WO2004013811A2 - Image segmentation using jensen-shannon divergence and jensen-renyi divergence - Google Patents

Image segmentation using jensen-shannon divergence and jensen-renyi divergence Download PDF

Info

Publication number
WO2004013811A2
WO2004013811A2 PCT/US2003/023685 US0323685W WO2004013811A2 WO 2004013811 A2 WO2004013811 A2 WO 2004013811A2 US 0323685 W US0323685 W US 0323685W WO 2004013811 A2 WO2004013811 A2 WO 2004013811A2
Authority
WO
WIPO (PCT)
Prior art keywords
value
contour
image
data
divergence
Prior art date
Application number
PCT/US2003/023685
Other languages
French (fr)
Other versions
WO2004013811A3 (en
Inventor
Lyndon S. Hibbard
Original Assignee
Computerized Medical Systems, Inc.
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Computerized Medical Systems, Inc. filed Critical Computerized Medical Systems, Inc.
Priority to AU2003256972A priority Critical patent/AU2003256972A1/en
Publication of WO2004013811A2 publication Critical patent/WO2004013811A2/en
Publication of WO2004013811A3 publication Critical patent/WO2004013811A3/en

Links

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00Image analysis
    • G06T7/0002Inspection of images, e.g. flaw detection
    • G06T7/0012Biomedical image inspection
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00Image analysis
    • G06T7/10Segmentation; Edge detection
    • G06T7/12Edge-based segmentation
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00Indexing scheme for image analysis or image enhancement
    • G06T2207/10Image acquisition modality
    • G06T2207/10072Tomographic images
    • G06T2207/10081Computed x-ray tomography [CT]
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00Indexing scheme for image analysis or image enhancement
    • G06T2207/30Subject of image; Context of image processing
    • G06T2207/30004Biomedical image processing

Definitions

  • the present invention relates to image segmentation. More specifically, the present invention addresses an improved technique for identifying the boundaries of distinct, discrete objects depicted in digital images, particularly medical images.
  • Image segmentation (also known as “autosegmentation”, “contouring”, and “autocontouring”) is a technique of processing a digital image to detect, classify, and enumerate discrete objects depicted therein.
  • Image segmentation presents a difficult problem because digital images generally lack sufficient information to constrain the possible solutions of the segmentation problem to a small set of solutions that includes the correct solution.
  • Image segmentation finds a popular application in the field of medical images, particularly computed tomography (CT) images, x-ray images, magnetic resonance (MR) images, and the like. It is highly desirable to accurately contour various anatomic objects (such as the prostate, kidneys, the liver, the pancreas, etc.) that are shown in such medical images .
  • CT computed tomography
  • MR magnetic resonance
  • the location of the anatomic object relative to its surroundings can be used to plan and execute medical procedures such as radiotherapy treatment for cancer.
  • medical procedures such as radiotherapy treatment for cancer.
  • the healthy tissues and organs surrounding cancerous tissue are highly sensitive to radiation, it is extremely important that the exact location of the anatomic object to be treated be known. That is, accurate placement, shaping, and intensity modulation of external X-ray beams for tumor irradiation (teletherapy) depends absolutely on detailed 3-D maps of the patient's anatomy, as do other forms of radiotherapy treatment such as brachytherapy and radioimmunotherapy.
  • Treatment planning for medical procedures begins by creating a detailed 3-D map of the patient's anatomy (including tumors). These maps are often created as a series of contours drawn on CT slices .
  • a set of external X-ray beam angles, beam cross-section shapes, and beam intensities can be designed to deliver a prescribed dose of radiation to the tumor while minimizing the harm done to nearby healthy tissues.
  • the appropriate location (s) to deliver radioactive seeds can be determined, also while minimizing the harm done to nearby healthy tissues .
  • a digital image of a target such as the human body is a data set comprising an array of data elements, each data element having a numerical data value corresponding to a property of the target, wherein the property is measured by an image sensor at regular intervals throughout the image sensor's field of view.
  • the property to which the data values correspond may be the light intensity of black and white photography, the separate RGB components of a color image, or other similar properties.
  • the image data set is an array of pixels, wherein each pixel has a value corresponding to intensity.
  • the leftmost image shown in Figure 1 is a CT digital image of an interior portion of a patient's body.
  • each data element can be modeled as a random variable.
  • a random variable is a measured quantity whose repeated observations cluster about some central value.
  • the "mean” is the most likely value and the "mode” is the most-often observed value. Random variables are most completely described by a histogram, rescaled as a probability distribution.
  • the data values of the data elements of the image data set constitute a random variable X which may take its values from a finite set of possible data values, such as the interval -1024 to 3071, typical for CT.
  • the data value of any given data element of the image data set may be equal to any Xi, wherein xi is a member of the set ⁇ x lf x 2 , ... x n ⁇ , and wherein "n" is 4096 for the CT example above.
  • Random variables whose values are limited to members of finite sets are termed discrete; random variables which can take any real value are continuous random variables .
  • Pr ⁇ is the probability that the expression in the braces ⁇ is true.
  • the full set of probabilities, written p (X) is the probability distribution. As is well-known, probabilities and their distribution have the following properties : p(Xi) > 0; 0 ⁇ p(xi) ⁇ 1; and
  • ⁇ n i- ⁇ P ( x_ ) To deal with 2 or more random variables at once, the concepts of joint probability and conditional probability are used.
  • a histogram corresponding to such a joint distribution can be represented by a two-dimensional plot with the x entries along the rows and the y entries along the columns .
  • conditional probability of Y, given X is defined as:
  • the relative sizes of the peaks shown in the histogram correspond to the probabilities that a pixel selected randomly will belong to one of the regions in the image.
  • two regions are easily discernible; interior region 100, which corresponds to the anatomical portion of the image, and exterior region 102, which corresponds to the non-anatomical portions of the image (the black background and the CT scanner couch and gantry ring depicted at the bottom of the image) .
  • the conditional probability (or likelihood) that a pixel with value i is in a particular region k, is p (X I region k ) . If the individual intensity probabilities in a region k are summed:
  • conditional probability expressions can be used to make decisions as to which pixels belong to which regions . For example, if two regions, region ! and region 2 , are equally probable, then a pixel with value x A most likely belongs to the region corresponding to the largest conditional probability p (xiI region k ) , wherein k is 1 or 2. Referring back to Fig. 2, it can be seen that more pixels with high intensity are found in the interior region than in the exterior region. Similarly, more pixels with a low intensity are found in the exterior region that the interior region.
  • LRT Likelihood ratio test
  • T is a threshold value determined from training data in which the regions corresponding to each pixel value are known exactly.
  • kidney, spleen, ribs, spinal cord, and vertebral body exhibit varying degrees of differences in pixel intensities . While the kidney, spleen, and spinal cord share similar pixel intensities, the pixel intensity for the ribs and vertebral body are markedly different therefrom. Further, the pixels within each object have more or less constant pixel intensities (regional coherence) . Moreover, it can be seen that there are often large steps in intensity as one passes out of one object and into a neighbor (edge step differences) . For example, note the dark region between the kidney and spleen. In fact, regional coherence and edge-steps are the two dominant structural features in images.
  • Region-based approaches assume that region pixels have almost-constant properties, and that these properties are different from region to region.
  • Pixel properties also called features
  • Region-based segmentation assumes that the means and variances of these properties are constant within a region, and differ among regions. See Rosenfeld, A. and Kak, A.C., Digital Picture Processing, Academic Press, 1982; Beveridge, JR, Griffith, J, Kohler, RR, Hanson, AR, and Riseman, EM, "Segmenting images using localized histograms and region merging," International Journal of Computer Vision, 2, 311-347, 1989; and Adams, R.
  • region-based approaches falter if the pixel properties vary across the region, or if the boundaries between regions with the same properties are blurred (by patient breathing motion in CT, for example) , or if neighboring regions' properties are similar or have overlapping distributions.
  • the success of a region-based segmentation operation is highly dependent upon the proper choice of a pixel homogeneity criterion. If the pixel homogeneity criterion is poorly chosen, the image segmentation operation results in ragged boundaries and holes in the segmented region. Moreover, even with a good choice of a pixel homogeneity criterion, the overlap of pixel distributions from adjoining regions constitutes a big problem for this approach.
  • Edge-based methods assume that all regions are surrounded by a significant step in the pixel property. Using this assumption, edge-based image segmentation extracts a segment by semantic analysis of the edges (see Ballard, DH and Brown, CM, Computer Vision, Prentice-Hall, Englewood Cliffs, NJ, 1982, the entire disclosure of which is hereby incorporated by reference) or by fitting a flexible curve or active contour to the actual boundary in the image (see Kass, M. , Wi kin, A., Terzopoulos, D., "Snakes: Active contour models," International Journal of Computer Vision, 1, 321-331, 1988.
  • Image-feature clustering segmentation identifies objects by looking in the space of image pixel properties for clusters that correspond to the feature combinations of object pixels. Each combination corresponds to a point in feature space, where the space dimension is equal to the number of features. Sets of object pixels with similar properties correspond to sets of points clustered in feature space. Finding such point clusters enables the analyst to go back to the image and identify the pixels inhabiting this concentration, and by implication, the object.
  • the nearest neighbor rule specifies that each pixel be assigned to the region corresponding to the feature space cluster nearest that pixel's feature space location.
  • Global optimization segmentation performs non-linear transformations on the image pixel values to maximize an objective function which is assumed to accurately describe the properties of the image, or the properties of the measurement process that formed the image.
  • the two most famous ideas are iterated conditional modes (See Besag, J., "On the statistical analysis of dirty pictures," Journal of the Royal Statistical Society, 48, 259--279, 1986, the entire disclosure of which is hereby incorporated by reference) and Markov-Gibbs refinement (See Geman, S. and Geman, D., "Stochastic relaxation, Gibbs distribution and the Bayesian restoration of images”.
  • Hybrid segmentation methods operate on several image properties at once.
  • a good example of these methods are those developed by Duncan and Staib. See Staib LH and Duncan JS, "Boundary finding with parametrically deformable models," IEEE Transactions on Pattern Analysis and Machine Intelligence, 14, 1061-1075, 1992, the entire disclosure of which is hereby incorporated by reference. Briefly, Staib and Duncan proposed applying rigorously correct statistical inference to automated image segmentation by matching the properties of a flexible-shape template to image objects. Later, Chakraborty added to the
  • the above-described Duncan/Staib/Chakraborty (DSC) method iteratively computes the value of the decision rule and adjusts the flexible curve so as to maximize the value of the decision-rule function.
  • the flexible curve is a finite trigonometric series whose coefficients, or parameters, fully define the location, size, and shape of the curve.
  • the actual "footprint" of the curve on the image can be quickly computed, and compared to an independently- segmented region boundary, to object edges in a map of all the image object edges, and to the shape of a similar object determined previously. The extent of these agreements is input to a decision rule function.
  • the ⁇ 951 patent describes a region-based approach using Bayesian classification of pixels operating on pixel intensity and variance features.
  • the ⁇ 594 patent describes a unique combination of Bayesian region classification with the DSC approach to create a useable tool .
  • the initial contour used by the programs described in the ⁇ 951 and ⁇ 594 patents are either generated manually by the operator or taken from the contour of a related image.
  • the ⁇ 951 technique operated on single image regions, and was capable of generating a refined contour faster than the operator could draw it.
  • the ⁇ 594 technique includes the functionality of the first program, and steps from CT section to CT section, automatically computing refined contours using previous-section contours as starting estimates.
