Speckle Noise Reduction in Ultrasound Images using Adaptive and Anisotropic Diffusion filter
The document discusses a modified anisotropic diffusion filter developed for reducing speckle noise in ultrasound images, with a focus on its theoretical background and application results. It highlights the effectiveness of the proposed filter in enhancing signal-to-noise ratio (SNR) and image quality indexes across various types of ultrasound images, while also noting limitations related to its application on different imaging modalities. The conclusion emphasizes the potential for future improvements and broader applications in noise reduction.
The presentation introduces speckle noise reduction in ultrasound images, supervised by Dr. Md. Motiur Rahman, focusing on adaptive and anisotropic diffusion filters.
Highlights the benefits of ultrasound (US) imaging and the necessity of reducing speckle noise using modified anisotropic filters.
Describes speckle noise in medical images, acquired through coherent imaging systems, showing its impact on image quality.
Discusses various metrics: SNR, RMSE, PSNR, IMGQ, and SSIM to evaluate ultrasound image quality and noise reduction effectiveness.
Covers Lee, Kuan, and Frost filters used for speckle noise reduction in ultrasound images, detailing their weighting functions and methodologies.
Explains anisotropic diffusion filter’s role in noise reduction while preserving edges, matching the work of Perona and Malik and further developments like SRAD.
Presents inputs and outputs of different ultrasound images processed for speckle noise reduction, showing results across various images including Abdomen and Prostate.
Introduces a modified anisotropic diffusion filter with detailed algorithms for speckle noise reduction and includes input-output examples.
Discusses the limitations of the proposed filter, highlighting its effectiveness in US images but not for others like CT or MRI.
Summarizes successful noise reduction capabilities of the proposed filter, suggesting future use in 3D and 4D ultrasound images.
Ends with a note of thanks and an invitation for questions.
Speckle Noise Reduction in Ultrasound Images using Adaptive and Anisotropic Diffusion filter
1.
1
Speckle Noise Reductionin Ultrasound Images using Adaptive and
Anisotropic Diffusion filter
Submitted By
Md. Shohel Rana
CE-06006
2005-06
Supervised By
Dr. Md. Motiur Rahman
Associate Professor
Dept. of CSE
MBSTU
2.
2
Motivation and Contribution
US Imaging Technique less cost
Nonlinear and Anisotropic filter for removing speckle noise can be
removed from US images
Proposed a modified Anisotropic filter which reduce speckle noises
3.
3
Speckle Noise
Medicalimages are usually corrupted by noise during their
acquisition and transmission.
Affects all coherent imaging systems, including medical
ultrasound.
Acquired image is corrupted by a random granular pattern.
A possible generalized model of the speckle imaging is
Where g , f , u and ξ stand for the observed image, original image,
multiplicative component and additive component of the speckle noise,
respectively.
( , ) ( , ) ( , ) ( , )g n m f n m u n m n mξ= +
4.
4
US Image MeasurementMetrics
SNR : Signal to Noise Ratio compares the level of desired signal to the level of
background noise
RMSE : The Root Mean square error is given by
PSNR : Peak Signal to Noise Ratio is computed by
22
, ,
1 1
10 2
1 1
, ,
( )
10log
( )
M N
i j i j
i j
M N
i j
i j i j
SNR
yx
yx
= =
= =
+
= =
∑∑
−∑∑
2
1 1
, ,
1
( )
M N
i j
i j i jRMSE
MN
yx
= =
= −∑∑
2
10
20 ( )logPSNR RMSEg=
5.
US Image Measurement
Metrics(Cont.)
IMGQ : The image quality index is given by
SSIM : The Structural Similarity Index between two images is
computed as
PT : To compare edge preservation performances of different
speckle reduction schemes, the parameter of transition
2 2 22
22
( ) ( )
xy y x
x y yx
y x
IMGQI
y x
σ σ σ
σ σ σσ
=
+ +
2 1
2 1
PT
µ µ
µ µ
−
=
+
1 2
2 2 2 2
1 2
(2 )(2 )
( )( )
xyx y
x yx y
SSIM
C C
C C
µ µ σ
µ µ σ σ
+ +
=
+ + + +
6.
6
Filtering On USImage
The Lee and Kuan filters produce the enhancement data
according to
Where W is the weighting function ranging between 0 for flat regions and
1 for regions with high signal activity, is the average of pixels in a
moving window and is the output of the filter.
ˆ( ) ( ) [ ( ) ( )] ( )R t I t I t I t W t= + − ×
I
ˆ( )R t
7.
7
Filtering On USImage(Cont.)
Lee filter
The weighting function for the Lee filter is
Kuan Filter
The weighting function of the kuan filter is defined as
Where and are the coefficients of variations of noise u and
image I
Frost Filter
The image Z[i, j] is modeled by frost as
Where h i j, is system impulse response and * denotes convolution
2
2
( ) 1
( )
u
I
W t
t
C
C
= −
2
2
2
1
( )
( )
1
u
I
u
t
W t
C
C
C
−
=
+
I
I
I
C
σ=u
u
u
C
σ=
, , , ,[ . ]*i j i j i j i jn hZ Z=
8.
Filtering On USImage (Cont.)
Frost Filter (Cont.)
Minimum mean square filter has the form
Where m(t ) function is an isotropic impulse response
Where K1 is a normalizing constant and α is the decay constant correlation
coefficient between adjacent pixels of the original is image x(t) and t
corresponds to the distance between pixels in the spatial domain.
ˆ( ) ( )* ( )x t z t m t=
1( ) exp( )m t tK α α= −
9.
