Brain perfusion weighted images acquired using dynamic contrast studies have an important clinical role in acute stroke diagnosis and treatment decisions. The purpose of my PhD research is to develop novel methodologies for improving the efficiency and quality of brain perfusion-imaging analysis so that clinical decisions can be made more accurately and in a shorter time. This thesis consists of three parts: My research investigates the possibility that parallel computing brings to make perfusion-imaging analysis faster in order to deliver results that are used in stroke diagnosis earlier. Brain perfusion analysis using local Arterial Input Functions (AIF) techniques takes a long time to execute due to its heavy computational load. As time is vitally important in the case of acute stroke, reducing analysis time and therefore diagnosis time can reduce the number of brain cells damaged and improve the chances for patient recovery. We present the implementation of a deconvolution algorithm for brain perfusion quantification on GPGPU (General Purpose computing on Graphics Processing Units) using the CUDA programming model. Our method aims to accelerate the process without any quality loss. Specific features of perfusion source images are also used to reduce noise impact, which consequently improves the accuracy of hemodynamic maps. The majority of existing approaches for denoising CT images are optimized for 3D (spatial) information, including spatial decimation (spatially weighted mean filters) and techniques based on wavelet and curvelet transforms. However, perfusion imaging data is 4D as it also contains temporal information. Our approach using Gaussian process regression (GPR) makes use of the temporal information in the perfusion source imges to reduce the noise level. Over the entire image, our noise reduction method based on Gaussian process regression gains a 99% contrast-to-noise ratio improvement over the raw image and also improves the quality of hemodynamic maps, allowing a better identification of edges and detailed information. At the level of individual voxels, GPR provides a stable baseline, helps identify key parameters from tissue time-concentration curves and reduces the oscillations in the curves. Furthermore, the results show that GPR is superior to the alternative techniques compared in this study. My research also explores automatic segmentation of perfusion images into potentially healthy areas and lesion areas, which can be used as additional information that assists in clinical diagnosis. Since perfusion source images contain more information than hemodynamic maps, good utilisation of source images leads to better understanding than the hemodynamic maps alone. Correlation coefficient tests are used to measure the similarities between the expected tissue time-concentration curves (from reference tissue) and the measured time-concentration curves (from target tissue). This information is then used to distinguish tissues at risk and dead tissues from healthy tissues. A correlation coefficient based signal analysis method that directly spots suspected lesion areas from perfusion source images is presented. Our method delivers a clear automatic segmentation of healthy tissue, tissue at risk and dead tissue. From our segmentation maps, it is easier to identify lesion boundaries than using traditional hemodynamic maps.

No description available.

Este projeto apresenta a pesquisa e projeto de um algoritmo computacional para realizar a avaliação volumétrica de estruturas cerebrais, a partir de imagens da Ressonância Magnética Nuclear (RMN). O algoritmo possibilita a seleção das estruturas a serem avaliadas nos vários planos do exame e, então realiza a integração das áreas calculadas para cada corte de modo a obter o volume de cada estrutura selecionada. O contorno da região de interesse em cada corte é definido pelo médico especialista de forma interativa, através da combinação de \"tresholding\" com uma função de edição manual de contorno. Os resultados obtidos indicaram a potencialidade de uso do algoritmo para o auxílio ao diagnóstico e estadiamento clínico de doenças como Mal de Alzheimer e Epilepsia do lobo temporal. O algoritmo foi considerado suficientemente preciso pelos especialistas que o testaram, mantendo a variação das medidas de volume abaixo de 1,5 mm3. O tempo médio para a realização de um estudo completo foi de aproximadamente 25 minutos. / This project presents the development of a computational algorithm focused on the volumetric measurement of cerebral structures in Nuclear Magnetic Resonance Images. The algorithm allows the selection of structures to be analysed on each exam slices and realizes the integration of calculated areas for each slice to get a volumetric measurement of each selected structure. The edge of interest region in each slice must be interatively defined for a specialist, based on a combination of thresholding with a manual segmentation function. The results points to the potenciality of the algorithm aplication to aid the diagnosis of diseases as Alzheimer\'s disease and temporal lobe epilepsy.

