmentation. Aim The aim of this thesis was to find a robust method for image segmentation of the thyroid gland that could adapt to various object sizes and contrasts. The aim was also to develop an accessible and flexible 3D thyroid phantom for measurements and optimisation of parameter settings. Materials and Methods Thyroid phantoms made from playdough loaded with 99mTcO4-, were placed in a neck phantom filled with 99mTcO4- solution of various concentration. SPECT and CT acquisitions of the phantoms were performed and the SPECT images were segmented using thresholding and region growing algorithms. The thresholds in the segmentation algorithms were optimised by minimisation of cost functions consisting of Dice score, against the CT-volume, and relative SPECT volume. To find thresholds that could be used on all phantom volumes and image backgrounds, two overall cost functions were optimised for high and low backgrounds respectively. The optimised thresholds were validated on another set of playdough phantoms. They were also used on a simpler plastic can phantom for comparison of the performance relative to the method used in the clinic today. Results The optimised thresholds showed a substantial divergence between the measurements, ranging from 40 to 58 % for the thresholding algorithm and from 8 to 19 % for the region growing algorithm. The overall optimised thresholds were 55 and 48 % for high and low image backgrounds for the thresholding algorithm which was selected for the validation measurements due to its lower overall cost function and high stability. The developed method indicated a higher accuracy in functional volume determination of the thyroid gland than the standard method used. Conclusions An image segmentation method for functional volume determination of the thyroid gland, that can adapt to image contrast, was developed in this thesis. The method indicates an improved accuracy for functional volume determination of thyroid glands, but more experiments would need to be conducted. The developed thyroid phantoms enable further optimisation of image segmentation parameters for various object sizes, contrasts and shapes. The results indicate that thresholds deduced from simpler phantoms may be too uncertain which might lead to overtreatment of hyperthyroidism with 131I. It was also indicated that thresholding is more suitable than region growing for image segmentation of SPECT images.

Autofluorescence retinal images are obtained with a confocal laser scanning ophthalmoscope, and used for the diagnostic of glaucoma. Glaucoma causes a gradual death of nerve cells and can cause blindness. Retina autofluorescence is caused by pigment lipofuscin, which causes cell damage. The aim of this work was to study methods suitable for segmentation of autofluorescence zones and method for tracking objects in an image. In this project was implemented algorithm of autofluorescence zone detection using method of region growing, designed and realized method for tracking autofluorescence regions.

The creation and classification of segments for object based urban land cover mapping is the key goal of this master thesis. An algorithm based on region growing and merging was developed, implemented and tested. The synergy effects of a fused data set of SAR and optical imagery were evaluated based on the classification results. The testing was mainly performed with data of the city of Beijing China. The dataset consists of SAR and optical data and the classified land cover/use maps were evaluated using standard methods for accuracy assessment like confusion matrices, kappa values and overall accuracy. The classification for the testing consists of 9 classes which are low density buildup, high density buildup, road, park, water, golf course, forest, agricultural crop and airport. The development was performed in JAVA and a suitable graphical interface for user friendly interaction was created parallel to the development of the algorithm. This was really useful during the period of extensive testing of the parameter which easily could be entered through the dialogs of the interface. The algorithm itself treats the pixels as a connected graph of pixels which can always merge with their direct neighbors, meaning sharing an edge with those. There are three criteria that can be used in the current state of the algorithm, a mean based spectral homogeneity measure, a variance based textural homogeneity measure and fragmentation test as a shape measure. The algorithm has 3 key parameters which are the minimum and maximum segments size as well as a homogeneity threshold measure which is based on a weighted combination of relative change due to merging two segments. The growing and merging is divided into two phases the first one is based on mutual best partner merging and the second one on the homogeneity threshold. In both phases it is possible to use all three criteria for merging in arbitrary weighting constellations. A third step is the check for the fulfillment of minimum size which can be performed prior to or after the other two steps. The segments can then in a supervised manner be labeled interactively using once again the graphical user interface for creating a training sample set. This training set can be used to derive a support vector machine which is based on a radial base function kernel. The optimal settings for the required parameters of this SVM training process can be found from a cross-validation grid search process which is implemented within the program as well. The SVM algorithm is based on the LibSVM java implementation. Once training is completed the SVM can be used to predict the whole dataset to get a classified land-cover map. It can be exported in form of a vector dataset. The results yield that the incorporation of texture features already in the segmentation is superior to spectral information alone especially when working with unfiltered SAR data. The incorporation of the suggested shape feature however doesn’t seem to be of advantage, especially when taking the much longer processing time into account, when incorporating this criterion. From the classification results it is also evident, that the fusion of SAR and optical data is beneficial for urban land cover mapping. Especially the distinction of urban areas and agricultural crops has been improved greatly but also the confusion between high and low density could be reduced due to the fusion. / Dragon 2 Project

