Center for Vision, Graphics and Medical Imaging

Date: Thursday November 15, 2007
Place: CSE E305
Time: 3:00 pm to 4:30 pm
Speaker: Nicholas Fisher
Topic: Seizure Detection via Spike Response Neuron Modeling
Source: Contact Neko
Abstract:
Epilepsy is one of the leading neurological disorders in the world effecting 1 in 100 people. Any technique that could detect a seizure in its early stages or predict one before it happens could vastly improve the quality of life of these people. Here we consider a method for seizure detection using spike response neurons driven by electroencephalogram data (EEG). By convolving random kernels with the eeg signal to drive the neurons, we create a Volterra series framework so that any time invariant system could be represented. We then use a support vector machine to try and classify spike trains generated by the seizure state versus those generated by the normal state.

Date: Thursday November 8, 2007
Place: CSE E305
Time: 3:00 pm to 4:30 pm
Speaker: Muhammad Rushdi
Topic: Speckle Reduction in Medical Ultrasound: A Novel Scatterer Density Weighted Nonlinear Diffusion Algorithm Implemented as a Neural-Network Filter
Source: Link to article
Abstract:
This paper proposes a novel algorithm for speckle reduction in medical ultrasound imaging while preserving the edges with the added advantages of adaptive noise filtering and speed. We propose a nonlinear image diffusion algorithm that incorporates two local parameters of image quality, namely, scatterer density and texture-based contrast in addition to gradient, to weight the nonlinear diffusion process. The scatterer density is proposed to replace the existing traditional measures of quality of the ultrasound diffusion process such as MSE, RMSE, SNR, and PSNR. This novel diffusion filter was then implemented using backpropagation neural network for fast parallel processing of volumetric images. The experimental results show that weighting the image diffusion with these parameters produces better noise reduction and produces a better edge detection quality with reasonable computational cost. The proposed filter can be used as a preprocessing phase before applying any ultrasound segmentation or active contour model processes

Date: Thursday November 1, 2007
Place: CSE E305
Time: 3:00 pm to 4:30 pm
Speaker: Nathan VanderKraats
Topic: Mutual Information Techniques for Spiketrain Analysis
Source: Contact Nathan
Abstract:
Through the years, information theory has provided an invaluable framework for research in a variety of areas ranging from electrical engineering to statistical physics. For this talk, we will explore the application of mutual information to a biological system: measuring transmission of information through feedforward networks of spiking neurons. The presentation will highlight both the benefits and difficulties inherent in this approach. We will also discuss some approximation techniques used in current research.

Date: Thursday October 25, 2007
Place: CSE E305
Time: 3:00 pm to 4:30 pm
Speaker: Niranjan Joshi
Topic: Nonparametric probability density function estimation for medical mage analysis
Source: Contact Niranjan
Abstract:
In this talk I will discuss a recently proposed nonparametric probability density function (PDF) estimation method for sampled signals called the non-parametric (NP) window method. Concept of this estimator relies upon interpolation of sampled signals to obtain better PDF estimates. I will present four different analytical expressions of obtaining the NP window estimate. These expressions will then be used to compare the NP window method with existing nonparametric PDF estimators, namely the histogram and the kernel density estimator. In the second part of my talk, I will present an image segmentation method based on the level set curve evolution. The NP window estimator described earlier will be used to drive evolution the level curves. In the final part of my talk, I will discuss the application of NP windows based level set method to segment colorectal cancer magnetic resonance (MR) images. The segmentation of these images is difficult due to complex anatomy of the colorectal region and various artifacts. I will present some of the latest results I have obtained for segmentation of the colorectal cancer images. I will conclude this talk by mentioning some future work we hope to undertake.

Date: Thursday October 11, 2007
Place: CSE E305
Time: 3:00 pm to 4:30 pm
Speaker: Angelos Barmpoutis
Topic:Symmetric Positive 4th Order Tensors & their Estimation from Diffusion Weighted MRI
Source: Contact Angelos
Abstract:
In Diffusion Weighted Magnetic Resonance Image (DW-MRI) processing a 2nd order tensor has been commonly used to approximate the diffusivity function at each lattice point of the DW-MRI data. It is now well known that this 2nd-order approximation fails to approximate complex local tissue structures, such as fibers crossings. In this paper we employ a 4th order symmetric positive semi-definite (PSD) tensor approximation to represent the diffusivity function and present a novel technique to estimate these tensors from the DW-MRI data guaranteeing the PSD property. There have been several published articles in literature on higher order tensor approximations of the diffusivity function but none of them guarantee the positive semi-definite constraint, which is a fundamental constraint since negative values of the diffusivity coefficients are not meaningful. In our methods, we parameterize the 4th order tensors as a sum of squares of quadratic forms by using the so called Gram matrix method from linear algebra and its relation to the Hilbert's theorem on ternary quartics. This parametric representation is then used in a nonlinear-least squares formulation to estimate the PSD tensors of order 4 from the data. We define a metric for the higher-order tensors and employ it for regularization across the lattice. Finally, performance of this model is depicted on synthetic data as well as real DW-MRI from an isolated rat hippocampus.