  • the inventor herein has developed an image segmentation technique using new decisional rules that do not require any model for the image data. Further, the image segmentation technique of the present invention is more capable than the prior art in segmenting objects having non-stationary data characteristics. Further still, the present invention treats both region data and edge data in a natural way.
  • the approach of the present invention is based on statistical pattern recognition, information theory, and numerical optimization of objective functions that capture the probabilistic dependencies among the different types of information contained in digital images .
  • JSD Jensen- Shannon divergence
  • JRD Jensen-Renyi divergence
  • a method of identifying the boundary of an object in an image the image being represented by a data set, the data set comprising a plurality of data elements, each data element having a data value corresponding to a feature of at least a part of the image, the method comprising adjusting a boundary estimate at least partially according to one of the group consisting of a Jensen-Shannon divergence value corresponding to the boundary estimate and a Jensen-Renyi divergence value corresponding to the boundary estimate such that the boundary estimate substantially matches the object boundary.
  • the divergence value represents a measure of the difference between the data values of the data elements located in a zone Z ⁇ and the data values of the data elements located in a zone Z 0 , Z ⁇ and Z 0 being defined by the boundary estimate and representing, respectively, the data elements of the image located inside and outside the boundary estimate.
  • the boundary estimate is preferably a contour C, and more preferably a parametric contour C comprising a plurality N of ordered contour coefficients c ⁇ through c N , each contour coefficient Ci having
  • the method may further comprise the steps of: (1) determining a contour that represents an approximation of the object boundary; (2) calculating one of the group consisting of the Jensen-Shannon divergence value and the Jensen-Renyi divergence for the contour; and (3) wherein the adjusting step comprises adjusting the approximate contour so as to increase its calculated divergence value.
  • the divergence value calculating step may comprise the steps of: (1) identifying which data elements are located in Z ⁇ ; (2) identifying which data elements are located Z 0 ; (3) calculating the conditional probability distribution p (x ⁇ z ⁇ ) ; (4) calculating the conditional probability distribution p(x
  • the adjusting step preferably further comprises adjusting the approximate contour so as to substantially maximize its calculated divergence value.
  • this adjustment process adjusts the most significant coefficients prior to adjusting less significant coefficients by individually adjusting the coefficient values of at least a plurality of approximate contour coefficients starting from c and continuing through successively higher order approximate contour coefficients until the approximate contour' s divergence value is substantially maximized.
  • the image is preferably a medical image of an interior portion of a body, wherein the data set comprises a plurality of pixels, each pixel having a pixel value corresponding to image intensity.
  • the image may comprise a plurality of objects, each object having a boundary, the method further comprising performing the adjusting step simultaneously for at least two of the image objects.
  • each data element may have a plurality of data values corresponding to different features of the image, and the segmentation process may use each of these data values when fitting the contour to the object boundary.
  • Fig. 1 is an exemplary CT digital image
  • Fig. 2 is an exemplary CT digital image accompanied by histograms of pixel intensities corresponding to the areas of the image inside and outside the patient;
  • FIGs . 3 (a) and 3 (b) are overviews of the apparatus for carrying out the present invention.
  • Fig. 3(c) depicts a multiprocessor computer that can perform the processing for the present invention
  • Fig. 4 is an illustration of the image segmentation problem
  • Fig. 5 is a flowchart for the Jensen-Shannon divergence (JSD) /Jensen-Renyi divergence (JRD) image segmentation process of the present invention when a single object is contoured
  • Fig. 6 is a flowchart for the JSD/JRD image segmentation process of the present invention when a multiple objects are contoured;
  • Fig. 7 depicts an example of JSD/JRD image segmentation performed on a random shape figure
  • Fig. 8 depicts examples of JSD/JRD image segmentation performed on a various random shape figures
  • Fig. 9 depicts examples of JSD/JRD image segmentation performed on a series of CT images
  • Fig. 10 depicts an example of JSD/JRD image segmentation performed on multiple objects simultaneously within a CT image.
  • Fig. 3(a) illustrates an overview of the system configured to carry out the image segmentation technique of the present invention.
  • An image sensor 148 is used to generate the image data set that is processed by processor 150.
  • Processor 150 performs the image segmentation.
  • the image data set may be passed directly from the sensor 148 to processor 150 (as shown in Fig. 3(a)) or passed via an image database 152 (as shown in Fig. 3(b)) .
  • the image sensor is preferably a CT scanner. However, any image sensor capable of producing a digital image, such as an MR scanner, X-ray scanner, or the like may be used in the practice of the present invention.
  • processor 150 can be a multi-processor computer.
  • Processor 150 is preferably a general purpose computer, such as a high performance UNIX workstation, executing software programmed to carry out the image segmentation technique of the present invention.
  • Fig. 4 illustrates the basic problem presented by image segmentation.
  • Fig. 4 shows an image 160 that includes an object 162 and a background 170 (background 170 encompasses the area of the image 160 outside the object 162) .
  • the image 160 is comprised of rows and columns of pixels that cover the image area. In image segmentation, the goal is to accurately identify object 162 from background 170.
  • a boundary estimate (segmentation estimate) 164 is determined.
  • the placement of boundary estimate 164 defines a zone Z ⁇ of pixels located inside the boundary estimate 164 and a zone Z 0 of pixels located outside the boundary estimate 164.
  • the boundary estimate 164 is refined such that it substantially matches the true boundary of the object 162.
  • Z ⁇ will correspond to the pixels making up the object 162
  • Z 0 will correspond to the pixels making up the background 170.
  • the boundary estimate 164 is preferably a flexible contour curve C.
  • Contour C has a length T and is a function of arc length t (wherein 0 ⁇ t ⁇ T) .
  • Contour C is defined by a vector of ordered real coefficients c x through c .
  • the contour coefficients c $ _ fully define the location, size, and shape of the contour C in the image space.
  • the contour coefficients serve as the independent variables whose numerical values are adjusted during the image segmentation process to fit the contour to the object boundary.
  • N 4K+2
  • the coefficients Ci are the coefficients of a finite trigonometric series truncated at K harmonics, known as a Fourier elliptic representation.
  • Giardina CR and Kuhl FP "Accuracy of curve approximation by harmonically related vectors with the elliptical loci," Computer Graphics and Image Processing, 21, 277-285, 1977
  • Kuhl FP and Giardina CR "Elliptic Fourier features of a closed contour, " Computer Graphics and Image Processing, 18, 236-258, 1982, the entire disclosures of which are incorporated herein by reference; and see also Duncan and Staib (1992) .
  • This Fourier elliptic representation is the coefficients of a finite trigonometric series truncated at K harmonics, known as a Fourier elliptic representation.
  • the number of harmonics K determines the accuracy with which C is capable of matching an object boundary. A trade-off exists when determining the number K. If the number of harmonics K is too small, the resolution of C may not be sufficient to accurately match the object boundary. A large value for K produces better boundary matches, but requires a proportionally longer time to compute. Further, if a very large value for K is used, the resulting contour C exhibits noise that appears as wiggles . The wiggles cause inaccuracy in the matching process.
  • H s (p) 0.
  • the relative entropy D is a measure of the differences between two distributions of X, p (X) and q(X) , and is defined as:
  • the present invention is concerned with deciding whether a series of samples ⁇ x 1 , X 2 , ...,X Q ⁇ drawn independently and identically from an unknown distribution corresponds to one or the other of two states of nature (anatomic regions) R x and H 2 which could have produced them. It is usual in statistics to cast the problem in terms of hypotheses :
  • H ! ⁇ X 1 ,...,X Q ⁇ d ⁇ av &omp(X ⁇ H 1 ) H 2 : ⁇ i ,-- ;X Q ⁇ drawn from p(X
  • Hi) is the probability of observing X given that it corresponds to the state of nature Hi-
  • L(X X ,..., X Q ) (20) where p x ( i) is the distribution of the Xi in the sequence of samples X X ,X 2 , ...,X Q , and where Qp ⁇ (Xi) weights the log-ratio by the number of times that Xi occurs in the sequence Xj. through X Q .
  • the relative entropy-equivalent to the LRT is: lfD(p x (x) ⁇ p(x ⁇ H 2 ))-D(p x (x) ⁇ p(x ⁇ H l ))>- ⁇ ogT, n (23) decide H,,else H 2 .
  • Formula 23 is a minimum-error classifier capable of assigning a sample distribution of a random variable to one of two known distributions, without reference to any model for the distributions . Renyi introduced a generalization of the Shannon entropy.
  • Renyi A "On measures of entropy and information, " Proceedings of the Fourth Berkeley Symposium on Mathematical Statistics and Probability, Berkeley, CA, June 20-July 30, 1960, Volume I, published 1961, pp. 547-561; Renyi A, "Some Fundamental Questions of Information Theory”. Selected Papers of Alfred Renyi, Volume 2, pp 526-552, Akademia Kiado, Budapest, 1976; and Renyi A, "On Measures of Entropy and Information”. Selected Papers of Alfred Renyi , Vol. 2. pp 565-580, Akademia Kiado, Budapest, 1976, the entire disclosures of which are hereby incorporated by reference.
  • the Renyi entropy of degree ⁇ , ⁇ R ⁇ (p) is defined as
  • JSD Jensen- Shannon divergence
  • JRD Jensen-Renyi divergence
  • JS ⁇ H s ( ⁇ x p(x ⁇ H ⁇ )+ ⁇ 2 p(x ⁇ H 2 ))- ⁇ ⁇ H s (p(x ⁇ H l ))- ⁇ 2 H s (p(x ⁇ H 2 )) ( 2 g )
  • JS ⁇ is a measure of the difference between two distributions p(x
  • ⁇ ⁇ represents the a priori probability that H x is true
  • ⁇ 2 represents the a priori probability that H 2 is true.
  • x could represent the prior probability that any pixel belongs to the region associated with H lf
  • ⁇ 2 could represent the prior probability that any pixel belongs to the region associated with H 2 .
  • the Jensen-Shannon divergence is symmetric. That is, JS ⁇ in equation (26) is unchanged if the subscripts 1 and 2 are reversed.
  • JS ⁇ can be defined for m hypotheses
  • JS ⁇ ( Pl ,...,p m ) H ,H(P,) (27) where the p x through p m are the conditional distributions p(x
  • the JS ⁇ is maximum when the distributions are all maximally different one from another.
  • the JRD is nonnegative for 0 ⁇ ⁇ ⁇ 1, symmetric to the interchange of subscripts (unlike the relative entropy) , and equals zero only when all the probability p(x
  • ⁇ l the JRD tends to the JSD.
  • the JRD is maximum when the distributions are all maximally different one from another.
  • contour C is fit to the object boundary by finding the vector of coefficients for C which defines the closed curve that best matches the actual boundary.
  • a family of such contours ..., c' 1"1 ', C (l) , C (l+1) , ... might be generated during the course of an iterative search for the best C.
  • the interior and exterior regions of the specific contour C (l) are designated Z ⁇ 1 ' and Z 0 ⁇ l) respectively.
  • Z,R) is the probability of observing X given both Z and R jointly
  • R) is the conditional probability of being at a point in zone Z given region R
  • p(R) is the prior probability of being located in region R.
  • the total distribution is formed over X alone by summing over the sets of zones and regions as :
  • the interior and exterior zones contain mixtures of both Rj and R 0 . That is, Z-, overlaps R so the conditional probability p(Z 3
  • Rewriting (35) with the sums over the sets of values the random variables X,X' may take: lOg L(X j , X 2 ,..., X Q , X , X 2 ,..., X Q )
  • the best contour C* is the one for which the distributions (x
  • the corresponding maximum contour expressions for the Jensen-Shannon and the Jensen-Renyi divergences are stated by analogy to equation (40) .
