Includes bibliographical references. / Three problems are addressed in this dissertation. They are intracranial volume extraction, noise suppression and automated segmentation of X-Ray Computerized Tomography (CT) images. The segmentation scheme is based on a Seeded Region Growing algorithm. The intracranial volume extraction is based on image symmetry and the noise suppression filter is based on the Gaussian nature of the tissue distribution. Both are essential in achieving good segmentation results. Simulated phantoms and real medical images were used in testing and development of the algorithms. The testing was done over a wide range of noise values, object sizes and mean object grey levels. All the methods were first implemented in two- and then three-dimensions. The 3-D implementation also included an investigation into volume formation and the advantages of 3-D processing. The results of the intracranial extraction showed that 9% of the data in the relevant grey level range consisted of unwanted scalp (The scalp is spatially not part of the intracranial volume, but has the same grey level values). This justified the extraction the intracranial volume for further processing. For phantom objects greater than 741.51mm³ (voxel resolution 0.48mm x 0.48mm x 2mm) and having a mean grey level distance of 10 from any other object, a maximum segmentation volume error of 15% was achieved.

No description available.

In recent decades, more types and quantities of medical data have been collected due to advanced technology. A large number of significant and critical information is contained in these medical data. High efficient and automated computational methods are urgently needed to process and analyze all available medical data in order to provide the physicians with recommendations and predictions on diagnostic decisions and treatment planning. Traumatic pelvic injury is a severe yet common injury in the United States, often caused by motor vehicle accidents or fall. Information contained in the pelvic Computed Tomography (CT) images is very important for assessing the severity and prognosis of traumatic pelvic injuries. Each pelvic CT scan includes a large number of slices. Meanwhile, each slice contains a large quantity of data that may not be thoroughly and accurately analyzed via simple visual inspection with the desired accuracy and speed. Hence, a computer-assisted pelvic trauma decision-making system is needed to assist physicians in making accurate diagnostic decisions and determining treatment planning in a short period of time. Pelvic bone segmentation is a vital step in analyzing pelvic CT images and assisting physicians with diagnostic decisions in traumatic pelvic injuries. In this study, a new hierarchical segmentation algorithm is proposed to automatically extract multiplelevel bone structures using a combination of anatomical knowledge and computational techniques. First, morphological operations, image enhancement, and edge detection are performed for preliminary bone segmentation. The proposed algorithm then uses a template-based best shape matching method that provides an entirely automated segmentation process. This is followed by the proposed Registered Active Shape Model (RASM) algorithm that extracts pelvic bone tissues using more robust training models than the Standard ASM algorithm. In addition, a novel hierarchical initialization process for RASM is proposed in order to address the shortcoming of the Standard ASM, i.e. high sensitivity to initialization. Two suitable measures are defined to evaluate the segmentation results: Mean Distance and Mis-segmented Area to quantify the segmentation accuracy. Successful segmentation results indicate effectiveness and robustness of the proposed algorithm. Comparison of segmentation performance is also conducted using both the proposed method and the Snake method. A cross-validation process is designed to demonstrate the effectiveness of the training models. 3D pelvic bone models are built after pelvic bone structures are segmented from consecutive 2D CT slices. Automatic and accurate detection of the fractures from segmented bones in traumatic pelvic injuries can help physicians detect the severity of injuries in patients. The extraction of fracture features (such as presence and location of fractures) as well as fracture displacement measurement, are vital for assisting physicians in making faster and more accurate decisions. In this project, after bone segmentation, fracture detection is performed using a hierarchical algorithm based on wavelet transformation, adaptive windowing, boundary tracing and masking. Also, a quantitative measure of fracture severity based on pelvic CT scans is defined and explored. The results are promising, demonstrating that the proposed method not only capable of automatically detecting both major and minor fractures, but also has potentials to be used for clinical applications.

