despeckle filtering in medical ultrasound imaging

29
Despeckle Filtering in Medical Ultrasound Imaging Hairong Shi (1) Xingxing Wu (2) (1) Department of Medical Physics, University of Wisconsin-Madison (2) Department of Electrical and Computer Engineering, University of Wisconsin- Madison

Upload: roden

Post on 06-Jan-2016

42 views

Category:

Documents


2 download

DESCRIPTION

Despeckle Filtering in Medical Ultrasound Imaging. Hairong Shi (1) Xingxing Wu (2). (1)    Department of Medical Physics, University of Wisconsin-Madison (2)    Department of Electrical and Computer Engineering, University of Wisconsin-Madison. Introduction. - PowerPoint PPT Presentation

TRANSCRIPT

Page 1: Despeckle Filtering in Medical Ultrasound Imaging

Despeckle Filtering in Medical Ultrasound Imaging

Hairong Shi (1) Xingxing Wu (2)

(1)    Department of Medical Physics, University of Wisconsin-Madison(2)    Department of Electrical and Computer Engineering, University of Wisconsin-Madison

Page 2: Despeckle Filtering in Medical Ultrasound Imaging

Introduction

The medical Ultrasound B-scan image is acquired by summation of the echo signals from locally correlated scatterers in beam range.

Locally correlated multiplicative noises from small scatterers corrupt ultrasound image. These noises are commonly called “speckles”.

In many cases the speckle noise degrades the fine details and edge definition, limits the contrast resolution, limits the detect ability of small, low contrast lesions in body. And it should be filtered out.

Page 3: Despeckle Filtering in Medical Ultrasound Imaging

Noise Statistics

For research purpose,Radio-frequency (RF) data are collected. To show B-mode image, RF data are first envelope detected, and then logarithm compressed.

The multiplicative speckle is converted into additive noise after logarithm compression, the noise is spatially correlated, and has a Rayleigh amplitude PDF:

For fully developed speckle magnitude, the mean to standard deviation-pointwise SNR=1.9 (5.58dB)

Page 4: Despeckle Filtering in Medical Ultrasound Imaging

Filtering Methods

In this project, we implement 4 filtering methods:

(1) Wiener Filter; (2) Anisotropic Diffusion Filter; (3) Wavelet Filter; (4) Adaptive Filter;

Page 5: Despeckle Filtering in Medical Ultrasound Imaging

Test Images

We use the following test images to evaluate the performance of the filters.

(1) 4 simulated inclusion phantoms with different contrast. Center frequency 3MHz, band width 40%, no attenuation. Contrast 10dB, 5dB, -5dB and -10dB.

cm

cm

10dB

0 2 4 6

0

1

2

3

4

5

6

7

cm

cm

5dB

0 2 4 6

0

1

2

3

4

5

6

7

cm

cm

-5dB

0 2 4 6

0

1

2

3

4

5

6

7

cm

cm

-10dB

0 2 4 6

0

1

2

3

4

5

6

7

Page 6: Despeckle Filtering in Medical Ultrasound Imaging

Test Images (Cont’d)

(2) An in-vitro B-mode image for a plaque from human carotid artery. The plaque is embedded in gelatin. From Aloka SSD2000 Medical Ultrasound system.

cm

cm

Carotid Artery Plaque

0 0.5 1 1.5 2 2.5 3

0

0.5

1

1.5

2

2.5

3

Page 7: Despeckle Filtering in Medical Ultrasound Imaging

Method 1: Wiener Filter

Since the input filter g=1 in frequency domain, the Wiener filter is:

The power spectrum of the underlying image is modeled as:

Where σs2 can be replaced by the mean

variance of the noised image σx2. μx and μy are

frequency coordinators, the range is [-π, π).

Page 8: Despeckle Filtering in Medical Ultrasound Imaging

Wiener Filter: Noise Power Spectrum

The Power Spectrum of speckle pattern Sww is averaged from 12 simulated speckle patterns with image size 128*128.