Date: Thursday October 4, 2007
Place: CSE E305
Time: 3:00 pm to 4:30 pm
Speaker: S. M. Shahed Nejhum
Topic: Progressive Finite Newton Approach To Real-time Nonrigid Surface Detection
Source: Paper
Abstract:
Nonrigid surface detection is usually regarded as a robust parameter estimation problem, which is typically solved iteratively from a good initialization in order to avoid local minima. This paper proposes a novel progressive finite Newton optimization scheme for the nonrigid surface detection problem, which is reduced to only solving a set of linear equations. The key approach is to formulate the nonrigid surface detection as an unconstrained quadratic optimization problem which has a closed-form solution for a given set of observations. Also, a progressive active-set selection scheme is employed, which takes advantage of the rank information of detected correspondences.

Date: Thursday September 27, 2007
Place: CSE E305
Time: 3:00 pm to 4:30 pm
Speaker: Karthik Gurumoorthy
Topic: Uniform Glivenko-Cantelli Theorems and Concentration of Measure in the Mathematical Modelling of Learning
Source: Link to paper.
Abstract:
This paper surveys certain developments in the use of probabilistic techniques for the modelling of generalization in machine learning.

Date: Thursday September 20, 2007
Place: CSE E305
Time: 3:00 pm to 4:30 pm
Speaker: Angelos Barmpoutis
Canceled:
Moved to October 11, 2007.

Date: Thursday September 13, 2007
Place: CSE E305
Time: 3:00 pm to 4:30 pm
Speaker: Adrian M. Peter
Topic:Information Geometry for Landmark Shape Ananlysis
Source: Applicable papers.
Abstract:
This talk was presented in October 2006 at the Mathematical Foundations of Computational Anatomy (workshop at MICCAI). We will discuss a new method for matching landmark shapes in which the deformation model between shapes comes directly from their representation as Gaussian mixtures.

Date: Thursday September 6, 2007
Place: CSE E305
Time: 3:00 pm to 4:30 pm
Speaker: Dr. Alireza Entezari
Topic:Optimal Sampling Lattices and Splines
Source: Contact Dr. Entezari
Abstract:
Faithful and accurate reconstruction of functions from a set of given data points is an integral part of many visualization, computer graphics and scientific computing algorithms. In this talk I will demonstrate how we exploit the optimal sphere packing lattices to advance the state-of-the-art visualization algorithms. Not only these dense packing lattices significantly increase the reconstruction accuracy, but also they minimize the computational cost of the reconstruction. This striking result is counter-intuitive to the general understanding about the traditionally popular Cartesian methods.

Discretization and reconstruction (interpolation) of data is also common to other computational fields such as medical imaging, vision, machine learning and numerical solutions to partial differential equations. I will discuss advantages of exploiting the optimal lattices in some of these areas.

Date: Wednesday 18th April, 2007
Place: CSE E305
Time: 4:00 pm
Speaker: Seniha Esen Yuksel
Topic:Trace Inference, Curvature Consistency, and Curve Detection (a paper by Pierre Parent and Steven W. Zucker, IEEE PAMI, 1989.)
Source: Click here
Abstract:
This paper describes a novel approach to curve inference based on curvature information. The inference procedure is divided into two stages: a trace inference stage, to which this paper is devoted, and a curve synthesis stage, which will be treated in a separate paper. It is shown that recovery of the trace of a curve requires estimating local models for the curve at the same time, and that tangent and curvature information are sufficient. These make it possible to specify powerful constraints between estimated tangents to a curve, in terms of a neighborhood relationship called co-circularity and between curvature estimates, in terms of a curvature consistency relation. Because all curve information is quantized, special care must be taken to obtain accurate estimates of trace points, tangents and curvatures. This issue is addressed specifically by the introduction of a smoothness constraint and a maximum curvature constraint. The procedure is applied to two types of images, artificial images designed to evaluate curvature and noise sensitivity, and natural images.