  • the maximum-contour Jensen-Renyi divergence expression can be similarly written. For the case of two regions (R x and R 0 ) to be separated, the Jensen-Renyi divergence is maximized for the boundary contour with parameter vector c* which most perfectly separates the two regions :
  • the single contour program is initialized with an image (step 5.1) and an initial boundary estimate, preferably a polygon created either by the operator or by the program itself (step 5.2) .
  • the polygon is converted to the initial pixel boundary vector b (0) with B 0 pixel-contiguous elements b ⁇ 01 ,b 2 (0) ,...,b B0 ⁇ 0) .
  • the outer loop (steps 5.6-5.12) iterates over successive versions of the parametric contour C (0) ,C (1) ,..., C (l) ,... , stopping when the increases between succeeding values of the JSD (or JRD) fall below a preselected, threshold minimum T. So, if JS (;+1) - JS w ⁇ T (45) or
  • This maximization can be done many ways (steepest descent, Quasi-Newton method, conjugate gradient descent, etc.), and preferably maximization is performed using Powell's direction set method or alternatively, the method of conjugate gradient descent with equivalently- accurate results.
  • Powell's direction set method or alternatively, the method of conjugate gradient descent with equivalently- accurate results.
  • BP Numerical Recipes in C++, Second Edition, Cambridge University Press, Cambridge, UK, 2002
  • Nocedal, J. and Wright, S. J. Numerical Optimization , Springer-Verlag, New York, 1999, the entire disclosures of which are incorporated herein by reference. It is worth noting that although the methods' names indicate minimization, minimizing the negative of the function locates the function's maximum.
  • the inner loop maximizes the JSD (or JRD) by systematically varying the numeric values of the individual parameters C ⁇ (l) ,c 2 (x) ,..., c 4K+2 !:L) , and noting the resulting values of JS ⁇ u) or R ⁇ ⁇ M , according to whichever of the maximization methods is used.
  • trial (*) contours c (1,:i> * (i-th iteration contour, j'-th component being varied) are computed for each of the several values of the parameter C !l> (step 5.7).
  • Each parametric contour is converted to its corresponding boundary b ,:i>* , pixels in the zones zX' ⁇ * and Zo' 1 '-'' * are enumerated, the conditional probability distributions and p(X
  • the largest maximum JS ⁇ ,:i)* or JR ⁇ (i,:i)* for all j is relabeled JS ⁇ ⁇ 1+1) or JR ⁇ (1+1) and submitted to the outer loop test, and the corresponding contour is saved as C (l+1) .
  • the random variable X used in the program correspond to pixel intensity
  • other pixel properties can be used.
  • the order in which the contour coefficients are adjusted during optimization has an important effect on optimization.
  • the low number coefficients affect the global properties while the higher coefficients represent the fine details.
  • contour coefficients 1 and 2 are the xy-centroid of the contour
  • coefficients 3-6 fix the overall size of the object
  • all higher coefficients add successively finer details to the contour shape.
  • Coefficient values for CT organs are largest for the centroid, and tend to have successively smaller values for increasing coefficient order.
  • step 5.9 the coefficient adjustments produce a series of new and monotonically increasing divergence values.
  • the divergence value resulting after all coefficients have been adjusted is the divergence value that will be maximum and passed to the outer loop test.
  • the program which computes multiple contours simultaneously using the objective functions (42) or (44) is initialized with an image and m-1 polygons created either by the operator or by the program itself (steps 6.1 and 6.2) .
  • m regions including the common background region.
  • Maximization of the JSD or the JRD for multiple object contours follows the same scheme as it does for single object contouring.
  • the main difference between the optimization of multiple contours and a single contour is the re-ordering of the multi-contour coefficients for the optimization step.
  • the polygons are converted to a set of initial pixel boundary vectors b ⁇ (0> , b 2 (0) , ... , b m-1 (0> (step 6.3) from which the initial set of parametric contours C ⁇ (0) , C 2 ⁇ 0) , ... , C m- i ⁇ 0> are computed (step 6.4).
  • the subscripts refer to the regions/objects being contoured.
  • the coefficient vectors for each contour are
  • each contour may have a different number of harmonics K .
  • optimization over a contour set has two nested loops.
  • the outer loop (steps 6.6-6.12) iterates over optimized versions o " f the contour set, c C j (1) ,cl- 2 (1) ,...c*-, repet_! (1)
  • step 6.7-6.9 traverses all the coefficients in all the contours in the set.
  • the convergence test for the outer loop is the same as that for the single contour program: stopping when the increases between successive values of the JSD (JRD) fall below a pre-selected, threshold minimum. So, if JS ⁇ ) - JS ⁇ ( ⁇ threshold (49) or
  • the inner loop maximizes the JSD (JRD) over all the contours' coefficients.
  • JRD JSD
  • M pamms ⁇ (AK i +2) (52) parameters in all contours . Because the Fourier elliptic representation orders the global parameters before the fine detail parameters, it is important to establish all the contours' global locations and sizes before determining the shape details for the contours. Therefore, we re-order the parameters in the set (51) by harmonic order as follows,
  • a multiprocessor computer can be used to carry out the contouring program of the present invention.
  • the image data set can be provided to a plurality of processors 150a.
  • increased processing speed can be obtained by splitting a problem into multiple independent parts and running each part on a different processor 150a. This would be especially the case where multiple object contouring is run on m regions in an image data set.
  • Independent functions here include enumerating the region pixels, assembling the pixel feature histograms and distributions, and computing the Jensen or Renyi entropies and their weights .
  • the processing required for each region can be performed on a separate processor 150a in parallel.
  • the results of the parallel processing can then be collated in another processor 150a that would perform the outer loop test.
  • the first set shows the action of the JRD contouring single figures in a series of random shapes with defined signal-to-noise ratios (SNRs) .
  • the second demonstration set shows the action of the JRD to contour anatomic objects in CT images . Because the Jensen-Renyi divergence method consistently produced lower errors than the Jensen-Shannon divergence method in tests on random figures (described below) only demonstrations of the Jensen-Renyi method will be shown here.
  • Figure 8 shows a series of randomly sized and shaped figures, with the JRD matching contours (white lines) computed from the circle initial estimate, and with SNR of about 3.0.
  • the error is defined as the average distance between corresponding points in the two curves.
  • the true, defined contour c ⁇ true) j . ag a corresponding pixel boundary b ltrue) which is a vector of B-pixel coordinates as defined b. (true) through b B (true) where bi ⁇ true> (ro , columni) .
  • the contour estimates C* obtained using the JSD or the JRD produce a trial pixel boundary b* with B* coefficients.
  • a comparison is made between intervals of points on the two curves, selected to sample both curves uniformly.
  • a W -pixel interval in b* is selected, offset by a varying number of pixels in order to find that interval in b* closest to the fixed b (true> interval.
  • the error is the average point-to-point distance for all points in all intervals
  • Figure 9 shows the action of JRD contouring the liver of a patient in a series of CT sections of the abdomen.
  • the thin lines indicate user-input polygons forming the initial estimates of the contours, and the thick white lines are the JRD matching contours.
  • These matching contours were computed using 6 Fourier harmonics . The maximization is that given in equation (43) .
  • Figure 10 shows examples of multiple objects contoured by the JRD, simultaneously in the same picture.
  • the maximization is that given in equation (44) .
  • JSD/JRD image segmentation of the present invention can be based on multiple random variables rather than only a single random variable.
  • Xi Xi
  • X d a random vector of properties
  • the extent to which several random variables contain more information than any one alone depends on whether they are drawn from independent distributions .
  • Extending the JSD/JRD methods to random vectors requires consideration of two cases: 1) the random variates Xi are conditionally dependent, or 2) the random variates Xi are mutually independent .
  • Dependence is the more general situation, with independence a special case of the conditionally dependent case.
  • any one of the individual random variables Xi is from a distribution independent of the distributions of other random variables X j , j ⁇ i
  • the best contour c * is the one for which the distributions p(X i
  • H s (p) - ⁇ p(X) ⁇ ogp(X) xsX (64) where the sum is taken over the domain of values on which X is defined.
  • the entropy is
  • H s (p(X)) -Y J Y j .- j p(X l )p(X 2 -p(X d ) ⁇ gp(X l )p(X 2 -p(X d )
  • Non-stationary applies to pixel data in which random variable means, variances, and distributions vary across the image of a single logical object.
  • Classifiers which depend on a fixed mean and variance (e.g., the multivariate Gaussian model described in Section 2) to sort pixels into one or another objects pictured can produce higher error rates for non-stationary image data.
  • JSD/JRD methods use functions of the random variables' distributions, it is possible that these methods may be more robust than parametric models since the JSD/JRD functions operate by portioning the data by distribution differences, regardless of any parameters that might be used to describe the distributions.
  • non- stationary data would produce small, continuous changes in the region data distributions, and such changes need not increase the error in the JSD/JRD autosegmentation.
  • the JSD/JRD image segmentation discussed above was a region-centered approach in which the decision rules operated on distributions of pixel properties .
  • the teachings of the present invention are also applicable to an edge-centered approach in which the decision rules operate on intensity gradients .
  • Edges represented by coheren high-valued gradient- image pixels should maximize a decision rule that superimposes the contour C* on gradient maxima pixels .
  • Edges contain the highest values in the gradient image so a rule which seeks to maximize the difference between edge-pixel distributions and all the other pixels should force the contour onto the gradient image maxima.
  • the present invention can be used in combination to (1) partition an object from a background (as previously described) and (2) separate the edge of the object, as observed in an image of the gray level gradient maxima (which is derived from the original image data set) by maximizing the JSD/JRD divergence between the pixels underneath contour C 1 and all other pixels in the gradient image.
  • the center panel of Fig. 1 illustrates an example of a gradient image. Because the set of pixels coincident with an edge will necessarily have a distribution different from that of the gradient image background plus other object edges, maximizing the divergence over these two sets of gradient image pixels should add to the specificity of the boundary estimate produced in the regional divergence estimation.
  • the "matching contour" can be defined as the contour that jointly maximizes the combined divergences. Because the JSD/JRD divergences take on arbitrary real values determined by the data at hand, the regional and edge divergences for any given region in any given image might have very different values, thereby requiring a weighting to make them equivalent . Such a function would be expressed as :
  • F(JSD) w region JSD region (C )+ w edge JSD edge (C ) ( 83 ) or
  • F(JRD) w region JRD region ⁇ c )+ w edge JRD edge ( ⁇ ) (84) wherein w region and w ege are real number coefficients that can be set to equalize the contributions of the region and edge to the total divergence .
  • the present invention is also applicable to images having more than 2 dimensions .
  • the present invention should work equally well in 3D as 2D, but would require flexible surfaces or volumes instead of a contour.

Abstract

A method of approximating the boundary of an object in an image, the image being represented by a data set, the data set comprising a plurality of data elements, each data element having a data value corresponding to a feature of the image, the method comprising determining which one of a plurality of contours most closely matches the object boundary at least partially according to a divergence value for each contour, the divergence value being selected from the group consisting of Jensen-Shannon divergence and Jensen-Renyi divergence. Each contour Ci defines a zone ZIi and a zone ZOi, ZIi representing the data elements inside the contour and ZOi representing the data elements outside the contour, each zone having a corresponding probability distribution of data values for the data elements therein, and wherein the divergence value for each contour Ci represents a measure of the difference between the probability distributions for the zones ZIi and ZOi. The boundary estimate is preferably a parametric contour. Further, the present invention supports the segmentation of multiple objects in a single data set simultaneously.

