Academic literature on the topic 'Ultrasound image segmentation'

The archetypal modern comic book superhero, Superman, has two superpowers of interest: the ability to see into objects and the ability to see distant objects. Now, humans possess these powers as well, due to the medical ultrasound imaging and sound navigation. Ultrasound, a type of sound we cannot hear, has enabled us to see a world otherwise invisible to us. Ultrasound medical imaging can be used to visualize and quantify anatomical and functional aspects of internal tissues and organs of the human body. Skeletal muscle tissue is functionally composed by so called motor units which are the smallest voluntarily activatable units and is of primary interest in this study. The major complexity in segmentation of motor units in skeletal muscle tissue in ultrasound image sequences is the aspect of overlapping objects. We propose a framework and evaluate the performance on simulated synthetic data. We have found that it is possible to segment motor units under an isometric contraction using high-end ultrasound scanners and we have proposed a framework which is robust when simulating up to 10 components when exposed to 20 dB Gaussian white noise. The framework is not satisfactory robust when exposed to significant amount of noise. In order to be able to segment a large number of components, decomposition is inevitable and together with development of a step including smoothing, the framework can be further improved.

This thesis outlines an algorithm for 2D and 3D semi-automatic segmentation of the prostate from B-mode trans-rectal ultrasound (TRUS) images. In semi-automatic segmentation, a computer algorithm outlines the boundary of the prostate given a few initialization points. The algorithm is designed for prostate brachytherapy and has the potential to: i) replace pre-operative manual segmentation, ii) enable intra-operative segmentation, and iii) be integrated into a visualization tool for training residents. The segmentation algorithm makes use of image warping to make the 2D prostate boundary elliptical. A Star Kalman based edge detector is then guided along the elliptical shape to find the prostate boundary in the TRUS image. A second ellipse is then fit to the edge detected measurement points. Once all 2D slices are segmented in this manner an ellipsoid is fit to the 3D cloud of points. Finally a reverse warping algorithm gives us the segmented prostate volume. In-depth 2D and 3D clinical studies show promising results. In 2D, distance based metrics show a mean absolute difference of 0.67 ± 0.18mm between manual and semi-automatic segmentation and area based metrics show average sensitivity and accuracy over 97% and 93% respectively. In 3D, i) the difference between manual and semi-automatic segmentation is on the order of interobserver variability, ii) the repeatability of the segmentation algorithm is consistently better than the intra-observer variability, and iii) the sensitivity and accuracy are 97% and 85% respectively. The 3D algorithm requires only 5 initialization points and can segment a prostate volume in less than 10 seconds (approximately 40 times faster than manual segmentation). The novelties of this algorithm, in comparison to other works, are in the warping and ellipse/ ellipsoid fitting steps. These two combine to provide a simple solution that works well even with non-ideal images to produce accurate, real-time results.<br>Applied Science, Faculty of<br>Electrical and Computer Engineering, Department of<br>Graduate

A level set-based segmentation procedure has been implemented to identify target object boundaries from 3D medical ultrasound images. Several test images (simulated, scanned phantoms, clinical) were subjected to various preprocessing methods and segmented. Two metrics of segmentation accuracy were used to compare the segmentation results to ground truth models and determine which preprocessing methods resulted in the best segmentations. It was found that by using an anisotropic diffusion filtering method to reduce speckle type noise with a 3D active contour segmentation routine using the level set method resulted in semi-automated segmentation on par with medical doctors hand-outlining the same images.