Date: Wednesday 11th April, 2007
Place: CSE E305
Time: 4:00 pm
Speaker: Bing Jian
Topic: Computational Diffusion MRI, Robust Point Set Registration, Multimodal Image Registration
Source: Click here

Date: Wednesday 4th April, 2007
Place: CSE E305
Time: 4:00 pm
Speaker: Dr. Jose C. Principe
Topic: Information Theoretic Learning
Source: ITL

Date: Wednesday 28th March, 2007
Place: CSE E305
Time: 4:00 pm
Speaker: Angelos Barmpoutis
Topic:Exponential Tensors: A Framework for Efficient Higher-order DT-MRI Computations (accepted at ISBI 2007)
Source: PDF
Time: 4:30 pm
Speaker: Bing Jian
Topic: A Continuous Mixture of Tensors Model for Diffusion-Weighted MR Signal Reconstruction (accepted at ISBI 2007)
Source: PDF

Date: Wednesday 21st, 28th February , 2007
Topic: Clustering by Passing Messages Between Data Points
Source: (A paper by Brendan J. Frey and Delbert Dueck)
Speaker: Venkat
Place: CSE E305
Time: 4 pm
Abstract:
Clustering data by identifying a subset of representative examples is important for processing sensory signals and detecting patterns in data. Such "exemplars" can be found by randomly choosing an initial subset of data points and then iteratively refining it, but this works well only if that initial choice is close to a good solution. We devised a method called "affinity propagation," which takes as input measures of similarity between pairs of data points. Real-valued messages are exchanged between data points until a high-quality set of exemplars and corresponding clusters gradually emerges. We used affinity propagation to cluster images of faces, detect genes in microarray data, identify representative sentences in this manuscript, and identify cities that are efficiently accessed by airline travel. Affinity propagation found clusters with much lower error than other methods, and it did so in less than one-hundredth the amount of time.

Date: Wednesday 1st November and 8th November, 2006.
Topic: A framework for efficient higher-order DT-MRI computations.
Speaker: Angelos Barmpoutis
Place: CSE E305
Time: 4 pm
Abstract:
In DT-MRI processing a 2nd order tensor has been commonly used to approximate the diffusivity profile at each lattice point of the 2D or 3D image. These tensors are symmetric positive definite matrices and the appropriate handling of this property increases significantly the complexity and the execution times of the algorithms. In this paper we present a novel parameterization of the diffusivity profile. Using this parameterization the positive definite property of the diffusivity is guaranteed without the need of any further computation. This parameterization can also be used for higher order approximations; we present 2, 4, 6 and 8th order tensor approximations. Furthermore we present an efficient framework for computing distances and geodesics on the space of the coefficients of our proposed diffusivity function. We validate and compare our method with other existing methods using synthetically generated data and real diffusion weighted MR data from two anatomical regions of excised, perfusion-fixed rat nervous tissue: a)optic chiasm and b) brain.

Date: Wednesday 11th October, 2006.
Topic: A Conic Section Classifier and its Application to Image Datasets.
Source: A Conic Section Classifier and its Application to Image Datasets. (accepted at CVPR 2006).
LINK TO PAPER
Speaker: Santhosh Kodipaka
Place: CSE E305
Time: 4 pm
Abstract:
Many problems in computer vision involving recognition and/or classification can be posed in the general framework of learning theory. There is however one aspect of image datasets, the high-dimensionality of the datapoints, that makes the direct application of off-the-shelf learning techniques problematic. In this paper, we present a novel concept class and a companion tractable algorithm for learning a suitable classifier from a given labeled dataset, that is particularly suited to high-dimensional sparse datasets. Each member class in the dataset is represented by a prototype conic section in the feature space, and new data points are classified based on a distance measure to each such representative conic section that is parameterized by its focus, directrix and eccentricity. Learning is achieved by altering the parameters of the conic section descriptor for each class, so as to better represent the data. We demonstrate the efficacy of the technique by comparing it to several well known classifiers on multiple public domain datasets.

Date: Wednesday 4th October, 2006.
Topic: A Novel Tensor Distribution Model for The Diffusion Weighted MR Signal.
Speaker: Bing Jian
Place: CSE E305
Time: 4:00 pm
Abstract:
Diffusion MRI is a non-invasive imaging technique that allows the measurement of water molecular diffusion through tissue \emph{in vivo}. In this paper, we present a novel statistical model which describes the diffusion-attenuated MR signal by the Laplace transform of a probability distribution over symmetric positive definite matrices. Using this new model, we analytically derive a Rigaut-type asymptotic fractal law for the MR signal decay behavior which has only been phenomenologically used before. We also develop an efficient scheme for reconstructing the multiple fiber bundles from the DW-MRI measurements. Experimental results on both synthetic and real data sets are presented to show the robustness and accuracy of the proposed algorithms.