Este trabalho apresenta uma metodologia de classificação de imagens multiespectrais aplicada a análise e mapeamento da evolução do uso da terra/cobertura vegetal em São Félix do Xingu, Sul do Pará. Imagens frações representando as proporções de sombra, vegetação e solo foram estimadas a partir das bandas 1 a 5 e 7 do Landsat-5 TM e relacionadas com as estruturas das classes de uso da terra/cobertura vegetal. As imagens frações geradas do modelo linear de mistura espectral foram importantes para reduzir a massa de dados e ao mesmo tempo realçar alvos de interesse na imagem. A banda do infravermelho próximo (TM-4) foi importante para realçar áreas de queimadas. A classificação adotada foi divida em etapas combinando técnica de segmentação por crescimento de regiões e uso de máscaras. Por meio da máscara foi possível restringir o processo de segmentação em regiões pré-estabelecidas com o intuito de adquirir um melhor particionamento da imagem. Adotando este procedimento, ao invés de realizar uma única segmentação para mapear todas as classes em uma única vez, foram realizadas várias segmentações ao longo das etapas. As regiões segmentadas foram agrupadas com um classificador não-supervionado batizado de ISOSEG. Os resultados mostram que a metodologia é bastante eficiente. A matriz de erro gerada para a classificação de 2008 apontou que as confusões mais freqüentes ocorreram entre as classes que apresentaram em certas localidades proporções de misturas parecidas: Capoeira e Campo/Pastagem-2; Campo/Pastagem-1 e Campo/Pastagem-2; Queimada-1 e Queimada-2; Solo Exposto e Campo/Pastagem-1. Considerando nove classes, o índice Kappa atingiu 0,58, o que representa um valor de concordância classificada como moderada. Quando o numero de classes foi reduzido para 6, agrupando as classes que apresentaram as maiores confusões, o índice Kappa subiu para 0,80, atingindo um valor de concordância quase perfeita. A comparação dos resultados das classificações de 1987, 1992, 2000 e 2008 juntamente com a analise de dados auxiliares permitiu traçar um modelo de evolução do desmatamento e do uso da terra para São Félix do Xingu. O intenso desmatamento observado principalmente a partir de 2000 foi relacionado com o incremento da atividade pecuária, sendo São Félix do Xingu o município que detém atualmente o segundo maior rebanho bovino do País. / In this work we present a methodological procedure for multi-spectral images classification to evaluate and map land-use and land-cover changes in São Félix do Xingu, Southern Pará (Brazilian Amazon). Fraction images representing shade, vegetation and soil abundance at the pixel scale were estimated using all six reflective bands of Thematic Mapper sensor (TM-1 to TM-5 and TM-7) and related to different types of land-use and land-cover classes. The linear spectral mixing analysis method was an alternative approach adopted to reduce the data-dimensionality while at the same time enhancing targets of interest. Also, the near-infrared band (TM-4) was employed to separate areas affected by burns (Queimadas in Portuguese). The classification routines were performed in stages by combining region-growing segmentation and use of masking techniques. For each stage, the segmentation process was directed to preselected areas by masking techniques in order to obtain a better image partitioning. This procedure resulted in more than one segmentation thereby reducing confusing errors during the classification routine. An unsupervised classifier by region named ISOSEG was employed to classify the segmented images. The analysis of classification results was mainly qualitative and visual except for the 2008 classification which was assessed through an error matrix. According to the error matrix analysis, misclassifications arose more frequently when a set of classes with similar mixture proportions were involved, such as: Capoeira and Campo/Pastagem-2; Campo/Pastagem-1 and Campo/Pastagem- 2; Queimada-1 and Queimada-2, and finally Bare Soil and Campo/Pastagem-1. As a robust measure of concordance for dichotomous data, the kappa statistic reached a value of 0.62 by considering nine land types of classes and it rose to 0.80 when the mapping classes were diminished to six. Theses kappa values represent moderate and strong agreements between the remotely sensed classification and the reference data, respectively. Making use of the classification results from 1987, 1992, 2000 and 2008 and auxiliary data, we tried to design a simple land evolution model to São Félix do Xingu. The deforestation process notably intensified since 2000 has been driven mainly by a continuous increase in cattle breeding, for wich São Félix do Xingu has the second-largest cattle herd of all Brazilian municipalities.

Advancement in Image Acquisition Equipment and progress in Image Processing Methods have brought the mathematicians and computer scientists into areas which are of huge importance for physicians and biologists. Early diagnosis of diseases like blindness, cancer and digestive problems have been areas of interest in medicine. Development of Laser Photon Microscopy and other advanced equipment already provides a good idea of very interesting characteristics of the object being viewed. Still certain images are not suitable to extract sufficient information out of that image. Image Processing methods have been providing good support to provide useful information about the objects of interest in these biological images. Fast computational methods allow complete analysis, in a very short time, of a series of images, providing a reasonably good idea about the desired characteristics. The thesis covers application of these methods in 3 series of images intended for 3 different types of diagnosis or inference. Firstly, Images of RP-mutated retina were treated for detection of rods, where there were no cones present. The software was able to detect and count the number of cones in each frame. Secondly, a gastrulation process in drosophila was studied to observe any mitosis and results were consistent with recent research. Finally, another series of images were treated where biological cells were observed to undergo mitosis. The source was a video from a photon laser microscope. In this video, objects of interest were biological cells. The idea was to track the cells if they undergo mitosis. Cell position, spacing and sometimes contour of the cell membrane are broadly the factors limiting the accuracy in this video. Appropriate method of image enhancement and segmentation were chosen to develop a computational method to observe this mitosis. Cases where human intervention may be required have been proposed to eliminate any false inference.

