Resolution Enhancement in MRI By: Eyal Carmi Joint work with: Siyuan Liu, Noga Alon, Amos Fiat & Daniel Fiat 10/06/2004
Download ReportTranscript Resolution Enhancement in MRI By: Eyal Carmi Joint work with: Siyuan Liu, Noga Alon, Amos Fiat & Daniel Fiat 10/06/2004
Resolution Enhancement in MRI By: Eyal Carmi Joint work with: Siyuan Liu, Noga Alon, Amos Fiat & Daniel Fiat 10/06/2004 1 Lecture Outline Introduction to MRI The SRR problem (Camera & MRI) Our Resolution Enhancement Algorithm Results Open Problems 2 Introduction to MRI Magnetic resonance imaging (MRI) is an imaging technique used primarily in medical settings to produce high quality images of the inside of the human body. 3 Introduction to MRI The nucleus of an atom spins, or precesses, on an axis. Hydrogen atoms – has a single proton and a large magnetic moment. 4 Magnetic Resonance Imaging Uniform Static Magnetic Field – Atoms will line up with the direction of the magnetic field. w0 B0 B0 MagneticField GyromagneticRatio w0 LarmorFrequency 5 Magnetic Resonance Imaging Resonance – A state of phase coherence among the spins. Applying RF pulse at Larmor frequency When the RF is turned off the excess energy is released and picked up. 6 Magnetic Resonance Imaging Gradient Magnetic Fields – Time varying magnetic fields (Used for signal localization) z x y 7 Magnetic Resonance Imaging Gradient Magnetic Fields: 1-D Y w0 B0 B0 MagneticField B0 B1 GyromagneticRatio B2 B3 B4 w0 LarmorFrequency X 8 Signal Localization slice selection Gradient Magnetic Fields for slice selection ω Gz,1 B1(t) FT Gz,2 ω2 ω1 B0 z1 z2 z z3 z4 9 Signal Localization frequency encoding Gradient Magnetic Fields for in-plane encoding x B=B0+Gx(t)x B0 t Gx(t) t 10 Signal Localization phase encoding Gradient Magnetic Fields for in-plane encoding x B=B0+Gx(t)x B0 t Gx(t) t 11 k-space interpretation y 1-D path Wy k-space x DFT Δkx Sampling Points Wx S k r e i 2 kr dr Object 12 Magnetic Resonance Imaging Collected Data (k-space) 2-D DFT 13 The Super Resolution Problem Definition: SRR (Super Resolution Reconstruction): The process of combining several low resolution images to create a high-resolution image. 14 SRR – Imagery Model The imagery process model: Yk Dk Bk Gk X Ek k 1 N Yk – K-th low resolution input image. Gk – Geometric trans. operator for the k-th image. Bk – Blur operator of the k-th image. Dk – Decimation operator for the k-th image. Ek – White Additive Noise. 15 SRR – Main Approaches Frequency Domain techniques Tsai & Huang [1984] Kim [1990] Frequency Domain 16 SRR – Main Approaches Iterative Algorithms Irani & Peleg [1993] : Iterative Back Projection Back Projection Current HR Best Guess Back Projected LR images Original LR images Iterative Refinement 17 SRR – Main Approaches Y HX E Patti, Sezan & Teklap [1994] POCS: Y H X 2 Elad & Feuer [1996 & 1997] ML: arg maxP Y X X MAP: arg maxPY X P X X POCS & ML 18 SRR – In MRI Peled & Yeshurun [2000] 2-D SRR, IBP, single FOV, problems with sub-pixel shifts. Greenspan, Oz, Kiryati and Peled [2002] 3-D SRR (slice-select direction), IBP. 19 Resolution Enhancement Alg. A Model for the problem Reconstruction using boundary values 1-D Algorithm 2-D Algorithm 20 Modeling The Problem Subject Area: 1x1 rectilinearly aligned square grid. (0,0) 21 Modeling The Problem True Image : A matrix of real values associated with a rectilinearly aligned grid of arbitrary high resolution. (0,0) 2 9 5 6 8 3 6 3 1 8 8 6 4 4 1 9 4 9 7 3 5 4 2 6 5 22 Modeling The Problem A Scan of the image: S (i, j ) F (m, x , y ) Pixel resolution – m m Offset – ( x , y ) True Image ( x , y ) 1 1 4 4 1 1 4 4 2 2 3 3 2 2 3 3 Image Scan, m=2 4 8 16 32 23 Modeling The Problem Definitions: Maximal resolution – n Pixel resolution = m m, m n Maximal Offset resolution – 1 n We can perform scans at offsets ( x , y ) where x , y k , k 1,2,3 with pixel resolution m 1 c , c Z 24 Modeling The Problem Goal: Compute an image of the subject area with pixel resolution 1 1 while the maximal measured pixel resolution is n n, n 1 δ Errors: 1. Errors ~ Pixel Size & Coefficients 2. Immune to Local Errors => localized errors should have localized effect. 