Date: Wednesday 20th, 27th September, 2006.
Topic: An Intrinsic Geometric Framework for Simultaneous Non-rigid Registration and Segmentation of Surfaces.
LINKS TO TECH REPORT
An Intrinsic Geometric Framework for Simultaneous Non-rigid Registration and Segmentation of Surfaces (CISE, UF TR 2006). (accepted at MFCA, 2006 )
Speaker: Nicholas Andrew Lord
Place: CSE E305
Time: 4:00 pm
Abstract:
In clinical applications where structural asymmetries between homologous shapes have been correlated with pathology, the questions of definition and quantification of 'asymmetry' arise naturally. When not only the degree but the position of deformity is thought relevant, asymmetry localization must also be addressed. Asymmetries between paired shapes can and have been formulated in the literature in terms of (nonrigid) diffeomorphisms between the shapes. For the infinity of such maps possible for a given pair, we define optimality as the minimization of total distortion, where 'distortion' is in turn defined as deviation from isometry. We thus propose a novel variational formulation for segmenting asymmetric regions from surface pairs based on the minimization of a functional of both the deformation map and the segmentation boundary, which controls gradient discontinuity of the map. This minimization is achieved via a quasi-simultaneous evolution of the map and curve. Our formulation is inherently intrinsic and parameterization-independent. We present examples using both synthetic data and pairs of left and right hippocampal structures, hippocampus malformation being linked with such neurological disorders as epilepsy and schizophrenia.

Date: Wednesday 13th September, 2006.
Topic:Image Processing Methods for the Detection of Acute Rejection after Kidney Transplantation.
LINKS TO PAPERS
Automatic detection of renal rejection after kidney transplantation (CARS 2005)
2D and 3D shape-based segmentation using deformable models (MICCAI 2005)
Speaker: Seniha Esen Yuksel
Place: CSE E305
Time: 4:00 pm
Abstract:
Acute rejection is the most common reason of graft failure after kidney transplantation, and early detection is crucial to survive the transplanted kidney function. In this study we introduce a new framework for the automatic classification of normal and acute rejection transplants from Dynamic Contrast Enhanced Magnetic Resonance Images (DCE-MRI). The proposed framework consists of three main steps. In the first step, mutual registration is employed to account for the motion of the kidney due to patient breathing. The second step isolates the kidney from the surrounding anatomical structures by evolving a deformable model based on two density functions; the first function describes the distribution of the gray level inside and outside the kidney region and the second function describes the prior shape of the kidney. In the third step, the perfusion curves that show the transportation of the contrast agent into the tissue are obtained from the segmented cortex of the whole image sequence of the patient. At the final step, we collect four features from these curves and use Bayesian classifiers to distinguish between acute rejection and normal transplants. Applications of the proposed approach yield promising results that may in the future be used to replace the use of current technologies such as nuclear imaging and ultrasonography, which are not specific enough to determine the type of kidney dysfunction.

Date: Wednesday 6th September, 2006.
Topic: A New Closed-Form Information Metric for Shape Analysis.
Source: A paper by Adrian Peter and Anand Rangarajan (accepted recently at MICCAI 2006).
LINKS TO PAPERS
Shape Analysis Using the Fisher-rao Riemmanian Metric: Unifying Shape Representation and Deformation (ISBI 2006)
Speaker: Adrian Peter
Place: CSE E305
Time: 3:30 pm
Abstract:
The purpose of this talk is to introduce a new technique for landmark-based shape matching. The unique aspect of this method is that is combines both the shape representation and deformation. The representation model uses parametric mixture models to describe landmark shapes. Given such a model, we will show how this leads to a geometric interpretation of a manifold on the parameter space of the mixture density. On this space, information matrices usually found in statistical analysis have an alternate interpretation as the natural metric tensor to intrinsically traverse the manifold. Consequently, this allows us to "walk" on the space to deform shapes and establish intrinsic distances between them. The metrics are directly derived from the parametric model -- thus coupling both the shape representation and deformation. We illustrate utility of this new method through pairwise matching of 9 corpora callosa landmark shapes.

Date: Monday 12th June, 2006.
Topic: A New Method Probability Density Estimation with Application to Mutual Information Based Image Registration.
Source: A paper by Ajit Rajwade, Arunava Banerjee and Anand Rangarajan (accepted recently at CVPR 2006). LINK TO PAPER
Speaker: Ajit Rajwade
Place: CSE E305
Time: 3 pm
Abstract:
We present a new, robust and computationally efficient method for estimating the probability density of the intensity values in an image. Our approach makes use of a continuous representation of the image and develops a relation between probability density at a particular intensity value and image gradients along the level sets at that value. Unlike traditional sample-based methods such as histograms, minimum spanning trees (MSTs), Parzen windows or mixture models, our technique expressly accounts for the relative ordering of the intensity values at different image locations and exploits the geometry of the image surface. Moreover, our method avoids the histogram binning problem and requires no critical parameter tuning. We extend the method to compute the joint density between two or more images. We apply our density estimation technique to the task of affine registration of 2D images using mutual information and show good results under high noise.

