System and Method For Feature Detection In Ultrasound Images

Abstract

A system and method for feature detection in ultrasound images is disclosed. The method estimates speckle distributions in windows on opposing sides of a pixel of an ultrasound image. The divergence is calculated for the pixel between the estimated speckle distributions in the windows. These steps are performed for each pixel in the ultrasound image, and a feature map is generated based on the divergence calculated between the estimated speckle distributions for each pixel.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1 is a block diagram illustrating ultrasound image formation;

FIG. 2 is a flowchart illustrating a method of feature detection in an ultrasound image according to an embodiment of the present invention;

FIG. 3( a) illustrates an exemplary cardiac ultrasound image;

FIG. 3( b) illustrates a feature map of the cardiac ultrasound image of FIG. 3(a) according to an embodiment of the present invention;

FIG. 3( c) illustrates a feature map of the cardiac ultrasound image of FIG. 3(a) generated using a conventional method;

FIG. 4( a)-4(d) illustrate an effect of window size on exemplary results of an embodiment of the present invention;

FIG. 5( a)-5(f) illustrate exemplary results of embodiments of the present invention compared with results of conventional methods; and

FIG. 6 a high level block diagram of a computer capable of implementing the present invention.

DETAILED DESCRIPTION OF EXEMPLARY EMBODIMENTS

The present invention is directed to a method for feature detection in ultrasound images using an information-theoretic approach based on speckle intensity distribution models derived from physical models of ultrasound image formation. FIG. 1 is a block diagram illustrating ultrasound image formation. An ultrasound transducer 102 transmits ultrasound waves into the body of a patient. The ultrasound waves are reflected by portions of the patient's body such as internal organs. The ultrasound transducer 102 measures the reflection of the ultrasound waves and generates RF data based on the reflection of the ultrasound waves. The RF data is then demodulated using a demodulation operator 104 to generate an in-phase/quadrature (IQ) image Q_I(x,y). The speckle in an IQ image Q_I(x,y) can be modeled as having a complex Gaussian distribution,

$\begin{array}{cc}p\left(Q_j\left(x,y\right)\right)=\frac{1}{2{\mathrm{\pi \sigma }}^2}{}^{-{Q_1\left(x,y\right)}^2/\left(2{\sigma }^2\right)},& \left(1\right)\end{array}$

where Q_I (x,y) is complex.

In order to produce a real image, envelope detection is performed by calculating the magnitude of the IQ image Q_I(x,y) using an absolute value operator 106, thus generating a magnitude IQ image M(x,y). The speckle in the magnitude IQ image M(x,y) can be modeled as having a Rayleigh distribution,

$\begin{array}{cc}p\left(M\left(x,y\right)\right)=\frac{M\left(x,y\right)}{{\sigma }^2}{}^{-{M\left(x,y\right)}^2/2{\sigma }^2},& \left(2\right)\end{array}$

where M(x,y) is real. The Rayleigh Distribution is described in detail in J. Goodman, Speckle Phenomena in Optics: Theory and Applications Version 5.0, 2005, which is incorporated herein by reference.

Since the magnitude IQ image M(x,y) has a large dynamic range, the magnitude IQ image M(x,y) can be logarithmically transformed using a log operator 108 to generate an image I(x,y) (Log Mag IQ image) suitable for display. The speckle in the image I(x,y) can be modeled as having a Fisher-Tippett (FT) distribution,

$\begin{array}{cc}p\left(l\left(x,y\right)\right)=2{}^{\frac{1}{2{\sigma }^2}}{}^{\left(2l\left(x,y\right)-\ln \left(2{\sigma }^2\right)-{}^{2/\left(x,y\right)\ln \left(2{\sigma }^2\right)}\right)}.& \left(3\right)\end{array}$

The FT distribution is described in detail in O. Michailovich and A. Tannenbaum, “Despeckling of Medical Ultrasound Images,” IEEE Transactions on Ultrasonics, Ferroelectrics, and Frequency Control, vol. 53, no. 1, January 2006, which is incorporated herein by reference.