Prostate segmentation in trans rectal ultrasound (TRUS) and magnetic resonance images (MRI) facilitates volume estimation, multi-modal image registration, surgical planing and image guided prostate biopsies. The objective of this thesis is to develop computationally efficient prostate segmentation algorithms in both TRUS and MRI image modalities. In this thesis we propose a probabilistic learning approach to achieve a soft classification of the prostate for automatic initialization and evolution of a deformable model for prostate segmentation. Two deformable models are developed for the TRUS segmentation. An explicit shape and region prior based deformable model and an implicit deformable model guided by an energy minimization framework. Besides, in MRI, the posterior probabilities are fused with the soft segmentation coming from an atlas segmentation and a graph cut based energy minimization achieves the final segmentation. In both image modalities, statistically significant improvement are achieved compared to current works in the literature.<br>La segmentació de la pròstata en imatge d'ultrasò (US) i de ressonància magnètica (MRI) permet l'estimació del volum, el registre multi-modal i la planificació quirúrgica de biòpsies guiades per imatge. L'objectiu d'aquesta tesi és el desenvolupament d'algorismes automàtics per a la segmentació de la pròstata en aquestes modalitats. Es proposa un aprenentatge automàtic inical per obtenir una primera classificació de la pròstata que permet, a continuació, la inicialització i evolució de diferents models deformables. Per imatges d'US, es proposen un model explícit basat en forma i informació regional i un model implícit basat en la minimització d'una funció d'energia. En MRI, les probalitats inicials es fusionen amb una imatge de probabilitat provinent d'una segmentació basada en atlas, i la minimització es realitza mitjançant tècniques de grafs. El resultat final és una significant millora dels algorismes actuals en ambdues modalitats d'imatge.

Ultrasound is one of the primary technologies used for clinical purposes. Ultrasound systems have favorable real-time capabilities, are fast and relatively inexpensive, portable and non-invasive. Recent interest in using ultrasound imaging for tongue motion has various applications in linguistic study, speech therapy as well as in foreign language education, where visual-feedback of tongue motion complements conventional audio feedback. Ultrasound images are known to be difficult to recognize. The anatomical structure in them, the rapidity of tongue movements, also missing segments in some frames and the limited frame rate of ultrasound systems have made automatic tongue contour extraction and tracking very challenging and especially hard for real-time applications. Traditional image processing-based approaches have many practical limitations in terms of automation, speed, and accuracy. Recent progress in deep convolutional neural networks has been successfully exploited in a variety of computer vision problems such as detection, classification, and segmentation. In the past few years, deep belief networks for tongue segmentation and convolutional neural networks for the classification of tongue motion have been proposed. However, none of these claim fully-automatic or real-time performance. U-Net is one of the most popular deep learning algorithms for image segmentation, and it is composed of several convolutions and deconvolution layers. In this thesis, we proposed a fully automatic system to extract tongue dorsum from ultrasound videos in real-time using a simplified version of U-Net, which we call sU-Net. Two databases from different machines were collected, and different training schemes were applied for testing the learning capability of the model. Our experiment on ultrasound video data demonstrates that the proposed method is very competitive compared with other methods in terms of performance and accuracy.

In the field of medical image analysis, ultrasound is a core imaging modality employed due to its real time and easy-to-use nature, its non-ionizing and low cost characteristics. Ultrasound imaging is used in numerous clinical applications, such as fetus monitoring, diagnosis of cardiac diseases, flow estimation, etc. Classical applications in ultrasound imaging involve tissue characterization, tissue motion estimation or image quality enhancement (contrast, resolution, signal to noise ratio). However, one of the major problems with ultrasound images, is the presence of noise, having the form of a granular pattern, called speckle. The speckle noise in ultrasound images leads to the relative poor image qualities compared with other medical image modalities, which limits the applications of medical ultrasound imaging. In order to better understand and analyze ultrasound images, several device-based techniques have been developed during last 20 years. The object of this PhD thesis is to propose new image processing methods allowing us to improve ultrasound image quality using postprocessing techniques. First, we propose a Bayesian method for joint deconvolution and segmentation of ultrasound images based on their tight relationship. The problem is formulated as an inverse problem that is solved within a Bayesian framework. Due to the intractability of the posterior distribution associated with the proposed Bayesian model, we investigate a Markov chain Monte Carlo (MCMC) technique which generates samples distributed according to the posterior and use these samples to build estimators of the ultrasound image. In a second step, we propose a fast single image super-resolution framework using a new analytical solution to the l2-l2 problems (i.e., $\ell_2$-norm regularized quadratic problems), which is applicable for both medical ultrasound images and piecewise/ natural images. In a third step, blind deconvolution of ultrasound images is studied by considering the following two strategies: i) A Gaussian prior for the PSF is proposed in a Bayesian framework. ii) An alternating optimization method is explored for blind deconvolution of ultrasound.

