05-07-2012, 09:55 AM
A View on Despeckling in Ultrasound Imaging
[attachment=26752]
Abstract
Ultrasound imaging is a widely used and safe medical diagnostic technique, due to its
noninvasive nature, low cost and capability of forming real time imaging. However the
usefulness of ultrasound imaging is degraded by the presence of signal dependant noise known
as speckle. The speckle pattern depends on the structure of the image tissue and various
imaging parameters. There are two main purposes for speckle reduction in medical ultrasound
imaging (1) to improve the human interpretation of ultrasound images (2) despeckling is the
preprocessing step for many ultrasound image processing tasks such as segmentation and
registration. A number of methods have been proposed for speckle reduction in ultrasound
imaging. While incorporating speckle reduction techniques as an aid for visual diagnosis, it
has to keep in mind that certain speckle contains diagnostic information and should be
retained. The objective of this paper is to give an overview about types of speckle reduction
techniques in ultrasound imaging
1. Importance of ultrasound Imaging
ULTRASOUND imaging application in medicine and other fields is enormous. It has
several advantages over other medical imaging modalities. The use of ultrasound in diagnosis
is well established because of its noninvasive nature, low cost, capability of forming real time
imaging and continuing improvement in image quality. It is estimated that one out of every
four medical diagnostic image studies in the world involves ultrasonic techniques. US waves
are characterized by frequency above 20 KHz which is the upper limit of human hearing. In
medical US applications, frequencies are used between 500 KHz and 30 MHz B-mode
imaging is the most used modality in medical US. An US transducer which is placed onto the
patient's skin over the imaged region sends an US pulse which travels along a beam into the
tissue. Due to interfaces some of the US energy is reflected back to the transducer which
converts it into echo signals. These signals are then sent into amplifiers and signal processing
circuits in the imaging machine's hardware to form a 2-D image. This process of sending
pulses launched in different directions is repeated in order to examine the whole region in the
body. Thus, US imaging involves signals which are obtained by coherent summation of echo
signals from scatterers in the tissue.
In many cases volume quantification is important in assessing the progression of
diseases and tracking progression of response to treatment. Thus, 3D ultrasound imaging has
drawn great attention in recent years.
2. Ultrasound Imaging System
International Journal of Signal Processing, Image Processing and Pattern Recognition
Vol.2, No.3, September 2009
86
Figure 1 shows a functional block diagram of an ultrasound imaging system. The
construction of ultrasound B-mode image involves capturing the echo signal returned from
tissue at the surface of piezoelectric crystal transducers. These transducers convert the
ultrasonic RF mechanical wave into electrical signal. Convex ultrasound probes collect the
echo from tissue in a radial form. Each group of transducers is simultaneously activated to
look at a certain spatial direction from which they generate a raw line signal (stick) to be used
later for raster image construction. These sticks are then demodulated and logarithmically
compressed to reduce their dynamic range to suit the commercial display devices. The final
Cartesian image is constructed from the sampled sticks in a process called scan conversion.
Speckle reduction techniques can be applied on envelope detected data, log compressed
data or on scan converted data. However, slightly different results will be produced for each
data. In the compression stage some useful information about the imaged object may be
deteriorated or even lost. However, any processing which works with envelope detected data
has more information at its disposal and preserves more useful information. Compared to
processing the scan converted image, envelope detected data has fewer pixels and thus incurs
lower computational cost.
For optimum result envelope detected data processing is preferred because some
information that lost after the compression stage cannot be recovered by working with log
compressed data or the scan converted image. However, the real time speckle reduction
methods are applied on the scan converted image, since the scan converted image is always
accessible where most commercial ultrasound systems do not output the envelope detected or
log compressed data.