Given a region Ω in a Mag IQ image M(x,y), it is possible to estimate the Rayleigh distribution in the region Ω based on the pixel intensities of pixels in the region Ω. Since the speckle in M(x,y) is modeled as having a Rayleigh distribution, the estimation of the Rayleigh distribution in the region Ω is an estimation of the speckle in the region Ω. The maximum likelihood of the Rayleigh distribution in the region Ω can be estimated using the log likelihood of Equation (2). The log likelihood of Equation (2) can be expressed as l(M(x,y),σ)=∫_Ωln p(M(x,y))dxdy, such that:

$\begin{array}{cc}\left(M\left(x,y\right),\sigma \right)={\int }_{\Omega }^{}\left(\ln \left(M\left(x,y\right)\right)-\ln {\sigma }^2-\frac{{M\left(x,y\right)}^2}{2{\sigma }^2}\right)xy.& \left(4\right)\end{array}$

In order to determine a maximum likelihood (ML) estimator for the Rayleigh distribution, l(M(x,y),σ) can be differentiated with respect to σ, and set equal to zero to determine the ML estimate of σ², such that:

$\begin{array}{cc}{\sigma }^2=\frac{{\int }_{\Omega }^{}{M\left(x,y\right)}^2xy}{2{\int }_{\Omega }^{}xy}.& \left(5\right)\end{array}$

Thus, given a region Ω in M(x,y), the ML estimate of the parameter σ² can be calculated from the pixel intensities in the region Ω assuming a Rayleigh distribution, and used to estimate the speckle in the region Ω.

Given a region Ω in a Log Mag IQ image I(x,y), it is possible to estimate the FT distribution in the region Ω based on the pixel intensities of pixels in the region Ω. Since the speckle in I(x,y) is modeled as having an FT distribution, the estimation of the FT distribution in the region Ω is an estimation of the speckle in the region Ω. An ML estimate for the FT distribution is derived similarly to the Rayleigh distribution described above. Accordingly, the parameter σ², which is the ML estimator for the FT distribution, can be determined based on the log likelihood of Equation (3), such that:

$\begin{array}{cc}{\sigma }^2=\frac{{\int }_{\Omega }^{}\left({}^{2l\left(x,y\right)}-1\right)xy}{2{\int }_{\Omega }^{}xy}.& \left(6\right)\end{array}$

Thus, given a region Ω in I(x,y), the ML estimate of the parameter σ² can be calculated from the pixel intensities in the region Ω assuming a FT distribution, and used to estimate the speckle in the region Ω.

Given two distributions, p and q, estimated from two different regions of an image, it is possible to measure the divergence or “distance” between the distributions p and q. For example, Kullback-Liebler (KL) divergence, or relative entropy, is an information-theoretic measure between two distributions p and q. KL divergence is described in detail in T. Cover and J. Thomas, Elements of Information Theory, New York: John-Wiley and Sons, 1991, which is incorporated herein by reference. The relative entropy D(p∥q) measures the inefficiency of assuming that a distribution is q when the true distribution is p. The KL divergence between two distributions p and q is defined as:

$\begin{array}{cc}D(pq)=\int p\left(x\right)\ln \frac{p\left(x\right)}{q\left(x\right)}x.& \left(7\right)\end{array}$

In the KL divergence defined in Equation (7), we assume that

$0\ln \frac{0}{0}=0$

and

$p\ln \frac{p}{0}=0.$

The KL divergence between two distributions p and q is asymmetric, that is, D(p∥q)≠D(q∥p). However, the KL divergence can be symmetrized using the J-divergence,

$J=\frac{D(pq)+D(qp)}{2}.$

The J-divergence can be thought of as a measure of “distance” between the probability distributions p and q.

