Download presentation

Presentation is loading. Please wait.

Published byLorena Leeming Modified over 2 years ago

1
Miguel Lourenço Rodrigues Master’s thesis in Biomedical Engineering December 2011 1

2
2 Outline 1. Introduction and Objectives 2. Methods: Problem Formulation, Simulations and Real Data 3.Results and Discussion 4. Conclusions

3
Outline 1. Introduction 2. Literature Review 3. Problem Formulation 4. Experimental Results and Discussion 5. Conclusions 3

4
4 Introduction -Cerebral Blood Flow (CBF): Volume of blood flowing per unit time [2] -Perfusion: CBF per unit volume of tissues Arterial Spin Labeling (ASL): -Non invasive technique for generating perfusion images of the brain [1] Se [1] e [2] são refs, deviam aparecer antes com nome e ano

5
5 Introduction Labeled acquisiton 1.Labeling of inflowing arterial blood 2. Image acquisition ASL: Este slide e o seguinte deviam ser 1 só

6
6 Introduction ASL Control acquisiton 3. No labeling 4. Image acquisition

7
7 Introduction ASL Control imageLabeled image CBF A number of control-label repetitions is required in order to achieve sufficient SNR to detect the magnetization difference signal, hence increasing scan duration. [C 1, L 1, C 2, L 2,…, C n/2, L n/2 ] n length vector C i – i th control image L i – i th labeled image P- perfusion

8
8 Introduction ASL signal processing methods Pair-wise subtraction: [P 1, P 2,…, P n/2 ]=[C 1 - L 1, C 2 - L 2,…, C n/2 -L n/2 ] Surround subtraction: [P 1, P 2,…, P n/2 ]=[C 1 - L 1, C 2 - (L 1 +L 2 ),…, C n/2 -(L (n/2)-1 -L n/2 )] 22 Sinc-interpolated subtraction: [P 1, P 2,…, P n/2 ]=[C 1 - L 1/2, C 2 - L 3/2,…, C n/2 -L n/2-1/2 ]

9
9 Objectives -Increase image Signal to Noise Ratio (SNR) -Reduce acquisition time Approach - New signal processing model - Bayesian approach - spatio-temporal priors No drastic signal variatons (except in organ boundaries)

10
10 Outline 1. Introduction 2. Literature Review 3. Problem Formulation 4. Experimental Results and Discussion 5. Conclusions

11
11 Problem Formulation Mathematical model Y(t)=F+D(t)+v(t)ΔM+Γ(t) Y (NxMxL) – Sequence of L PASL images F (NxM) – Static magnetization of the tissues D (NxM x L) – Slow variant image (baseline fluctuations of the signal – Drift) v (L x 1) - Binary signal indicating labeling sequences ΔM (NxM ) - Magnetization difference caused by the inversion Γ (NxM xL) – Additive White Gaussian Noise ~ N (0,σ y 2 ) (1)

12
12 Problem Formulation Mathematical model Y(t)=F+D(t)+v(t)ΔM+Γ(t) (1)

13
13 Problem Formulation Algorithm implementation Y(t)=F+D(t)+v(t)ΔM+Γ(t) (1) Vectorization Y=fu T +D+Δmv T +Γ Y (NM x L) f (NM x1) u (L x 1) D (NM x L) v (L x 1) Δm (NM x 1) Γ (NM x 1) (2)

14
14 Problem Formulation Algorithm implementation Since noise is AWGN, p(Y)~ N (μ, σ y 2 ), whereμ=fu T +D+Δmv T Maximum likelihood (ML) estimation of unknown images, θ={f,D, Δm} θ=arg min E y (Y,v,θ) θ Ill-posed problem (3)

15
15 Problem Formulation Algorithm implementation Using the Maximum a posteriori (MAP) criterion, regularization is introduced by the prior distribution of the parameters θ=arg min E y (Y,v,θ) θ (3) θ=arg min E (Y,v,θ) θ (4) E (Y,v,θ)=E y (Y,v, θ) + E θ (θ) (5) Data – fidelity termPrior term