Description

METHOD AND APPARATUS FOR IMAGE SEGMENTATION USING JENSEN-SHANNON DIVERGENCE AND JENSEN-RENYI DIVERGENCE
FIELD OF THE INVENTION:
The present invention relates to image segmentation. More specifically, the present invention addresses an improved technique for identifying the boundaries of distinct, discrete objects depicted in digital images, particularly medical images.
BACKGROUND OF THE INVENTION:
Image segmentation (also known as "autosegmentation", "contouring", and "autocontouring") is a technique of processing a digital image to detect, classify, and enumerate discrete objects depicted therein. Image segmentation presents a difficult problem because digital images generally lack sufficient information to constrain the possible solutions of the segmentation problem to a small set of solutions that includes the correct solution. Image segmentation finds a popular application in the field of medical images, particularly computed tomography (CT) images, x-ray images, magnetic resonance (MR) images, and the like. It is highly desirable to accurately contour various anatomic objects (such as the prostate, kidneys, the liver, the pancreas, etc.) that are shown in such medical images . By accurately determining the boundary of such anatomic objects, the location of the anatomic object relative to its surroundings can be used to plan and execute medical procedures such as radiotherapy treatment for cancer. Because it is often the case that the healthy tissues and organs surrounding cancerous tissue are highly sensitive to radiation, it is extremely important that the exact location of the anatomic object to be treated be known. That is, accurate placement, shaping, and intensity modulation of external X-ray beams for tumor irradiation (teletherapy) depends absolutely on detailed 3-D maps of the patient's anatomy, as do other forms of radiotherapy treatment such as brachytherapy and radioimmunotherapy.
Treatment planning for medical procedures such as teletherapy and brachytherapy begins by creating a detailed 3-D map of the patient's anatomy (including tumors). These maps are often created as a series of contours drawn on CT slices . In teletherapy procedures, once the map is obtained, a set of external X-ray beam angles, beam cross-section shapes, and beam intensities can be designed to deliver a prescribed dose of radiation to the tumor while minimizing the harm done to nearby healthy tissues. For brachytherapy procedures, once the map is obtained, the appropriate location (s) to deliver radioactive seeds can be determined, also while minimizing the harm done to nearby healthy tissues .
Image segmentation operates on medical images in their digital form. A digital image of a target such as the human body is a data set comprising an array of data elements, each data element having a numerical data value corresponding to a property of the target, wherein the property is measured by an image sensor at regular intervals throughout the image sensor's field of view. The property to which the data values correspond may be the light intensity of black and white photography, the separate RGB components of a color image, or other similar properties. Typically the image data set is an array of pixels, wherein each pixel has a value corresponding to intensity. The leftmost image shown in Figure 1 is a CT digital image of an interior portion of a patient's body. As can be seen, depicted in the image is the left kidney along with nearby anatomical objects such the spleen, the vertebral body, the spinal cord, and the ribs. The usefulness of digital images derives partly from the capacity of digital images to be transformed and enhanced by computer programs so that meaning can be extracted therefrom. The data value of each data element can be modeled as a random variable. A random variable is a measured quantity whose repeated observations cluster about some central value. The "mean" is the most likely value and the "mode" is the most-often observed value. Random variables are most completely described by a histogram, rescaled as a probability distribution. Typically, the data values of the data elements of the image data set constitute a random variable X which may take its values from a finite set of possible data values, such as the interval -1024 to 3071, typical for CT. In mathematical notation, the data value of any given data element of the image data set may be equal to any Xi, wherein xi is a member of the set {xlf x2, ... xn} , and wherein "n" is 4096 for the CT example above. Random variables whose values are limited to members of finite sets are termed discrete; random variables which can take any real value are continuous random variables . The probability that a particular data element takes the value x± for random variable X, is expressed as p ( i) = xi} . Pr{} is the probability that the expression in the braces {} is true. The full set of probabilities, written p (X) , is the probability distribution. As is well-known, probabilities and their distribution have the following properties : p(Xi) > 0; 0 < p(xi) < 1; and
ni-ιP(x_) = To deal with 2 or more random variables at once, the concepts of joint probability and conditional probability are used. For two discrete random variables (X and Y) , the joint distribution is the set of joint probabilities p (x,y) =Pr{X=x, Y=y} for all possible values of x and y. A histogram corresponding to such a joint distribution can be represented by a two-dimensional plot with the x entries along the rows and the y entries along the columns .
The conditional probability of observing any X given that Y=y is written as p ( | Y=y) , which corresponds to the column of histogram entries for Y=y. The conditional and joint probabilities are related by the definition: p(X)
The conditional probability of Y, given X, is defined as:
(r|χ) = ( )
Referring to Fig. 2, it can be seen that the relative sizes of the peaks shown in the histogram correspond to the probabilities that a pixel selected randomly will belong to one of the regions in the image. In the CT image shown in Fig. 2, two regions are easily discernible; interior region 100, which corresponds to the anatomical portion of the image, and exterior region 102, which corresponds to the non-anatomical portions of the image (the black background and the CT scanner couch and gantry ring depicted at the bottom of the image) . The conditional probability (or likelihood) that a pixel with value i is in a particular region k, is p (X I regionk) . If the individual intensity probabilities in a region k are summed:
∑"=1 P(x t I regionk) = p(regionk) wherein p (regionk) is the probability of hitting region k when a pixel is randomly selected from the image. If there are R regions in the image, the sum of probabilities over all regions is
∑* lp(regionk) = l ^ because the regions cover the entire image area. Thus, the total probability over all regions and all pixel values is :
∑ ∑!lι-p(χi' i res/o"*) = 1 (5) These conditional probability expressions can be used to make decisions as to which pixels belong to which regions . For example, if two regions, region! and region2, are equally probable, then a pixel with value xA most likely belongs to the region corresponding to the largest conditional probability p (xiI regionk) , wherein k is 1 or 2. Referring back to Fig. 2, it can be seen that more pixels with high intensity are found in the interior region than in the exterior region. Similarly, more pixels with a low intensity are found in the exterior region that the interior region. Therefore, if a pixel with value -1024 (the minimum pixel value) is observed, a reasonable decision can be made that such a pixel belongs in the exterior region rather than the interior region. Various decision-making techniques for processing digital images to perform image segmentation have been developed from this basic concept.
For example, one known method of assigning pixels to a particular region is known as the "likelihood ratio test" (LRT) . The LRT is defined as : p(X: \ region,) _
If > 1 , then decide regιonx, else decide reg on2 p(x, I regionf)
(6) wherein T is a threshold value determined from training data in which the regions corresponding to each pixel value are known exactly.
The conditional probability of regionk being the true region given that a pixel has a pixel value X=x± is proportional to the product of the conditional likelihood and the prior (region) probability. This conditional probability is called the a posteriori or posterior probability, and the maximum a posteriori
(MAP) decision rule is derived therefrom as :
If pjregion, ^,) = p(xi \ regi0nx)p(regior ) > ^ ^.^ ^.^ p(region2 \ xt) p(xt \ region2)p(region2) (7) else region2
Referring back to Fig. 1, it can be seen that the different anatomic objects (kidney, spleen, ribs, spinal cord, and vertebral body) exhibit varying degrees of differences in pixel intensities . While the kidney, spleen, and spinal cord share similar pixel intensities, the pixel intensity for the ribs and vertebral body are markedly different therefrom. Further, the pixels within each object have more or less constant pixel intensities (regional coherence) . Moreover, it can be seen that there are often large steps in intensity as one passes out of one object and into a neighbor (edge step differences) . For example, note the dark region between the kidney and spleen. In fact, regional coherence and edge-steps are the two dominant structural features in images. Five basic image segmentation techniques are known in the art: (1) region-based methods, (2) edge-based methods, (3) image feature clustering methods, (4) global optimization methods, and (5) hybrid methods combining region- and edge-based ideas. See Haralick RM and Shapiro LG, "Image segmentation techniques,"
Computer Vision Graphics and Image Processing, 29:100-132, 1985; Pal, N.R. and Pal, S.K., "A review on image segmentation techniques," Pattern Recogni tion, 26:1277-1294, 1993; and Tang, M. and Ma, S., "General scheme of region competition based on scale space," IEEE Transactions on Pattern Analysis and Machine Intelligence, 23(12), 1366-1378, 2001, the entire disclosures of which are hereby incorporated by reference.
Region-based approaches assume that region pixels have almost-constant properties, and that these properties are different from region to region. Pixel properties (also called features) include pixel intensity, and mean, variance, range, moments, etc. computed over a neighborhood of surrounding pixels. Region-based segmentation assumes that the means and variances of these properties are constant within a region, and differ among regions. See Rosenfeld, A. and Kak, A.C., Digital Picture Processing, Academic Press, 1982; Beveridge, JR, Griffith, J, Kohler, RR, Hanson, AR, and Riseman, EM, "Segmenting images using localized histograms and region merging," International Journal of Computer Vision, 2, 311-347, 1989; and Adams, R. and Bischof, L., "Seeded region growing," IEEE Trans. Pattern Analysis and Machine Intelligence, 16(6), 641-647, 1994, the entire disclosures of which are hereby incorporated by reference . Because of this assumption, region-based approaches falter if the pixel properties vary across the region, or if the boundaries between regions with the same properties are blurred (by patient breathing motion in CT, for example) , or if neighboring regions' properties are similar or have overlapping distributions. The success of a region-based segmentation operation is highly dependent upon the proper choice of a pixel homogeneity criterion. If the pixel homogeneity criterion is poorly chosen, the image segmentation operation results in ragged boundaries and holes in the segmented region. Moreover, even with a good choice of a pixel homogeneity criterion, the overlap of pixel distributions from adjoining regions constitutes a big problem for this approach.
Edge-based methods assume that all regions are surrounded by a significant step in the pixel property. Using this assumption, edge-based image segmentation extracts a segment by semantic analysis of the edges (see Ballard, DH and Brown, CM, Computer Vision, Prentice-Hall, Englewood Cliffs, NJ, 1982, the entire disclosure of which is hereby incorporated by reference) or by fitting a flexible curve or active contour to the actual boundary in the image (see Kass, M. , Wi kin, A., Terzopoulos, D., "Snakes: Active contour models," International Journal of Computer Vision, 1, 321-331, 1988. ,- Cohen, L.D., On active contour models and balloons," CVGIP: Image Understanding, 53(2), 211-218, 1991; and Malladi, R., Sethia, J.A., and Vemuri, B.C., "Shape modeling with front propagation: A level set approach," IEEE Trans . Pattern Analysis and Machine Intelligence, 17(2), 158-175, 1995, the entire disclosures of which are hereby incorporated by reference) . Like the region approach, active contour segmentation may fail if the actual pixel data does not exhibit sharp steps everywhere around the object. Successful segmentation also depends on the kind of flexible model used to match the boundary. Flexible models such as snakes and balloons show troublesome sensitivity to initial placement in the image, and the image segmentation goes awry if that placement is not sufficiently close to the true boundary. Semantic methods can assemble contour fragments discovered in the map of edges by a set of rules to produce the object form, but this is a very complex task, and except for some highly regular applications, has not found wide use in medical image analysis .