For an image M(x,y) (i.e., a Mag IQ image) that can be modeled locally with Rayleigh distributions, a distribution p_M can be estimated in one window of pixels, and another distribution q_M can be estimated in another window of pixels, such that:

$\begin{array}{cc}{p}_M\left(M\right)=\frac{M\left(x,y\right)}{{\sigma }_1^2}{}^{-{M\left(x,y\right)}^2/\left(2{\sigma }_1^2\right)}& \left(8\right)\\ q_M\left(M\right)=\frac{M\left(x,y\right)}{{\sigma }_2^2}{}^{-{M\left(x,y\right)}^2/\left(2{\sigma }_2^2\right)},& \left(9\right)\end{array}$

where M(x,y) is the intensity at pixel (x,y) in the magnitude IQ image and σ₁² and σ₂² are the parameters of the distributions p_M and q_M, respectively.

The J-divergence can be computed between the distributions p_M and q_M as a measure of how “different” the regions are. The derivation of the expression for D(p_M∥q_M) is shown below, and D(q_M∥p_M) can be similarly derived to obtain the J-divergence. In the derivation below M is used to represent M(x,y) for convenience. Based on Equations (7)-(9), D(p_M∥q_M) can be expressed as:

$\begin{array}{cc}D(p_Mq_M)={\int }_0^{\infty }\frac{M}{{\sigma }_1^2}{}^{-M^2/\left(2{\sigma }_1^2\right)}\ln \left(\frac{\left(\frac{M}{{\sigma }_1^2}{}^{-M^2/\left(2{\sigma }_1^2\right)}\right)}{\left(\frac{M}{{\sigma }_2^2}{}^{-M^2/\left(2{\sigma }_2^2\right)}\right)}\right)M.& \left(10\right)\end{array}$

By expanding the ln term in Equation (10), performing the integration, and simplifying, Equation (10) can be re-written as:

$\begin{array}{cc}D(p_Mq_M)=\ln \left(\frac{{\sigma }_2^2}{{\sigma }_1^2}\right)-1+\frac{{\sigma }_1^2}{{\sigma }_2^2}.& \left(11\right)\end{array}$

Therefore, the J-divergence can be expressed as:

$\begin{array}{cc}{J}_M=-1+\frac{{\sigma }_1^2}{2{\sigma }_2^2}+\frac{{\sigma }_2^2}{2{\sigma }_1^2},& \left(12\right)\end{array}$

where σ₁² and σ₂² are estimated parameters of the Rayleigh distributions determined from Equation (5).

For an image I(x,y) (i.e., a Log Mag IQ image) that can be modeled locally with FT distributions, distributions p_FT and q_FT can be estimated in different windows of I(x,y) as FT distributions having parameters σ₁²and σ₂² respectively. The J-divergence can be computed between the distributions p_FT and q_FT as a measure of how “different” the windows are. The derivation of the expression for D(p_FT∥q_FT) is shown below, and D(q_{FT ∥qFT}) can be similarly derived to obtain the J-divergence. In the derivation below, I is used to represent I(x,y) for convenience. Based on Equations(3) and (7), D(p_FT∥q_FT) can be expressed as:

$\begin{array}{cc}D(p_{\mathrm{FT}}q_{\mathrm{FT}})={\int }_0^{\infty }2{}^{\frac{1}{2{\sigma }_1^2}}{}^{\left(2l-\ln \left(2{\sigma }_1^2\right)-{}^{2l-\ln \left(2{\sigma }_1^2\right)}\right)}\ln \left(\frac{\left(2{}^{\frac{1}{2{\sigma }_1^2}}{}^{\left(2l-\ln \left(2{\sigma }_1^2\right)-{}^{2l-\ln \left(2{\sigma }_1^2\right)}\right)}\right)}{\left(2{}^{\frac{1}{2{\sigma }_2^2}}{}^{\left(2l-\ln \left(2{\sigma }_2^2\right)-{}^{2l-\ln \left(2{\sigma }_2^2\right)}\right)}\right)}\right)l& \left(13\right)\end{array}$

