Selection of an appropriate structure for storage and analysis of complex datasets is a vital but often overlooked decision in the design of data mining and machine learning experiments. Most present techniques impose a matrix structure on the dataset, with rows representing observations and columns representing features. While this assumption is reasonable when features are scalar and do not exhibit co-dependence, the matrix data model becomes inappropriate when dependencies between non-target features must be modeled in parallel, or when features naturally take the form of higher-order multilinear structures. Such datasets particularly abound in functional medical imaging modalities, such as fMRI, where accurate integration of both spatial and temporal information is critical. Although necessary to take full advantage of the high-order structure of these datasets and built on well-studied mathematical tools, tensor analysis methodologies have only recently entered widespread use in the data mining community and remain relatively absent from the literature within the biomedical domain. Furthermore, naive tensor approaches suffer from fundamental efficiency problems which limit their practical use in large-scale high-order mining and do not capture local neighborhoods necessary for accurate spatiotemporal analysis. To address these issues, a comprehensive framework based on wavelet analysis, tensor decomposition, and the WaveCluster algorithm is proposed for addressing the problems of preprocessing, classification, clustering, compression, feature extraction, and latent concept discovery on large-scale high-order datasets, with a particular emphasis on applications in computer-assisted diagnosis. Our framework is evaluated on a 9.3 GB fMRI motor task dataset of both high dimensionality and high order, performing favorably against traditional voxelwise and spectral methods of analysis, discovering latent concepts suggestive of subject handedness, and reducing space and time complexities by up to two orders of magnitude. Novel wavelet and tensor tools are derived in the course of this work, including a novel formulation of an r-dimensional wavelet transform in terms of elementary tensor operations and an enhanced WaveCluster algorithm capable of clustering real-valued as well as binary data. Sparseness-exploiting properties are demonstrated and variations of core algorithms for specialized tasks such as image segmentation are presented. / Computer and Information Science

<p>We present a new x-ray imaging technique, Correlation Imaging (CI), for improved breast and lung cancer detection. In CI, multiple low-dose radiographic images are acquired along a limited angular arc. Information from unreconstructed angular projections is directly combined to reduce the effect of overlying anatomy - the largest bottleneck in diagnosing cancer with projection imaging. In addition, CI avoids reconstruction artifacts that otherwise limit the performance of tomosynthesis. This work involved assessing the feasibility of the CI technique, its optimization, and its implementation for breast and chest imaging.</p><p>First a theoretical model was developed to determine the diagnostic information content of projection images using a mathematical observer. The model was benchmarked for a specific application in assessing the impact of reduced dose in mammography. Using this model, a multi-factorial task-based framework was developed to optimize the image acquisition of CI using existing low-dose clinical data. The framework was further validated using a CADe processor. Performance of CI was evaluated on mastectomy specimens at clinically relevant doses and further compared to tomosynthesis. Finally, leveraging on the expected improvement in breast imaging, a new hardware capable of CI acquisition for chest imaging was designed, prototyped, evaluated, and experimentally validated.</p><p>The theoretical model successfully predicted diagnostic performance on mammographic backgrounds, indicating a possible reduction in mammography dose by as much as 50% without adversely affecting lesion detection. Application of this model on low-dose clinical data showed that peak CI performance may be obtained with 15-17 projections. CAD results confirmed similar trends. Mastectomy specimen results at higher dose revealed that the performance of optimized breast CI may exceed that of mammography and tomosynthesis by 18% and 8%, respectively. Furthermore, for both CI and tomosynthesis, highest dose setting and maximum angular span with an angular separation of 2.75o was found to be optimum, indicating a threshold in the number of projections per angular span for optimum performance. </p><p>Finally, for the CI chest imaging system, the positional errors were found to be within 1% and motion blur to have minimal impact on the system MTF. The clinical images had excellent diagnostic quality for potentially improved lung cancer detection. The system was found to be robust and scalable to enable advanced applications for chest radiography, including novel tomosynthesis trajectories and stereoscopic imaging.</p> / Dissertation

<p>Numerous studies have shown that the combination of radiation therapy and hyperthermia, when delivered at moderate temperatures (40°-45°C) for sustained times (30-90 minutes), can help to provide palliative relief and augment tumor response, local control, and survival. However, the dependence of treatment success on achieved temperature highlights the need for accurate thermal dosimetry, so that the prescribed thermal dose can be delivered to the tumor. This can be achieved noninvasively with MR thermometry. However, there are many challenges to performing MR thermometry in the breast, where hyperthermia of locally advanced breast cancer can provide a benefit. These include magnetic field system drift, fatty tissue, and breathing motion.</p> <p>The purpose of this research was to develop a system for the hyperthermia treatment of LABC while performing MR thermometry. A hardware system was developed for performing the hyperthermia treatment within the MR bore. Methods were developed to correct for magnetic field system drift and to correct for breath hold artifacts in MR thermometry of the tumor using measurement of field changes in fat references. Lastly, techniques were developed for measuring temperature in the fatty tissue using multi-echo fat water separation methods, reducing the error of performing MR thermometry in such tissues. All of these methods were characterized with phantom and in vivo experiments in a 1.5T MR system. </p> <p>The results of this research can provide the means for successful hyperthermia treatment of LABC with MR thermometry. With this thermometry, accurate thermal doses can be obtained, potentially providing improved outcomes. However, these results are not only applicable in the breast, but can also be used for improved MR thermometry in other areas of the body, such as the extremities or abdomen.</p> / Dissertation