Date: Monday 12th June, 2006.
Topic: Groupwise point pattern registration using a novel CDF-based Jensen-Shannon Divergence.
Source: A paper by Fei Wang, Baba Vemuri and Anand Rangarajan (accepted recently at CVPR 2006). LINK TO PAPER
Speaker: Fei Wang
Place: CSE E305
Time: 3 pm
Abstract:
In this paper, we propose a novel and robust algorithm for the groupwise non-rigid registration of multiple unlabeled point-sets with no bias toward any of the given point-sets. To quantify the divergence between multiple probability distributions each estimated from the given point sets, we develop a novel measure based on their cumulative distribution functions that we dub the CDF-JS divergence. The measure parallels the well known Jensen-Shannon divergence (defined for probability density functions) but is more regular than the JS divergence since its definition is based on CDFs as opposed to density functions. As a consequence, CDF-JS is more immune to noise and statistically more robust than the JS. We derive the analytic gradient of the CDF-JS divergence with respect to the non-rigid registration parameters for use in the numerical optimization of the groupwise registration leading a computationally efficient and accurate algorithm. The CDF-JS is symmetric and has no bias toward any of the given point-sets, since there is NO fixed reference data set. Instead, the groupwise registration takes place between the input data sets and an evolving target dubbed the pooled model. This target evolves to a fully registered pooled data set when the CDF-JS defined over this pooled data is minimized. Our algorithm is especially useful for creating atlases of various shapes (represented as point distribution models) as well as for simultaneously registering 3D range data sets without establishing any correspondence. We present experimental results on non-rigid registration of 2D/3D real point set data.

Date: Thursday 20th April, 2006.
Topic: Matching 3-D anatomical surfaces with non-rigid deformations using octree-splines.
Source: A paper by R. Szeliski and S. Lavallée from IJCV, 1996.
Speaker: Ritwik Kumar
Place: CSE E305
Time: 4 pm
Abstract:
This paper presents a new method for determining the minimal non-rigid deformation between two 3-D surfaces, such as those which describe anatomical structures in 3-D medical images. Although we match surfaces, we represent the deformation as a volumetric transformation. Our method performs a least squares minimization of the distance between the two surfaces of interest. To quickly and accurately compute distances between points on the two surfaces, we use a precomputed distance map represented using an octree spline whose resolution increases near the surface. To quickly and robustly compute the deformation, we use a second octree spline to model the deformation function. The coarsest level of the deformation encodes the global (e.g., affine) transformation between the two surfaces, while finer levels encode smooth local displacements which bring the two surfaces into closer registration. We present experimental results on both synthetic and real 3-D surfaces.

LINK TO PAPER

Date: Thursday 23rd March and 13th April, 2006.
Topic: A Conic Section Classifier and its Application to Image Datasets.
Source: A paper by Arunava Banerjee, Santhosh Kodipaka and Baba Vemuri (accepted recently at CVPR 2006).
Speaker: Santhosh Kodipaka
Place: CSE E305
Time: 4 pm
Abstract:
Many problems in computer vision involving recognition and/or classification can be posed in the general framework of learning theory. There is however one aspect of image datasets, the high-dimensionality of the datapoints, that makes the direct application of off-the-shelf learning techniques problematic. In this paper, we present a novel concept class and a companion tractable algorithm for learning a suitable classifier from a given labeled dataset, that is particularly suited to high-dimensional sparse datasets. Each member class in the dataset is represented by a prototype conic section in the feature space, and new data points are classified based on a distance measure to each such representative conic section that is parameterized by its focus, directrix and eccentricity. Learning is achieved by altering the parameters of the conic section descriptor for each class, so as to better represent the data. We demonstrate the efficacy of the technique by comparing it to several well known classifiers on multiple public domain datasets.

LINK TO PAPER

Date: Thursday 9th March, 2006
Topic: Simultaneous Nonrigid Registration of Multiple Point Sets and Atlas Construction.
Source: A paper by Fei Wang, Baba Vemuri and Anand Rangarajan (accepted recently at ECCV 2006).
Speaker: Fei Wang
Place: CSE E404
Time: 4 pm
Abstract:
Estimating a meaningful average or mean shape from a set of shapes represented by unlabeled point-sets is a challenging problem since, usually this involves solving for point correspondence under a non-rigid motion setting. In this paper, we propose a novel and robust algorithm that is capable of simultaneously computing the mean shape from multiple unlabeled point-sets (represented by finite mixtures) and registering them nonrigidly to this emerging mean shape. This algorithm avoids the correspondence problem by minimizing the Jensen-Shannon (JS) divergence between the point sets represented as finite mixtures. We derive the analytic gradient of the cost function namely, the JS-divergence, in order to efficiently achieve the optimal solution. The cost function is fully symmetric with no bias toward any of the given shapes to be registered and whose mean is being sought. Our algorithm can be especially useful for creating atlases of various shapes present in images as well as for simultaneously (rigidly or non-rigidly) registering 3D range data sets without having to establish any correspondence. We present experimental results on non-rigidly registering 2D as well as 3D real data (point sets).

