Reconstruction History. Image Reconstruction. Radon Transform. Central Slice Theorem (I)

Similar documents
A linear imaging system with white additive Gaussian noise on the observed data is modeled as follows:

Xin Li Department of Information Systems, College of Business, City University of Hong Kong, Hong Kong, CHINA

8/25/17. Data Modeling. Data Modeling. Data Modeling. Patrice Koehl Department of Biological Sciences National University of Singapore

Dr. Shalabh Department of Mathematics and Statistics Indian Institute of Technology Kanpur

Quality-of-Service Routing in Heterogeneous Networks with Optimal Buffer and Bandwidth Allocation

Boundary Value Problems. Lecture Objectives. Ch. 27

A total variation approach

A MODIFIED METHOD FOR SOLVING SYSTEM OF NONLINEAR EQUATIONS

Chapter 7 Generalized and Weighted Least Squares Estimation. In this method, the deviation between the observed and expected values of

Line Drawing and Clipping Week 1, Lecture 2

Topology optimization of plate structures subject to initial excitations for minimum dynamic performance index

DECOUPLING THEORY HW2

6. Hamilton s Equations

Linear system of the Schrödinger equation Notes on Quantum Mechanics

Georgia Tech PHYS 6124 Mathematical Methods of Physics I

Research on Complex Networks Control Based on Fuzzy Integral Sliding Theory

Module 3 LOSSY IMAGE COMPRESSION SYSTEMS. Version 2 ECE IIT, Kharagpur

The Fourier Transform

Structure from Motion. Forsyth&Ponce: Chap. 12 and 13 Szeliski: Chap. 7

MARKOV CHAIN AND HIDDEN MARKOV MODEL

Fourier Transform. Additive noise. Fourier Tansform. I = S + N. Noise doesn t depend on signal. We ll consider:

Linear Classification, SVMs and Nearest Neighbors

Report on Image warping

Maximum Likelihood Estimation of Binary Dependent Variables Models: Probit and Logit. 1. General Formulation of Binary Dependent Variables Models

Computational Astrophysics

Cyclic Codes BCH Codes

COMPLEX NUMBERS AND QUADRATIC EQUATIONS

Lossy Compression. Compromise accuracy of reconstruction for increased compression.

WAVELET-BASED IMAGE COMPRESSION USING SUPPORT VECTOR MACHINE LEARNING AND ENCODING TECHNIQUES

IV. Performance Optimization

Machine Learning. Classification. Theory of Classification and Nonparametric Classifier. Representing data: Hypothesis (classifier) Eric Xing

Outline and Reading. Dynamic Programming. Dynamic Programming revealed. Computing Fibonacci. The General Dynamic Programming Technique

2-Adic Complexity of a Sequence Obtained from a Periodic Binary Sequence by Either Inserting or Deleting k Symbols within One Period

Orthogonal Functions and Fourier Series. University of Texas at Austin CS384G - Computer Graphics Spring 2010 Don Fussell

Hongyi Miao, College of Science, Nanjing Forestry University, Nanjing ,China. (Received 20 June 2013, accepted 11 March 2014) I)ϕ (k)

OUTLINE. A model for the Container Stowage and Routing Problem. Container Stowage and Routing Problem. Container Stowage and Routing Problem

The line method combined with spectral chebyshev for space-time fractional diffusion equation

Perfect Competition and the Nash Bargaining Solution

MACHINE APPLIED MACHINE LEARNING LEARNING. Gaussian Mixture Regression

A General Column Generation Algorithm Applied to System Reliability Optimization Problems

Support Vector Machines. Vibhav Gogate The University of Texas at dallas

Example: Suppose we want to build a classifier that recognizes WebPages of graduate students.

CS 2750 Machine Learning. Lecture 5. Density estimation. CS 2750 Machine Learning. Announcements

Delay tomography for large scale networks

Neuro-Adaptive Design - I:

Lecture Notes on Linear Regression

Digital PI Controller Equations

Integral Transforms and Dual Integral Equations to Solve Heat Equation with Mixed Conditions

2. PROBLEM STATEMENT AND SOLUTION STRATEGIES. L q. Suppose that we have a structure with known geometry (b, h, and L) and material properties (EA).