The use of beamlets as a dose calculation tool in Intensity Modulated Radiation Therapy (IMRT) treatment planning is widespread and well documented. A beamlet can simply be defined as the contribution of radiation passing through a particular geometrically defined subdivision of a given linear accelerator's emerging radiation field. The most common classes of algorithms used today to calculate the dose distributions deposited by beamlets are the pencil beam convolution and col-lapsed cone classes of algorithms. Using BEAMnrc [1], a Monte Carlo (MC) based radiation transport simulation software package, this thesis presents a novel method of calculating MC beamlet dose distributions with a level of accuracy not achievable using the above analytic dose calculation methods. In a first application, the MC beamlet dose distributions generated in this thesis are used to fine-tune the output of the MC or "virtual" linear accelerator from which they are produced. This is achieved through the adjustment of individual beamlet weights to align the output of the virtual accelerator to the experimentally measured output of the modeled accelerator in water. In a second application, MC beamlets are used to derive corrections to particular Multileaf Collimator (MLC) leaf sequences of IMRT treatment plans that have been miscalculated by a convolution-based dose calculation algorithm. These calculation inaccuracies (up to as much as 15%) arise due to the well known fact that convolution-based algorithms do not accurately model dose deposition in inhomoge¬neous media, such as lung [2] [3] [4]. In a final application, the MC beamlet generation method described in this thesis is implemented into a direct aperture optimization (DAO) algorithm. The implementation of MC beamlet generation in DAO forms the basis for a purely MC based inverse treatment planning system.

L'un des principaux défis de l'interprétation radiographique réside dans la compréhension de l’anatomie radiographique, laquelle est intrinsèquement liée à la disposition tridimensionnelle des structures anatomiques et à l’impact du positionnement du tube radiogène vis-à-vis de ces structures lors de l'acquisition de l'image. Traditionnellement, des radiographies obtenues selon des projections standard sont employées pour enseigner l'anatomie radiographique en médecine vétérinaire. La tomodensitométrie − ou communément appelée CT (Computed Tomography) − partage plusieurs des caractéristiques de la radiographie en ce qui a trait à la génération des images. À l’aide d'un plug-in spécialement développé (ORS Visual ©), la matrice contenant les images CT est déformée pour reproduire les effets géométriques propres au positionnement du tube et du détecteur vis-à-vis du patient radiographié, tout particulièrement les effets de magnification et de distorsion. Afin d'évaluer le rendu des images simulées, différentes régions corporelles ont été imagées au CT chez deux chiens, un chat et un cheval, avant d'être radiographiées suivant des protocoles d'examens standards. Pour valider le potentiel éducatif des simulations, dix radiologistes certifiés ont comparé à l'aveugle neuf séries d'images radiographiques simulées aux séries radiographiques standard. Plusieurs critères ont été évalués, soient le grade de visualisation des marqueurs anatomiques, le réalisme et la qualité radiographique des images, le positionnement du patient et le potentiel éducatif de celles-ci pour différents niveaux de formation vétérinaire. Les résultats généraux indiquent que les images radiographiques simulées à partir de ce modèle sont suffisamment représentatives de la réalité pour être employées dans l’enseignement de l’anatomie radiographique en médecine vétérinaire. / Understanding radiographic anatomy, which is intimately linked to the comprehension of tridimensional anatomy and the impact of patient, radiographic tube and x-ray detector positioning, represents a challenge for students. Traditionally, radiographs obtained under specific angles of projection have been used for teaching radiographic anatomy. Computed tomography (CT) shares several features with radiography with regard to image production. A plug-in was developed for a DICOM viewer (ORS visual ©) simulating radiographs using CT datasets. This plug-in distorts the CT image matrix to reproduce the magnification and distortion effects that take place in radiographs due to the variations in radiographic tube, patient and detector positioning and angulation. In order to test this model, specific body parts of two dogs, one cat and one horse were radiographed and CT-scanned. The CT datasets were used to generate a total of nine series of radiographic simulations that could be compared to corresponding standard radiographic projections. Ten board-certified veterinary radiologists blindly scored several parameters in these image series, including the visualization of specific anatomical landmarks, image realism and quality, patient positioning, and the educational potential for students and veterinarians of variable degree of veterinary training Overall results indicate that simulated radiographs are representative enough to be used to teach several concepts of image formation and radiographic anatomy in veterinary radiology.