Sacral insufficiency fractures are an under-diagnosed source of acute lower back pain. A polymethylmethacrylate (PMMA) cement injection procedure called sacroplasty has recently been utilized as a treatment for sacral insufficiency fractures. It is believed that injection of cement reduces fracture micromotion, thus relieving pain. In this study, finite element models were used to examine the mechanical effects of sacroplasty. Finite element models were constructed from CT images of cadavers on which sacroplasties were performed. The images were used to create the mesh geometry, and to apply non-homogeneous material properties to the models. Models were created with homogeneous and non-homogeneous material properties, normal and osteoporotic bone, and with and without cement. The results indicate that the sacrum has a 3D multi-axial state of strain. While compressive strains were the largest, tensile and shear strains were significant as well. It was found that a homogeneous model can account for around 80% of the variation in strain seen in a non-homogeneous model. Thus, while homogeneous models provide a reasonable estimate of strains, non-homogeneous material properties have a significant effect in modeling bone. A reduction in bone density simulating osteoporosis increased strains nearly linearly, even with non-homogeneous material properties. Thus, the non-homogeneity was modeled similarly in both density cases. Cement in the sacrum reduced strains 40-60% locally around the cement. However, overall model stiffness only increased 1-4%. This indicates that the effects of sacroplasty are primarily local. / Master of Science

The research investigated whether a Latent Semantic Analysis (LSA)-based approach to image retrieval can map pixel intensity into a smaller concept space with good accuracy and reasonable computational cost. From a large set of computed tomography (CT) images, a retrieval query found all images for a particular patient based on semantic similarity. The effectiveness of the LSA retrieval was evaluated based on precision, recall, and F-score. This work extended the application of LSA to high-resolution CT radiology images. The images were chosen for their unique characteristics and their importance in medicine. Because CT images are intensity-only, they carry less information than color images. They typically have greater noise, higher intensity, greater contrast, and fewer colors than a raw RGB image. The study targeted level of intensity for image features extraction. The focus of this work was a formal evaluation of the LSA method in the context of large number of high-resolution radiology images. The study reported on preprocessing and retrieval time and discussed how reduction of the feature set size affected the results. LSA is an information retrieval technique that is based on the vector-space model. It works by reducing the dimensionality of the vector space, bringing similar terms and documents closer together. Matlab software was used to report on retrieval and preprocessing time. In determining the minimum size of concept space, it was found that the best combination of precision, recall, and F-score was achieved with 250 concepts (k = 250). This research reported precision of 100% on 100% of the queries and recall close to 90% on 100% of the queries with k=250. Selecting a higher number of concepts did not improve recall and resulted in significantly increased computational cost.

The overall aim of my PhD research work is to design, develop, and evaluate a new practical environment to generate separated representations of arterial and venous trees in non-contrast pulmonary CT imaging of human subjects and to extract quantitative measures at different tree-levels. Artery/vein (A/V) separation is of substantial importance contributing to our understanding of pulmonary structure and function, and immediate clinical applications exist, e.g., for assessment of pulmonary emboli. Separated A/V trees may also significantly boost performance of airway segmentation methods for higher tree generations. Although, non-contrast pulmonary CT imaging successfully captures higher tree generations of vasculature, A/V are indistinguishable by their intensity values, and often, there is no trace of intensity variation at locations of fused arteries and veins. Patient-specific structural abnormalities of vascular trees further complicate the task. We developed a novel multi-scale topo-morphologic opening algorithm to separate A/V trees in non-contrast CT images. The algorithm combines fuzzy distance transform, a morphologic feature, with a topologic connectivity and a new morphological reconstruction step to iteratively open multi-scale fusions starting at large scales and progressing towards smaller scales. The algorithm has been successfully applied on fuzzy vessel segmentation results using interactive seed selection via an efficient graphical user interface developed as a part of my PhD project. Accuracy, reproducibility and efficiency of the system are quantitatively evaluated using computer-generated and physical phantoms along with in vivo animal and human data sets and the experimental results formed are quite encouraging. Also, we developed an arc-skeleton based volumetric tree generation algorithm to generate multi-level volumetric tree representation of isolated arterial/venous tree and to extract vascular measurements at different tree levels. The method has been applied on several computer generated phantoms and CT images of pulmonary vessel cast and in vivo pulmonary CT images of a pig at different airway pressure. Experimental results have shown that the method is quite accurate and reproducible. Finally, we developed a new pulmonary vessel segmentation algorithm, i.e., a new anisotropic constrained region growing method that encourages axial region growing while arresting cross-structure leaking. The region growing is locally controlled by tensor scale and structure scale and anisotropy. The method has been successfully applied on several non-contrast pulmonary CT images of human subjects. The accuracy of the new method has been evaluated using manually selection of vascular and non-vascular voxels and the results found are very promising.

