Archived Weekly Agendas for Fall 2010
From Utah Center for Neuroimage Analysis
Contents |
Dates
August 23rd
Speaker: Xiang Hao, Bo Wang
Presentation Title: Review of Computer Vision and Pattern Recognition (CVPR) 2010
Description/Abstract:
Xiang Hao and Bo Wang went to Computer Vision and Pattern Recognition (CVPR) 2010 (June 13 - 18), they will give a review of CVPR 2010.
Supplemental material:
The Oral Presentations and Poster Spotlights of CVPR 2010 video
List of Tutorials/Short Courses of CVPR 2010 slides
August 30th
Speaker: Stephanie Allassonniere
Presentation Title: A Stochastic Algorithm for Probabilistic Independent Component Analysis
Description/Abstract:
The decomposition of a sample of images on a relevant subspace is a recurrent problem in many different fields from Computer Vision to medical image analysis. We propose in this paper a new learning principle and implementation of the generative decomposition model generally known as noisy ICA (for independent component analy- sis) based on the SAEM algorithm, which is a versatile stochastic approximation of the standard EM algorithm. We demonstrate the applicability of the method on a large range of decomposition models and illustrate the developments with experimental results on various data sets.
Additional information: Dr. Stephanie Allassonniere is a visiting professor at SCI, she is doing research in statistical analysis of images and deformations. Dr. Stephanie Allassonniere's SCI page is http://www.sci.utah.edu/people/sallassonniere.html
September 6th
Labor Day holiday
September 13rd
Speaker: Marcel Prastawa - Presentation of MICCAI 2010.
Presentation Title: Image Registration Driven by Combined Probabilistic and Geometric Descriptors
Description/Abstract:
Deformable image registration in the presence of considerable contrast differences and large-scale size and shape changes represents a significant challenge for image registration. A representative driving application is the study of early brain development in neuroimaging, which requires co-registration of images of the same subject across time or building 4-D population atlases. Growth during the first few years of development involves significant changes in size and shape of anatomical structures but also rapid changes in tissue properties due to myelination and structuring that are reflected in the multi-modal Magnetic Resonance (MR) contrast measurements. We propose a new registration method that generates a mapping between brain anatomies represented as a multi-compartment model of tissue class posterior images and geometries. We transform intensity patterns into combined probabilistic and geometric descriptors that drive the matching in a diffeomorphic framework, where distances between geometries are represented using currents which does not require geometric correspondence. We show preliminary results on the registrations of neonatal brain MRIs to two-year old infant MRIs using class posteriors and surface boundaries of structures undergoing major changes. Quantitative validation demonstrates that our proposed method generates registrations that better preserve the consistency of anatomical structures over time
Supplemental material: Paper
Speaker: Fangxiang Jiao - Presentation of MICCAI-MIAR 2010.
Presentation Title: Metrics for Uncertainty Analysis and Visualization of Diffusion Tensor Images
Description/Abstract:
In this paper, we propose three metrics to quantify the differences between the results of diffusion tensor magnetic resonance imaging (DT-MRI) fiber tracking algorithms: the area between corresponding fibers of each bundle, the Earth Mover's Distance (EMD) between two fiber bundle volumes, and the current distance between two fiber bundle volumes. We also discuss an interactive fiber track comparison visualization toolkit we have developed based on the three proposed fiber difference metrics and have tested on six widely-used fiber tracking algorithms. To show the effectiveness and robustness of our metrics and visualization toolkit, we present results on both synthetic data and high resolution monkey brain DT-MRI data. Our toolkit can be used for testing the noise effects on fiber tracking analysis and visualization and to quantify the difference between any pair of DT-MRI techniques, compare single subjects within an image atlas.
Supplemental material: Paper
Speaker: Nikhil Singh - Presentation of MICCAI 2010.