By expanding the ln term in Equation (13), performing the integration, and simplifying, Equation (13) can be re-written as:

$\begin{array}{cc}D(p_{\mathrm{FT}}q_{\mathrm{FT}})=\ln \left(\frac{{\sigma }_2^2}{{\sigma }_1^2}\right)-1+\frac{{\sigma }_1^2}{{\sigma }_2^2}.& \left(14\right)\end{array}$

Therefore, the J-divergence can be expressed as:

$\begin{array}{cc}{J}_{\mathrm{FT}}=-1+\frac{{\sigma }_1^2}{2{\sigma }_2^2}+\frac{{\sigma }_2^2}{2{\sigma }_1^2},& \left(15\right)\end{array}$

where σ₁² and σ₂² are estimated parameters of the FT distributions determined from Equation (6).

As described above, speckle in a region of an image can be estimated using a Rayleigh distribution or a FT distribution, and the divergence between distributions estimated in two regions can be calculated using the J-divergence. According to embodiments of the present invention, the above-described operations can be used for feature detection in ultrasound images. FIG. 2 is a flowchart illustrating a method of feature detection in an ultrasound image according to an embodiment of the present invention. According to various embodiments of the present invention, the ultrasound image may be a magnitude IQ image or a Log Mag IQ image.

At step 202, for a pixel in the ultrasound image, adjacent windows to the pixel are defined. An adjacent window to a particular pixel is a region of the ultrasound image including at least one adjacent pixel to the particular pixel. According to a preferred embodiment of the present invention, adjacent windows to the pixel can be defined on opposing sides of the pixel in first and second directions (i.e., x and y directions). The first and second directions can be orthogonal with respect to each other, such that the pixel is surrounded by adjacent windows. The windows can have any shape or size, but it is preferable that windows on opposing sides of the pixel have the same shape and size. FIG. 3(a) illustrates an exemplary cardiac ultrasound image 302. As illustrated in FIG. 3(a), the windows w₁ and w₂ can be defined on opposing sides of a pixel in the horizontal (x) direction and the vertical (y) direction. FIG. 3(a) illustrates windows w₁ and w₂ on opposing sides of pixel 301 in the horizontal (x) direction and windows w₃ and w₄ on opposing sides of pixel 303 in the vertical (y) direction. According to a preferred embodiment of the present invention, a first set of windows w₁ and w₂ are defined on opposing sides of a pixel in the x direction and another set of windows w₃ and w₄ are defined in the y direction for the same pixel. It is also possible that the windows be defined in any other direction relative to the pixel.

Returning to FIG. 2, at step 204, the distribution of the speckle is estimated in each of the windows. For example, the speckle distribution in each window can be estimated as a Rayleigh distribution or an FT distribution, as described above. According to an embodiment of the present invention in which the ultrasound image is a magnitude IQ image M(x,y), the speckle distribution is estimated in each window as a Rayleigh distribution by calculating the ML estimation of σ² for each window using Equation (5). According to an embodiment of the present invention in which the ultrasound image is a Log Mag IQ image I(x,y), the speckle distribution is estimated in each window as an FT distribution by calculating the ML estimation of σ² for each window using Equation (6).

At step 206, the divergence is calculated between the speckle distributions estimated in windows on opposing sides of the pixel. The calculated divergence is a measure of how “different” the opposing windows are. For example, the J-divergence can be calculated between the speckle distributions estimated in windows on opposing sides of the pixel using Equation (12) or (15) for Rayleigh distributions or FT distributions, respectively. The J-divergence is calculated for each set of opposing windows. For example, when windows are defined on opposing sides of the pixel in the x and y directions, a J-divergence J_x(x,y) is calculated between the speckle distributions estimated in the windows on opposing sides of the pixel in the x direction, and a J-divergence J_y(x,y) is calculated between the speckle distributions estimated in the windows on opposing sides of the pixel in the y direction.