Image-feature clustering segmentation identifies objects by looking in the space of image pixel properties for clusters that correspond to the feature combinations of object pixels. Each combination corresponds to a point in feature space, where the space dimension is equal to the number of features. Sets of object pixels with similar properties correspond to sets of points clustered in feature space. Finding such point clusters enables the analyst to go back to the image and identify the pixels inhabiting this concentration, and by implication, the object. Well-established methods exist for identifying point density maxima and classifying the feature space points/image pixels . See Devroye L, Gyorfi L, and Lugosi G, A Probabilistic Theory of
Pattern Recognition, Springer, New York, Springer, 1996; and Duda RO, Hart PE, and Stork DG, Pattern Classification, 2nd Ed., New York: Wiley-Interscience, 2001, the entire disclosures of which are hereby incorporated by reference. For example, the nearest neighbor rule specifies that each pixel be assigned to the region corresponding to the feature space cluster nearest that pixel's feature space location. Many other clustering rules exist and are used in, for example, commercial data mining applications. See Witten, I.H. and Frank, E., Data Mining, Morgan Kaufmann, San Francisco, CA, 1999 (the entire disclosure of which is hereby incorporated by reference) ; and Duda et al. (2001) . The problems which can confound density rules for image clustering are the same problems described above in connection with region-based techniques; property densities overlap making classification error inevitable. Estimating cluster point densities requires that feature space be "binned", which means that a smoothed version of the space must be computed so that areas of low frequency are given some reasonable probability. If the number of features is large, the data is spread out over more dimensions, making the pixel classification more computationally expensive, and exacerbating the data sparsity.
Global optimization segmentation performs non-linear transformations on the image pixel values to maximize an objective function which is assumed to accurately describe the properties of the image, or the properties of the measurement process that formed the image. The two most famous ideas are iterated conditional modes (See Besag, J., "On the statistical analysis of dirty pictures," Journal of the Royal Statistical Society, 48, 259--279, 1986, the entire disclosure of which is hereby incorporated by reference) and Markov-Gibbs refinement (See Geman, S. and Geman, D., "Stochastic relaxation, Gibbs distribution and the Bayesian restoration of images". IEEE Transactions on Pattern Analysis and Machine Intelligence, 6 (6) :721--741, 1984, the entire disclosure of which is hereby incorporated by reference) . In both cases, the image is transformed by small, iterative steps to versions in which the regions have constant features different from the surrounding regions. While built on powerful ideas, these methods assume simple behaviors for the pixel properties, and these assumed behaviors do not seem to correspond well to those of real images. One problem, already mentioned, is that the pixel properties are not stationary-that is, their means and variances change across a region. Non-stationarity is a complex problem not well modeled by any current theories . A second problem is that the pixel properties do not always exhibit normal distributions (Gaussian) or other distributions that such theories often employ. As such, these methods are less well-equipped to operate on images with arbitrary property distributions .
Hybrid segmentation methods operate on several image properties at once. A good example of these methods are those developed by Duncan and Staib. See Staib LH and Duncan JS, "Boundary finding with parametrically deformable models," IEEE Transactions on Pattern Analysis and Machine Intelligence, 14, 1061-1075, 1992, the entire disclosure of which is hereby incorporated by reference. Briefly, Staib and Duncan proposed applying rigorously correct statistical inference to automated image segmentation by matching the properties of a flexible-shape template to image objects. Later, Chakraborty added to the
Duncan/Staib method by integrating the flexible contour model with region, edge, and prior shape information to find a region boundary. See Chakraborty A, Staib LH, Duncan JS, "Deformable boundary finding in medical images by integrating gradient and region information," IEEE Transactions on Medical Imaging, 15, 859-870, 1996, the entire disclosure of which is hereby incorporated by reference. So, in addition to the region and edge properties of the current image, one could add a bias in favor of the shapes of the same organ determined previously on neighboring CT sections. The inference methods used are either the likelihood ratio decision rule or the maximum a posteriori decision rule (when prior boundary shapes are known) . These decision rules guarantee minimum average errors given the data.
Because the correct boundary is not computable directly, the above-described Duncan/Staib/Chakraborty (DSC) method iteratively computes the value of the decision rule and adjusts the flexible curve so as to maximize the value of the decision-rule function. The flexible curve is a finite trigonometric series whose coefficients, or parameters, fully define the location, size, and shape of the curve. The actual "footprint" of the curve on the image can be quickly computed, and compared to an independently- segmented region boundary, to object edges in a map of all the image object edges, and to the shape of a similar object determined previously. The extent of these agreements is input to a decision rule function. If the value of the decision rule function is higher (more probable) than any previous value, this curve is saved, and the program searches for another curve that gives even higher values to the decision rule function. After satisfying a stopping criterion, the program reports the curve corresponding to the largest value of the decision rule function. In addition to the five segmentation approaches described above, there have been a small number of information theoretic image applications. Information theory since its start has been applied to problems of data compression, coding, and model estimation. Kullback and Leibler discovered information theoretic equivalents to the minimum-error maximum likelihood and Bayesian decision rules . These rules use the data distributions directly, giving results that are independent of the origin of the data or any model of the data.
Image segmentation techniques previously developed by the inventor herein are disclosed in U.S. Patent Nos. 5,859,951 and 6,249,594. The Λ951 patent describes a region-based approach using Bayesian classification of pixels operating on pixel intensity and variance features. The Λ594 patent describes a unique combination of Bayesian region classification with the DSC approach to create a useable tool . The initial contour used by the programs described in the Λ951 and Λ594 patents are either generated manually by the operator or taken from the contour of a related image. The Λ951 technique operated on single image regions, and was capable of generating a refined contour faster than the operator could draw it. The Λ594 technique includes the functionality of the first program, and steps from CT section to CT section, automatically computing refined contours using previous-section contours as starting estimates.
While the above-described segmentation techniques have contributed to the art, a need still exists for a segmentation technique that is more accurate and less sensitive to the effects of modeling errors or the like.
SUMMARY OF THE INVENTION: Toward this end, the inventor herein has developed an image segmentation technique using new decisional rules that do not require any model for the image data. Further, the image segmentation technique of the present invention is more capable than the prior art in segmenting objects having non-stationary data characteristics. Further still, the present invention treats both region data and edge data in a natural way. The approach of the present invention is based on statistical pattern recognition, information theory, and numerical optimization of objective functions that capture the probabilistic dependencies among the different types of information contained in digital images .
The inventor herein has found that using either the Jensen- Shannon divergence (JSD) or the Jensen-Renyi divergence (JRD) as the criterion for optimizing the accuracy of a boundary estimate greatly improves the accuracy of image segmentation. Accordingly, disclosed herein is a method of identifying the boundary of an object in an image, the image being represented by a data set, the data set comprising a plurality of data elements, each data element having a data value corresponding to a feature of at least a part of the image, the method comprising adjusting a boundary estimate at least partially according to one of the group consisting of a Jensen-Shannon divergence value corresponding to the boundary estimate and a Jensen-Renyi divergence value corresponding to the boundary estimate such that the boundary estimate substantially matches the object boundary. The divergence value represents a measure of the difference between the data values of the data elements located in a zone Zτ and the data values of the data elements located in a zone Z0, Zτ and Z0 being defined by the boundary estimate and representing, respectively, the data elements of the image located inside and outside the boundary estimate. Preferably, each data value xi of the random variable X is any member of the data value set X' ={xl f x2, ... xn}, X' comprising a plurality of possible data values, and wherein each data value is the value of a random variable X drawn from the data value set X' according to the probability law p (xi) =Pr{x=Xi} , and wherein the probability laws in the several image zones may be different such that the probability of observing X with the specific value Xi in the zone Z0 is the conditional probability p {x | Z0) and the probability of observing X with the specific value xA in the zone Zτ is the conditional probability p ixi l Zj) , and wherein the divergence value represents a measure of the difference between a conditional probability distribution p(x|Zj) and a conditional probability distribution p(x|z0). Also, the boundary estimate is preferably a contour C, and more preferably a parametric contour C comprising a plurality N of ordered contour coefficients cλ through cN, each contour coefficient Ci having a numeric value.
Moreover, the method may further comprise the steps of: (1) determining a contour that represents an approximation of the object boundary; (2) calculating one of the group consisting of the Jensen-Shannon divergence value and the Jensen-Renyi divergence for the contour; and (3) wherein the adjusting step comprises adjusting the approximate contour so as to increase its calculated divergence value.
The divergence value calculating step may comprise the steps of: (1) identifying which data elements are located in ZΣ; (2) identifying which data elements are located Z0; (3) calculating the conditional probability distribution p (x \ zτ) ; (4) calculating the conditional probability distribution p(x|z0); and (5) calculating the divergence value between p(x|Zϊ) and p(x|z0) •
Also, the adjusting step preferably further comprises adjusting the approximate contour so as to substantially maximize its calculated divergence value. Preferably this adjustment process adjusts the most significant coefficients prior to adjusting less significant coefficients by individually adjusting the coefficient values of at least a plurality of approximate contour coefficients starting from c and continuing through successively higher order approximate contour coefficients until the approximate contour' s divergence value is substantially maximized.
The image is preferably a medical image of an interior portion of a body, wherein the data set comprises a plurality of pixels, each pixel having a pixel value corresponding to image intensity.
Furthermore, the image may comprise a plurality of objects, each object having a boundary, the method further comprising performing the adjusting step simultaneously for at least two of the image objects.
Further still, each data element may have a plurality of data values corresponding to different features of the image, and the segmentation process may use each of these data values when fitting the contour to the object boundary. These and other features and advantages of the present invention will be in part apparent and in part pointed out in the following description and referenced figures .
BRIEF DESCRIPTION OF THE DRAWINGS: Fig. 1 is an exemplary CT digital image;
Fig. 2 is an exemplary CT digital image accompanied by histograms of pixel intensities corresponding to the areas of the image inside and outside the patient;
Figs . 3 (a) and 3 (b) are overviews of the apparatus for carrying out the present invention;
Fig. 3(c) depicts a multiprocessor computer that can perform the processing for the present invention;
Fig. 4 is an illustration of the image segmentation problem; Fig. 5 is a flowchart for the Jensen-Shannon divergence (JSD) /Jensen-Renyi divergence (JRD) image segmentation process of the present invention when a single object is contoured; Fig. 6 is a flowchart for the JSD/JRD image segmentation process of the present invention when a multiple objects are contoured;
Fig. 7 depicts an example of JSD/JRD image segmentation performed on a random shape figure;
Fig. 8 depicts examples of JSD/JRD image segmentation performed on a various random shape figures;
Fig. 9 depicts examples of JSD/JRD image segmentation performed on a series of CT images; and Fig. 10 depicts an example of JSD/JRD image segmentation performed on multiple objects simultaneously within a CT image.
DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENT:
Fig. 3(a) illustrates an overview of the system configured to carry out the image segmentation technique of the present invention. An image sensor 148 is used to generate the image data set that is processed by processor 150. Processor 150 performs the image segmentation. The image data set may be passed directly from the sensor 148 to processor 150 (as shown in Fig. 3(a)) or passed via an image database 152 (as shown in Fig. 3(b)) . The image sensor is preferably a CT scanner. However, any image sensor capable of producing a digital image, such as an MR scanner, X-ray scanner, or the like may be used in the practice of the present invention. To speed processing, processor 150 can be a multi-processor computer. As will become more apparent below, many of the computations made in the preferred implementation of the invention can be performed independently. As such, these independent computations can be made on separate hardware processors in parallel to speed the operation of the invention. Processor 150 is preferably a general purpose computer, such as a high performance UNIX workstation, executing software programmed to carry out the image segmentation technique of the present invention. Fig. 4 illustrates the basic problem presented by image segmentation. Fig. 4 shows an image 160 that includes an object 162 and a background 170 (background 170 encompasses the area of the image 160 outside the object 162) . The image 160 is comprised of rows and columns of pixels that cover the image area. In image segmentation, the goal is to accurately identify object 162 from background 170. To do this, a boundary estimate (segmentation estimate) 164 is determined. The placement of boundary estimate 164 defines a zone Zτ of pixels located inside the boundary estimate 164 and a zone Z0 of pixels located outside the boundary estimate 164. During image segmentation, the boundary estimate 164 is refined such that it substantially matches the true boundary of the object 162. When the refined boundary estimate 164 substantially matches the object's true boundary, Zτ will correspond to the pixels making up the object 162 and Z0 will correspond to the pixels making up the background 170.