Este projeto apresenta a pesquisa e projeto de um algoritmo computacional para realizar a avaliação volumétrica de estruturas cerebrais, a partir de imagens da Ressonância Magnética Nuclear (RMN). O algoritmo possibilita a seleção das estruturas a serem avaliadas nos vários planos do exame e, então realiza a integração das áreas calculadas para cada corte de modo a obter o volume de cada estrutura selecionada. O contorno da região de interesse em cada corte é definido pelo médico especialista de forma interativa, através da combinação de \"tresholding\" com uma função de edição manual de contorno. Os resultados obtidos indicaram a potencialidade de uso do algoritmo para o auxílio ao diagnóstico e estadiamento clínico de doenças como Mal de Alzheimer e Epilepsia do lobo temporal. O algoritmo foi considerado suficientemente preciso pelos especialistas que o testaram, mantendo a variação das medidas de volume abaixo de 1,5 mm3. O tempo médio para a realização de um estudo completo foi de aproximadamente 25 minutos. / This project presents the development of a computational algorithm focused on the volumetric measurement of cerebral structures in Nuclear Magnetic Resonance Images. The algorithm allows the selection of structures to be analysed on each exam slices and realizes the integration of calculated areas for each slice to get a volumetric measurement of each selected structure. The edge of interest region in each slice must be interatively defined for a specialist, based on a combination of thresholding with a manual segmentation function. The results points to the potenciality of the algorithm aplication to aid the diagnosis of diseases as Alzheimer\'s disease and temporal lobe epilepsy.

Arterial spin labeling (ASL) perfusion Magnetic Resonance Imaging (MRI) is a noninvasive technique for measuring quantitative cerebral blood flow (CBF) but subject to an inherently low signal-to-noise-ratio (SNR), resulting in a big challenge for data processing. Traditional post-processing methods have been proposed to reduce artifacts, suppress non-local noise, and remove outliers. However, these methods are based on either implicit or explicit models of the data, which may not be accurate and may change across subjects. Deep learning (DL) is an emerging machine learning technique that can learn a transform function from acquired data without using any explicit hypothesis about that function. Such flexibility may be particularly beneficial for ASL denoising. In this dissertation, three different machine learning-based methods are proposed to improve the image quality of ASL MRI: 1) a learning-from-noise method, which does not require noise-free references for DL training, was proposed for DL-based ASL denoising and BOLD-to-ASL prediction; 2) a novel deep learning neural network that combines dilated convolution and wide activation residual blocks was proposed to improve the image quality of ASL CBF while reducing ASL acquisition time; 3) a prior-guided and slice-wise adaptive outlier cleaning algorithm was developed for ASL MRI. In the first part of this dissertation, a learning-from-noise method is proposed for DL-based method for ASL denoising. The proposed learning-from-noise method shows that DL-based ASL denoising models can be trained using only noisy image pairs, without any deliberate post-processing for obtaining the quasi-noise-free reference during the training process. This learning-from-noise method can also be applied to DL-based ASL perfusion prediction from BOLD fMRI as ASL references are extremely noisy in this BOLD-to-ASL prediction. Experimental results demonstrate that this learning-from-noise method can reliably denoise ASL MRI and predict ASL perfusion from BOLD fMRI, result in improved signal-to-noise-ration (SNR) of ASL MRI. Moreover, by using this method, more training data can be generated, as it requires fewer samples to generate quasi-noise-free references, which is particularly useful when ASL CBF data are limited. In the second part of this dissertation, we propose a novel deep learning neural network, i.e., Dilated Wide Activation Network (DWAN), that is optimized for ASL denoising. Our method presents two novelties: first, we incorporated the wide activation residual blocks with a dilated convolution neural network to achieve improved denoising performance in term of several quantitative and qualitative measurements; second, we evaluated our proposed model given different inputs and references to show that our denoising