Figure 1. Block diagram of Ultrasound Imaging System
3. Speckle in Ultrasound Imaging
Speckle in US B-scans is seen as a granular structure which is caused by the constructive and
destructive coherent interferences of back scattered echoes from the scatterers that are typically
much smaller than the spatial resolution of medical ultrasound system. This phenomenon is
common to laser, sonar and synthetic aperture radar imagery (SAR). Speckle pattern is a form
of multiplicative noise and it depends on the structure of imaged tissue and various imaging
parameters.
Ultra sound
Transducer
Puls
Envelop
e
Scan
conversi
Transmi
t/
Receiv Receiv
e beam
formin
Receive
focusing
Transm
it
TG
Logarithm
ic
International Journal of Signal Processing, Image Processing and Pattern Recognition
Vol. 2, No.3, September 2009
87
Speckle degrades the target delectability in B-scan images and reduces the contrast,
resolutions which affect the human ability to identify normal and pathological tissue. It also
degrades the speed and accuracy of ultrasound image processing tasks such as segmentation and
registration.
The nature of the speckle pattern can be categorized into one of three classes according to the
number of scatterers per resolution cell or the so called scatterer number density (SND), spatial
distribution and the characteristics of the imaging system itself. These classes are described as
follows:
1. FFS (Fully formed speckle) pattern, which occurs when many fine randomly distributed
scattering sites exist within the resolution cell of the pulse-echo system. In this case, the
amplitude of the backscattered signal can be modeled as a Rayleigh distributed random
variable with a constant SNR of 1.92. Under such conditions, the textural features of the
speckle pattern represent a multivariate signature of the imaging instrument and its point
spread function. Blood cells are typical examples of this type of scatterers.
2. Non randomly distributed with long-range order (NRLR). Examples of this type are the
lobules in liver parenchyma. It contributes a coherent or specular backscattered intensity that
is in itself spatially varying. Due to the correlation between scatterers, the effective number of
scatterers is finite. This situation can be modeled by the K-distribution. This type is
associated with SNR below 1.92. It can also be modeled by the Nakagami distribution.
3. Non randomly distributed with short-range order (NRSR). Examples of this type include
organ surfaces and blood vessels. When a spatially invariant coherent structure is present
within the random scatterer region, the probability density function (PDF) of the
backscattered signals becomes close to the Rician distribution. This class is associated with
SNR above 1.92
4. Need for despeckling
Thus, speckle is considered as the dominant source of noise in ultrasound imaging and should
be processed without affecting important image features.The main purposes for speckle
reduction in medical ultrasound imaging are:
1. To improve the human interpretation of ultrasound images – speckle reduction makes an
ultrasound image cleaner with clearer boundaries.
2. Despeckling is a preprocess step for many ultrasound image processing tasks such as
segmentation and registration – speckle reduction improves the speed and accuracy of
automatic and semiautomatic segmentation & registration.
5. Key issues in developing an efficient and robust denoising method
One has to take into account the following factors in developing an efficient and robust
denoising method for ultrasound images
5.1. Adaptation to features of interest
For an experienced radiologist speckle noise (Also referred as “texture” in medical
literature) may present diagnostic information. The degree of speckle smoothing depends on
the expert’s knowledge & the application at hand like enhancement for visual inspection or
preprocessing for automatic segmentation
International Journal of Signal Processing, Image Processing and Pattern Recognition
Vol.2, No.3, September 2009
88
For automatic segmentation it is usually preferred to keep the sharpness of the boundaries
between different image regions and to smooth out the speckle texture. For visual
interpretation the texture smoothing may be less preferable.
5.2. Adaptation to spatial content
The medical ultrasound images have significant spatial correlation. A spatially adoptive
denoising can be based on statistical content models or on adopting certain filter parameters
based on measurements from a local window around each pixel
5.3. Proposal of noise models
The basic assumption in majority of speckle filters is that the speckle is fully developed
and is modeled as multiplicative noise. Logarithmic operation transforms speckle into
additive white Gaussian Noise.