25 Multiple offsets of a single resolution scan? 2x2 26 Multiple offsets of a single resolution scan? 3x3 27 Multiple offsets of a single resolution scan? 4x2 28 Using boundary value conditions Assumption: 0 0 0 0 0000 or C C C C CCC Reconstruct using multiple scans with the same m m pixel resolution. Introduce a variable for each HR pixel of physical dimension . Algorithm: Perform c2 scans at all offsets & Solve linear equations (Gaussian elimination). 29 Using boundary value conditions Example: 4 Scans, PD=2x2 Second sample 30 Using boundary value conditions Example: 4 Scans, PD=2x2 0 0 0 0 31 Using boundary value conditions Example: 4 Scans, PD=2x2 32 Problems using boundary values Example: 4 Scans, PD=2x2 (0,0) 33 Problems using boundary values Example: 4 Scans, PD=2x2 (-1,-1) 34 Problems using boundary values Example: 4 Scans, PD=2x2 Add more information •Solve using LS •Propagation problem 35 Demands On the algorithm ???? No assumptions on the values of the true image. ? ? ? ? ? ? ???? Over determined set of equations Use LS to reduce errors: min Ax b l k where A , l k & b A l x 2 b Error propagation will be localized. 36 The One dimensional algorithm Input: Pixel of dimension 1 x & 1 y x, y N Notation: v( x, i), i - valueof the1 x pixelat offseti gcd(x,y) – greatest common divisor of x & y. a, b. ax by gcd( x, y) (Extended Euclidean Algorithm) a y, b x 37 The One dimensional algorithm (Extended Euclidean Algorithm) a, b. ax by gcd( x, y) Given all v( x, i) & v( y, i) We can compute all v(gcd( x, y), i) gcd( x, y) 1 v(1, i) 38 One dimensional reconstruction 39 One dimensional reconstruction 40 The One dimensional algorithm Algorithm: Given v( x, j ),0 j m x v( y, i),0 j m x w.l.g, let: a>0 & b<0 To compute v(gcd( x, y), i) , compute: a 1 b 1 j 0 j 0 v x, i xj v y, i yj 41 The One dimensional algorithm Localized Reconstruction ax by 1, a 0, b 0 ax 1 ( mod y) Localized Reconstruction Effective Area: x+y high-resolution pixels 42 One dimensional reconstruction 43 Two and More Dimensions Given pixels of size: x x, y y & z z Where x,y & z are relatively prime. Reconstruct 1x1 pixels. Error Propagation is limited to an area of O(xyz) HR pixels. x x y y xy x xy y xy 1 44 Two and More Dimensions xy 1 xz 1 x 1 x 1 y 1 xy 1 zy 1 y 1 1 1 45 Example Two dimensional reconstruction PD=5x5 PD=3x3 PD=15x1 46 Example Two dimensional reconstruction PD=4x4 PD=3x3 PD=12x1 47 Example Two dimensional reconstruction PD=5x5 PD=4x4 PD=20x1 48 Example Two dimensional reconstruction 49 Larger Dimensions Generalize to Dimension k Using k+1 relatively prime Low-Resolution pixels 50 Results Model Results Experiment Results Problems… 51 Model Design Noise HR Scene Blur HR Image Sampling LR LR LR Image Image Image SRR Algorithm 52 Results 53 Experiment GE clinical 1.5T MRI scanner was used. Phantom: - plastic frames - filled with water Three FOV: 230.4, 307.2 & 384 mm. 54 Experiment 55 Experiment Results MRI Data Modeled Data 56 Experiment Results MRI Data Modeled Data 57 Problems Homogeneity of the phantom Phantom Orientation 58 Problems Homogeneity of the phantom Phantom Orientation Rectangular Blur Vs. Gaussian-like Blur 59 Problems Homogeneity of the phantom Phantom Orientation Rectangular Blur Vs. Gauss-like Blur Truncated 60 Problems Homogeneity of the phantom Phantom Orientation Rectangular Blur Vs. Gauss-like Blur k-space and Fourier based MRI 61 k-space and Fourier based MRI y 1-D path Wy k-space x DFT Δk Δkx x Sampling Points Wx S k r e i 2 kr dr Object 62 Problems # Samples Infinite Noise No Problem One scan is enough Infinite Yes SNR too low Finite Yes No perfect reconstruction Apply manual shifts → Different experiment 63 Open Problems Optimization problems: “What is the smallest number of scans we can do to reconstruct the high resolution image?“ 64 Scan Selection Problem 65 Scan Selection Problem S5x and… 66 Scan Selection Problem S3x and… 67 Open Problems Optimization problems: “What is the smallest number of scans we can do to reconstruct the high resolution image?“ Decision/Optimization problem: Given a set of scans, what can we reconstruct? Design problem: Plan a set of scans for “good” error localization. 68 Questions 69