model can be generalized to input with different levels of SNR and yields images with better quality than other methods. In the final part of this dissertation, a prior-guided and slice-wise adaptive outlier cleaning (PAOCSL) method is proposed to improve the original Adaptive Outlier Cleaning (AOC) method. Prior information guided reference CBF maps are used to avoid bias from extreme outliers in the early iterations of outlier cleaning, ensuring correct identification of the true outliers. Slice-wise outlier rejection is adapted to reserve slices with CBF values in the reasonable range even they are within the outlier volumes. Experimental results show that the proposed outlier cleaning method improves both CBF quantification quality and CBF measurement stability. / Electrical and Computer Engineering

Made available in DSpace on 2015-03-26T13:39:50Z (GMT). No. of bitstreams: 1 texto completo.pdf: 1545911 bytes, checksum: 374fbc9951e52fdc2397c84a434ed6a2 (MD5) Previous issue date: 2012-08-27 / Conselho Nacional de Desenvolvimento Científico e Tecnológico / The macaúba [Acrocomia aculeata (Jacq.) Lodd. Ex Mart.] is a rustic oilseed, with high productivity and multiple potential, which is strongly demanded in the food, cosmetic, and especially the energy industries, due of its suitability for biodiesel production. However, despite the enormous potential, the exploitation of macaw palm still boils down to foraging. Studies aiming at their sustainable use and the resulting domestication of the species are ignorant, because of the lack of practical and efficient methods to support breeding programs on genotypes selections. The main objective of this study was to develop and propose methods to estimate the oil content and volume of the mesocarp in macaw palm fruit using optical characteristics. For methodologies evaluation two assays were carried out. In the first, 420 samples were used to develop calibration model to predict oil content of mesocarp using visible and near infrared spectrometry. The soxhlet method was used as reference. In the second, consisted of 20 samples, were developed imaging algorithms to estimate the mesocarp volume, and as a reference was used water displacement method. In the first assay the developed model showed consistent results in the calibration and validation sets, and it s potentially feasible for preliminary selections and characterization of genotypes intended macaw palm improvement. In the second, were developed two algorithms, one considering each fruit as sphere and another one using ellipsoid approximation. The results show that the algorithms can be used in the pre and post-harvest sector and to make genotypes selection in macaw palm breeding programs. / A macaúba [Acrocomia aculeata (Jacq.) Lodd. Ex Mart.] é uma oleaginosa rústica, de alta produtividade e múltiplas potencialidades, com grande demanda nas indústrias alimentícia, cosmética e, principalmente, a energética, em função da adequação para produção do biodiesel. Contudo, apesar do enorme potencial, a exploração da macaúba ainda resume-se ao extrativismo. Os estudos visando a sua domesticação e o consequente uso sustentável da espécie são insipientes, em função da falta de métodos práticos e eficientes que possam subsidiar os programas de melhoramento na seleção de genótipos de interesse. Assim, o objetivo deste trabalho foi desenvolver e propor métodos para estimar o teor de óleo e volume do mesocarpo no fruto da macaúba utilizando características ópticas. Para avaliação das metodologias foram conduzidos dois ensaios. No primeiro foram utilizadas 420 amostras para desenvolver um modelo de predição do teor de óleo do mesocarpo usando a espectrometria do visível e infravermelho próximo. Como referência foi usado o método soxhlet. No segundo, composto por 20 amostras, foram desenvolvidos algoritmos do processamento de imagens digitais para estimar o volume do mesocarpo, e como referência empregou-se o método do deslocamento da coluna da água. No primeiro ensaio, o modelo desenvolvido apresentou resultados consistentes na calibração e validação, sendo potencialmente viável para a caracterização e pré-seleção de genótipos visando o melhoramento da macaúba. No segundo ensaio, foram desenvolvidos dois algoritmos, um considerando cada fruto como uma esfera e outro por aproximação a um elipsóide. Os resultados permitem concluir que os algoritmos podem ser usados nas áreas da pré e pós-colheita e na seleção de genótipos em programas do melhoramento da macaúba.