But for different reasons such a speckle model seems to be too simplistic in the case of
medical ultrasound images. Speckle is not necessarily fully developed and there exists a
pronounced spatial correlation. Moreover the ultrasound devices themselves usually perform
a preprocessing of the raw data including even logarithmic compression. Thus in the
displayed medical ultrasound images the noise differs significantly from often assumed
multiplicative method.
6. Speckle reduction methods
Several techniques have been proposed for despeckling in medical ultrasound imaging. In
this section we present the classification and theoretical overview of existing despeckling
techniques
6.1. Compounding Methods
Number of papers have been proposed based on compounding technique [1]-[3].In
this method a series of ultrasound images of the same target are acquired from different scan
directions and with different transducer frequencies or under different strains. Then the
images are averaged to form a composite image.
The compounding method can improve the target detectability but they suffer from
degrade spatial resolution and increased system complexity.
6.2. Post Acquisition Methods
This method do not require many hardware modification .The post acquisition image
processing technique falls under two categories (1) Single scale spatial filtering (2) Multiscale
Methods
6.2.1. Single scale spatial filtering Methods
A speckle reduction filter that changes the amount of smoothing according to the
ratio of local variance to local mean was developed [4].in that method smoothing is
increased in homogeneous region where speckle is fully developed and reduced or
even avoided in other regions to preserve details
International Journal of Signal Processing, Image Processing and Pattern Recognition
Vol. 2, No.3, September 2009
89
An unsharp masking filter was suggested [5] in which the smoothing level is adjusted
depending on the statistics of log compressed images
The above mentioned filters have difficulty in removing speckle near or on image edges
Recently proposed filter utilizing short line segments in different angular orientations
and selecting the orientation that is most likely to represent a line in the image [6]
This technique poses a tradeoff between effective line enhancement and speckle
reduction
Numbers of Region growing based spatial filtering methods [7-9] have been proposed
.In these methods it is assumed that pixels that have similar gray level and
connectivity are related and likely to belong to the same object or region. After all
pixels are allocated to different groups, spatial filtering is performed based on the
local statistics of adaptive regions whose sizes and shapes are determined by the
information content of the image.
The main difficulty in applying region growing based methods is how to design
appropriate similarity criteria for region growing. Different types of filters are used in the
application of despeckling in ultrasound imaging. The most commonly used types of filters
are:
a. Kaun &Lee Filters [10] - Lee filter form an output image by computing a linear
combination of the center pixel intensity in a filter window with the average intensity
of the window.Kaun and Lee filter have the same formulation although signal model
assumption and derivations are different. These two filters achieve a balance between
straight forward averaging in homogeneous regions and identity filter where edges
and point features exist. This balance depends on the coefficient of variation inside
the moving window.
b. Frost Filter [11] achieves a balance between averaging and all pass filter by forming
an exponentially shaped filter kernel. The response of the filter varies locally with the
coefficient of variation
c. Enhanced Lee & Frost filter are used to alter the performance locally according to
the threshold value. Pure averaging is induced when the local coefficient of variation
is below a lower threshold. The filter performs as a strict all pass filters when the
local coefficient of variation is above a higher threshold. When the coefficient of
variation is in between the two thresholds, a balance between averaging and identity
operation is computed.
d. Mean Filter [12] has the property of locally reducing the variance thus reducing
SNR and it requires the user to specify only the size of the window. However it has
the effect of potentially blurring the image. This filter is optimal for additive
Gaussian noise whereas the speckled image obeys a multiplicative model with non
Gaussian noise. Therefore simple mean is not the optimal choice.
International Journal of Signal Processing, Image Processing and Pattern Recognition
Vol.2, No.3, September 2009
90
e. Median Filters [13] are utilized for despeckling due to their robustness against
impulsive type noise and edge preserving characteristics. The median filter produces
less blurred images. The compounding procedure uses both the mean and median
filters.
f. Maximum a posterior (MAP) filter requires assumption about the distribution of the
true process and the degradation model. Different assumptions lead to different MAP
estimators and different complexities.