0.5

1

1.5

2

2.5

x 106

x

y

Power Spectrum of Noise

-3 -2 -1 0 1 2 3

-3

-2

-1

0

1

2

3

Page 9: Despeckle Filtering in Medical Ultrasound Imaging

Wiener Filter Results

40

60

80

100

120

140

Weiner Filter restored image, inclusion 10dB

cm

cm

0 2 4 6

0

1

2

3

4

5

6

730

40

50

60

70

80

90

100

110

120

Weiner Filter restored image, inclusion 5dB

cm

cm

0 2 4 6

0

1

2

3

4

5

6

7

30

40

50

60

70

80

90

100

110Weiner Filter restored image, inclusion -5dB

cm

cm

0 2 4 6

0

1

2

3

4

5

6

730

40

50

60

70

80

90

100

110Weiner Filter restored image, inclusion -10dB

cmcm

0 2 4 6

0

1

2

3

4

5

6

7

10dB 5dB

-5dB -10dB

The restored images by Wiener filter are excellent:

(1) Most speckles are removed;

(2) Inclusions are clearly seen. even for 5dB contrast cases

(3) The background is uniform as we simulated.

The main reason is that the averaged power spectrum of the noise is very close to the noise power in the noised images, so we can restore images well.

Page 10: Despeckle Filtering in Medical Ultrasound Imaging

Wiener Filter Results (Cont’d)

Plaque Sample

cm

cm

Carotid Artery Plaque After Wiener Filter

0 0.5 1 1.5 2 2.5 3

0

0.5

1

1.5

2

2.5

3

Page 11: Despeckle Filtering in Medical Ultrasound Imaging

Wiener Filter Results (Cont’d)

The power spectrum of simulated noise can be applied well onto the real B-mode images:

(1) The speckles are also removed efficiently (2)The structure of the materials are restored. There are still some speckles in restored images, which

means the simulated noise power spectrum is not perfectly matched with the real ones.

The rest speckles can be removed by median filters. The image qualities can be improved by unsharp mask

and histogram stretch.

Page 12: Despeckle Filtering in Medical Ultrasound Imaging

Method 2: Anisotropic Diffusion Filter

Anisotropic diffusion is an efficient nonlinear technique for simultaneously performing contrast enhancement and noise reduction. It smoothes homogeneous image regions and retains image edges.

The main concept of Anisotropic diffusion is diffusion coefficient. Perona and Malik (1990) proposed 2 options:

Or

Page 13: Despeckle Filtering in Medical Ultrasound Imaging

Method 2: Anisotropic Diffusion Filter (Cont’d)

The anisotropic diffusion method can be iteratively applied to the output image:

Parameter k~[20,100], step sizeλ<=0.25.

Page 14: Despeckle Filtering in Medical Ultrasound Imaging

Anisotropic Diffusion Results:

20

40

60

80

100

120

140

cm

cm

Inclusion 10dB. Iteration Step: 100 k= 50 l= 0.25 method: 1

0 2 4 6

0

1

2

3

4

5

6

720

40

60

80

100

120

cm

cm

Inclusion 5dB. Iteration Step: 100 k= 50 l= 0.25 method: 1

0 2 4 6

0

1

2

3

4

5

6

7

10

20

30

40

50

60

70

80

90

100

110

cm

cm

Inclusion -5dB. Iteration Step: 100 k= 50 l= 0.25 method: 1

0 2 4 6

0

1

2

3

4

5

6

710

20

30

40

50

60

70

80

90

100

110

cm

cm

Inclusion -10dB. Iteration Step: 100 k= 50 l= 0.25 method: 1

0 2 4 6

0

1

2

3

4

5

6

7

10dB 5dB

-5dB -10dB

The anisotropic diffusion filter can restore noised image well:

(1)Speckles are removed and inclusions show clearly.

(2)In Anisotropic diffusion method, we don’t need know the noise pattern or power spectrum, this is the advantage over Wiener filter.

The anisotropic diffusion method needs more computation time than Wiener Filter method.

Parameter selection, iteration loop selection all affect the final results.