1. Flexible Parametric Contours
The boundary estimate 164 is preferably a flexible contour curve C. Contour C has a length T and is a function of arc length t (wherein 0 < t < T) . Contour C is defined by a vector of ordered real coefficients cx through c .
C(t) = f(c,t);c = (cl ,c2 ,- - - ,cN );c e RN (g)
The contour coefficients c$_ fully define the location, size, and shape of the contour C in the image space. The contour coefficients serve as the independent variables whose numerical values are adjusted during the image segmentation process to fit the contour to the object boundary.
Preferably, N = 4K+2 , and the coefficients Ci are the coefficients of a finite trigonometric series truncated at K harmonics, known as a Fourier elliptic representation. See Giardina CR and Kuhl FP, "Accuracy of curve approximation by harmonically related vectors with the elliptical loci," Computer Graphics and Image Processing, 21, 277-285, 1977; and Kuhl FP and Giardina CR, "Elliptic Fourier features of a closed contour, " Computer Graphics and Image Processing, 18, 236-258, 1982, the entire disclosures of which are incorporated herein by reference; and see also Duncan and Staib (1992) . This Fourier elliptic representation:
Figure imgf000016_0001
is defined by the curve coordinate formulas 10 and 11 for the coefficients as :
Figure imgf000016_0002
Figure imgf000017_0001
wherein : c = Cj , c2 , • • • , c +2 ) = (A0 , __>0 , βj , Dγ , Cj , Ωj , a2 , o2 , c2 , u2 , • • • , ικ , oκ , cκ , aκ )
(12) where parameter t e[0,T) for the closed contour C of length T.
The number of harmonics K determines the accuracy with which C is capable of matching an object boundary. A trade-off exists when determining the number K. If the number of harmonics K is too small, the resolution of C may not be sufficient to accurately match the object boundary. A large value for K produces better boundary matches, but requires a proportionally longer time to compute. Further, if a very large value for K is used, the resulting contour C exhibits noise that appears as wiggles . The wiggles cause inaccuracy in the matching process. Preferably, to resolve this tradeoff, the rule disclosed in Hibbard, ., "Maximum a posteriori Segmentation for Medical Visualization", IEEE Workshop on Biomedical Image Analysis (WBIA98) , Santa Barbara, CA, June 26-27, 1998, pp. 93-102 (the entire disclosure of which is hereby incorporated by reference) , is followed. The transformation of C from x,y coordinates to parameters Ci (and vice versa) is straightforward and known in the art. See Kuhl and Giardina (1982) . Because the computation of coordinates from coefficients (and from coefficients to coordinates) is fast, a computer program can explore a large space of different contours by systematically varying the values of the individual coefficients ci. The question then becomes how a decision-making process can be designed to determine how the coefficients should be adjusted to result in a matching contour C and how such a process will recognize when a substantial match has occurred.
2. Jensen- Shannon and Jensen-Renyi Divergences
Consider the possible values for pixel intensity to be a discrete random variable X which may take a specific value X = x from a set of n possible values xe{x!,x2,...,xn} . The probability distribution for X is p (x±) =Pr{x=Xi} where p(xi)>0 and the sum of all p (Xi) for all n values of X is 1. The Shannon entropy Hs (p) for the distribution p is defined as : n
Hs (p) =-∑ p( i) logp(Xi)
''=1 (13) where Hs(p) > 0. See Shannon CE, "A mathematical theory of communication," The Bell System Technical Journal , Vol. 27, pp. 379-423, October 1948; and Cover TM and Thomas JA, Elements of Information Theory. New York: Wiley-Interscience, 1991, the entire disclosures of which are incorporated herein by reference. The Shannon entropy for a distribution is a measure of the degree of randomness of the distribution. Hs (p) =0 exactly if all the probability is associated with one outcome, such as X always equals Xi such that p(xi)=l, with all the other elements of the distribution p(xk)=0, k≠i. Hs(p) is a maximum when all the x± are equally probable, or p (Xi) =l/n for all i.
For the discrete random variable X, there may exist two (or more) distributions associated with X in different states of nature. The relative entropy D is a measure of the differences between two distributions of X, p (X) and q(X) , and is defined as:
Figure imgf000018_0001
where, by convention 0log(0/q)=0 and plog(p/0) = oo. see Kullback S and Leibler RA, "On information and sufficiency," Annals of Mathematical Statistics, 22, 79-86, 1951; and Kullback S,
Information Theory and Statistics, New York: Wiley, 1959, the entire disclosures of which are incorporated herein by reference; and Cover and Thomas (1991) . The relative entropy is also known as the Kullback-Leibler divergence, the J-divergence, the cross entropy, etc. Relative entropy is always non-negative, is exactly zero when p( i)=q( i) for all i, and increases in value as the distributions are more different. Further, relative entropy is not symmetric since D(p| |q) ≠ D(q| |p) in general.
The present invention is concerned with deciding whether a series of samples {x1, X2, ...,XQ} drawn independently and identically from an unknown distribution corresponds to one or the other of two states of nature (anatomic regions) Rx and H2 which could have produced them. It is usual in statistics to cast the problem in terms of hypotheses :
H! : {X1,...,XQ} dιav &omp(X \ H1) H2 : { i ,-- ;XQ} drawn from p(X | H2) (ιs ) where p(x|Hi) is the probability of observing X given that it corresponds to the state of nature Hi- These are the class- conditional distributions, or data likelihoods, and if they can be estimated from labeled training data or other information, one can use the LRT to decide between Hx and H2. See Van Trees HL,
Detection, Estimation, and Modulation Theory, Part I, Wiley, New York, 1969; Duda RO and Hart PE, Pattern Classification and Scene Analysis, Wiley, New York, 1973; and Fukunaga K, Introduction to Statistical Pattern Recognition, Academic Press, New York, 1990, the entire disclosures of which are incorporated herein by reference; and Devroye (1996) ; and Duda (2001) .
The ratio of the likelihoods can be rewritten as the log- quantity:
Figure imgf000019_0001
Since the samples are independent, their joint conditional probability is equal to the product of the individual conditional probabilities ,
p(Xl,...,XQ \ Hi) : ■■I!**, \ Ht) q=\ (17) and the log of the joint probability is :
logp(Xr,...,XQ \ Hi) = ∑logp(Xq \ Hi) . (is) q=\
Therefore, the log-product of probability-ratios in (16) can be written directly as the sum of log-ratio terms:
Figure imgf000019_0002
Changing from summation over the samples to summation over the set of values each sample may take:
L(XX ,..., XQ) = (20)
Figure imgf000019_0003
where px( i) is the distribution of the Xi in the sequence of samples XX,X2, ...,XQ, and where Qpχ(Xi) weights the log-ratio by the number of times that Xi occurs in the sequence Xj. through XQ. Now, multiplying each term by a factor of one, px( ) /px(Xi) , one can obtain the difference of sums :
Figure imgf000020_0001
which can be written as the difference of relative entropies : L(X1,...,XQ) = n[D(px(x)\\p(x\H2))
-D(px(x)\\p(x\H1))].
Therefore, the relative entropy-equivalent to the LRT is: lfD(px(x)\\p(x\H2))-D(px(x)\\p(x\Hl))>-\ogT, n (23) decide H,,else H2.
Formula 23 is a minimum-error classifier capable of assigning a sample distribution of a random variable to one of two known distributions, without reference to any model for the distributions . Renyi introduced a generalization of the Shannon entropy.
See Renyi A, "On measures of entropy and information, " Proceedings of the Fourth Berkeley Symposium on Mathematical Statistics and Probability, Berkeley, CA, June 20-July 30, 1960, Volume I, published 1961, pp. 547-561; Renyi A, "Some Fundamental Questions of Information Theory". Selected Papers of Alfred Renyi, Volume 2, pp 526-552, Akademia Kiado, Budapest, 1976; and Renyi A, "On Measures of Entropy and Information". Selected Papers of Alfred Renyi , Vol. 2. pp 565-580, Akademia Kiado, Budapest, 1976, the entire disclosures of which are hereby incorporated by reference. The Renyi entropy of degree α, Η(p) , is defined as
Figure imgf000020_0002
That is, the Renyi entropy tends to the Shannon entropy as cc-l.
The Shannon and Renyi entropies are related as :
H≥Hs≥H, 0<α<l, β>l. (2g) See Principe JC, Xu D, and Fisher JW III, "Information-theoretic learning," in Unsupervised Adaptive Filtering, S. Haykin, Ed. New York: Wiley, 1999, 265-319, the entire disclosure of which is incorporated herein by reference. Jensen- Shannon divergence (JSD) and Jensen-Renyi divergence (JRD) are two relative entropy measures . For the case of two hypotheses Hi and H2, the JSD is defined as :
JSπ = Hsxp(x\ Hϊ)+π2p(x \ H2))-πϊHs(p(x \ Hl))-π2Hs(p(x \ H2)) ( 2 g ) See Lin J, "Divergence measures based on the Shannon entropy, " IEEE Transactions on Information Theory, 37, 145-151, 1991, the entire disclosure of which is incorporated herein by reference. JSπ is a measure of the difference between two distributions p(x|Hι) and p(x|H2), where the variables πx and π2 are weights for the probability distributions . Preferably, πλ represents the a priori probability that Hx is true , and π2 represents the a priori probability that H2 is true. In image segmentation, x could represent the prior probability that any pixel belongs to the region associated with Hlf and π2 could represent the prior probability that any pixel belongs to the region associated with H2. Unlike the relative entropy, the Jensen-Shannon divergence is symmetric. That is, JSπ in equation (26) is unchanged if the subscripts 1 and 2 are reversed.
In addition, JSπ can be defined for m hypotheses
JSπ (Pl,...,pm) = H ,H(P,) (27)
Figure imgf000021_0001
where the px through pm are the conditional distributions p(x|Hi) through p(x|Hm) and the itj. are the distribution weights where πx > 0 and where the sum of πx for each value of i is 1. The JSπ is maximum when the distributions are all maximally different one from another.
For the case of m-distributions p(x|Hi) through p(x|Hm) of X, the JRD is defined as:
JRα π (Pl,...,pm) H ,)) (28)
Figure imgf000021_0002
where π through πm are the weights attached to each distribution such that π ≥ 0 and the sum of πx for each value of i is 1. See
He Y, Hamza AB, and Krim H, "An information divergence measure for inverse synthetic aperture radar focusing, " IEEE Workshop on
Signals and Signal Processing, Singapore, August 2001, the entire disclosure of which is incorporated herein by reference. The JRD is nonnegative for 0 < α < 1, symmetric to the interchange of subscripts (unlike the relative entropy) , and equals zero only when all the probability p(x|Hχ) through p(x|Hm) are equal for all . Finally, as α→l the JRD tends to the JSD. And like the JSD, the JRD is maximum when the distributions are all maximally different one from another.
3. Divergence and Optimal Contours
The present invention is addressed to minimum error contouring based on either the Jensen-Shannon or Jensen-Renyi divergences. Referring back to Fig. 4, when object 162 (or region Rτ) is to be segmented from the background 170 (or region R0) , contour C is fit to the object boundary by finding the vector of coefficients for C which defines the closed curve that best matches the actual boundary. A family of such contours ..., c'1"1', C(l), C(l+1), ... might be generated during the course of an iterative search for the best C. The interior and exterior regions of the specific contour C(l) are designated Z^1' and Z0 {l) respectively.
