70
Rician Noise Removal in Diffusion Tensor - MRI Thesis Defense Saurav Basu School of Computing University of Utah 1

Rician Noise Removal in Diffusion Tensor MRI with speaker notes

Embed Size (px)

DESCRIPTION

Masters Thesis Defense: University of Utah Rician Noise Removal in Diffusion Tensor MRI

Citation preview

Page 1: Rician Noise Removal in Diffusion Tensor MRI with speaker notes

Rician Noise Removal in Diffusion

Tensor - MRIThesis Defense

Saurav BasuSchool of Computing

University of Utah

1

Page 2: Rician Noise Removal in Diffusion Tensor MRI with speaker notes

2

Organization• Brief overview of DT MRI

• Goals for this thesis

• Motivation :

- why noise removal ?

- why Rician noise ?

- previous DT-MRI filtering methods

• Rician Bias Correction Filter

• Results and discussion

• Conclusion: summary, future work

• Questions ?

Page 3: Rician Noise Removal in Diffusion Tensor MRI with speaker notes

March-April 2006

DT-MRI is themost recent in aseries of astonishingbreakthroughs in brain imaging

3

Page 4: Rician Noise Removal in Diffusion Tensor MRI with speaker notes

4

HyperStreamLines used to Visualize White Matter Fibres in the brain

Page 5: Rician Noise Removal in Diffusion Tensor MRI with speaker notes

5

Diffusion Tensor

Brief overview of DT-MRI

Imaging technique to compute a 3x3 matrix (D)Characterizes diffusion of water across brain tissue

•Symmetric

•Positive Definite

•All eigenvalues are positive

Used to study structure of brain fibresKey: More diffusion along fibres than across

fibres

Page 6: Rician Noise Removal in Diffusion Tensor MRI with speaker notes

6

‣ Visualize the eigen values of D over the volume to infer connectivity and structure

Tensor Orientation: Principle Eigen VectorTensor Anisotropy: directional characteristics

Page 7: Rician Noise Removal in Diffusion Tensor MRI with speaker notes

7

How is the tensor computed ?

A0

Ai

gi

Stejskal Tanner equation

Known: b, gi

Measured: A0, Ai

Find DMost Common:Linear Least Squares on

Page 8: Rician Noise Removal in Diffusion Tensor MRI with speaker notes

8

Goals for this Thesis:

• What is the best way to filter DT-MRI data?

• How do current filtering methods compare ?

• Is there a better way of doing filtering?

Answer these questions.

Page 9: Rician Noise Removal in Diffusion Tensor MRI with speaker notes

9

Motivation

• Why is DT - MRI filtering important?

• Why is it important to account for Rician noise in the filtering process?

Page 10: Rician Noise Removal in Diffusion Tensor MRI with speaker notes

10

• DT MRI plagued by low SNR

‣ Multiple Scans needed to increase SNR

‣ Issues: long acquisition time, patient comfort system throughput

‣ Mis-registration issues (motion artifacts)

‣ Partial voluming (volume averaging) : voxel covers a non homogeneous tissue region

Why DT - MRI filtering?

Page 11: Rician Noise Removal in Diffusion Tensor MRI with speaker notes

11

• what is Rician noise? how does it arise in DT MRI?

• how does it effect tensors?

• previous filtering methods

Why Rician noise removal?

Page 12: Rician Noise Removal in Diffusion Tensor MRI with speaker notes

12

Rician noise in DT MRI ?

•DWI images are magnitudes of complex valued signals.

•If the real and imaginary components of the signal are assumed to have a Gaussian noise, the resulting magnitude image will have Rician distributed noise.

gaussian

magnitude

where is zero mean , stationary Gaussian noise with standard deviation

Page 13: Rician Noise Removal in Diffusion Tensor MRI with speaker notes

13

Rician Noise

A signal is said to be corrupted with Rician noise if the pdf of the noisy signal has a Rice distribution

Page 14: Rician Noise Removal in Diffusion Tensor MRI with speaker notes

14

p(x|A)

ARice Distribution

Page 15: Rician Noise Removal in Diffusion Tensor MRI with speaker notes

15

p(x)

ANormal Distribution

Page 16: Rician Noise Removal in Diffusion Tensor MRI with speaker notes

16

10000 samples , sigma=20

Page 17: Rician Noise Removal in Diffusion Tensor MRI with speaker notes

17

How does Rician noise affect estimated tensors?

previous studies show

noise trace and FA

“However, when we performed Monte Carlo simulations with Rician noise with diffusion tensors characteristic of those in the human brain we found FA and trace can be incorrectly estimated when tensors are aligned with gradient directions.”

aligned tensors: noise FA trace

Page 18: Rician Noise Removal in Diffusion Tensor MRI with speaker notes

18

Tensor Splitting Gradient direction

Tensor aligned with gradient direction

Page 19: Rician Noise Removal in Diffusion Tensor MRI with speaker notes

19

This tells us FA can be overestimated or underestimated depending on how a person sits inside the scanner !

Bottom Line!

It is important to consider Rician noise in filtering process

Can Seriously affect the validity of clinical studies using these FA estimates.

