System and method for ultrasound specific segmentation using speckle distributions
View Patent ↗A method for segmenting a digitized ultrasound image includes providing a digitized in-phase/quadrature ultrasound image comprising a plurality of intensities defined on an N-dimensional grid, decorrelating the ultrasound image wherein spatial correlations are substantially reduced in the intensity data, modeling the decorrelated image intensities with a statistical distribution, propagating, an active contour in the image where the contour segments the image, where the contour is propagated based on the statistical distributions of intensity data inside and outside the active contour.
1. A method of segmenting a digitized ultrasound image comprising the steps of:
providing a digitized in-phase/quadrature (IQ) ultrasound image comprising a plurality of intensity data values defined on a N-dimensional grid of points;
decorrelating said ultrasound image wherein spatial correlations are removed in said intensity data;
modeling said decorrelated image intensities with a statistical distribution;
propagating an active contour in said image by initializing a contour to a closed hypercurve that divides the image into two regions, determining for each region a statistical distribution that maximizes a likelihood function of said distribution in each said region, and propagating said contour in a normal direction using an Euler-Lagrange equation derived from said likelihood function, wherein said contour is a level set of a scalar function whose values are negative in one region and positive in the other region, wherein said contour segments said image, wherein said statistical distribution is a Fisher-Tippet distribution, and wherein said likelihood function includes the terms
∫ Ω (ln2+2 l ( {right arrow over (x)} )−ln σ i 2 −exp(2 l ( {right arrow over (x)} )−ln σ i 2 )) {right arrow over (d)}x
∫ Ω o (ln2+2 l ( {right arrow over (x)} )−ln σ o 2 −exp(2 l ( {right arrow over (x)} )−ln σ o 2 )){right arrow over (d)}x
wherein l({right arrow over (x)}) is the image intensity for point ({right arrow over (x)}), i and o denote regions inside and outside the contour, respectively, and σ i and σ o are the Fisher-Tippett parameters inside regions Ω i and Ω o .
2. The method of claim 1 , further comprising determining a magnitude of said in-phase/quadrature image and log-compressing said magnitude image.
3. The method of claim 1 , wherein decorrelation is performed using a whitening filter.
4. The method of claim 3 , wherein decorrelation comprises removing outliers from a Fourier transformed IQ image to obtain a reduced IQ image, filtering said reduced IQ image with a wavelet transform to obtain a spectrum of a point spread function, and applying a Wiener deconvolution filter to said IQ image and said point spread function spectrum to obtain a de-correlated image.
5. The method of claim 4 , wherein said outliers are defined by computing residuals R({right arrow over (x)}) of said IQ image G({right arrow over (x)}),
R
(
x
→
)
=
{
sign
(
Δ
G
(
x
→
)
)
·
(
Δ
G
(
x
→
)
-
λ
)
,
0
,
{
if
Δ
G
(
x
→
)
≥
λ
if
Δ
G
(
x
→
)
≤
λ
,
wherein ({right arrow over (x)}) represent a point, ΔG is a difference between G and a median-filtered version of G, and λ is a predefined threshold wherein λ is dynamically adapted to a level such that a pre-defined percentage of the differences ΔG is preserved, and wherein said outliers are removed by subtracting said residuals from said IQ image.
6. The method of claim 4 , wherein said Wiener deconvolution filter includes a term proportional to
DFT
-
1
{
DFT
{
g
(
x
→
)
}
S
2
+
ɛ
}
,
wherein DFT represents a discrete Fourier transform, S represents said spectrum of said point spread function, g({right arrow over (x)}) represents said IQ image, and 0<ε<<1.