First, the total distribution of pixel property X is defined as :
p(X) = ^p(X \ RJ)p( j) (29) j={l,0}
where p(x|Rj) is the conditional probability of observing data X given that it is associated with region Rj and p(Rj) is the probability of encountering region Rj (the prior probability) . See Grimmett GR and Stirzaker DR, Probability and Random Processes, 2nd Edition, Oxford University Press, 1992, the entire disclosure of which is incorporated herein by reference.
Likewise, the total distribution for the zones for an approximate contour C(1) can be written as:
p(X) = ∑p(X \ ZJ)p(Zj). (30)
J=V,0}
The full and complete description of the variation in the random variable X with respect to the set of zones {Zι,Z0} and regions {Rι,R0} is expressed in the total joint probability distribution p(X,Z,R) which may be written as the product of conditional probabilities . p(X,Z,R) = p(X\Z,RMZ\R)p(R) {31)
where p(x|Z,R) is the probability of observing X given both Z and R jointly, p(z|R) is the conditional probability of being at a point in zone Z given region R, and p(R) is the prior probability of being located in region R. The total distribution is formed over X alone by summing over the sets of zones and regions as :
p(X)= ∑ ∑p(X\ZJ,R,)p(ZJ\R,)p(Rl). (32) j={I, ) /={/,0}
For any contour C that is not equal to the true region boundary, the interior and exterior zones contain mixtures of both Rj and R0. That is, Z-, overlaps R so the conditional probability p(Z3|Rx) captures the extent of overlap of zone Z-, with region Ri. With improvement of the C approximation, the zones will approach the regions in size, shape, and pixel distribution, or as ZD-RX, j={l,0}, it will be true that p(Z-|R-) >> p(Z3|Rx) for i≠j . Since p(R_) is constant, the total distribution can be approximated by
PiX)" ∑p(X\ZJ)p(ZJ\ J)p( J) (33)
J={I,0)
As the contour C is fit to the true region boundary, the changes in the distribution p (X) will be a function only of the factors dependent on Z, p(x|Z-) and p(Z-|R:). Next, the LRT for two separate distributions (inside X!,X2,...,XQ e p(x|Zi) and outside X'ι,X'2,...,X'Q e p(x|z0) simultaneously as the joint probability distribution p(Xi,X2,...,XQ,X'i,X'2,...,X'Q) . The LRT for the combined distributions versus their opposing region-hypotheses is written:
Figure imgf000023_0001
A P(X,\ Zj )p(ZI\ R7 R ) Q p(X: \Z0)p(Z0\ R0 )p(R0 ) f p(X, I Z0)p(Z0 I R0)i?(R0)if PX I ZJ)p(Z1 I RI)p(R1) r QrP(X,\ZI) Q p(Xl'\Z0) p(Xl\Z0)i, p(Xl'\ZI)
(34) since the probabilities not involving X or X' can be factored out to give a product equal to 1. In log form,
Figure imgf000024_0001
(35) which is a function only of the zones and not the regions .
By derivation, the difference between relative entropies can be obtained. Rewriting (35) with the sums over the sets of values the random variables X,X' may take: lOg L(X j , X2 ,..., XQ , X , X2 ,..., XQ )
Figure imgf000024_0002
where the distributions px(x),px(x') are those of the random variables with respect the finite set of values they may assume. Multiplying each term by a factor of 1:
\ogL(Xr,X2,...,X Q,X1,X2,...,X Q)
Figure imgf000024_0003
(37) and rearranging:
\ogL(Xλ,X2,...,X Q,X1,X2,...,X Q)
Figure imgf000024_0004
YQpx,(x,')log P W -YQpx,(x,')log Pr(χ,, ^ ' *P&,\Zt) ^x 'J P(x,'\Z0) we obtain the sums and differences of relative entropy:
\ogL(X,X2,...,X Q,X,X ,...,X Q
= Q[D(px(x)\\p(x\Z0))-D(Px(x)\\p(x\ZI)) (39)
+ D(px,(x)\\ p(x\Z1))~D(px,(x)\\ p(x\Z0))]
The log-likelihood logL(.)is maximized when the distribution of X inside the contour C is most like the distribution X conditioned on being in the zone Zτ and simultaneously when the distribution of X' outside the contour C is most like the distribution X' conditioned on being in the zone Z0. This achieved for the simpler expression: c* =axgmaxD(p(X\ZI)\\p(X'\Z0)) . (4o) c
That is, the best contour C* is the one for which the distributions (x|zτ) and p(X' | Z0) are most different by the relative entropy measure. The corresponding maximum contour expressions for the Jensen-Shannon and the Jensen-Renyi divergences are stated by analogy to equation (40) . The Jensen-Shannon divergence corresponding to the best contour for the case of two regions is c*= argmaxJSΛ. c
= argmax[Hs(.rn 7 ' | Z0) + πxp(X | Z )) (4i) c
0Hs(p(X' \ Zo)) -%lHs(p(X \ ZI)) where Ηs ( . ) is the Shannon entropy defined in (13) . For m regions, or m-1 objects plus the background, the set of best contours Cx* through Cm-1* are those whose collective coefficients maximize the Jensen-Shannon divergence: {c.*,c2*- •c m-i*} = ars, maχ ysπ(Pl,...,pm)]
Z,))]
Figure imgf000025_0001
(42) where the πA are the weights on the probability distributions of the m regions . The contour subscripts match the index of the corresponding region.
The maximum-contour Jensen-Renyi divergence expression can be similarly written. For the case of two regions (Rx and R0) to be separated, the Jensen-Renyi divergence is maximized for the boundary contour with parameter vector c* which most perfectly separates the two regions :
Figure imgf000025_0002
= arg max[HΛα0p(X | Z0 ) + π p(X | Z7 )) (43) c
0HRa(p(X \ Zo))-πlHRa (p(X \ ZI))] where ΗRoc(.) is the Renyi entropy defined in (24) . For m regions, or (m-1) objects plus the background, the set of best contours Ci* through Cm-ι* are those whose collective coefficients maximize the Jensen-Renyi divergence:
{c1*,c2*,...,cffl_1
Zi))
Figure imgf000025_0003
(44) where the 7ti are the weights on the probability distributions of the m regions . The contour subscripts match the index of the corresponding region.
4. Autocontouring Program Based On JSD/JRD
The program using the objective functions (41) or (43) for single object contouring is described in Fig. 5. Multiple object contouring using objective functions (42) or (44) is described in Fig. 6. Both the JSD and the JRD can maximized by the same schemes, given in detail here.
The single contour program is initialized with an image (step 5.1) and an initial boundary estimate, preferably a polygon created either by the operator or by the program itself (step 5.2) . The polygon is converted to the initial pixel boundary vector b(0) with B0 pixel-contiguous elements b^01 ,b2 (0),...,bB0 <0) . Each element contains the row and column coordinates of the contour pixel as bj (0'=(row j, column j) . From these pixels, the coefficients for the initial parametric contour C(0> are computed (see (8) through (11)). Contour determination proceeds through two loops . The outer loop (steps 5.6-5.12) iterates over successive versions of the parametric contour C(0) ,C(1),..., C(l),... , stopping when the increases between succeeding values of the JSD (or JRD) fall below a preselected, threshold minimum T. So, if JS (;+1) - JS w < T (45) or
J TROα*(,'+1) ~ J TROα π( <_- T T ( ,46) then the program reports the contour ctl+1) and stops (step 5.13) . As would be appreciated by those of ordinary skill in the art, the choice of a value for T is up to the practitioner of the invention, with smaller values of T providing higher accuracy but requiring more time to compute . The inventor herein has found that a value in the range of 0.001 to 0.1 is preferable for T. The inner loop (steps 5.7-5.9) maximizes the JSD (or JRD) over the components of the current contour C(l) . This maximization can be done many ways (steepest descent, Quasi-Newton method, conjugate gradient descent, etc.), and preferably maximization is performed using Powell's direction set method or alternatively, the method of conjugate gradient descent with equivalently- accurate results. See Press, WH, Teukolsky, SA, Vettering, WT, and Flannery, BP, Numerical Recipes in C++, Second Edition, Cambridge University Press, Cambridge, UK, 2002; and Nocedal, J. and Wright, S. J. , Numerical Optimization , Springer-Verlag, New York, 1999, the entire disclosures of which are incorporated herein by reference. It is worth noting that although the methods' names indicate minimization, minimizing the negative of the function locates the function's maximum.
Through the argmax function (step 5.7), the inner loop maximizes the JSD (or JRD) by systematically varying the numeric values of the individual parameters Cι(l) ,c2 (x),..., c4K+2 !:L) , and noting the resulting values of JSπ u) or R π M , according to whichever of the maximization methods is used. At each inner-loop substep, trial (*) contours c(1,:i>* (i-th iteration contour, j'-th component being varied) are computed for each of the several values of the parameter C !l> (step 5.7). Each parametric contour is converted to its corresponding boundary b ,:i>*, pixels in the zones zX'^* and Zo'1'-''* are enumerated, the conditional probability distributions
Figure imgf000027_0001
and p(X| Zo<"j>*) are computed, and the JSD or JRD entropy terms and divergences JSπ (l,j)* or JR (1,;i>* are computed. The largest maximum JSπ ,:i)* or JRαπ (i,:i)* for all j is relabeled JSπ <1+1) or JRαπ (1+1) and submitted to the outer loop test, and the corresponding contour is saved as C(l+1) . While it is preferred that the random variable X used in the program correspond to pixel intensity, other pixel properties can be used. The order in which the contour coefficients are adjusted during optimization has an important effect on optimization. Like other Fourier representations, the low number coefficients affect the global properties while the higher coefficients represent the fine details. In the Fourier elliptic representation, contour coefficients 1 and 2 are the xy-centroid of the contour, coefficients 3-6 fix the overall size of the object, and all higher coefficients add successively finer details to the contour shape. Coefficient values for CT organs are largest for the centroid, and tend to have successively smaller values for increasing coefficient order. By optimizing the divergences over the coefficients in order, the global properties are fixed first which thereby produces an efficient and accurate approach to the maximum. Thus, as the program loops from step 5.7 through step 5.9 iteratively, the coefficient adjustments produce a series of new and monotonically increasing divergence values. The divergence value resulting after all coefficients have been adjusted is the divergence value that will be maximum and passed to the outer loop test.
Referring to Fig. 6, the program which computes multiple contours simultaneously using the objective functions (42) or (44) is initialized with an image and m-1 polygons created either by the operator or by the program itself (steps 6.1 and 6.2) . There are m regions including the common background region. Maximization of the JSD or the JRD for multiple object contours follows the same scheme as it does for single object contouring. The main difference between the optimization of multiple contours and a single contour is the re-ordering of the multi-contour coefficients for the optimization step.
The polygons are converted to a set of initial pixel boundary vectors bι (0>, b2 (0), ... , bm-1 (0> (step 6.3) from which the initial set of parametric contours Cι (0), C2 <0), ... , Cm-i<0> are computed (step 6.4). The subscripts refer to the regions/objects being contoured. The coefficient vectors for each contour are
C,. - (Ci ,-, Ci ι+2 ) (47) where the first subscript for each coefficient identifies the contour and the second subscript identifies the coefficient' s order. Also, each contour may have a different number of harmonics K .