Presentation Title: Multivariate Statistical Analysis of Deformation Momenta Relating Anatomical Shape to Neuropsychological Measures
Description/Abstract: The purpose of this study is to characterize the neuroanatomical variations observed in neurological disorders such as dementia. We do a global statistical analysis of brain anatomy and identify the relevant shape deformation patterns that explain corresponding variations in clinical neuropsychological measures. The motivation is to model the inherent relation between anatomical shape and clinical measures and evaluate its statistical significance. We use Partial Least Squares for the multivariate statistical analysis of the deformation momenta under the Large Deformation Diffeomorphic framework. The statistical methodology extracts pertinent directions in the momenta space and the clinical response space in terms of latent variables. We report the results of this analysis on 313 subjects from the Mild Cognitive Impairment group in the Alzheimer’s Disease Neuroimaging Initiative (ADNI).
Supplemental material: Paper
September 20th
Speaker: Emilee Minalga
Presentation Title: Radio Frequency Coils for Magnetic Resonance Guided High Intensity Focused Ultrasound
Description/Abstract: The purpose of this work was to design and construct a magnetic resonance radio frequency (RF) coil that had better performance for imaging the breast than a single loop chest coil. The specific goal was to increase SNR in the breast for improved temperature measurement, cancer verification, and tissue characterization for MR guided high intensity focused ultra-sound (HIFU).
Supplemental material: Paper
Speaker: Emmanuel Bitaud
Presentation Title: Effects of Prenatal Cocain exposure on neonatal brain structure
Description/Abstract: The purposed of this project is to study effects of prenatal cocain exposure on neonatal brain development, based on MRI and DTI imaging techniques.
September 27th
Speaker: Yaniv Gur
Presentation Title: Recent trends in High Angular Resolution Diffusion Imaging (HARDI)
Description/Abstract: Diffusion Tensor Imaging (DTI) has been successfully used over the last two decades to infer the neural fiber connectivity in the brain. However, the classical DTI uses a single second-order tensor to model the diffusion profile within a brain voxel, hence, it fails to detect complex fiber structures such as crossing fibers. Over the last decade many attempts to overcome the limitations of DTI have been made. These attempts led to High Angular Resolution Diffusion Imaging (HARDI) which is a very active research area. In this image lunch talk I will present the progress has been recently made in SCI in this field, and I will discuss some interesting mathematical tools which are used to make this progress. Some preliminary results will be presented.
Additional information: Dr. Yaniv Gur is a postdoctoral fellow at SCI. His research interests are PDE-based methods in computer vision, differential geometry and Lie-group methods in computer vision, and medical imaging (DW-MRI). Dr. Yaniv Gur's SCI page is http://www.sci.utah.edu/people/yanivg.html
October 4th
Speaker: William Thompson
Presentation Title: Brief overview of CS 6650 -- Visual Perception from a Computer Graphics and Visualization Perspective
Description/Abstract: This course, to be offered in Spring 2010, provides an introduction to human visual perception intended for those studying or working in the fields of computer graphics and visualization. Much of the material is also relevant in motivating methods used in computer vision and image analysis. Visual neuroscientists will find that the emphsis in the course on visual performance complements studies of biological mechanisms.
Additional information: http://www.eng.utah.edu/~cs6650/
Speaker: Stanley Durrleman
Presentation Title: Comparison of the endocast growth of chimpanzees and bonobos via temporal regression and spatiotemporal registration
Description/Abstract: In this paper, we aim at characterizing and quantifying the differences between the growth of bonobos (Pan paniscus) and chimpanzees (Pan troglodytes). We use a collection of endocasts of wild-shot animals of both species. Each sample has been associated with a dental age, as a common temporal marker. To compare the endocasts, we used the current-based metric which allows us to quantify the shape differences without the need to find homologous landmarks on the surfaces. First, we perform a temporal shape regression, which estimates a typical growth scenario of the endocast for the bonobos and the chimpanzees. Then, a spatiotemporal registration scheme is used to quantify the differences between these two growth scenarios. The variations are decomposed into one morphological deformation and one time warp. The morphological deformation accounts for the anatomical differences independently of the age. The time warp accounts for the change of the dynamics of growth. It shows that the growth speed of the bonobos at juvenility is more than twice less than the one of the chimpanzees. This estimation gives more insights into the developmental delay observed in the bonobos growth.