[attachment=26752]
Abstract
Ultrasound imaging is a widely used and safe medical diagnostic technique, due to its
noninvasive nature, low cost and capability of forming real time imaging. However the
usefulness of ultrasound imaging is degraded by the presence of signal dependant noise known
as speckle. The speckle pattern depends on the structure of the image tissue and various
imaging parameters. There are two main purposes for speckle reduction in medical ultrasound
imaging (1) to improve the human interpretation of ultrasound images (2) despeckling is the
preprocessing step for many ultrasound image processing tasks such as segmentation and
registration. A number of methods have been proposed for speckle reduction in ultrasound
imaging. While incorporating speckle reduction techniques as an aid for visual diagnosis, it
has to keep in mind that certain speckle contains diagnostic information and should be
retained. The objective of this paper is to give an overview about types of speckle reduction
techniques in ultrasound imaging
1. Importance of ultrasound Imaging
ULTRASOUND imaging application in medicine and other fields is enormous. It has
several advantages over other medical imaging modalities. The use of ultrasound in diagnosis
is well established because of its noninvasive nature, low cost, capability of forming real time
imaging and continuing improvement in image quality. It is estimated that one out of every
four medical diagnostic image studies in the world involves ultrasonic techniques. US waves
are characterized by frequency above 20 KHz which is the upper limit of human hearing. In
medical US applications, frequencies are used between 500 KHz and 30 MHz B-mode
imaging is the most used modality in medical US. An US transducer which is placed onto the
patient's skin over the imaged region sends an US pulse which travels along a beam into the
tissue. Due to interfaces some of the US energy is reflected back to the transducer which
converts it into echo signals. These signals are then sent into amplifiers and signal processing
circuits in the imaging machine's hardware to form a 2-D image. This process of sending
pulses launched in different directions is repeated in order to examine the whole region in the
body. Thus, US imaging involves signals which are obtained by coherent summation of echo
signals from scatterers in the tissue.
In many cases volume quantification is important in assessing the progression of
diseases and tracking progression of response to treatment. Thus, 3D ultrasound imaging has
drawn great attention in recent years.
2. Ultrasound Imaging System
International Journal of Signal Processing, Image Processing and Pattern Recognition
Vol.2, No.3, September 2009
86
Figure 1 shows a functional block diagram of an ultrasound imaging system. The
construction of ultrasound B-mode image involves capturing the echo signal returned from
tissue at the surface of piezoelectric crystal transducers. These transducers convert the
ultrasonic RF mechanical wave into electrical signal. Convex ultrasound probes collect the
echo from tissue in a radial form. Each group of transducers is simultaneously activated to
look at a certain spatial direction from which they generate a raw line signal (stick) to be used
later for raster image construction. These sticks are then demodulated and logarithmically
compressed to reduce their dynamic range to suit the commercial display devices. The final
Cartesian image is constructed from the sampled sticks in a process called scan conversion.
Speckle reduction techniques can be applied on envelope detected data, log compressed
data or on scan converted data. However, slightly different results will be produced for each
data. In the compression stage some useful information about the imaged object may be
deteriorated or even lost. However, any processing which works with envelope detected data
has more information at its disposal and preserves more useful information. Compared to
processing the scan converted image, envelope detected data has fewer pixels and thus incurs
lower computational cost.
For optimum result envelope detected data processing is preferred because some
information that lost after the compression stage cannot be recovered by working with log
compressed data or the scan converted image. However, the real time speckle reduction
methods are applied on the scan converted image, since the scan converted image is always
accessible where most commercial ultrasound systems do not output the envelope detected or
log compressed data.