Η παρούσα διπλωματική εργασία, πραγματεύεται την εύρεση της κυματομορφής της μεταβολής της φωτεινότητας φωτονίων που διέρχονται από το χέρι ανθρώπου και δίνουν πληροφορία για την αρτηριακή πίεση και κατ’ επέκταση τον υπολογισμό του καρδιακού ρυθμού ενός ατόμου με τη χρήση τεχνικών επεξεργασίας εικόνας. Χρησιμοποιώντας μια σειρά από διόδους εκπομπής, στο ορατό και υπέρυθρο φάσμα, κατευθύνουμε το φως προς ένα δίκτυο ιστών όπου αυτό είναι λεπτό και το διαπερνά (δάκτυλο, λοβίο αυτιού κλπ). Στη συνέχεια, μέσω μιας βιντεοκάμερας παίρνουμε τα υπό εξέταση δεδομένα. Συγκρίνοντας την απορρόφηση του φωτός στις διαδοχικές εικόνες και ύστερα από κατάλληλη επεξεργασία των εικόνων με τη βοήθεια του Matlab οδηγούμαστε στην εύρεση του καρδιακού ρυθμού. / The present thesis deals with the determination of the waveform that depicts the fluctuation of the brightness of photons which pass through the hand of a person and provides information for the arterial pressure. Exploiting the results through digital image processing techniques, subject’s cardiac rhythm can be conclusively calculated. Using a series of diodes emitting in the visible spectrum along with a second series of diodes emitting in the infrared spectrum, we direct their light to a part of the human tissue which is thin (finger, earlobe etc) and can be easily penetrated. Afterwards via a CCD video camera we capture picture data of the light that is not absorbed. Cardiac rhythm can be calculated by comparing the absorption of light in successive pictures processed by digital imaging processing tools of Matlab.

<p>    </p> <p>Cerebrovascular diseases, such as stroke, constitute the most common life-threatening neurological disease in the United States. To support normal brain function, maintaining adequate brain perfusion (i.e., cerebral blood flow (CBF)) is important. Therefore, it is crucial to assess the brain perfusion so that early intervention in cerebrovascular diseases can be applied if abnormal perfusion is observed. The goal of my study is to develop metrics to measure the brain perfusion through modeling brain physiology using resting-state and task-based blood-oxygenation-level- dependent (BOLD) functional MRI (fMRI). My first and second chapters focused on deriving the blood arrival time using the resting-state BOLD signal. In the first chapters, we extracted the systemic low-frequency oscillations (sLFOs) in the fMRI signal from the internal carotid arteries (ICA) and the superior sagittal sinus (SSS). Consistent and robust results were obtained across 400 scans showing the ICA signals leading the SSS signals by about 5 seconds. This delay time could be considered as an effective perfusion biomarker that is associate with the cerebral circulation time (CCT). To further explore sLFOs in assessing dynamic blood flow changes during the scan, in my second chapter, a “carpet plot” (a 2-dimensional plot time vs. voxel) of scaled fMRI signal intensity was reconstructed and paired with a developed slope-detection algorithm. Tilted vertical edges across which a sudden signal intensity change took place were successfully detected by the algorithm and the averaged propagation time derived from the carpet plot matches the cerebral circulation time. Given that CO₂ is a vasodilator, controlling of inhaled CO₂ is able to modulate the BOLD signal, therefore, as a follow-up study, we focused on investigating the feasibility of using a CO₂ modulated sLFO signal as a “natural” bolus to track CBF with the tool developed from the second chapter. Meaningful transit times were derived from the CO₂-MRI carpet plots. Not only the timing, the BOLD signal deformation (the waveform change) under CO₂ challenge also reveals very useful perfusion information, representing how the brain react to stimulus. Therefore, my fourth chapter focused on characterizing the brain reaction to the CO₂ stimulus to better measure the brain health using BOLD fMRI. Overall, these studies deepen our understanding of fMRI signal and the derived perfusion parameters can potentially be used to assess some cerebrovascular diseases, such as stroke, ischemic brain damage, and steno-occlusive arterial disease in addition to functional activations. </p>