Date: Thursday 2nd March, 2006 (POSTPONED)
Topic: Curve Evolution Implementation of the Mumford-Shah Functional for Image Segmentation, Denoising, Interpolation and Magnification.
Source: A paper by Tsai, Yezzi and Willsky published in IEEE Transactions on Image Processing, August 2001 (see link below).
Speaker: Nicholas Andrew Lord
Place: CSE E404
Time: 4 pm
Abstract:
In this work, Tsai, Yezzi and Willsky first address the problem of simultaneous image segmentation and smoothing by approaching the Mumford.Shah paradigm from a curve evolution perspective. In particular, we let a set of deformable contours define the boundaries between regions in an image where we model the data via piecewise smooth functions and employ a gradient flow to evolve these contours. Each gradient step involves solving an optimal estimation problem for the data within each region, connecting curve evolution and the Mumford.Shah functional with the theory of boundary-value stochastic processes. The resulting active contour model offers a tractable implementation of the original Mumford.Shah model (i.e., without resorting to elliptic approximations which have traditionally been favored for greater ease in implementation) to simultaneously segment and smoothly reconstruct the data within a given image in a coupled manner. Various implementations of this algorithm are introduced to increase its speed of convergence.We also outline a hierarchical implementation of this algorithm to handle important image features such as triple points and other multiple junctions. Next, by generalizing the data fidelity term of the original Mumford. Shah functional to incorporate a spatially varying penalty, we extend our method to problems in which data quality varies across the image and to images in which sets of pixel measurements are missing. This more general model leads us to a novel PDE-based approach for simultaneous image magnification, segmentation, and smoothing, thereby extending the traditional applications of the Mumford.Shah functional which only considers simultaneous segmentation and smoothing.

Date: Thursday 23rd February, 2006
Topic: Tensor Splines for Interpolation of Diffusion Tensor MRI Data
Speaker: Angelos Barmpoutis
Place: CSE E404
Time: 4 pm
Abstract:
In this paper, we present a novel and robust spline interpolation algorithm given a noisy symmetric positive definite (SPD) tensor field. Such tensor fields commonly arise in the field of Medical Imaging in the form of Diffusion Tensor (DT) MRI data sets. We develop a statistically robust algorithm for constructing a tensor product of B-splines -- for interpolating these data -- using the Riemannian metric of the manifold of SPD tensors. Our method involves a two step procedure wherein the first step uses Riemannian distances in order to evaluate a tensor spline by computing a weighted intrinsic average of diffusion tensors and the second step involves minimization of the Riemannian distance between the evaluated spline curve and the given data. These two steps are alternated to achieve the desired tensor spline approximation to the given tensor field. We present comparisons of our algorithm with three existing methods of tensor interpolation applied to DT-MRI data from fixed heart slices of a rabbit, and show significantly improved results in the presence of noise and outliers. We also present validation results for our algorithm using synthetically generated noisy tensor field data with outliers. This interpolation work has many applications e.g., in DT-MRI registration, in DT-MRI Atlas construction etc.

Date: Thursday 16th February, 2006
Topic: Solving Eikonal Equation to get Depths from Normal Vector Fields
Speaker: Dr. Jeffrey Ho
Place: CSE E404
Time: 4 pm
Abstract:
Integration of surface normal vectors is a vital component in many shape reconstruction algorithms that require integrating surface normals to yield their final outputs, the depth values. In this paper, we introduce a fast and efficient method for computing the depth values from surface normal vectors. The method is based on solving the Eikonal equation using the Fast Marching Method. We introduce two ideas. First, while it is not possible to solve for the depths, $Z$, directly using Fast Marching Method, we solve the Eikonal equation for a function $W$ of the form $W=Z+\lambda f$. With appropriately chosen values for $\lambda$, we can ensure that the Eikonal equation for $W$ can be solved using Fast Marching Method. Second, we solve for $W$ in two stages with two different $\lambda$ values, first in a small neighborhood of the given initial point with large $\lambda$, and then for the rest of the domain with a small $\lambda$. The proposed method is very easy to implement, and extremely efficient and fast.

Date: Thursday 9th February, 2006
Topic: Epileptic Seizure Prediction Using Model Networks of Spiking Neurons
Speaker: Nicholas Fisher
Place: CSE E404
Time: 4 pm
Abstract:
Background and Significance
Epilepsy, broadly defined, is a brain disorder in which certain groups of neurons display "abnormal" spiking behavior over brief intervals of time. These episodes, called ictal (seizure) states, can occur spontaneously and are usually interspersed by prolonged normal (interictal) states. During an ictal state, the normal pattern of neuronal activity becomes disturbed, causing strange sensations, emotions, and behavior, and in certain cases convulsions, muscle spasms, and loss of consciousness. Epilepsy ranks amongst the most widespread of brain disorders in the world today. According to some estimates, 7 out of every 1000 individuals (approximately 40 million individuals worldwide) are afflicted by this chronic disease. Moreover, incidence rates range from 24 to 53 per 100, 000. For those that suffer from it, socially mandated restrictions such as the withholding of driving privileges, impose serious limitations to their way of life. Although epilepsy is known to occur in all age groups, incidence rates are higher in children and the elderly. Since epilepsy can result from a large number of causes, including brain insults such as craniofacial trauma, nervous system infections, brain tumors, hypoxia, febrile convulsions, ischemia, as well as genetic and developmental anomalies, it afflicts members of all socio-economic, racial, as well as geographical groups. Given the widespread incidence of epilepsy, any device/technique that could accurately predict an oncoming seizure by more than five to ten minutes, from continuously recorded Electroencephalogram (EEG) signals, or even from intracranial field potentials recorded using chronically implanted electrodes, could immensely benefit society. In this talk, we present a novel technique for on-line seizure prediction that is based on a phase-space analysis of the dynamics of a model system of spiking neurons driven by the EEG/intracranial field potential signal.