Page 20: Rician Noise Removal in Diffusion Tensor MRI with speaker notes

Previous filtering approaches2

categories

DWI spaceTensor Space

1) Non linear smoothing for reduction of systematic errors. Parker(2000)

2) Constrained Variational approach Wang, Vemuri (2004)

2) Bayesian regularization using Gaussian markov random fields. Martin (2004)

1) Riemannian Space filtering Pennec (2004)

Very effective techniques, but do not explicitly handling Rician noise as part of

the filtering process.

20

Others: Median filtering, K- Space(Fourier Domain) methods

Page 21: Rician Noise Removal in Diffusion Tensor MRI with speaker notes

21

• DWI Space filter

• Based on maximum a posteriori (MAP) approach to image reconstruction

( In statistics MAP estimation is used to obtain a point estimate of an unobserved quantity based on empirical data )

Rician Bias Correction Filter

Page 22: Rician Noise Removal in Diffusion Tensor MRI with speaker notes

22

MAP Image Reconstruction

•A Prior Model

•A Likelihood or Noise Model

3.Optimization Scheme (maximize posterior)

3 Key components

Page 23: Rician Noise Removal in Diffusion Tensor MRI with speaker notes

23

➡To estimate the clean value we want to maximize p(u|u0)

From Baye’s Rule:

constant for a given noisy image u0

MAP Formulation

Given: Noisy Image u0

estimate

Output: Clean/Filtered Image u

Known: p(u0|u) has a Rician distribution

Page 24: Rician Noise Removal in Diffusion Tensor MRI with speaker notes

24

posterior likelihood prior

maximizewith

gradient ascent

Capture some prior knowledge about the filtered image.Example: enforce smoothing criteria on the image

Captures the noise model on the data

Essentially says : What is the probabilityof the clean image given that i have a particularnoisy image.

For gradient ascent we need to take derivatives!

Page 25: Rician Noise Removal in Diffusion Tensor MRI with speaker notes

25

Likelihood Term

Taking derivative w.r.t u ,

Rician attachment term or

Bias correction term

After Substituting for Rice pdf

The Likelihood Term:

Page 26: Rician Noise Removal in Diffusion Tensor MRI with speaker notes

26

We use a Gibb’s prior with an energy functional which enforces a smoothness without blurring edges

The Prior Term:

Gibb’s prior Energy functional

conductance

weighing factor

edge preserving smoothing prior

Page 27: Rician Noise Removal in Diffusion Tensor MRI with speaker notes

27

Combining the Rician correction term with the variational of the energy functional we get the

update equation for the filtered image

Derivative of likelihood term

Variational of energy functional

Implementation:

Modify PDE Diffusion Filtering to use this modified update term

1. u = vector image of 7 or more DWIs2 Bias correction term term computed independently for each component of the vector

Note:

Page 28: Rician Noise Removal in Diffusion Tensor MRI with speaker notes

28

ResultsWe compared 4 different filtering methods

on both synthetic and real data sets

DWI SpaceTensor Space

1. Anisotropic Diffusion without Rician attachment

2. Rician Bias Correction filter

1. Anisotropic Diffusion in euclidean space. 2. Anisotropic Diffusion on the Riemannian manifold

Page 29: Rician Noise Removal in Diffusion Tensor MRI with speaker notes

29

Error Metrics

•Tensor Components

•Fractional Anisotropy (FA)

3.Trace

Page 30: Rician Noise Removal in Diffusion Tensor MRI with speaker notes

30

Synthetic Data -1•10x10x4 volume of tensors

• 2 tensor orientations (along gradient and splitting the gradient directions)

• Synthetic rician noise, baseline image intensity=250

Clean Noisy (SNR=15)

Gradient Directions:

( 1 0 1 )(-1 0 1 )( 0 1 1 )( 0 1 -1)( 1 1 0 )(-1 1 0 )

Page 31: Rician Noise Removal in Diffusion Tensor MRI with speaker notes

31

Aniso DWI Rician DWI

DWI Space Filters

Page 32: Rician Noise Removal in Diffusion Tensor MRI with speaker notes

32

Euclidean Riemannian

Tensor Space Filters

Page 33: Rician Noise Removal in Diffusion Tensor MRI with speaker notes

33

Page 34: Rician Noise Removal in Diffusion Tensor MRI with speaker notes

34

Page 35: Rician Noise Removal in Diffusion Tensor MRI with speaker notes

35

Page 36: Rician Noise Removal in Diffusion Tensor MRI with speaker notes

36

• To check whether variability in directions affects results we generated a torus with tensors oriented in all possible directions.