Centre of Rotation Determination Using Projection Data in X-ray Micro Computed Tomography

Mechanics Physics 151

Lecture 12: Discrete Laplacian

Research Report. Eiichi Ono, Yoshikazu Hattori, Yuji Muragishi. Abstract. Special Issue Estimation and Control of Vehicle Dynamics for Active Safety

Lecture 5 Decoding Binary BCH Codes

In this section is given an overview of the common elasticity models.

Image Classification Using EM And JE algorithms

SIO 224. m(r) =(ρ(r),k s (r),µ(r))

Introduction to the R Statistical Computing Environment R Programming

p(z) = 1 a e z/a 1(z 0) yi a i x (1/a) exp y i a i x a i=1 n i=1 (y i a i x) inf 1 (y Ax) inf Ax y (1 ν) y if A (1 ν) = 0 otherwise

The Geometry of Logit and Probit

Fall 2012 Analysis of Experimental Measurements B. Eisenstein/rev. S. Errede

CME 302: NUMERICAL LINEAR ALGEBRA FALL 2005/06 LECTURE 13

1 Derivation of Rate Equations from Single-Cell Conductance (Hodgkin-Huxley-like) Equations

PHYS 705: Classical Mechanics. Canonical Transformation II

VECTORS AND MATRICES:

Chapter 4 The Wave Equation

Tracking with Kalman Filter

Some Comments on Accelerating Convergence of Iterative Sequences Using Direct Inversion of the Iterative Subspace (DIIS)

Expectation Maximization Mixture Models HMMs

Adjoint Methods of Sensitivity Analysis for Lyapunov Equation. Boping Wang 1, Kun Yan 2. University of Technology, Dalian , P. R.

Primer on High-Order Moment Estimators

Originated from experimental optimization where measurements are very noisy Approximation can be actually more accurate than

Salmon: Lectures on partial differential equations. Consider the general linear, second-order PDE in the form. ,x 2

Yong Joon Ryang. 1. Introduction Consider the multicommodity transportation problem with convex quadratic cost function. 1 2 (x x0 ) T Q(x x 0 )

On an Extension of Stochastic Approximation EM Algorithm for Incomplete Data Problems. Vahid Tadayon 1

LINEAR REGRESSION ANALYSIS. MODULE IX Lecture Multicollinearity

COXREG. Estimation (1)

Robust Norm Equivalencies and Preconditioning

Matching Dyadic Distributions to Channels

A Computational Viewpoint on Classical Density Functional Theory

AERODYNAMICS I LECTURE 6 AERODYNAMICS OF A WING FUNDAMENTALS OF THE LIFTING-LINE THEORY

Managing Capacity Through Reward Programs. on-line companion page. Byung-Do Kim Seoul National University College of Business Administration

Composite Hypotheses testing

Nested case-control and case-cohort studies

4DVAR, according to the name, is a four-dimensional variational method.

NMT EE 589 & UNM ME 482/582 ROBOT ENGINEERING. Dr. Stephen Bruder NMT EE 589 & UNM ME 482/582

AN ASYMMETRIC GENERALIZED FGM COPULA AND ITS PROPERTIES

Supervised Learning. Neural Networks and Back-Propagation Learning. Credit Assignment Problem. Feedforward Network. Adaptive System.

Fuzzy approach to solve multi-objective capacitated transportation problem

G : Statistical Mechanics

we have E Y x t ( ( xl)) 1 ( xl), e a in I( Λ ) are as follows:

A FAST HEURISTIC FOR TASKS ASSIGNMENT IN MANYCORE SYSTEMS WITH VOLTAGE-FREQUENCY ISLANDS

ANSWERS. Problem 1. and the moment generating function (mgf) by. defined for any real t. Use this to show that E( U) var( U)

CSC321 Tutorial 9: Review of Boltzmann machines and simulated annealing

MD. LUTFOR RAHMAN 1 AND KALIPADA SEN 2 Abstract

Linear Approximation with Regularization and Moving Least Squares

8 : Learning in Fully Observed Markov Networks. 1 Why We Need to Learn Undirected Graphical Models. 2 Structural Learning for Completely Observed MRF