Uncertainty in the sizing of breast cancer masses is a major issue in breast screening programs, as there is a tendency to severely underestimate the sizing of malignant masses, especially with ultrasound imaging as part of the standard triple assessment. Due to this issue about 20% of all surgically treated women have to undergo a second resection, therefore the aim of this thesis is to address this issue by developing novel image analysis methods. Ultrasound elasticity imaging has been proven to have a better ability to differentiate soft tissues compared to standard B-mode. Thus a novel segmentation algorithm is presented, employing elasticity imaging to improve the sizing of malignant breast masses in ultrasound. The main contributions of this work are the introduction of a novel filtering technique to significantly improve the quality of the B-mode image, the development of a segmentation algorithm and their application to an ongoing clinical trial. Due to the limitations of the employed ultrasound device, the development of a method to improve the contrast and signal to noise ratio of B-mode images was required. Thus, an autoregressive model based filter on the radio-frequency signal is presented which is able to reduce the misclassification error on a phantom by up to 90% compared to the employed device, achieving similar results to a state-of-the art ultrasound system. By combining the output of this filter with elasticity data into a region based segmentation framework, a computationally highly efficient segmentation algorithm using Graph-cuts is presented. This method is shown to successfully and reliably segment objects on which previous highly cited methods have failed. Employing this method on 18 cases from a clinical trial, it is shown that the mean absolute error is reduced by 2 mm, and the bias of the B-Mode sizing to underestimate the size was overcome. Furthermore, the ability to detect widespread DCIS is demonstrated.

Maternal diet can have a great impact on the health and development of the fetus. Poor fetal nutrition has been linked to the development of a set of conditions in later life, such as coronary heart disease, type 2 diabetes and hypertension, while restricted growth can result in hypogylcemia, hypocalcemia, hypothermia, polycythemia, hyperbilirubinemia and cerebral palsy. High alcohol consumption during pregnancy can result in Fetal Alcohol Syndrome, a condition that can cause growth retardation, lowered intelligence and craniofacial defects. Current biometric assessment of the fetus involves size-based measures which may not accurately portray the state of fetal development, since they cannot differentiate cases of small-but-healthy or large-but-unhealthy fetuses. This thesis aims to outline a set of more appropriate measures of accurately capturing the state of fetal development. Specifically, soft tissue area and liver volume measurement are examined, followed by facial shape characterisation. A number of tools are presented which aim to allow clinicians to achieve accurate segmentations of these landmark regions. These are modifications on the Live Wire algorithm, an interactive segmentation method in which the user places a number of anchor points and a minimum cost path is calculated between the previous anchor point and the cursor. This focuses on giving the clinician intuitive control over the exact position of the segmented contour. These modifications are FA-S Live Wire, which utilises Feature Asymmetry and a weak shape constraint, ASP Live Wire, which is a 3D expansion of Live Wire, and FA-O Live Wire, which uses Feature Asymmtery and Local Orientation to guide the segmentation process. These have been designed with each of the specific biometric landmarks in mind. Finally, a method of characterising fetal face shape is proposed, using a combination of the segmentation methods described here and a simple shape model with a parameterised b-spline meshing approach to facial surface representation.