• Ran the filtering on the torus data set

Synthetic Data Set -2 : Hollow Torus

Page 37: Rician Noise Removal in Diffusion Tensor MRI with speaker notes

37

Clean Tensor

Page 38: Rician Noise Removal in Diffusion Tensor MRI with speaker notes

38

Noisy Tensor (sigma=10)

Page 39: Rician Noise Removal in Diffusion Tensor MRI with speaker notes

39

Euclidean Filtering

Page 40: Rician Noise Removal in Diffusion Tensor MRI with speaker notes

40

Riemannian Filtering

Page 41: Rician Noise Removal in Diffusion Tensor MRI with speaker notes

41

Aniso DWI Filtering

Page 42: Rician Noise Removal in Diffusion Tensor MRI with speaker notes

42

Rician DWI Filtering

Page 43: Rician Noise Removal in Diffusion Tensor MRI with speaker notes

43

Page 44: Rician Noise Removal in Diffusion Tensor MRI with speaker notes

44

Real Data Results

Issue: No ground truth data available for DT-MRI !

How do we evaluate filtering performance quantitatively?

➡ use repeated scans of the same subject.

Solution:

p(x/A) is the Rician pdf

Maximize (Brent’s| Golden search method)

ML Estimate:

LIKELIHOOD FUNCTION

Page 45: Rician Noise Removal in Diffusion Tensor MRI with speaker notes

45

ML Estimator versus Averaging for generating Ground truth

Page 46: Rician Noise Removal in Diffusion Tensor MRI with speaker notes

46

Page 47: Rician Noise Removal in Diffusion Tensor MRI with speaker notes

47

Page 48: Rician Noise Removal in Diffusion Tensor MRI with speaker notes

48

Page 49: Rician Noise Removal in Diffusion Tensor MRI with speaker notes

49

Page 50: Rician Noise Removal in Diffusion Tensor MRI with speaker notes

50

Page 51: Rician Noise Removal in Diffusion Tensor MRI with speaker notes

•5 scans of healthy volunteer•Resolution: 2 mm x 2 mm x 2 mm 3.3T scanner , scan time 12 mins.

Real Data Filtering Results

About the Real Data:

Added Rician noise SNR levels of 10,15 and 20 with respect to white matter signal level.and ran our filtering methods.

Page 52: Rician Noise Removal in Diffusion Tensor MRI with speaker notes

52

Clean

Page 53: Rician Noise Removal in Diffusion Tensor MRI with speaker notes

53

Sigma=10

Euclidean

Riemannian

Aniso

Rician

Noisy Image

Page 54: Rician Noise Removal in Diffusion Tensor MRI with speaker notes

54

Noisy Image

Euclidean

Riemannian

Aniso

Rician

Page 55: Rician Noise Removal in Diffusion Tensor MRI with speaker notes

Noisy Image

Euclidean

Riemannian

Aniso

Rician

55

Page 56: Rician Noise Removal in Diffusion Tensor MRI with speaker notes

Noisy Image

Euclidean

Riemannian

Aniso

Rician

56

Page 57: Rician Noise Removal in Diffusion Tensor MRI with speaker notes

Noisy Image

Euclidean

Riemannian

Aniso

Rician

57

Page 58: Rician Noise Removal in Diffusion Tensor MRI with speaker notes

58

Page 59: Rician Noise Removal in Diffusion Tensor MRI with speaker notes

59

Page 60: Rician Noise Removal in Diffusion Tensor MRI with speaker notes
Page 61: Rician Noise Removal in Diffusion Tensor MRI with speaker notes

Discussion

• Rician Filter : best RMS error performance.

• Real Data: Filtering DWIs better.

• Riemannian filtering: overall performs poorly

61

Page 62: Rician Noise Removal in Diffusion Tensor MRI with speaker notes

Conclusions

• Bias effects of Rician noise

• New Rician-bias correction filter

• Systematic comparison

Summary & Contributions

62

Page 63: Rician Noise Removal in Diffusion Tensor MRI with speaker notes

• ML method: low noise DWIs

• Filtering tools:

✦ Rician Bias Correction Filter

✦ Riemannian Space Tensor Filter

✦ Anisotropic Diffusion Filter on tensors

✦ Anisotropic Diffusion Filter on DWIs

63

Page 64: Rician Noise Removal in Diffusion Tensor MRI with speaker notes

• DT-MRI Ground Truth: investigate ML methods

• Noise effects:

•fiber-tractography,

•diagnostic decisions.

• Rician Noise model : Tensor estimation

Future Work:

64

Page 65: Rician Noise Removal in Diffusion Tensor MRI with speaker notes

Acknowledgments

Advisor:Dr. P. T. Fletcher

Committee: Dr. Ross T. Whitaker, Dr. Tolga Tasdizen

Gordon for help with Deft and teem

Josh for help with ITK

Dr. Guido Gerig, Dr. Wei Lin from UNC for providing us the real DT-MRI data

VIPER, NAMIC : Funding.

65

Page 66: Rician Noise Removal in Diffusion Tensor MRI with speaker notes

66

Thanks, Question?

Page 67: Rician Noise Removal in Diffusion Tensor MRI with speaker notes

67

Euclidean Space

Gradient

neighbors

Page 68: Rician Noise Removal in Diffusion Tensor MRI with speaker notes

68

Riemannian Space:

Gradient neighbors

Page 69: Rician Noise Removal in Diffusion Tensor MRI with speaker notes

69

Page 70: Rician Noise Removal in Diffusion Tensor MRI with speaker notes

70