Biofidelity of finite element head model (FEHM) includes geometric and material aspects. A FEHM with inhomogeneous material properties was proposed to improve material biofidelity. The proposed FEHM was validated against experimental data and good agreements were observed. The capability of the proposed model in simulating large tissue deformation was also demonstrated. Influences of inhomogeneous material properties on the mechanical responses of head were investigated by comparing with homogeneous material model. The inhomogeneous material properties induce large peak strains in head constituents, which are probably the cause of various brain injuries. Helmets are effective in preventing head injuries. Parametric studies were conducted to investigate how changes in helmet shell stiffness, foam density and pad thickness influence the performance of a helmet in protecting the brain. Results showed that strain energy absorbed by foam component, contact stress on the interfaces and intracranial responses are significantly affected by foam density and pad thickness.

Biofidelity of finite element head model (FEHM) includes geometric and material aspects. A FEHM with inhomogeneous material properties was proposed to improve material biofidelity. The proposed FEHM was validated against experimental data and good agreements were observed. The capability of the proposed model in simulating large tissue deformation was also demonstrated. Influences of inhomogeneous material properties on the mechanical responses of head were investigated by comparing with homogeneous material model. The inhomogeneous material properties induce large peak strains in head constituents, which are probably the cause of various brain injuries. Helmets are effective in preventing head injuries. Parametric studies were conducted to investigate how changes in helmet shell stiffness, foam density and pad thickness influence the performance of a helmet in protecting the brain. Results showed that strain energy absorbed by foam component, contact stress on the interfaces and intracranial responses are significantly affected by foam density and pad thickness.