Figure 1. Block diagram of Ultrasound Imaging System
3. Speckle in Ultrasound Imaging
Speckle in US B-scans is seen as a granular structure which is caused by the constructive and
destructive coherent interferences of back scattered echoes from the scatterers that are typically
much smaller than the spatial resolution of medical ultrasound system. This phenomenon is
common to laser, sonar and synthetic aperture radar imagery (SAR). Speckle pattern is a form
of multiplicative noise and it depends on the structure of imaged tissue and various imaging
parameters.
Ultra sound
Transducer
Puls
Envelop
e
Scan
conversi
Transmi
t/
Receiv Receiv
e beam
formin
Receive
focusing
Transm
it
TG
Logarithm
ic
International Journal of Signal Processing, Image Processing and Pattern Recognition
Vol. 2, No.3, September 2009
87
Speckle degrades the target delectability in B-scan images and reduces the contrast,
resolutions which affect the human ability to identify normal and pathological tissue. It also
degrades the speed and accuracy of ultrasound image processing tasks such as segmentation and
registration.
The nature of the speckle pattern can be categorized into one of three classes according to the
number of scatterers per resolution cell or the so called scatterer number density (SND), spatial
distribution and the characteristics of the imaging system itself. These classes are described as
follows:
1. FFS (Fully formed speckle) pattern, which occurs when many fine randomly distributed
scattering sites exist within the resolution cell of the pulse-echo system. In this case, the
amplitude of the backscattered signal can be modeled as a Rayleigh distributed random
variable with a constant SNR of 1.92. Under such conditions, the textural features of the
speckle pattern represent a multivariate signature of the imaging instrument and its point
spread function. Blood cells are typical examples of this type of scatterers.
2. Non randomly distributed with long-range order (NRLR). Examples of this type are the
lobules in liver parenchyma. It contributes a coherent or specular backscattered intensity that
is in itself spatially varying. Due to the correlation between scatterers, the effective number of
scatterers is finite. This situation can be modeled by the K-distribution. This type is
associated with SNR below 1.92. It can also be modeled by the Nakagami distribution.
3. Non randomly distributed with short-range order (NRSR). Examples of this type include
organ surfaces and blood vessels. When a spatially invariant coherent structure is present
within the random scatterer region, the probability density function (PDF) of the
backscattered signals becomes close to the Rician distribution. This class is associated with
SNR above 1.92
4. Need for despeckling
Thus, speckle is considered as the dominant source of noise in ultrasound imaging and should
be processed without affecting important image features.The main purposes for speckle
reduction in medical ultrasound imaging are:
1. To improve the human interpretation of ultrasound images – speckle reduction makes an
ultrasound image cleaner with clearer boundaries.
2. Despeckling is a preprocess step for many ultrasound image processing tasks such as
segmentation and registration – speckle reduction improves the speed and accuracy of
automatic and semiautomatic segmentation & registration.
5. Key issues in developing an efficient and robust denoising method
One has to take into account the following factors in developing an efficient and robust
denoising method for ultrasound images
5.1. Adaptation to features of interest
For an experienced radiologist speckle noise (Also referred as “texture” in medical
literature) may present diagnostic information. The degree of speckle smoothing depends on
the expert’s knowledge & the application at hand like enhancement for visual inspection or
preprocessing for automatic segmentation
International Journal of Signal Processing, Image Processing and Pattern Recognition
Vol.2, No.3, September 2009
88
For automatic segmentation it is usually preferred to keep the sharpness of the boundaries
between different image regions and to smooth out the speckle texture. For visual
interpretation the texture smoothing may be less preferable.
5.2. Adaptation to spatial content
The medical ultrasound images have significant spatial correlation. A spatially adoptive
denoising can be based on statistical content models or on adopting certain filter parameters
based on measurements from a local window around each pixel
5.3. Proposal of noise models
The basic assumption in majority of speckle filters is that the speckle is fully developed
and is modeled as multiplicative noise. Logarithmic operation transforms speckle into
additive white Gaussian Noise.