Speckle noise appears when coherent illumination is employed, as for example Laser, Synthetic Aperture Radar (SAR), Sonar, Magnetic Resonance, X-ray and Ultrasound imagery. Backscattered echoes from the randomly distributed scatterers in the microscopic structure of the medium are the origin of speckle phenomenon, which characterizes coherent imaging with a granular appearance. It can be shown that speckle noise is of multiplicative nature, strongly correlated and more importantly, with non-Gaussian statistics. These characteristics differ greatly from the traditional assumption of white additive Gaussian noise, often taken in image segmentation, filtering, and in general, image processing; which leads to reduction of the methods effectiveness for final image information extraction; therefore, this kind of noise severely impairs human and machine ability to image interpretation. Statistical modeling is of particular relevance when dealing with speckled data in order to obtain efficient image processing algorithms; but, additionally, clinical ultrasound imaging systems employ nonlinear signal processing to reduce the dynamic range of the input echo signal to match the smaller dynamic range of the display device and to emphasize objects with weak backscatter. This reduction in dynamic range is normally achieved through a logarithmic amplifier i.e. logarithmic compression, which selectively compresses large input signals. This kind of nonlinear compression totally changes the statistics of the input envelope signal; and, a closed form expression for the density function of the logarithmic transformed data is usually hard to derive. This thesis is concerned with the statistical distributions of the Log-compressed amplitude signal in coherent imagery, and its main objective is to develop a general statistical model for log-compressed ultrasound B-scan images. The developed model is adapted, making the pertinent physical analogies, from the multiplicative model in Synthetic Aperture Radar (SAR) context. It is shown that the proposed model can successfully describe log-compressed data generated from different models proposed in the specialized ultrasound image processing literature. Also, the model is successfully applied to model in-vivo echo-cardiographic (ultrasound) B-scan images. Necessary theorems are established to account for a rigorous mathematical proof of the validity and generality of the model. Additionally, a physical interpretation of the parameters is given, and the connections between the generalized central limit theorems, the multiplicative model and the compound representations approaches for the different models proposed up-to-date, are established. It is shown that the log-amplifier parameters are included as model parameters and all the model parameters are estimated using moments and maximum likelihood methods. Finally, three applications are developed: speckle noise identification and filtering; segmentation of in vivo echo-cardiographic (ultrasound) B-scan images and a novel approach for heart ejection fraction evaluation<br>El ruido Speckle aparece cuando se utilizan sistemas de iluminación coherente, como por ejemplo Láser, Radar de Apertura Sintética (SAR), Sonar, Resonancia Magnética, rayos X y ultrasonidos. Los ecos dispersados por los centros dispersores distribuidos al azar en la estructura microscópica del medio son el origen de este fenómeno, que caracteriza las imágenes coherentes con un aspecto granular. Se puede demostrar que el ruido Speckle es de carácter multiplicativo, fuertemente correlacionados y lo más importante, con estadística no Gaussiana. Estas características son muy diferentes de la suposición tradicional de ruido aditivo gaussiano blanco, a menudo asumida en la segmentación de imágenes, filtrado, y en general, en el procesamiento de imágenes; lo cual se traduce en la reducción de la eficacia de los métodos para la extracción de información de la imagen final. La modelización estadística es de particular relevancia cuando se trata con datos Speckle, a fin de obtener algoritmos de procesamiento de imágenes eficientes. Además, el procesamiento no lineal de señales empleado en sistemas clínicos de imágenes por ultrasonido para reducir el rango dinámico de la señal de eco de entrada de manera que coincida con el rango dinámico más pequeño del dispositivo de visualización y resaltar así los objetos con dispersión más débil, modifica radicalmente la estadística de los datos. Esta reducción en el rango dinámico se logra normalmente a través de un amplificador logarítmico es decir, la compresión logarítmica, que comprime selectivamente las señales de entrada y una forma analítica para la expresión de la función de densidad de los datos transformados logarítmicamente es por lo general difícil de derivar. Esta tesis se centra en las distribuciones estadísticas de la amplitud de la señal comprimida logarítmicamente en las imágenes coherentes, y su principal objetivo es el desarrollo de un modelo estadístico general para las imágenes por ultrasonido comprimidas logarítmicamente en modo-B. El modelo desarrollado se adaptó, realizando las analogías físicas relevantes, del modelo multiplicativo en radares de apertura sintética (SAR). El Modelo propuesto puede describir correctamente los datos comprimidos logarítmicamente a partir datos generados con los diferentes modelos propuestos en la literatura especializada en procesamiento de imágenes por ultrasonido. Además, el modelo se aplica con éxito para modelar ecocardiografías en vivo. Se enuncian y demuestran los teoremas necesarios para dar cuenta de una demostración matemática rigurosa de la validez y generalidad del modelo. Además, se da una interpretación física de los parámetros y se establecen las conexiones entre el teorema central del límite generalizado, el modelo multiplicativo y la composición de distribuciones para los diferentes modelos propuestos hasta a la fecha. Se demuestra además que los parámetros del amplificador logarítmico se incluyen dentro de los parámetros del modelo y se estiman usando los métodos estándar de momentos y máxima verosimilitud. Por último, tres aplicaciones se desarrollan: filtrado de ruido Speckle, segmentación de ecocardiografías y un nuevo enfoque para la evaluación de la fracción de eyección cardiaca.