At step 208, it is determined whether all of a plurality of pixels in the ultrasound image have been processed. If any of the plurality of pixels have not been processed, the method proceeds to a next pixel at step 210, and loops back to step 202 for the next pixel. If it is determined that all of the plurality of pixels have been processed, the method proceeds to step 212.

At step 212, a feature map of the ultrasound images in generated. The feature map is generated by assigning each pixel in the ultrasound image an intensity value based on the divergence calculated for each pixel. For example, if J_x(x,y) represents the J-divergence calculated for a pixel (x,y) in the x direction, and J_y(x,y) represents the J-divergence calculated for a pixel (x,y) in the y direction, the feature map F_J(x,y) can be defined as:

F _J(x,y)=√{square root over (J_x(x,y)²+J_y(x,y)²)}{square root over (J_x(x,y)²+J_y(x,y)²)}.   (16)

Accordingly, pixels in the ultrasound image at which a large divergence is detected between speckle distributions estimated surrounding the pixels will have higher intensity values in the feature map F_J(x,y). Thus, the feature map F_J(x,y) detects edges and features in the ultrasound image. The feature map F_J(x,y) can be output to a display device in order to be displayed thereon.

FIG. 3( b) illustrates a feature map 304 of the cardiac ultrasound image 302 of FIG. 3(a) according to an embodiment of the present invention. The feature map 304 was generated using FT distributions to estimate the speckle distributions according to an embodiment of the present invention. As illustrated in FIG. 3(b), the feature map 304 detects the most salient features of the cardiac ultrasound image 302. The feature map 304 of FIG. 3(b) can be compared with a conventional feature map 306 of FIG. 3(c), which was generated using a conventional gradient magnitude operator, as described above in the Background of the Invention section. As illustrated in FIG. 3(c), the conventional feature map 306 has significantly more speckle 308 than the feature map 304 generated according to an embodiment of the present invention.

Sensitivity of the feature detection method described above can be adjusted by varying the window size of the windows. For example, increasing the window size may give a better statistical modeling of a distribution's parameter in the window, and adjusts the scale of the features detected. FIGS. 4(a)-4(d) illustrate feature maps of the cardiac ultrasound image 302 of FIG. 3(a) generated according to an embodiment of the present invention using different window sizes. As illustrated in FIGS. 4(a)-4(d), feature maps 402, 404, 406, and 408 were generated using window sizes of 3×3, 5×5, 7×7, and 9×9, respectively. It can be observed from FIGS. 4(a)-4(d) that the size of features detected by the feature maps 402, 404, 406, and 408 is proportional to the window size.

FIG. 5( a)-5(f) illustrate exemplary results of embodiments of the present invention compared with results of conventional methods. FIG. 5(a) illustrates a lesion phantom image 502. FIG. 5(b) illustrates a J-divergence feature map 504 applied to the magnitude IQ image of image 502 using the Rayleigh distribution according to an embodiment of the present invention. The feature map 504 of FIG. 5(b) was generated using a window size of 7×7. FIG. 5(c) illustrates a J-divergence feature map 506 applied to the Log Mag IQ image of image 502 using the FT distribution according to an embodiment of the present invention. The feature map 506 of FIG. 5(c) was generated using a window size of 7×7. FIG. 5(d) illustrates a conventional feature map 508 generated using the conventional derivative of Gaussian feature detector, which was performed using σ=5. As illustrated in FIG. 5(d), even with a large standard deviation, there is still a large amount of speckle prevalent. FIG. 5(e) illustrates a conventional feature map 510 generated using a Canny edge detector, which was optimized to detect the salient edges while minimizing false detection of speckle edges. FIG. 5(f) illustrates a feature detector 512 generated using the J-divergence non-parametrically, i.e., calculating

$J=\frac{D(pq)+D(qp)}{2}\mathrm{for}$

Parzen-windowed histograms of the windows without estimating the speckle distributions in the windows. As illustrated in FIG. 5(f), the feature map 512 of FIG. 5(f) results in false detections due to speckle. Compared to other methods, the feature maps 504 and 506 generated using embodiments of the present invention have strong feature response while mitigating false responses due to speckle.