But for different reasons such a speckle model seems to be too simplistic in the case of
medical ultrasound images. Speckle is not necessarily fully developed and there exists a
pronounced spatial correlation. Moreover the ultrasound devices themselves usually perform
a preprocessing of the raw data including even logarithmic compression. Thus in the
displayed medical ultrasound images the noise differs significantly from often assumed
multiplicative method.
6. Speckle reduction methods
Several techniques have been proposed for despeckling in medical ultrasound imaging. In
this section we present the classification and theoretical overview of existing despeckling
techniques
6.1. Compounding Methods
Number of papers have been proposed based on compounding technique [1]-[3].In
this method a series of ultrasound images of the same target are acquired from different scan
directions and with different transducer frequencies or under different strains. Then the
images are averaged to form a composite image.
The compounding method can improve the target detectability but they suffer from
degrade spatial resolution and increased system complexity.
6.2. Post Acquisition Methods
This method do not require many hardware modification .The post acquisition image
processing technique falls under two categories (1) Single scale spatial filtering (2) Multiscale
Methods
6.2.1. Single scale spatial filtering Methods
A speckle reduction filter that changes the amount of smoothing according to the
ratio of local variance to local mean was developed [4].in that method smoothing is
increased in homogeneous region where speckle is fully developed and reduced or
even avoided in other regions to preserve details
International Journal of Signal Processing, Image Processing and Pattern Recognition
Vol. 2, No.3, September 2009
89
An unsharp masking filter was suggested [5] in which the smoothing level is adjusted
depending on the statistics of log compressed images
The above mentioned filters have difficulty in removing speckle near or on image edges
Recently proposed filter utilizing short line segments in different angular orientations
and selecting the orientation that is most likely to represent a line in the image [6]
This technique poses a tradeoff between effective line enhancement and speckle
reduction
Numbers of Region growing based spatial filtering methods [7-9] have been proposed
.In these methods it is assumed that pixels that have similar gray level and
connectivity are related and likely to belong to the same object or region. After all
pixels are allocated to different groups, spatial filtering is performed based on the
local statistics of adaptive regions whose sizes and shapes are determined by the
information content of the image.
The main difficulty in applying region growing based methods is how to design
appropriate similarity criteria for region growing. Different types of filters are used in the
application of despeckling in ultrasound imaging. The most commonly used types of filters
are:
a. Kaun &Lee Filters [10] - Lee filter form an output image by computing a linear
combination of the center pixel intensity in a filter window with the average intensity
of the window.Kaun and Lee filter have the same formulation although signal model
assumption and derivations are different. These two filters achieve a balance between
straight forward averaging in homogeneous regions and identity filter where edges
and point features exist. This balance depends on the coefficient of variation inside
the moving window.
b. Frost Filter [11] achieves a balance between averaging and all pass filter by forming
an exponentially shaped filter kernel. The response of the filter varies locally with the
coefficient of variation
c. Enhanced Lee & Frost filter are used to alter the performance locally according to
the threshold value. Pure averaging is induced when the local coefficient of variation
is below a lower threshold. The filter performs as a strict all pass filters when the
local coefficient of variation is above a higher threshold. When the coefficient of
variation is in between the two thresholds, a balance between averaging and identity
operation is computed.
d. Mean Filter [12] has the property of locally reducing the variance thus reducing
SNR and it requires the user to specify only the size of the window. However it has
the effect of potentially blurring the image. This filter is optimal for additive
Gaussian noise whereas the speckled image obeys a multiplicative model with non
Gaussian noise. Therefore simple mean is not the optimal choice.
International Journal of Signal Processing, Image Processing and Pattern Recognition
Vol.2, No.3, September 2009
90
e. Median Filters [13] are utilized for despeckling due to their robustness against
impulsive type noise and edge preserving characteristics. The median filter produces
less blurred images. The compounding procedure uses both the mean and median
filters.
f. Maximum a posterior (MAP) filter requires assumption about the distribution of the
true process and the degradation model. Different assumptions lead to different MAP
estimators and different complexities.