7. The method of claim 1 wherein said likelihood function includes a term to regularize said contour.
8. A method of segmenting a digitized ultrasound image comprising the steps of:
providing a digitized in-phase/quadrature ultrasound image comprising a plurality of intensity data values defined on a N-dimensional grid of points;
determining a magnitude of said in-phase/quadrature image and log-compressing said magnitude image;
initializing a contour to a closed hypercurve that divides the image into two regions, determining for each region a Fisher-Tippett statistical distribution that maximizes a likelihood function of said distribution in each said region,
propagating said contour in a normal direction using an Euler-Lagrange equation derived from said likelihood function, wherein said contour is a level set of a scalar function whose values are negative in one region and positive in the other regions, wherein said contour segments said image, and
wherein said likelihood function includes the terms
∫ Ω (ln2+2 l ( {right arrow over (x)} ) −ln σ i 2 −exp(2 l ( {right arrow over (x)} )−ln σ i 2 )) {right arrow over (d)}x
∫ Ω o (ln2+2 l ( {right arrow over (x)} )−ln σ o 2 −exp(2 l ( {right arrow over (x)} ) −ln σ o 2 )) {right arrow over (d)}x
wherein l({right arrow over (x)}) is the image intensity for point ({right arrow over (x)}), i and o denote regions isnide and outside the contour, respectively,. and σ i and σ o are the Fisher-Tippett parameters inside regions Ω i and Ω o .
9. The method of claim 8 , further comprising decorrelating said in-phase/quadrature image by applying a whitening filter on a Fourier transform of said image and deconvolving said filtered image, wherein spatial correlations are removed in said data.
10. A program storage device readable by a computer, tangibly embodying a program of instructions executable by the computer to perform the method steps for segmenting digitized ultrasound images, said method comprising the steps of:
providing a digitized in-phase/quadrature (IQ) ultrasound image comprising a plurality of intensity data values defined on a N-dimensional grid of points;
decorrelating said ultrasound image wherein spatial correlations are removed in said intensity data;
modeling said decorrelated image intensities with a statistical distribution;
propagating an active contour in said image by initializing a contour to a closed hypercurve that divides the image into two regions, determining for each region a statistical distribution that maximizes a likelihood function of said distribution in each region, and propagating said contour in a normal direction using an Euler-Lagrange equation derived from said likelihood function, wherein said contour is a level set of a scalar function whose values are negative in one region and positive in the other regions, wherein said contour segments said image, wherein said statistical distribution is a Fisher-Tippett distribution, and wherein said likelihood function includes the terms
∫ Ω (ln2+2 l ({right arrow over ( x )})−ln σ i 2 −exp(2 l ({right arrow over ( x )})−ln σ i 2 )) {right arrow over (d)}x
∫ Ω o (ln2+2 l ({right arrow over ( x )})−ln σ o 2 −exp(2 l ({right arrow over ( x )})−ln σ o 2 )) {right arrow over (d)}x
wherein l({right arrow over (x)}) is image intensity for point ({right arrow over (x)}), i and o denote regions inside and outside the contour, respectively, and σ i and σ o are Fisher-Tippet paramaters inside regions Ω i and Ω o .
11. The computer readable program storage device of claim 10 , the method further comprising determining a magnitude of said in-phase/quadrature image and log-compressing said magnitude image.
12. The computer readable program storage device of claim 10 , wherein decorrelation is performed using a whitening filter.
13. The computer readable program storage device of claim 12 , wherein decorrelation comprises removing outliers from a Fourier transformed IQ image to obtain a reduced IQ image, filtering said reduced IQ image with a wavelet transform to obtain a spectrum of a point spread function, and applying a Wiener deconvolution filter to said IQ image and said point spread function spectrum to obtain a de-correlated image.
14. The computer readable program storage device of claim 13 , wherein said outliers are defined by computing residuals R({right arrow over (x)}) of said IQ image G({right arrow over (x)}),
R
(
x
→
)
=
{
sign
(
Δ
G
(
x
→
)
)
·
(
Δ
G
(
x
→
)
-
λ
)
,
0
,
{
if
Δ
G
(
x
→
)
≥
λ
if
Δ
G
(
x
→
)
≤
λ
,
wherein ({right arrow over (x)}) represent a point, ΔG is a difference between G and a median-filtered version of G, and λ is a predefined threshold wherein λ is dynamically adapted to a level such that a pre-defined percentage of the differences ΔG is preserved, and wherein said outliers are removed by subtracting said residuals from said IQ image.
15. The computer readable program storage device of claim 13 , wherein said Wiener deconvolution filter includes a term proportional to
DFT
-
1
{
DFT
{
g
(
x
→
)
}
S
2
+
ɛ
}
,
wherein DFT represents a discrete Fourier transform, S represents said spectrum of said point spread function, g({right arrow over (x)}) represents said IQ image, and 0<ε<<1.
16. The computer readable program storage device of claim 10 , wherein said likelihood function includes a term to regularize said contour.