L'un des principaux défis de l'interprétation radiographique réside dans la compréhension de l’anatomie radiographique, laquelle est intrinsèquement liée à la disposition tridimensionnelle des structures anatomiques et à l’impact du positionnement du tube radiogène vis-à-vis de ces structures lors de l'acquisition de l'image. Traditionnellement, des radiographies obtenues selon des projections standard sont employées pour enseigner l'anatomie radiographique en médecine vétérinaire. La tomodensitométrie − ou communément appelée CT (Computed Tomography) − partage plusieurs des caractéristiques de la radiographie en ce qui a trait à la génération des images. À l’aide d'un plug-in spécialement développé (ORS Visual ©), la matrice contenant les images CT est déformée pour reproduire les effets géométriques propres au positionnement du tube et du détecteur vis-à-vis du patient radiographié, tout particulièrement les effets de magnification et de distorsion. Afin d'évaluer le rendu des images simulées, différentes régions corporelles ont été imagées au CT chez deux chiens, un chat et un cheval, avant d'être radiographiées suivant des protocoles d'examens standards. Pour valider le potentiel éducatif des simulations, dix radiologistes certifiés ont comparé à l'aveugle neuf séries d'images radiographiques simulées aux séries radiographiques standard. Plusieurs critères ont été évalués, soient le grade de visualisation des marqueurs anatomiques, le réalisme et la qualité radiographique des images, le positionnement du patient et le potentiel éducatif de celles-ci pour différents niveaux de formation vétérinaire. Les résultats généraux indiquent que les images radiographiques simulées à partir de ce modèle sont suffisamment représentatives de la réalité pour être employées dans l’enseignement de l’anatomie radiographique en médecine vétérinaire. / Understanding radiographic anatomy, which is intimately linked to the comprehension of tridimensional anatomy and the impact of patient, radiographic tube and x-ray detector positioning, represents a challenge for students. Traditionally, radiographs obtained under specific angles of projection have been used for teaching radiographic anatomy. Computed tomography (CT) shares several features with radiography with regard to image production. A plug-in was developed for a DICOM viewer (ORS visual ©) simulating radiographs using CT datasets. This plug-in distorts the CT image matrix to reproduce the magnification and distortion effects that take place in radiographs due to the variations in radiographic tube, patient and detector positioning and angulation. In order to test this model, specific body parts of two dogs, one cat and one horse were radiographed and CT-scanned. The CT datasets were used to generate a total of nine series of radiographic simulations that could be compared to corresponding standard radiographic projections. Ten board-certified veterinary radiologists blindly scored several parameters in these image series, including the visualization of specific anatomical landmarks, image realism and quality, patient positioning, and the educational potential for students and veterinarians of variable degree of veterinary training Overall results indicate that simulated radiographs are representative enough to be used to teach several concepts of image formation and radiographic anatomy in veterinary radiology.

<p>The ability to image ventilation and perfusion enables pulmonary researchers to study functional metrics of gas exchange on a regional basis. There is a huge interest in applying imaging methods to study the large number of genetic models of pulmonary diseases available in small animals. Existing techniques to image ventilation and perfusion are often associated with low spatial resolution and ionizing radiation. Magnetic Resonance Imaging (MRI) has been demonstrated successfully for ventilation and perfusion studies in humans. Translating these techniques in small animals remains challenging. This work addresses the ventilation and perfusion imaging in small animals using MRI. </p><p>Qualitative ventilation imaging in rats and mice is possible and has been demonstrated using MRI, however perfusion imaging remains a challenge. In humans and large animals perfusion can be assessed using dynamic contrast-enhanced (DCE) MRI with a single bolus injection of a gadolinium (Gd)-based contrast agent. But the method developed for the clinic cannot be translated directly to image the rat due to the combined requirements of higher spatial and temporal resolution. This work describes a novel image acquisition technique staggered over multiple, repeatable bolus injections of contrast agent using an automated microinjector, synchronized with image acquisition to achieve dynamic first-pass contrast enhancement in the rat lung. This allows dynamic first-pass imaging that can be used to quantify pulmonary perfusion. Further improvements are made in the spatial and temporal resolution by combining the multiple injection acquisition method with Interleaved Radial Imaging and 'Sliding window-keyhole' reconstruction (IRIS). The results demonstrate a simultaneous increase in spatial resolution (<200>um) and temporal resolution (<200>ms) over previous methods, with a limited loss in signal-to-noise-ratio. </p><p>While is it possible to create high resolution images of ventilation in rats using hyperpolarized ³He, extracting meaningful quantitative information indicative of changes in ventilation is difficult. In this work, we also present a signal calibration technique used to normalize the signal of ³He to volume of ³He which can then be used to extract quantitative information of changes in ventilation via normalized difference maps. Combining the techniques for quantitative ventilation and quantitative perfusion we perform studies of change in ventilation/perfusion (V/Q) before and after airway obstruction in rats. The technique is sensitive in detecting statistically significant differences in the heterogeneity of the distribution of V/Q ratio.</p> / Dissertation