Additional information: Dr. Stanley Durrleman is a postdoctoral fellow at SCI. His research interests are statistical analysis of 3D shapes in the context of Computational Anatomy, correspondence-free metric between shapes like currents, and registration using large diffeomorphic deformation. Dr. Stanley Durrleman's SCI page is http://www.sci.utah.edu/people/stanley.html
October 11st
Fall break
October 18th
Speaker: Firdaus Janoos
Presentation Title: Spatio-temporal Models of Cognitive Processes with fMRI
Description/Abstract: Understanding the highly complex, spatially distributed and temporally organized phenomena entailed by mental processes using functional MRI is an important research problem in cognitive and clinical neuroscience. Classically, the analysis of functional Magnetic Resonance Imaging (fMRI) has focused either on the creation of static maps localizing the metabolic fingerprints of neural processes or on studying their temporal evolution in a few pre-selected regions in the human brain. However, it widely acknowledged that cognition recruits the entire brain and that the underlying mental processes are fundamentally spatio-temporal in nature. By neglecting either the temporal dimension or the spatial entirety of brain function, such methods must necessarily compromise on extracting and representing all the information contained in the data.
In this work, I present a new paradigm to facilitate a multivariate spatio-temporal model that allows a time-resolved exploration of mental processes as captured by fMRI. Using a state-space formalism that models the brain transitioning through a sequence of cognitive states as it solves a mental task, we are able to study the spatial distribution of activity along with its temporal structure.
In addition to revealing the mental patterns of an individual subject, such a generative model enables group-level inferences in terms of information-theoretic properties such as complexity and mutual information. Efficient algorithms for estimating the parameters, state-sequence and the hemodynamic behavior of the brain have been developed.
This method was applied to a multi-subject fMRI study for developmental disorders such as dyslexia and dyscalculia (i.e. math learning disability). I shall show the kind of inferences possible with this method in analyzing and differentiating the groups and the neuro-scientific conclusions that it provides.
Additional information: Firdaus Janoos is a PhD candidate in the Computer Science and Engineering Dept at Ohio State University. His reseach interests are building models for the spatio-temporal analysis of functional neuro-imaging data, specifically fMRI. Firdaus Janoos's home page is http://www.cse.ohio-state.edu/~janoos/
October 25th
Speaker: Dafang Wang
Presentation Title: Simulating Inverse Problem in Electrocardiography: Finite Element Strategy and Variational-Formed Regularization
Description/Abstract: Inverse problem in electrocardiography (ECG) aims to non-invasively identify electrical activities of the heart from the electric potentials measured on the body surface. It is essentially an inverse source problem based on the elliptic equation, which models the bioelectric field within the human body. Computer simulation of ECG involves solving the mathematical equations over a domain approximating the realistic human anatomy. The challenge to solve the inverse problem lies in its ill-posed nature – small input errors may result in unbounded errors in the solution.
In this talk I will present two studies. First, I will present a finite element simulation study of the inverse ECG problem in terms of recovering epicardial potentials. I will discuss what is an optimal numerical approximation for the inverse problem, and how to achieve this goal with finite element discretization. The discretization strategy leads to two numerical methods, a) the use of hybrid-shaped finite elements, and b) a linear component truncation scheme from high-order finite elements.
To tackle the ill-conditioning of the inverse problem, regularization is often needed. The second topic is the proposition of a new family of regularizers formulated by the variational principle underlying finite element methods. These variational-formed regularizers work within the classic Tikhonov regularization framework. Compared with traditional Tikhonov regularizers, the variational regularizers are independent of discretization resolution and therefore achieve consistent regularization under multi-scale simulations. The variational formulation also enables a simple construction of a discrete gradient operator over irregular meshes, which is difficult to obtain from traditional discretization.[RM1] Since the formulation provides a new approach to evaluating a scalar field, it may bring benefits to other potential-based problems.
All simulation studies are validated with data obtained in a clinical experiment involving a live canine heart suspended in a human-torso-shaped electrolytic tank.