CoordenaÃÃo de AperfeiÃoamento de Pessoal de NÃvel Superior / Sistemas computacionais vÃm desempenhando papel importante em vÃrias Ãreas da medicina, notadamente no auxÃlio ao diagnÃstico mÃdico por imagem. Neste sentido, estudos na Ãrea de VisÃo Computacional sÃo realizados para desenvolver tÃcnicas e sistemas capazes de detectar automaticamente diversas doenÃas. Dentre os exames existentes que permitem o auxÃlio ao diagnÃstico e a aplicaÃÃo de sistemas computacionais em conjunto, destaca-se a Tomografia Computadorizada (TC) que possibilita a visualizaÃÃo de ÃrgÃos internos, como por exemplo, o pulmÃo e suas estruturas. Sistemas de Vis~ao Computacional utilizam estas imagens obtidas por exames de TC para extrair informaÃÃo por meio de tÃcnicas com a finalidade de segmentar, reconhecer e identificar detalhes da regiÃo de interesse nestas imagens. Este trabalho centraliza seus esforÃos na etapa de segmentaÃÃo dos pulmÃes a partir de imagens de TC, empregando-se, para tanto, tÃcnicas baseadas em MÃtodo de Contorno Ativo (MCA), tambÃm conhecido como emph{snake}. Este mÃtodo consiste em traÃar uma curva inicial, em torno ou dentro de um objeto de interesse, deformando-a conforme algumas forÃas que atuam sobre a mesma, deslocando-a atà as bordas do objeto. Este processo à realizado por iteraÃÃes sucessivas de minimizaÃÃo de uma dada funÃÃo energia, associada à curva. Neste contexto, esta tese propÃe um novo mÃtodo para a segmentaÃÃo dos pulmÃes em imagens de TC do tÃrax denominado MÃtodo de Contorno Ativo Crisp Adaptativo. Este MCA à o aperfeiÃoamento do MCA Crisp desenvolvido em um estudo anterior, que visa aumentar a precisÃo, diminuir o tempo de anÃlise e reduzir a subjetividade na segmentaÃÃo e anÃlise dos pulmÃes dessas imagens pelos mÃdicos especialistas. Este mÃtodo à proposto para a segmentaÃÃo de uma imagem isolada ou do exame completo, sendo primeiramente em 2D e expandido para 3D. O MCA Crisp Adaptativo 2D à comparado com os MCAs THRMulti, THRMod, GVF, VFC, Crisp e tambÃm com o sistema SISDEP, sendo esta avaliaÃÃo realizada utilizando como referÃncia 36 imagens segmentadas manualmente por um pneumologista. Jà o MCA Crisp Adaptativo 3D à aplicado na segmentaÃÃo dos pulmÃes em exames de TC e comparado com o mÃtodo Crescimento de RegiÃes 3D, cujos resultados das segmentaÃÃes sÃo avaliados por 2 mÃdicos pneumologistas. Os resultados obtidos demonstram que os mÃtodos propostos sÃo superiores aos demais na segmentaÃÃo dos pulmÃes em imagens de TC do tÃrax, tanto em uma imagem pelo MCA Crisp Adaptativo 2D, como em exames completos pelo MCA Crisp Adaptativo 3D. Deste modo, pode-se concluir que estes mÃtodos podem integrar sistemas de auxÃlio ao diagnÃstico mÃdico na Ãrea de Pneumologia. / Computer systems have been playing a very important role in many areas of medicine, particularly, on medical diagnosis through image processing. Therefore, studies on the field of Computer Vision are made to develop techniques and systems to perform automatic detection of several diseases. Among the existing tests that enable the diagnosis and the application of computational system together, there is the Computed Tomography (CT), which allows the visualization of internal organs, such as the lung and its structures. Image analysis techniques applied to CT scans are able to extract important information to segment and recognize details on regions of interest on these images. This work focuses its e↵orts on the stage of lungs segmentation through CT images, using techniques based on Active Contour Method (ACM), also known as snake. This method consists in tracing an initial curve, around or inside the object of interest, wich deform itself according to forces that act over the same, shifting to the object edge. This process is performed by successive iterations of minimization of a given energy, associated to the curve. In this context, this work proposes a new aproach for lung segmentation of chest CT images, which is called Adaptative Crisp Active Contour Method. This ACM is an improvement the previous developed Crisp ACM. The purpose of this new ACM is to increase accuracy, decrease analysis time and reduce segmentation subjectivity in the manual analysis of specialized doctors. This method is proposed to isolated images segmentation or the complete exam, being first in 2D, then expanding to 3D. The 2D Adaptative Crisp ACM is compared to ACMs THRMulti, THRMod, GVF, VFC, Crisp and also with the system SISDEP, being this evaluation performed by using a set of 36 manually segmented images by one pulmonologist. The 3D Adaptative Crisp ACM is applied on lung segmentation in CT exams and compared with the 3D Region Growing method, which segmentation results were evaluated by two pulmonologists. The obtained results shows that the proposed methods are superior to the other methods on lung segmentation in chest CT images, both as in one image by 2D Adaptative Crisp ACM as in full exam by the 3D Adaptative Crisp ACM. Thus, it is possible to conclude that these method can integrate systems to aid medical diagnosis in the field of pulmonology.

Medical imaging is a key tool used in healthcare to diagnose and prognose patients by aiding the detection of a variety of diseases and conditions. In practice, medical image screening must be performed by clinical practitioners who rely primarily on their expertise and experience for disease diagnosis. The ability of convolutional neural networks (CNNs) to extract hierarchical features and determine classifications directly from raw image data makes CNNs a potentially useful adjunct to the medical image analysis process. A common challenge in successfully implementing CNNs is optimizing hyperparameters for training. In this study, we propose a method which utilizes scheduled hyperparameters and Bayesian optimization to classify cancerous and noncancerous tissues (i.e., segmentation) from head and neck computed tomography (CT) and positron emission tomography (PET) scans. The results of this method are compared using CT imaging with and without PET imaging for 2D and 3D image segmentation models.