The above-described method for feature detection in ultrasound images can be implemented on a computer using well-known computer processors, memory units, storage devices, computer software, and other components. A high level block diagram of such a computer is illustrated in FIG. 6. Computer 602 contains a processor 604 which controls the overall operation of the computer 602 by executing computer program instructions which define such operation. The computer program instructions may be stored in a storage device 612 (e.g., magnetic disk) and loaded into memory 610 when execution of the computer program instructions is desired. Thus, applications for estimating the speckle distributions in regions of the image and calculating divergence between estimated speckle distributions can be defined by the computer program instructions stored in the memory 610 and/or storage 612 and controlled by the processor 604 executing the computer program instructions. The computer 602 also includes one or more network interfaces 606 for communicating with other devices via a network. The computer 602 also includes other input/output devices 608 that enable user interaction with the computer 602 (e.g., display, keyboard, mouse, speakers, buttons, etc.) For example, a display can display the feature map generated using the above described method. One skilled in the art will recognize that an implementation of an actual computer could contain other components as well, and that FIG. 6 is a high level representation of some of the components of such a computer for illustrative purposes.

Embodiments of the present invention are described herein to give a visual understanding of the feature detection method. It is to be understood that these embodiments may be performed within a computer system using data stored within the computer system. Accordingly, some steps of the method can occur as internal representations within the computer system.

The foregoing Detailed Description is to be understood as being in every respect illustrative and exemplary, but not restrictive, and the scope of the invention disclosed herein is not to be determined from the Detailed Description, but rather from the claims as interpreted according to the full breadth permitted by the patent laws. It is to be understood that the embodiments shown and described herein are only illustrative of the principles of the present invention and that various modifications may be implemented by those skilled in the art without departing from the scope and spirit of the invention. Those skilled in the art could implement various other feature combinations without departing from the scope and spirit of the invention

Claims

1. A method for feature detection in an ultrasound image, comprising: (a) estimating speckle distribution in each of first and second windows adjacent to a pixel of said ultrasound image; and(b) calculating for said pixel a first divergence value between the estimated speckle distribution in the first window and the estimated speckle distribution in the second window.

2. The method of claim 1, wherein said first and second windows are on opposing sides of said pixel in a first direction

3. The method of claim 2, further comprising: (c) estimating speckle distribution in each of third and fourth windows adjacent to said pixel on opposing sides of said pixel in a second direction; and(d) calculating a second divergence value for said pixel between the estimated speckle distribution in the third window and the estimated speckle distribution in the fourth window.

4. The method of claim 3, further comprising: (e) repeating steps (a)-(d) for each of a plurality of pixels in said ultrasound image.

5. The method of claim 4, further comprising: (f) generating a feature map of each of said plurality of pixels, wherein an intensity value for each pixel in the feature map is based on the first and second divergence values calculated for the pixel.

6. The method of claim 5, further comprising: (g) outputting said feature map to a display device to be displayed thereon.

7. The method of claim 3, wherein said first and second directions are orthogonal with respect to each other.

8. The method of claim 1, wherein said ultrasound image is a magnitude in-phase/quadrature image, and step (a) comprises: estimating a Rayleigh distribution in each of the first and second windows based on intensity values of pixels within each of the first and second windows, respectively.

9. The method of claim 1, wherein said ultrasound image is a logarithmic transform of a magnitude in-phase/quadrature image, and step (a) comprises: estimating a Fisher-Tippett distribution in each of the first and second windows based on intensity values of pixels within each of the first and second windows.

10. The method of claim 1, wherein step (b) comprises: calculating J-divergence between the estimated speckle distribution in the first window and the estimated speckle distribution in the second window based on Kullback-Liebler divergences calculated between the estimated speckle distributions in the first and second windows.