16
16 Problem Formulation Algorithm implementation Figure from [11]

17
17 Problem Formulation Algorithm implementation E (Y,v,θ)=E y (Y,v, θ) + E θ (θ) (5) ½ Trace [(Y-fu T -D-Δmv T ) T (Y-fu T -D-Δmv T )] E (Y,v,θ)= +αTrace[(φ h D) T (φ h D)+(φ v D) T (φ v D)+(φ t D) T (φ t D)] +β(φ h f) T (φ h f)+(φ v f) T (φ v f) +γ(φ h Δm) T (φ h Δm)+(φ v Δm) T (φ v Δm) (6)

18
18 Problem Formulation Algorithm implementation -In equation (6), the matrices φ h,v,t are used to compute the horizontal, Vertical and temporal first order differences, respectively 10 0. 1 0.0 0 1 0..............0 00. 1 Φ=Φ= -α, β and γ are the priors.

19
19 Problem Formulation Algorithm implementation -MAP solution as a global mininum -Stationary points of the Energy Function – equation (6) - Equations implemented in Matlab and calculated iteratively

20
20 Outline 1. Introduction 2. Literature Review 3. Problem Formulation 4. Experimental Results and Discussion 5. Conclusions

21
21 Experimental Results and Discussion Synthetic data -Brain mask (64x64) -Axial slice -White matter (WM) and Gray matter (GM) ISNR=SNR f -SNR i ∑ 100 NxM N,M i=1,j=1 |x i,j -x i,j | x i,j ^ Mean error(%)= SNR= A signal A noise 2 - ; -

22
22 Experimental Results and Discussion Synthetic data Control acquisitionLabeled acquisition Parameters: σ=1 Δm(GM)=1 Δm(WM)=0.5 D=[-1,1] F=10000 α=0 β=0 γ=0

23
23 Experimental Results and Discussion Synthetic data Proposed algorithm Pair-wise subtraction Surround Subtraction Parameters: σ=1 Δm(GM)=1 Δm(WM)=0.5 D=[-1,1] F=10000 α=0 β=0 γ=0

24
24 Experimental Results and Discussion Synthetic data MethodISNR(dB)Mean Error (%) Proposed algorithm13.90624.658 Pair-wise subtraction13.90624.658 Surround Subtraction13.99924.393

25
25 Experimental Results and Discussion Synthetic data Prior optimization

26
26 Experimental Results and Discussion Synthetic data Prior optimization Incresasing prior value

27
27 Experimental Results and Discussion Synthetic data Prior optimization

28
28 Experimental Results and Discussion Synthetic data Prior optimization β=1 γ=5

29
29 Experimental Results and Discussion Synthetic data Parameters: σ=1 Δm(GM)=1 Δm(WM)=0.5 D=[-1,1] F=10000 α=1 β=1 γ=5 Proposed algorithm Pair-wise subtraction Surround Subtraction

30
30 Experimental Results and Discussion Synthetic data Parameters: σ=1 Δm(GM)=1 Δm(WM)=0.5 D=[-1,1] F=10000 α=1 β=1 γ=5

31
31 Experimental Results and Discussion Synthetic data MethodISNR(dB)Mean Error (%) Proposed algorithm16.99017.807 Pair-wise subtraction14.02624.492 Surround Subtraction14.10324.269

32
32 Experimental Results and Discussion Synthetic data MethodISNR(dB)Mean Error (%) Proposed algorithm16.99017.807 Pair-wise subtraction14.02624.492 Surround Subtraction14.10324.269 3dB 7% 23% -30%

33
33 Experimental Results and Discussion Synthetic data Monte Carlo Simulation for different noise levels

34
34 Experimental Results and Discussion Real data -One healthy subject -3T Siemens MRI system (Hospital da Luz, Lisboa) -PICORE-Q2TIPS PASL sequence -TI1/TI1s/TI2=750ms/900ms/1700ms -GE-EPI -TR/TE=2500ms/19ms -201 repetitions -spatial resolution: 3.5x3.5x7.0 mm 3 -Matrix size: 64x64x9