INTRODUÇÃO: Os transtornos mentais psicóticos são condições frequentes na população em geral e estão associados à grande morbidade e elevadas taxas de comprometimento funcional, tornando-os um grave problema de saúde pública. O desenvolvimento de novos métodos de auxílio diagnóstico e prognóstico a pratica clínica psiquiátrica possibilitando que intervenções efetivas sejam feitas precocemente na história natural da doença são, dessa forma, desejáveis. A classificação de padrões neuroanatômicos é uma robusta técnica para processamento e análise de imagens médicas que permite tanto a realização de comparações voxel-a-voxel entre grupos com alta dimensionalidade de variáveis, como a classificação individualizada das imagens. OBJETIVOS: Avaliar o desempenho diagnóstico de um classificador de padrões morfológicos complexos baseado em support vector machine (SVM) na discriminação entre diferentes transtornos psicóticos no momento do primeiro episódio, utilizando-se uma abordagem epidemiológica para a seleção de casos e controles, bem como na determinação de prognóstico de 1 ano em pacientes com primeiro episódio de esquizofrenia. MÉTODOS: Uma amostra de 62 pacientes com primeiro episódio de esquizofrenia/ transtorno esquizofreniforme, 23 casos de primeiro episódio de mania psicótica (transtorno bipolar tipo I, TB-I), e 19 indivíduos com depressão maior (DM) psicótica foram estudados com ressonância magnética (RM) estrutural de 1.5T, assim como um total de 89 controles residentes na mesma região dos casos. As imagens T1 foram inicialmente registradas a uma imagem molde comum através de um método com preservação de massa, permitindo a obtenção de volumes cerebrais regionais. Um classificador neuroanatômico multivariado baseado em redução de dimensionalidade e SVM foi utilizado para identificar o melhor conjunto de características morfológicas que diferencia cada transtorno psicótico (esquizofrenia/ transtorno esquizofreniforme, TB-I e DM psicótica) de subgrupos de controles saudáveis pareados por idade, gênero e anos de escolaridade. Os resultados obtidos pelo classificador foram, então, analisados com o auxílio de uma curva ROC, e um mapa espacial de alta dimensionalidade daquelas regiões cerebrais que constituem um padrão de distribuição tecidual cerebral característico de cada transtorno psicótico em relação aos controles foi gerado. RESULTADOS: O classificador obteve uma discriminação apenas modesta entre pacientes com primeiro episódio de esquizofrenia/ transtorno esquizofreniforme e controles saudáveis, com uma medida de área sob a curva (AUC) de 0,75 e acurácia de 73,4%. O mapa espacial discriminatório resultante mostrou um padrão complexo de alterações volumétricas comprometendo regiões fronto-límbicas tanto de substância cinzenta como de substância branca cerebral bilateralmente, fascículos cerebrais associativos, terceiro ventrículo e o ventrículo lateral esquerdo. Um desempenho diagnóstico pobre foi observado nas comparações entre pacientes com TB-I e MD psicótica e controles. Além disso, o classificador baseado em SVM não conseguiu predizer satisfatoriamente o prognóstico de 1 ano (evolução de remissão versus não remissão) dos pacientes com primeiro episódio de esquizofrenia. CONCLUSÃO: Utilizando uma amostra de pacientes com psicoses afetivas e não afetivas com características clínicas semelhantes aos pacientes vistos na nossa prática psiquiátrica (comorbidade com transtornos de uso de substâncias e curso clínico variável) e selecionados através de uma abordagem epidemiológica populacional, o classificador de padrões neuroanatômicos não obteve bom desempenho diagnóstico na discriminação entre as formas esquizofreniformes e afetivas de primeiro episódio psicótico, e também não conseguiu predizer satisfatoriamente o prognóstico de 1 ano em primeiro episódio de esquizofrenia, utilizando apenas imagens estruturais de RM / INTRODUCTION: Psychotic disorders are prevalent medical conditions in the general population, and are usually associated with high morbidity and functional impairment rates, which make them a major concern for public health. The development of new methods aiming to aid diagnostic and prognostic value in clinical psychiatric practice thus allowing