Technique
The majority of the current, state-of-the-art, techniques used to detect or predict an epileptic seizure involve linearly/non-linearly transforming the EEG signal using one of several mathematical black boxes, and subsequently trying to predict or detect the seizure based off the output of the black box. These black boxes include some purely mathematical transformations, like the Fourier transform, or some class of machine learning techniques, like artificial neural networks, or some combination of the two. The fact that an EEG signal is generated in a particular biological context, and is representative of a particular physical aspect of the system, does not play a significant role in these techniques. In this talk, we present a novel technique for seizure detection and prediction that is based on the core observation that the spread of seizures in brains are the outcome of the interactions between the dynamics of normally operating systems of neurons and systems that have already evolved to a state of seizure. We consider the EEG signal as an impoverished representation of the total activity in a group of interconnected neurons in the brain. Our system drives a model system of spiking neurons using this EEG signal and subsequently uses the resulting dynamics of the system to detect and predict oncoming seizures. In effect, we attempt to mimic the process of the spread of a seizure described above. By analyzing the substantially richer information available in the spatio-temporal spike dynamics of the model system of neurons, we hope to improve the predictive process. Our system is based off an abstract dynamical system where the spiking nature of the neuron assumes center-stage. The system is based on a limited set of realistic assumptions, and hence accommodates a wide range of neuronal models. The system has been shown to be chaotic in nature. The chaotic dynamics of our system has significant implications for the class of analyses that may be applied to detect or predict oncoming seizures in the system. This allows us to consider various signatures of chaotic systems, that could be an indication of an imminent seizure. The simplest signature that we shall consider is the total number of excitatory and inhibitory spikes in the successive state descriptions of the system as its dynamics evolves. We will also consider other signatures like the sensitivity index, the oscillation index, and the velocity index of the trajectories created by the system.

Date: Thursday 2nd February, 2006
Topic: Sparse LU factorization: looking left, looking right
Speaker: Dr. Tim Davis
Place: CSE E404
Time: 4 pm
Abstract:
The two most commonly used methods for sparse LU factorization are the left-looking method and the right-looking method. Both are used internally in MATLAB. In this talk, specific examples of these two methods will be discussed and compared. In the left-looking method, the kth step of factorization computes the kth column of L and U from the kth column of A and columns 1 to k-1 of L and U. In the sparse case, it has a particularly elegant implementation (due to Gilbert and Peierls), and is the only known algorithm whose run time is O(flop count). The right-looking method is based on outer products. In the multifrontal method, these outer products are held in small square or rectangular dense submatrices. The method is much more complex than the left-looking method, and it is not guaranteed to run in time O(flop count). However, it can exploit dense matrix operations (the BLAS) better than the left-looking method. For each method, specific classes of problems generate matrices that are best factorized with that method.

Date: Thursday 17th November, 2005
Topic: The Role of Biologically Inspired Spikes in the Design of Ultra-Wide Dynamic Range Imaging Systems.
Speaker: Dr. John G. Harris
Place: CSE E404
Time: 4 pm
Abstract:
There is growing interest in using biological inspiration to improve the design of computation systems, particularly in the areas of sensory processing and pattern recognition where biological systems far outperform the best man-made systems. In this talk, we consider the role of biologically-inspired spike representations in the design of a new generation of ultra-wide dynamic range CMOS imagers. These spike-based designs more than double the dynamic range of conventional imaging systems.

Date: Thursday 13th October, 2005
Topic: Simultaneous Registration and Segmentation of Anatomical Structures from Brain MRI.
Speaker: Fei Wang
Place: CSE E404
Time: 4 pm
Abstract:
In this paper, we present a novel variational formulation of the registration assisted image segmentation problem which leads to solving a coupled set of nonlinear PDEs that are solved using efficient numerical schemes. Our work is a departure from earlier methods in that we have a unified variational principle wherein non-rigid registration and segmentation are simultaneously achieved; unlike previous methods of solution for this problem, our algorithm can accommodate for image pairs having very distinct intensity distributions. We present examples of performance of our algorithm on synthetic and real data sets along with quantitative accuracy estimates of the registration.