The objective of this thesis is to develop a general obstacle segmentation algorithm for use on board a ground based unmanned vehicle (GUV). The algorithm processes video data captured by a single monocular camera mounted on the GUV. We make the assumption that the GUV moves on a locally planar surface, representing the ground plane. We start by deriving the equations of the expected motion field (observed by the camera) induced by the motion of the robot on the ground plane. Given an initial view of a presumably static scene, this motion field is used to generate a predicted view of the same scene after a known camera displacement. This predicted image is compared to the actual image taken at the new camera location by means of an optical flow calculation. Because the planar assumption is used to generate the predicted image, portions of the image which mismatch the prediction correspond to salient feature points on objects which lie above or below the ground plane, we consider these objects obstacles for the GUV. We assume that these salient feature points (called seed pixels ) capture the color statistics of the obstacle and use them to initialize a Bayesian region growing routine to generate a full obstacle segmentation. Alignment of the seed pixels with the obstacle is not guaranteed due to the aperture problem, however successful segmentations were obtained for natural scenes. The algorithm was tested off line using video captured by a camera mounted on a GUV.

Avancement dans l'acquisition d'image et le progrès dans les méthodes de traitement d'image ont apporté les mathématiciens et les informaticiens dans les domaines qui sont d'une importance énorme pour les médecins et les biologistes. Le diagnostic précoce de maladies (comme la cécité, le cancer et les problèmes digestifs) ont été des domaines d'intérêt en médecine. Développement des équipements comme microscope bi-photonique à balayage laser et microscope de fluorescence par réflexion totale interne fournit déjà une bonne idée des caractéristiques très intéressantes sur l'objet observé. Cependant, certaines images ne sont pas appropriés pour extraire suffisamment d'informations sur de cette image. Les méthodes de traitement d'image ont été fournit un bon soutien à extraire des informations utiles sur les objets d'intérêt dans ces images biologiques. Rapide méthodes de calcul permettent l'analyse complète, dans un temps très court, d'une série d'images, offrant une assez bonne idée sur les caractéristiques souhaitées. La thèse porte sur l'application de ces méthodes dans trois séries d'images destinées à trois différents types de diagnostic ou d'inférence. Tout d'abord, Images de RP-muté rétine ont été traités pour la détection des cônes, où il n'y avait pas de bâtonnets présents. Le logiciel a été capable de détecter et de compter le nombre de cônes dans chaque image. Deuxièmement, un processus de gastrulation chez la drosophile a été étudié pour observer toute la mitose et les résultats étaient cohérents avec les recherches récentes. Enfin, une autre série d'images ont été traités où la source était une vidéo à partir d'un microscopie photonique à balayage laser. Dans cette vidéo, des objets d'intérêt sont des cellules biologiques. L'idée était de suivre les cellules si elles subissent une mitose. La position de la cellule, la dispersion spatiale et parfois le contour de la membrane cellulaire sont globalement les facteurs limitant la précision dans cette vidéo. Des méthodes appropriées d'amélioration de l'image et de segmentation ont été choisies pour développer une méthode de calcul pour observer cette mitose. L'intervention humaine peut être requise pour éliminer toute inférence fausse. / Advancement in Image Acquisition Equipment and progress in Image Processing Methods have brought the mathematicians and computer scientists into areas which are of huge importance for physicians and biologists. Early diagnosis of diseases like blindness, cancer and digestive problems have been areas of interest in medicine. Development of Laser Photon Microscopy and other advanced equipment already provides a good idea of very interesting characteristics of the object being viewed. Still certain images are not suitable to extract sufficient information out of that image. Image Processing methods have been providing good support to provide useful information about the objects of interest in these biological images. Fast computational methods allow complete analysis, in a very short time, of a series of images, providing a reasonably good idea about the desired characteristics. The thesis covers application of these methods in 3 series of images intended for 3 different types of diagnosis or inference. Firstly, Images of RP-mutated retina were treated for detection of rods, where there were no cones present. The software was able to detect and count the number of cones in each frame. Secondly, a gastrulation process in drosophila was studied to observe any mitosis and results were consistent with recent research. Finally, another series of images were treated where biological cells were observed to undergo mitosis. The source was a video from a photon laser microscope. In this video, objects of interest were biological cells. The idea was to track the cells if they undergo mitosis. Cell position, spacing and sometimes contour of the cell membrane are broadly the factors limiting the accuracy in this video. Appropriate method of image enhancement and segmentation were chosen to develop a computational method to observe this mitosis. Cases where human intervention may be required have been proposed to eliminate any false inference.