COMPUTATIONALLY EFFICIENT WAVELET AFFINE INVARIANT FUNCTIONS FOR SHAPE RECOGNITION. Erdem Bala, Dept. of Electrical and Computer Engineering,

Estimation: Part 2. Chapter GREG estimation

SELECTION OF MIXED SAMPLING PLANS WITH CONDITIONAL DOUBLE SAMPLING PLAN AS ATTRIBUTE PLAN INDEXED THROUGH MAPD AND LQL USING IRPD

Transcription:

Reconstructon Hstory wth Bomedca Acatons EEG-475/675 Prof. Barner Reconstructon methods based on Radon s wor 97 cassc mage reconstructon from roectons aer 97 Hounsfed deveo the frst commerca x-ray CT scanner Hounsfed and Cormac receve the 979 obe Prze for ther CT contrbutons Cassca reconstructon s based on the radon transform Method nown as bacroecton Aternatve aroaches: Fourer transform and teratve seres-exanson methods Statstca estmaton methods Waveet and other mutresouton methods Unversty of Deaware Radon Transform Centra Sce Theorem (I) Radon transform: P(, ) R{ f( x, y) } f( x, y) d s defned aong the ath such that x cos + ysn Convertng the oar (,) system to a rotated coordnate system (,q) x cos + ysn xsn + ycos q Aternatve Radon exresson R{ f( x, y) } J ( ) f( cos qsn, sn + qcos ) dq Images are reconstructed by bacroectng the roectons A mted number of roectons eads to aasng and/or geometrc artfacts Theorem reates Fourer transforms of obect functon and Fourer transform of roecton Fourer transform the of the (-D) roecton: S( ω) J( ) e d ω f ( cos q sn, sn q cos ) e dqd ω Utzng xcos + ysn q xsn + ycos yeds S ( ω) f( x, y) e dxdy F( ω, ) + x y ω( cos sn ) Ths s a -D FT of f(x,y), denoted as F(u,v), u ω cos v ω sn u and v are the frequency comonents n the x- and y-drectons Unversty of Deaware 3 Unversty of Deaware 4

Centra Sce Theorem (II) Inverse Transform S (ω) s the Fourer transform of the roecton J () J () s taen that ange n the sace doman wth rotated coordnate system (,q) The sectrum S (ω) s ocated on a ne of ange n the frequency doman F(u,v) S (ω) obtaned at dfferent anges are set aong the corresondng rada nes n the F(u,v) frequency doman Reconstructon can be acheved by nverse Fourer transformaton Samng ssues? Inverse FT reconstructon yeds: (, ) { (, )} (, ) Changng to oar coordnates: ω ( xcos 0 + v ) sn Reacng F(ω,)wth S (ω) yeds: f x y F u v F u v e dudv ( xu+ vy ) f (, r ) F( ω, ) e ω dωd f r ω S ω e dωd J d (, ) 0 ( ) ( ') x v ω ( + ) cos sn 0 ( cos sn ) J ( ') S ( ) e x + v ω ω dω Modfed roectons Reconstructon s the bacroected ntegra of modfed roectons Unversty of Deaware 5 Unversty of Deaware 6 Bacroecton Method (I) Bacroecton Method (II) The modfed roecton can be exressed as: J( ') S( ) e d + x v ω ( cos sn ) ω ω ω ω ω S ( ω) e dω I { ω S ( ω) } I { ω } J ( ) Convouton of roecton and ω fter Genera exresson for reconstructon: f ( xy, ) J ( ') h ( dd ') ' 0 Genera fter functon: h() Infnte bandwdth fters are robematc why? H(ω) s generay bandmted: H(ω) ω B(ω) f ω Ω B( ω) 0 otherwse Bandmted fter n the frequency doman Imuse resonse for samng nterva t sn( / τ) sn( / τ) h ( ) τ / τ 4 τ /τ Ftered roecton: J( ') J( ') h( ') d' Reconstructon: * f ( xy, ) J ( ') Quaty of reconstructon deends Resouton of the roectons umber of roectons Unversty of Deaware 7 Unversty of Deaware 8