9
Anisotropic Diffusion Filter(Cont.)
In Image Procesing and Computer Vision, Anisotropic
Diffusion, also called Perona–Malik diffusion.
A technique aiming at reducing image noise without removing
significant parts of the image content, typically edges.
Anisotropic diffusion is an efficient nonlinear technique for
simultaneously performing contrast enhancement and noise
reduction. It smoothes homogeneous image regions and retains
image edges.
( )[ ]
( )
==
∇⋅∇=
∂
∂
00 ItI
IIcdiv
t
I
10.
10
Anisotropic Diffusion Filter(Cont.)
Pietro Perona and Jitendra Malik pioneered the idea of
anisotropic diffusion in 1990 and proposed two functions for
the diffusion coefficient
and
The anisotropic diffusion method can be iteratively
applied to the output image:
( ) ( )
( )
( ) ( ) ( )
( ) ( )
( )
( ) ( ) ( )
( ) ( )
∇⋅∇+∇⋅∇+
∇⋅∇+∇⋅∇
×+=+
n
South
n
South
n
West
n
West
n
East
n
East
n
North
n
Northnn
IIcIIc
IIcIIc
II λ1
( )
( )2
/1
1
kI
Ic
∇+
=∇( ) ( )[ ]2
/||exp|| kIIc ∇−=∇
11.
11
Anisotropic Diffusion Filter(Cont.)
SRAD
Allows the generation of an image scale space without bias due to filter
window size and shape.
Preserves edges and enhances edges by inhibiting diffusion across edges and
allowing diffusion on either side of the edge.
Related directly to the Lee and Frost window-based filters.
The edge sensitive extension of conventional adaptive speckle filter
The automatic determination of q0(t) is desired in real applications to
eliminate heuristic parameter choice
Where is a constant, and q0(t) is the speckle coefficient of variation in the observed
image.
*
0 0
( ) t
tq q e
ρ−
≈
ρ
12.
Anisotropic Diffusion Filter(Cont.)
SRAD (Cont.)
The discrete isotropic diffusion update is
Assuming that pixels in the region are statistically independent and identically
distributed and the local means remains the same before and after an iteration
So the final updating equation is
Where is the divergence.
ρ
, , 1, 1, , 1 , 1 ,( 4 )
4
t t t t t t t t
i j i j i j i j i j i j i j
t
I I I I I I I
+∆
+ − + −
∆
= + + + + −
2
2
( ) ( )
4
( )(1 )o o
t t t
tq q t+∆ = +
∆−∆
1
, , ,
4
nn n
i j i j i j
t
dI I
+ ∆
= +
,
n
i jd
13.
13
Experiment and Result(Cont.)
Input and Output
Input and Output of
Abdomen image for
speckle noise (0.04) in
various filtering
14.
Experiment and Result(Cont.)
Input and Output
Input and Output of
Ortho image for
speckle noise (0.04) in
various filtering
15.
Experiment and Result(Cont.)
Input and Output
Input and Output of
Liver_GB image for
speckle noise (0.04) in
various filtering
16.
Experiment and Result(Cont.)
Input and Output
Input and Output of
Kidney image for
speckle noise (0.04) in
various filtering
17.
Experiment and Result(Cont.)
Input and Output
Input and Output of
Brest image for
speckle noise (0.04) in
various filtering
18.
Experiment and Result(Cont.)
Input and Output
Input and Output of
Prostrate image for
speckle noise (0.04) in
various filtering
20
Proposed Filter
In thecase of proposed filtering is the modified version of
anisotropic diffusion filtering the direction and strength of the
diffusion are controlled by an edge detection function
Algorithm
1. Input image with or without speckle noise
2. Choose a kernel or window of size 5*5 or 3*3.
3. Set the kernel or window to the noisy image and replace each
pixel value of image by the following equation
0[1: ] [1: ]
( )
( )
n
sort n
mid sort
F MN
sort
mid
W I
W W
I I
=
=
=
21.
21
Proposed Filter (Cont.)
Algorithm(Cont.)
4. Calculate gradient in all directions (N,S,E,W) of Processed image Imid .
5. Calculate Diffusion coefficients in all directions according to the method
using proposed modified equation.
6. Finally follow the following equation
2
1
(|| ||)
1
|| ||
mid
mid
C I
I
K
∆ =
+
∆
( ) ( )
( ) ( )
0
North mid North mid East mid East mid
out mid
West mid West mid South mid South mid
c c
c c
I I I I
I I I
I I I I
λ
∇ ×∇ + ∇ ×∇
= + + ×
+ ∇ ×∇ + ∇ ×∇
Limitation
My proposedmodified filter gives best result of SNR, SSIM and IMGQ for
all types of ultrasound images but not of RMSE, PSNR, PT for all images.
It has been demonstrated that proposed modified filter is capable of
reducing speckle noise but not for other types of image.
It has been demonstrated that proposed modified filter is used for only
Ultrasound images but not for CT, MRI, X-RAY and so on.
It has been demonstrated that proposed modified filter is capable of
reducing speckle noise on 2D images but not 3D, 4D.
Used images’ extension was “.tif”.
25.
25
Conclusion And FutureWork
It has been demonstrated that proposed filter is capable of reducing
speckle noise.
My proposed modified filter gives Signal to Noise Ratio (SNR) value is
best for all six (Abdomen, Ortho, Liver_GB, Kidney, Brest and Prostrate)
types of US images as well as gives the best result for Image Quality
Index (IMGQI) and Structural Similarity Index (SSIM).
It can be used in 3D,4D Ultrasound images.
Trying to remove all types of noises from images.