11. The method of claim 1, further comprising: (e) assigning an intensity value to said pixel equal based on the first divergence value for said pixel; and(f) repeating steps (a)-(e) for each of a plurality of pixels of said ultrasound image.

12. A computer readable medium encoded with computer executable instructions for performing a method for feature detection in an ultrasound image, the computer executable instructions defining steps comprising: (a) estimating speckle distribution in each of first and second windows adjacent to a pixel of said ultrasound image; and(b) calculating for said pixel a first divergence value between the estimated speckle distribution in the first window and the estimated speckle distribution in the second window.

13. The computer readable medium of claim 12, wherein said first and second windows are on opposing sides of said pixel in a first direction.

14. The computer readable medium of claim 13 wherein the computer executable instructions further define the steps of: (c) estimating speckle distribution in each of third and fourth windows adjacent to said pixel on opposing sides of said pixel in a second direction; and(d) calculating a second divergence value for said pixel between the estimated speckle distribution in the third window and the estimated speckle distribution in the fourth window.

15. The computer readable medium of claim 14, wherein the computer executable instructions further define the step of: (e) repeating steps (a)-(d) for each of a plurality of pixels in said ultrasound image.

16. The computer readable medium of claim 15, wherein the computer executable instructions further define the step of: (f) generating a feature map of each of said plurality of pixels, wherein an intensity value for each pixel in the feature map is based on the first and second divergence values calculated for the pixel.

17. The computer readable medium of claim 12, wherein said ultrasound image is a magnitude in-phase/quadrature image, and the computer executable instructions defining step (a) comprise computer executable instructions defining the step of: estimating a Rayleigh distribution in each of the first and second windows based on intensity values of pixels within each of the first and second windows, respectively.

18. The computer readable medium of claim 12, wherein said ultrasound image is a logarithmic transform of a magnitude in-phase/quadrature image, and the computer executable instructions defining step (a) comprise computer executable instructions defining the step of: estimating a Fisher-Tippeft distribution in each of the first and second windows based on intensity values of pixels within each of the first and second windows.

19. The computer readable medium of claim 12, wherein the computer executable instructions defining step (b) comprise computer executable instructions defining the step of: calculating J-divergence between the estimated speckle distribution in the first window and the estimated speckle distribution in the second window based on Kullback-Liebler divergences calculated between the estimated speckle distributions in the first and second windows.

20. The computer readable medium of claim 12, wherein the computer executable instructions further define the steps of: (e) assigning an intensity value to said pixel based on the first divergence value for said pixel; and(f) repeating steps (a)-(e) for each of a plurality of pixels of said ultrasound image.

21. An apparatus for detecting features in an ultrasound image, comprising: means for estimating speckle distribution in each of first and second windows adjacent to a pixel of said ultrasound image; andmeans for calculating for said pixel a divergence value between the estimated speckle distribution in the first window and the estimated speckle distribution in the second window.

22. The apparatus of claim 21, further comprising: means for generating a feature map of said ultrasound image, by assigning an intensity value for said pixel in the feature map based on the divergence value calculated for said pixel.

23. The apparatus of claim 22, further comprising: means for displaying said feature map.

24. The apparatus of claim 21, wherein said ultrasound image is a magnitude in-phase/quadrature image, and said means for estimating speckle distribution comprises: means for estimating a Rayleigh distribution in each of the first and second windows based on intensity values of pixels within each of the first and second windows, respectively.

25. The apparatus of claim 21, wherein said ultrasound image is a logarithmic transform of a magnitude in-phase/quadrature image, and said means for estimating speckle distribution comprises: means for estimating a Fisher-Tippett distribution in each of the first and second windows based on intensity values of pixels within each of the first and second windows.

26. The apparatus of claim 21 wherein said means for calculating a divergence value comprises: means for calculating J-divergence between the estimated speckle distribution in the first window and the estimated speckle distribution in the second window based on Kuliback-Liebler divergences between the estimated speckle distributions in the first and second windows.

27. The apparatus of claim 21, further comprising: means for selecting a size of said first and second windows.