Link to Paper

Date: Thursday 6th October, 2005
Topic: Multimodality Image Registration Using an Extensible Information Metric and High Dimensional Histogramming.
Speaker: Dr. Anand Rangarajan
Place: CSE E404
Time: 4 pm
Abstract:
We extend an information metric from intermodality (2- image) registration to multimodality (multiple-image) registration so that we can simultaneously register multiple images of different modalites. And we also provide the normalized version of the extensible information metric, which has better performance in high noise situations. Compared to mutual information which can even become negative in the multiple image case, our metric can be easily and naturally extended to multiple images. After using a new technique to efficiently compute high dimensional histograms, the extensible information metric can be efficiently computed even for multiple images. To showcase the new measure, we compare the results of direct multimodality registration using high-dimensional histogramming with repeated intermodality registration. We find that registering 3 images simultaneously with the new metric is more accurate than pair-wise registration on 2D images obtained from synthetic magnetic resonance (MR) proton density (PD), MR T2 and MR T1 3D volumes from Brain Web. We perform the unbiased registration of 5 multimodality images of anatomy, CT, MR PD, T1 and T2 from Visible Human Male Data with the normalized metric and high-dimensional histogramming. Our results demonstrate the efficacy of the metrics and high-dimensional histogramming in affine, multimodality image registration.

Link to Paper

Date: Thursday 29th September, 2005
Topic: Computing Geodesics and Minimal Surfaces via Graph Cuts
Speaker: NA (Recorded Presentation)
Place: CSE E404
Time: 4 pm
Abstract:
We shall be seeing an ICCV (2003) presentation by Yuri Boykov and Vladimir Kolmogorov, the abstract of which is here below:

Geodesic active contours and graph cuts are two standard image segmentation techniques. We introduce a new segmentation method combining some of their benefits. Our main intuition is that any cut on a graph embedded in some continuous space can be interpreted as a contour (in 2D) or a surface (in 3D). We show how to build a grid graph and set its edge weights so that the cost of cuts is arbitrarily close to the length (area) of the corresponding contours (surfaces) for any anisotropic Riemannian metric. There are two interesting consequences of this technical result. First, graph cut algorithms can be used to find globally minimum geodesic contours (minimal surfaces in 3D) under arbitrary Riemannian metric for a given set of boundary conditions. Second, we show how to minimize metrication artifacts in existing graph-cut based methods in vision. Theoretically speaking, our work provides an interesting link between several branches of mathematics - differential geometry, integral geometry, and combinatorial optimization. The main technical problem is solved using Cauchy-Crofton formula from integral geometry.

Link to Paper

Date: Thursday 22nd September, 2005
Topic: Genetic Programming
Speaker: Nathan VanderKraats
Place: CSE E404
Time: 4 pm
Abstract:
The field of Evolutionary Computing (EC) borrows the ideas of natural selection and evolution from biology and applies them to scientific problems. It is a fascinating one to many computer scientists, largely because its random nature has the potential to yield unorthodox solutions to problems. However, beyond its intuitive biological motivation, most people have only a minimal understanding of how such an idea is implemented in practice. This talk will begin with an overview of Genetic Algorithms, EC's most popular subarea. Genetic Algorithms have a wide range of applications from physics to function optimization and illustrate well how evolution may be simulated computationally. We will then focus on a particular subfield called Genetic Programming (GP), which deals with evolving entire programs. Some of the topics to be discussed include:

(1) Evolutionary constructs such as mutation, crossover, and probabilistic selection of future generations of programs.
(2)How to represent a program in a GP framework so as to be compatible with these operations.
(3)How to handle advanced issues such as recursion and dynamic memory.

The talk will reference Nathan's work implementing a genetic programming system to play the game of checkers, and the unique situations that must be dealt with to run such a system efficiently.

Date: Thursday 8th September, 2005
Topic: Robust Nonrigid Multimodal Image Registration using Local Frequency Maps.
Speaker: Bing Jian
Place: CSE E404
Time: 4 pm
Abstract:
Automatic multi-modal image registration is central to numerous tasks in medical imaging today and has a vast range of applications, e.g., image guidance, atlas construction, etc. In this paper, we present a novel multi-modal 3D non-rigid registration algorithm where in 3D images to be registered are represented by their corresponding local frequency maps efficiently computed using the Riesz transform as opposed to popular Gabor filters. The non-rigid registration between these local frequency maps is formulated in a statistically robust framework involving the minimization of the integral squared error, a.k.a. $L_2E$ ($L_2$ error). This error is expressed as the squared difference between the true density of the residual (which is the squared difference between the non-rigidly transformed reference and the target local frequency representations) and a Gaussian or mixture of Gaussians density approximation of the same. The non-rigid transformation is expressed in a B-spline basis to achieve the desired smoothness in the transformation as well as computational efficiency.
Link to Paper