Additional information: Dafang Wang is a sixth year PhD candidate in computer science. His research is inverse problems in ECG/EEG, finite element methods and scientific computing. Dafang Wang's home page is http://www.sci.utah.edu/people/dfwang.html
November 1st
Speaker: Sam Preston
Presentation Title: AtlasWerks: Efficient large-scale atlas building
Description/Abstract: AtlasWerks is an open-source software package being developed here at SCI for creating 3D medical image atlases. This talk will give a high-level overview of the atlas building process, as well as a (hopefully) easy-to-follow explanation of the Large Deformation Diffeomorphic Metric Mapping equations, aimed at those without experience with nonlinear image registration techniques. I will also cover the tools available in the AtlasWerks package, and necessary information for running these tools on the GPU and on clusters of machines.
Additional information: Sam Preston is a research stuff at Scientific Computing and Imaging Institute (SCI), University of Utah. His research interests are Scientific visualization and Image processing. Sam Preston's home page is http://www.sci.utah.edu/people/jsam.html
November 8th
Speakers: Nikhil Singh, Fangxiang Jiao
Presentation Title: MICCAI 2010 review
Description/Abstract:
Nikhil Singh and Fangxiang Jiao went to the 13th Medical Image Computing and Computer-Assisted Intervention (MICCAI) 2010 (Sept. 20 - 24), they will give a review of MICCAI 2010.
Supplemental material:
Proceedings of MICCAI 2010 are available in Springer papers
November 15th
Speaker: Robert Oakes
Presentation Title: The Contribution of MEMS in Current Imaging Research
Description/Abstract:
Biomedical engineering is one of the fastest growing academic and industrial fields; thanks primarily to the research boons that have taken place during the recent decades. Two of the prominent research sectors in the field are the advancement of bio-imaging techniques and development of micro-scale functional devices, or micro-electrical-mechanical systems (MEMS). Current research is underway to bring these two aspects of biomedical engineering together; using MEMS components to develop devices designed for in-vivo and in-vitro purposes. MEMS fabrication techniques and components are currently being employed to create varying systems including an imaging catheter using spectral-domain optical coherence tomography (SD-OCT) for 3D endoscopic imaging, a micro-resonant device (MRD) as a solid-state contrast agent for magnetic resonance imaging (MRI), a two-photon microscope developed in a 2.9 g implantable package, and an implantable device enabling fluorescence imaging and electrophysiological experiments with built in CMOS sensor.
Speaker: Prasanna Muralidharan
Presentation Title: Rotation groups
Description/Abstract:
One might have come across rotation as an important recurring theme in Image analysis. The objective of this discussion would be to understand rotation matrices and its Lie group structure. I will introduce the notion of a Lie group and that of a Lie algebra. I will go on to discuss the exponential map associated with the Lie algebra. We will try and understand how these notions translate in the rotation groups SO(2) and SO(3).
November 22nd
Speaker: Yiming Kang
Presentation Title: Using, validating and comparing different image processing softwares to process diffusion DT-MRI tractography
Description/Abstract:
Yiming's summer research aims at using, validating and comparing different image processing softwares to process diffusion DT-MRI tractography. Fundamental procedure of forming myocardial fiber bundles includes converting data via inverse Fourier transform, estimating tensor field, performing tractography with various algorithms, and post-tractographic analysis of fiber bundles. The software SCIRun, MedINRIA, 3D Slicer and DTIStudio show individual advantageous features and disadvantages during the study. Among them, MedINRIA ranks the highest, because of its convenient operation, capability in performing tensor estimation and tractography, various algorithm choice and threshold, fast rendering, etc. At the same time, the different algorithms of estimating tensor field and processing DTI tractography, and setup for threshold are explored. This achievement will provide a better understanding of image processing, specially focused on tractographic techniques. It is considered to be an helpful and efficient approach to improve my computational skills for senior project. Converting data into required format also promote my programing skills. Some physiologic explorations are done towards myocardial fiber orientation and cardiomyopathy.