effective interventions at an early course of the illness are, therefore, desirable. Neuroanatomical pattern classification is a powerful technique for image processing and analysis which allows both high-dimensional voxelwise group comparisons and classification of images at an individual basis. OBJECTIVES: To evaluate the diagnostic performance of a support vector machine (SVM)-based complex morphological pattern classifier was used to discriminate different non-affective and affective psychotic disorders at the first episode using a population-based approach to recruit both cases and healthy controls, and also to predict 1-year prognosis (i.e., remitting versus non-remitting course) in a group of patients with first-episode schizophrenia. METHODS: A sample of 62 patients with first-episode schizophrenia/ schizophreniform disorder, 23 cases presenting with their first-episode of psychotic mania (bipolar I disorder, BD-I) and 19 individuals with psychotic major depressive disorder (MDD) was studied with 1.5T structural magnetic resonance imaging (MRI), as well as a pool of 89 epidemiologically recruited controls. T1-weighted images were first registered to a common template through a robust mass-preserving routine allowing regional volumetric analysis. A high-dimensional multivariate classification method based on dimensionality reduction and SVM was employed to identify the best and most parsimonious set of morphological features that discriminate each psychotic group (schizophrenia/ schizophreniform disorder, BD-I & psychotic MDD) from subgroups of age, gender and educationally-matched healthy controls. The abnormalities scores generated by the classifier were analyzed with a ROC curve analysis and a high-dimensional spatial map of the brain regions that constitute a pattern of brain tissue distribution characteristic of each of the non-affective and affective groups relative to controls was created. RESULTS: The SVM-classifier afforded modest discrimination between subjects with first-episode schizophrenia/ schizophreniform disorder and controls, with an area under the curve (AUC) value of 0.75 and overall accuracy of 73.4%. The resulting discriminative spatial map revealed a complex pattern of regional volumetric abnormalities affecting both gray and white matter fronto-limbic regions bilaterally, long associative fasciculi, besides the third and lateral ventricles. A poor diagnostic performance was observed in the pairwise comparisons between BD-I and psychotic MDD versus controls. Also, the SVM-classifier failed to predict 1-year prognosis (remitting versus non-remitting course) in the first-episode schizophrenia group. CONCLUSION: The present results suggest that at the population level and using a real world sample of affective and non-affective psychotic patients with comorbid substance use disorders and variable disease course, we failed to achieve good discrimination between schizophreniform and affective forms of first-episode psychosis, and also in predicting 1-year prognosis of first-episode schizophrenia patients, using structural images

Whereas manually drawn contours could contain artifacts related to hand tremor, automatically detected contours could contain noise and inaccuracies due to limitations or errors in the procedures for the detection and segmentation of the related regions. To improve the further step of description, modeling procedures are desired to eliminate the artifacts in a given contour, while preserving the important and significant details present in the contour. In this work, are presented a couple of polygonal modeling methods, first a method applied direct on the original contour and other derived from the turning angle function. Both methods use the following parametrization Smin e µmax to infer about removing or maintain a given segment. By the using of the mentioned parameters the proposed methods could be configured according to the application problem. Both methods have been shown eficient to reduce the influence of noise and artifacts while preserving relevant characteristic for further analysis. Systems to support the diagnosis by images (CAD) and retrieval of images by content (CBIR) use shape descriptor methods