35
35 Control imageLabeled image Experimental Results and Discussion Real data

36
36 Experimental Results and Discussion Real data Proposed algorithm Pair-wise subtraction Surround Subtraction

37
37 Experimental Results and Discussion Real data -Influence of the number of iterations

38
38 Proposed algorithm Pair-wise subtraction Surround Subtraction Experimental Results and Discussion Real data

39
39 Experimental Results and Discussion Real data

40
40 Outline 1. Introduction 2. Literature Review 3. Problem Formulation 4. Experimental Results and Discussion 5. Conclusions

41
41 Conclusion -The proposed bayesian algorithm showed improvement of SNR and ME -SNR increased by 3db (23%) -ME decreased by 7% (30%) -Applied to real data Future work: -Automatic prior calculation -Reducing the number of control acquisitions -Validation tests on empirical data

42
42 [1] T.T. Liu and G.G. Brown. Measurement of cerebral perfusion with arterial spin labeling: Part 1. Methods. Journal of the International neuropsychological Society, 13(03):517-525, 2007. [2]A.C. Guyton and J.E. Hall. Textbook of medical physiology. WB Saunders (Philadelphia),1995. [4]ET Petersen, I. Zimine, Y.C.L. Ho, and X. Golay. Non-invasive measurement of perfusion: a critical review of arterial spin labeling techniques. British journal of radiology, 79(944):688, 2006. [3]D.S. Williams, J.A. Detre, J.S. Leigh, and A.P. Koretsky. Magnetic resonance imaging of perfusion using spin inversion of arterial water. Proceedings of the National Academy of Sciences, 89(1):212, 1992. [5]R.R. Edelman, D.G. Darby, and S. Warach. Qualitative mapping of cerebral blood flow and functional localization with echo-planar mr imaging and signal targeting with alternating radio frequency. Radiology, 192:513-520, 1994. Bibliography [6]DM Garcia, C. De Bazelaire, and D. Alsop. Pseudo-continuous ow driven adiabatic inversion for arterial spin labeling. In Proc Int Soc Magn Reson Med, volume 13, page 37, 2005. [7]E.C. Wong, M. Cronin, W.C. Wu, B. Inglis, L.R. Frank, and T.T. Liu. Velocity-selective arterial spin labeling. Magnetic Resonance in Medicine, 55:1334{1341, 2006. [8]W.C. Wu and E.C. Wong. Feasibility of velocity selective arterial spin labeling in functional mri. Journal of Cerebral Blood Flow & Metabolism, 27(4):831{838, 2006 [9]GK Aguirre, JA Detre, E. Zarahn, and DC Alsop. Experimental Design and the Relative Sensitivity of BOLD and Perfusion fMRI. NeuroImage, 15:488{500, 2002. [10]E.C. Wong, R.B. Buxton, and L.R. Frank. Implementation of Quantitative Perfusion Imaging Techniques for Functional Brain Mapping using Pulsed Arterial Spin Labeling. NMR in Biomedicine, 10:237{249, 1997. [11] J.M. Sanches, J.C. Nascimento, and J.S. Marques. Medical image noise reduction using the Sylvester-Lyapunov equation. IEEE transactions on image processing, 17(9), 2008.

43
43 Questions

Similar presentations

OK

T.T. Liu, fBIRN AHM 2006 Status of Arterial Spin Labeling for the fBIRN Thomas T. Liu UCSD Center for Functional MRI.

T.T. Liu, fBIRN AHM 2006 Status of Arterial Spin Labeling for the fBIRN Thomas T. Liu UCSD Center for Functional MRI.

© 2017 SlidePlayer.com Inc.

All rights reserved.

Ads by Google

Ppt on interest rate futures Ppt on diffusion in chemistry Ppt on acid-base indicators colors Ppt online compressors Projector view ppt online Ppt on teachers day in english Ppt on graph theory in electrical engineering Ppt on emotional intelligence Ppt on campus recruitment system problem Ppt on acid-base indicators animations free