Speaker: Nishith Tirpankar
Presentation Title: Heuristic framework for the evaluation of video motion tracking algorithms
Description/Abstract:
Video tracking is a complex problem because the environment, in which video motion needs to be tracked, is widely varied based on the application and poses several constraints on the design and performance of the tracking system. Current datasets, which are used to evaluate and compare video motion tracking algorithms, use a cumulative performance measure without thoroughly analyzing the effect of these different constraints imposed by the environment. There is need to build a heuristic framework which analyzes these constraints as parameters of the framework and their effect on performance of tracking algorithm. The emphasis in this paper is to identify these parameters and define subjective measures for the comparison of datasets and performance evaluation of tracking algorithms.
November 29th
Speaker: Alan Cannaday II
Presentation Title: Regularization methods for the inverse Laplace transform
Description/Abstract: In many applications such as transistors, solar cells, LEDs, and diode lasers it is important to control the spontaneous photon emissions from atoms. One way of quantifying this effect is to embed quantum dots in the photonic crystal. An isolated quantum dot absorbs light at a given frequency and emits it with an exponentially decaying intensity with known decay rate. When many quantum dots are embedded in a photonic crystal and observed while absorbing and emitting light, the sum of their emissions or emission intensity I(t) at time t can be modeled using the Laplace transform, i.e. I(t)/I(0) = L[φ(γ)](t). W here φ(γ) is the distribution of concentration of emitters for a certain decay rate γ. The problem of determining φ(γ) from its Laplace transform I(t) is known to be a severely ill-posed linear inverse problem, which means that small per- turbations in I(t) can lead to large errors in the estimation of φ(γ). For my project I explored different regularization methods for the inverse Laplace transform proposed by Epstein and Schotland (2008) that allows us to compute the inverse Laplace transform without requiring prior information about φ(γ).
Speaker: Jihwan Kim
Presentation Title: Optical Flow based on Horn and Schunk's paper.
Description/Abstract: Optical flow gives important information about the spacial arrangement of the objects viewed and the rate of change of this arrangement. Horn's paper "Determining Optical Flow" is one of the fundamental papers about optical flow. By reviewing Horn's paper and implementing a Matlab code, I have tried to understand the optical flow.
December 6th
Speaker: James Fishbaugh
Presentation Title: Force Driven Longitudinal Shape Regression
Description/Abstract: A generic non-parametric approach to longitudinal shape regression involves finding the optimal flow of diffeomorphisms, parameterized by time-varying momentum vectors, that strikes a balance between fidelity to data and regularity of deformation. In the current implementation of this framework, a baseline shape is continuously deformed over time to match target shapes, with the total kinetic energy of the deformation measuring regularity. By minimizing the trade-off between shape similarity and total kinetic energy, the resulting flow of diffeomorphisms is piecewise geodesic in the space of diffeomorphisms. While this piecewise geodesic property provides a continuous shape evolution, the resulting flow is not differentiable everywhere, as shape evolution can change instantaneously at observation points due to discontinuities in the time-varying velocity field. We propose an extension to this framework that ensures the flow of diffeomorphisms is differentiable everywhere. This is accomplished by parameterizing the flow of diffeomorphisms by "force" (acceleration), the derivative of momentum with respect to time. This parameterization guarantees that the velocity field is differentiable everywhere, resulting in smoother shape evolution across time.
Speaker: Miaomiao Zhang
Presentation Title: Single image dehazing and denoising
Description/Abstract: We will present a model to easily remove the effects of haze and noise which is introduced by high ISO from a single real color photo simultaneously. We construct an energy minimization functional, and make use of dark channel prior to accelerate the convergence of the cost functional. In addition, soft matting algorithm is also used to refine the results.
Speaker: Caleb Rottman
Presentation Title: Denoising Fluoroscopy Data
Description/Abstract: SCI has been working with GE Healthcare in attempt to reduce the ionizing radiation that a patient receives during a fluoroscopy procedure. Real-time denoising techniques must be used to achieve a higher signal-to-noise ratio. This presentation will be covering the current attempts by SCI to denoise these images using bilateral filtering, PCA filtering, and non-local means filtering.