to make possible to infer about factors existing in a given contour or as base to classify groups with dierent patterns. Shape factors methods should represent a value that is aected by the shape of an object, thus it is possible to characterize the presence of a factor in the contour or identify similarity among contours. Shape factors should be invariant to rotation, translation or scale. In the present work there are proposed the following shape features: index of the presence of convex region (XRTAF ), index of the presence of concave regions (V RTAF ), index of convexity (CXTAF ), two measures of fractal dimension (DFTAF e DF1 TAF ) and the index of spiculation (ISTAF ). All derived from the smoothed turning angle function. The smoothed turning angle function represent the contour in terms of their concave and convex regions. The polygonal modeling and the shape descriptors methods were applied on the breast masses classification issue to evaluate their performance. The polygonal modeling procedure proposed in this work provided higher compression and better polygonal fitness. The best classification accuracies, on discriminating between benign masses and malignant tumors, obtain for XRTAF , V RTAF , CXTAF , DFTAF , DF1 TAF and ISTAF , in terms of area under the receiver operating characteristics curve, were 0:92, 0:92, 0:93, 0:93, 0:92 e 0:94, respectively. / Contornos obtidos manualmente podem conter ruídos e artefatos oriundos de tremores da mão bem como contornos obtidos automaticamente podem os conter dado a problemas na etapa de segmentação. Para melhorar os resultados da etapa de representação e descrição, são necessários métodos capazes de reduzir a influência dos ruídos e artefatos enquanto mantém características relevantes da forma. Métodos de aproximação poligonal têm como objetivo a remoção de ruídos e artefatos presentes nos contornos e a melhor representação da forma com o menor número possível de segmentos de retas. Nesta disserta ção são propostos dois métodos de aproximação poligonal, um aplicado diretamente no contorno e outro que é obtido a partir da função tangencial do contorno original. Ambos os métodos fazem uso dos parâmetros Smin e µmax para inferirem sobre a permanência ou remoção de um dado segmento. Com a utilização destes parâmetros os métodos podem ser configurados para serem utilizados em vários tipos de aplicações. Ambos os métodos mostram-se eficientes na remoção de ruídos e artefatos, enquanto que características relevantes para etapas de pós-processamento são mantidas. Sistemas de apoio ao diagnóstico por imagens e de recuperação de imagens por conte údo fazem uso de métodos descritores de forma para que seja possível inferir sobre características presentes em um dado contorno ou ainda como base para medir a dissimilaridade entre contornos. Métodos descritores de características são capazes de representar um contorno por um número, assim é possível estabelecer a presença de uma característica no contorno ou ainda identificar uma possível similaridade entre os contornos. Métodos para extração de características devem ser invariantes a rotação, translação e escala. Nesta dissertação são propostos os seguintes métodos descritores de características: índice de presença de regiões convexas (XRTAF ), índice da presença de regiões côncavas (V RTAF ), índice de convexidade (CXTAF ), duas medidas de dimensão fractal (DFTAF e DF1 TAF ) e o índice de espículos (ISTAF ). Todos aplicados sobre a função tangencial suavizada. A função tangencial suavizada representa o contorno em termos de suas regiões côncavas e regiões convexas. Os métodos de aproximação poligonal e descritores de características foram aplicados para o problema de classificação de lesões de mama. Os resultados obtidos, mostraram que os métodos de aproximação poligonal propostos neste trabalho resultam em polígonos mais compactos e com melhor representação do contorno original. Os melhores resultados de classificação, na discriminação entre lesões benignas e tumores malignos, obtidos por XRTAF , V RTAF , CXTAF , DFTAF , DF1 TAF e ISTAF , em termos da área sob a curva ROC, foram 0:92, 0:92, 0:93, 0:93, 0:92 e 0:94, respectivamente. / Mestre em Ciência da Computação