Page 15: Despeckle Filtering in Medical Ultrasound Imaging

Anisotropic Diffusion Results: (Cont’d)

Plaque Samplecm

cm

Carotid Artery Plaque After Anisotropic Diffusion Filter

0 0.5 1 1.5 2 2.5 3

0

0.5

1

1.5

2

2.5

3

Page 16: Despeckle Filtering in Medical Ultrasound Imaging

Anisotropic Diffusion Results: (Cont’d)

The anisotropic diffusion method gives better contrast while removing speckles effectively.

In fact, because the parameters in anisotropic diffusion method are adjustable, we can control parameters and

choose the best image.

Page 17: Despeckle Filtering in Medical Ultrasound Imaging

Comparison of Image Profile

50 100 150 200 250

20

40

60

80

100

120

140

Beam Profile at line 350

Original ProfileWiener Filter ResultAnisotropic Diffusion Result

Image profiles before and after Wiener filter, and anisotropic diffusion are plotted.

Image becomes smoother after filtering.

Page 18: Despeckle Filtering in Medical Ultrasound Imaging

Method 3: K-distribution Based Adaptive Filter

The K distribution model is a model for speckle statistics of ultrasound echo speckle.

The K distribution is a good model for the echo envelope signal statistics when the scatter number densities are low.

The model can accurately predict variations in the statistics with varying scatterer number.

Page 19: Despeckle Filtering in Medical Ultrasound Imaging

K Distribution

The K distribution as a function of a

K distribution pdf

Page 20: Despeckle Filtering in Medical Ultrasound Imaging

Adaptive filter for uncompressed images

The restored image Y can be calculated by

Where X is the original image, is the image averaged value

Page 21: Despeckle Filtering in Medical Ultrasound Imaging

Adaptive filter for log-compressed images

The restored image Y can be calculated by

Where X is the original image, is the image averaged value. is the compensation coefficient.

Page 22: Despeckle Filtering in Medical Ultrasound Imaging

Adaptive Filter results:

Original image for 5dB inclusion phantom

Image of 5dB inclusion Phantom after filtering

Page 23: Despeckle Filtering in Medical Ultrasound Imaging

Adaptive Filter results: (Cont’d)

The filter can smooth image locally based on some local statistics.

This filter is easy to implement and the statistics is easy to estimate.

There is no need to find an optimal solution.

Page 24: Despeckle Filtering in Medical Ultrasound Imaging

Method 4: Wavelet Filter

The wavelet techniques are widely used in the image processing, such as the image compression, image denoising.

The wavelet filter has good image processing performance.

We use thresholding method to despeckle.

Page 25: Despeckle Filtering in Medical Ultrasound Imaging

Wavelet FilterImage decomposition Equation:

Decomposed image

Page 26: Despeckle Filtering in Medical Ultrasound Imaging

Wavelet Filter Results:

Original image for Carotid Artery Plaque

Image of plaqueAfter filtering

Page 27: Despeckle Filtering in Medical Ultrasound Imaging

Wavelet Filter Results: (Cont’d)

Original image for 5dB inclusion phantom

Image of 5dB inclusion Phantom after filtering

Page 28: Despeckle Filtering in Medical Ultrasound Imaging

Comparison of Filter Performance

Original Wiener Anisotropic diffusion

Adaptive filter

Wavelet

SNR 4.295 26.885 16.841 18.866 4.655

To evaluate the performance of 4 different filters, we we take the same small region with pixel size 64*64, and calculate the mean-standard deviation ratio, i.e. pixel-wised SNR.

Wiener filter, Anisotropic diffusion filter and k distribution based adaptive filter improve

the SNR. Wavelet filter doesn’t improve the SNR very much.

Page 29: Despeckle Filtering in Medical Ultrasound Imaging

Conclusion

The Wiener filter can improve the image qualities well and simulated power spectrum of speckle can be applied on many situations.

The Anisotropic diffusion filter can also despeckle well as long as we choose reasonable parameters, and it doesn’t need extra information of noise pattern.

The K-distribution based adaptive filter can improve the image quality, the method is easy to implement and the statistics is easy to estimate and characterize.

The wavelet filter is not highly suitable for removing the speckle in ultrasound images.