Bacroecton Method (III) Agebrac Reconstructon Technque (ART) Shar fter cut-off resuts n artfacts Cut-off avods amfcaton of hghfrequency nose Cut-off ntroduces rngng Aternatve fter: generazed Hammng wndow Smoother cut-of o rngng n muse resonse Proecton data s dstrbuted over the mage reconstructon Dstrbuton crtera: mnmze error between Actua roectons Proectons comuted from the reconstructed obect System mode: f vaue of the th mage (sce) xe roecton of a secfc ray assng through the mage w, weght (fracton) of f contrbutng to w, f for,..., M M equatons and unnowns Unversty of Deaware 9 Unversty of Deaware 0 ART Souton Estmaton Methods Souton methods Matrx reresentaton and nverson Comutatona and memory ntensve Dynamc rogrammng Iteratve souton (ART) Iteratve souton stes Intazaton a xes assgned a redetermned vaue (average roecton vaue) Comute the ray sums for the reconstructed mage Comute the true ray sum and roected ray sum errors Dstrbute errors to corresondng reconstructon xes Addtve correcton: correcton vaue s added to the current xe vaue Muty to correcton: correcton vaue scaes current xe vaue th ray sum at the th teraton q f w, Iteraton ste + q f f + w, w, Agorthm varatons Addtve ART Ray contrbuton weghts w, are bnary Yeds an effcent mementaton Mutcatve ART Other varatons desgned to mrove convergence, comutatona cost, and fna resuts Statstca estmaton methods often rovde sueror reconstructons Detector hoton count s Posson dstrbuted defnes the ntegraton ne (hoton ray) Vector of detector measurements: J[J, J,, J ] Rewrte ne ntegra ( xyzd,, ) a attenuaton coeffcent n voxe E[ J] me a contrbuton of voxe to ne Set weght matrx: A{a } ( x, yzd, ) Rewrte measurement vector as [ A] J( ) me [ A] a Unversty of Deaware Unversty of Deaware 3

Maxmum ehood Souton Roughness Penaty The M souton s gven by: ˆ arg max ( ); 0 ( ) og P J ; ; Returns the (voxe) attenuaton coeffcent vector that maxmzes the robabty of observng the measured hoton counts Usng the Posson dstrbuton mode for the hoton counts ( ) [ ( )] e P J ; P[ J ; ]! If the measurements are obtaned (ndeendenty) as ray sums ( ) h ([ A ] ) h( ) og( ) me me The souton to the above may not be unque A soutons may not be arorate Addressed through the ncuson of a roughness enaty Genera roughness enaty functon K R( ) ψ([ C] ) [ C] c C s a enaty matrx ψ( ) s a convex, symmetrc, nonnegatve, dfferentabe, otenta functon Exame: ψ(t)w t / (quadratc) Quadratc roughness enaty functon: K R( ) w ([ C] ) Penazed M obectve functon arg max Φ( ) Φ ( ) ( ) βr( ) β contros eve of smoothness n the fna reconstructon Unversty of Deaware 3 Unversty of Deaware 4 Surrogate Functon Exectaton Maxmzaton Soutons c 0 and (n) [A (n) ] ( th ne ntegra at teraton n) Reformuatng the estmaton robem usng the surrogate functon: φ( ; ) Q( ; ) βr( ) a Q e ( M ) ( e ) a ( ; ) n + n ar r r S [ A ] M + me me a ( n ) r S {} r r [ A ] + me me S s the set of xes between the source and xe aong the th ray. M and are the current exected number of hotons enterng and eavng xe aong the th ray. E-Ste: M-Ste: Fnd Q( ; ) and φ ( ; ) arg max ( ; ) ( n+ ) φ Combnng the stes yeds the M-EM agorthm ( n ) + ( M ) ( + ) a M Unversty of Deaware 5 Unversty of Deaware 6 4

Reconstructon Exames Eght Proecton Case Orgna Phantom Bacroecton ART Reconstructon MAP Reconstructon Unversty of Deaware 7 5