<p>Mice have become the preferred model system for studying brain function and disease. With the powerful genetic tools available, mouse models can be created to study the underlying molecular basis of neurobiology in vivo. Just as magnetic resonance imaging is the dominant tool for evaluating the human brain, high-resolution MRI--magnetic resonance microscopy (MRM)--is a useful tool for studying the brain of mouse models. However, the need for high spatial resolution limits the signal-to-noise ratio (SNR) of the MRM images. To address this problem, T1-shortening contrast agents can be used, which not only improve the tissue contrast-to-noise ratio (CNR) but also increase SNR by allowing the MR signal to recover faster between pulses. By "actively staining" the tissue with these T1-shortening agents, MRM can be performed with higher resolution, greater contrast, and shorter scan times. In this work, active staining with T1-shortening agents was used to enhance three types of in vivo mouse brain MRM: (1) angiographic imaging of the neurovasculature, (2) anatomical imaging of the brain parenchyma, and (3) functional imaging of neuronal activity.</p> <p></p> <p>For magnetic resonance angiography (MRA) of the mouse, typical contrast agents are not useful because they are quickly cleared by the body and/or extravasate from the blood pool before a high-resolution image can be acquired. To address these limitations, a novel contrast agent--SC-Gd liposomes--has been developed, which is cleared slowly by the body and is too large to extravasate from the blood pool. In this work, MRA protocols were optimized for both the standard technique (time-of-flight contrast) and SC-Gd liposomes. When the blood was stained with SC-Gd liposomes, small vessel CNR improved to 250% that of time-of-flight. The SC-Gd liposomes could also be used to reduce scan time by 75% while still improving CNR by 32%.</p> <p>For MRM of the mouse brain parenchyma, active staining has been used to make dramatic improvements in the imaging of ex vivo specimens. However for in vivo imaging, the blood-brain barrier (BBB) prevents T1-shortening agents from entering the brain parenchyma. In this work, a noninvasive technique was developed for BBB opening with microbubbles and ultrasound (BOMUS). Using BOMUS, the parenchyma of the brain could be actively stained with the T1-shortening contrast agent, Gd-DTPA, and MRM images could be acquired in vivo with unprecedented resolution (52 x 52 x 100 micrometers3) in less than 1 hour.</p> <p>Functional MRI (fMRI), which uses blood oxygen level dependant (BOLD) contrast to detect neuronal activity, has been a revolutionary technique for studying brain function in humans. However, in mice, BOLD contrast has been difficult to detect and thus routine fMRI in mice has not been feasible. An alternative approach for detecting neuronal activity uses manganese (Mn2+). Mn2+ is a T1-shortening agent that can enter depolarized neurons via calcium channels. Thus, Mn2+ is a functional contrast agent with affinity for active neurons. In this work, Mn2+ (administered with the BOMUS technique) was used to map the neuronal response to stimulation of the vibrissae. The resultant activation map showed close agreement to published maps of the posterior-lateral and anterior-medial barrel field of the primary sensory cortex.</p> <p>The use of T1-shortening agents to actively stain tissues of interest--blood, brain parenchyma, or active neurons--will facilitate the use of MRM for studying mouse models of brain development, function, and disease.</p> / Dissertation

Image guided radiation therapy (IGRT) is a new type of radiotherapy used to deliver lethal doses of radiation to mobile tumors, while preventing surrounding healthy structures from receiving high doses of radiation. It relies on image guidance to track the tumor and ensure its prescribed position in the radiation beam. The main goal of this work was to determine if a new three-dimensional ultrasound (3D US) image guidance device, called the Restitu System, could safely replace (or be used interchangeably with) an existing method involving x-ray images of implanted fiducial markers (FMs) for prostate IGRT. Using comparison statistics called 95 % limits of agreement (LOA), it was found that the new 3D US system did not produce measurements that agreed sufficiently closely to those made using the FM technique, and therefore, could not safely replace FMs for prostate IGRT. Ultrasound image quality and user variability were determined to have a significant impact on the agreement between the two methods. It was further shown that using the Restitu System offered no significant clinical advantages over a conventional patient re-positioning technique.

No description available.