Like the single contour program, optimization over a contour set has two nested loops. The outer loop (steps 6.6-6.12) iterates over optimized versions o"f the contour set,
Figure imgf000028_0001
c Cj (1) ,cl-2 (1) ,...c*-,„_!(1)
(2) _ (2)
'1 ..c (2) m-1
(48)
c Cj w ,c2 w ,...c*.„._!(
and the inner loop (steps 6.7-6.9) traverses all the coefficients in all the contours in the set. The convergence test for the outer loop is the same as that for the single contour program: stopping when the increases between successive values of the JSD (JRD) fall below a pre-selected, threshold minimum. So, if JSπ ) - JSπ ( < threshold (49) or
JR a 0+1) -JR« aπ (° < threshold (50) then the program reports the contour set C1 +1> through Cm-1 (l+1> as the matching contour set and stops (steps 6.11, 6.13).
For each pass of the outer loop, the inner loop maximizes the JSD (JRD) over all the contours' coefficients. The entire set of coefficients, ordered by contour, is
r c2,l(0 -> r c2AK2+2(0 > (51)
c/w-l,l -Xm-l Km-i+2 where the subscripts are indexes for contour and coefficient, respectively, and there are
Mpamms = ∑^ (AKi +2) (52) parameters in all contours . Because the Fourier elliptic representation orders the global parameters before the fine detail parameters, it is important to establish all the contours' global locations and sizes before determining the shape details for the contours. Therefore, we re-order the parameters in the set (51) by harmonic order as follows,
Figure imgf000029_0001
r C (0> r c2,3(0 >—> r cm-l,3(0'
cl,4K\+2 '->c2AK2+2 '-Xm-lAKm_ι+2 S
(53) and where the contours have different numbers of coefficients, the order is filled in harmonic order by the remaining coefficients in the largest contour sets. If we re-label the re-ordered contours c -( as singly-indexed parameters J in the set, ιrc cι(') 'c c 2« '--. ....-CΛ,-(') ,-, £cM (')ι i ,(5_4,
Figure imgf000029_0002
) we can describe the inner loop optimization just like that of the single contour program.
With reference to Fig. 3(c) and as noted above, a multiprocessor computer can be used to carry out the contouring program of the present invention. For example, as shown in Fig. 3(c), the image data set can be provided to a plurality of processors 150a. For computations that do not depend on each other, increased processing speed can be obtained by splitting a problem into multiple independent parts and running each part on a different processor 150a. This would be especially the case where multiple object contouring is run on m regions in an image data set. Independent functions here include enumerating the region pixels, assembling the pixel feature histograms and distributions, and computing the Jensen or Renyi entropies and their weights . The processing required for each region can be performed on a separate processor 150a in parallel. The results of the parallel processing can then be collated in another processor 150a that would perform the outer loop test.
5. Demonstrations Two sets of demonstrations are shown in Figs. 7-10. The first set shows the action of the JRD contouring single figures in a series of random shapes with defined signal-to-noise ratios (SNRs) . The second demonstration set shows the action of the JRD to contour anatomic objects in CT images . Because the Jensen-Renyi divergence method consistently produced lower errors than the Jensen-Shannon divergence method in tests on random figures (described below) only demonstrations of the Jensen-Renyi method will be shown here.
In the first demonstration of the action of the JRD, a estimation is performed for the contours in a series of random- shape figures defined by curves C(true) = {cι (true), c2 (true), ... , c4K+2 (true)} with known, but random, coefficients. This "true" boundary is designed to fill the center of a 256x256 image array, and to have somewhat sharper direction changes than usual in medical anatomy in CT. Figure 7 shows the process: panel A is a figure template corresponding to a random shape figure, and panel B shows the shape of panel A smoothed with a 3x3 moving window and with Gaussian noise added, to the level of a SNR of about 3.0. Panel C shows a circle that serves as the initial contour estimate, and panel D shows the matching contour (white line) computed by the JRD and overlaid on the figure.
Figure 8 shows a series of randomly sized and shaped figures, with the JRD matching contours (white lines) computed from the circle initial estimate, and with SNR of about 3.0.
To compare the computed estimate contours with the true curves, the error is defined as the average distance between corresponding points in the two curves. The true, defined contour c {true) j. ag a corresponding pixel boundary bltrue) which is a vector of B-pixel coordinates as defined b.(true) through bB (true) where bi<true> = (ro , columni) . The contour estimates C* obtained using the JSD or the JRD produce a trial pixel boundary b* with B* coefficients. To make the two curves b(true) and b* commensurate, a comparison is made between intervals of points on the two curves, selected to sample both curves uniformly. The true boundary is divided into W equal intervals with W = BIW pixels per interval.
For each W -pixel interval in b(true), a W -pixel interval in b* is selected, offset by a varying number of pixels in order to find that interval in b* closest to the fixed b(true> interval. The error is the average point-to-point distance for all points in all intervals
Figure imgf000031_0001
In tests, the b( rue> - b* errors associated with the JSD technique, the JRD technique, and the known relative entropy technique consistently follow the order: Error (JRD) < Error (JSD) < Error (Relative Entropy). For example, the images in Figures 7 and 8 have SNR ~ 3.5 and for the JSD/JRD contours have distance errors of 4-5 pixels. The relative entropy error distance for the same images is 5-7 pixels. Fuller quantization of these methods performance will follow shortly.
Figure 9 shows the action of JRD contouring the liver of a patient in a series of CT sections of the abdomen. The thin lines indicate user-input polygons forming the initial estimates of the contours, and the thick white lines are the JRD matching contours. These matching contours were computed using 6 Fourier harmonics . The maximization is that given in equation (43) .
Figure 10 shows examples of multiple objects contoured by the JRD, simultaneously in the same picture. The maximization is that given in equation (44) .
6. Additional Embodiments
The JSD/JRD image segmentation of the present invention can be based on multiple random variables rather than only a single random variable. JSD/JRD image segmentation can be applied to a random vector of properties, =(Xi, Xx, ... , Xd) in which each component Xi is a discrete random variable. Better estimates of contours might be obtained by using more information encoded in multiple random variables than using just one random variable. The extent to which several random variables contain more information than any one alone depends on whether they are drawn from independent distributions . Extending the JSD/JRD methods to random vectors requires consideration of two cases: 1) the random variates Xi are conditionally dependent, or 2) the random variates Xi are mutually independent . Dependence is the more general situation, with independence a special case of the conditionally dependent case.
The probability distribution of a random vector is simply the joint distribution over the separate component random variables (see Grimmett and Stirzaker (1992) ) p(X) = p(X1,X2,...,Xd)
= p(X - x ,X2 = x ,...,Xd = xd) (5g) where d is the dimension of the property vector. The distributions from which the individual variates are drawn may exhibit some inter-dependence, and this is expressed using the conditional probability, defined in ((1), (2)). The joint probability distribution may be written as a product of conditional probabilities, illustrated for these examples:
p(X1,X2) = p(X1 \ X2)p(X2) p(X1,X2,X3) = p(Xl \ X2,X3)p(X2,X3)
= p(Xl \ X2,X3)p(X2 \ X3)p(X3)
p(Xl,X2,..,Xd) = p(Xl \ X2,...,Xd)p(X2,.,Xd)
= p(Xλ \ X2,...,Xd)p(X2 \ X3,...,Xd)p(X3,...,Xd)
= p(Xl \ X2,...,Xd)p(X2 \ X3,...,Xd)-p(Xd). The j oint probability distribution can be written more compactly as the product formula d p(Xi,X2,...,Xd) = Y p(Xi l Xi+l,Xi+2,...,Xd)
*=1 . (57)
If any one of the individual random variables Xi is from a distribution independent of the distributions of other random variables Xj , j ≠ i, the conditional distribution of Xi relative to other variables is simply equal to its probability, without conditions. That is, p(Xi \ ...,XJ,...) = p(Xi) (58) for all Xj , j ≠ i. In the case where all the random variables are mutually independent, the joint distribution is the product of the individual variates' probabilities, d p(X) = p(X1,X2,...,Xd) = Y[p(Xi)
'=1 . (59)
If the variates are truly independent, not conditionally dependent, the decision rules for the relative entropy/LRT ((23), (40)5.12), the JSD ((41), (42) 5.14) and the JRD ((43), (44) 5.15, 5.16) could simply be implemented as d parallel classifications with a rule to combine the separate decisions (see Devroye, 1996) . The multivariate LRT-equivalent relative energy decision rule from (23) is for the i-th variate
If D(px(Xt)|| p(Xt I H2 )-D(P (Xi)|| p(Xt \ H j) > T, decide Hl5 else H2. and the overall decision for Ηlf H2 going to the outcome having the largest number of individual variate-decisions. In the case of even numbers of variates, tie votes would be always given to one outcome, say Hx. The multivariate relative entropy contour function (40) is rewritten c* = argmaxE>(^(X| Z1) || X' | Z0)) c . (61) where X, X' are the random property vectors for pixels inside and outside the contour c* . Rewriting the expression for the relative entropy, c = arg max c Σ X x)io_- Pg( j If z0 i)
Figure imgf000034_0001
which is the sum of individual-variate relative entropies,
Figure imgf000034_0002
So the best contour c* is the one for which the distributions p(Xi|z1), p(Xi|Z0) are most different by their contributions to the sum of the individual-variate relative entropies.
The Jensen-Shannon divergence for multiple independent variates begins with a reconsideration of the equation for estimating a single contour (41) in which the multivariate random vectors X, X' have been substituted, c* = argma JS^. c
= arg max[Hs0p(X | Z0) + πlP(X | Z^) c
-.τ0HsQ.(XΗ Z0)) -πιH5(/.(X| Z1)) ( 63 )
Recall that the Shannon entropy for the distribution of a random variable X is
Hs(p) = - ∑p(X)\ogp(X) xsX (64) where the sum is taken over the domain of values on which X is defined. For the random vector X with independent random variable components, the entropy is
Hs(p(X)) = -YJYj.- jp(Xl)p(X2 -p(Xd)\ gp(Xl)p(X2 -p(Xd)
Xχ X2 Xd
(65) where the sums are taken over the domains of the respective component random variables . Since the logarithm of a product is equal to the sum of the logs of the factors,
Hs(p(X)) = -∑∑-∑p(X )p(X2)-p(Xd)∑\ogp(Xi)
X\ *2 Xd 1=1 (66) and since independent variates sum only over their own domains, equation (66) can be rearranged to give
Hs(p(X)) = ∑ -∑p(Xi)logp(Xi)
1=1 Xi
= ∑Hs(p(Xt)) ι=l (67) where the entropy of the random vector is the sum of the independent, component variables' entropies. We can use (67) to rewrite equation (64) as follows,
c* =
Figure imgf000035_0001
0∑Hs(p(Xi' I Z0))-π1∑Hs(p(Xi \ Zx)) i=\ i=\
(68) That is, the best contour estimate is obtained by the parameter set c for which the expression in the brackets is a maximum.
By analogy, we can write the equation corresponding to the multi-contour estimate given by equation (42) as
{c1*,c2*,...cM_1*} = arg, max [JSff (j.j (X)-...,j!.M (X))]
Figure imgf000035_0002
(69) where the M regions are the background and the interiors for the M-1 contours. Using (67) we obtain an expression in terms of the individual random variables Xi
{c1*,c2*...-cM_1*} = arg
Figure imgf000035_0003
(70) The Jensen-Renyi divergence has a similar development . Starting with the definition of the Renyi entropy of degree α from (24)
H Ra (p) = - 1 ^—l gΫp(x - a sv ir α>0, α≠l.
I *—•
'=» (71) we substitute the random vector for the scalar random variable to obtain
Figure imgf000036_0001
the log of a product of sums since the variables Xi are independent. This can be simplified to d
HRa(p(X)) = -^-∑log∑p(Xi)a l~a xt
Figure imgf000036_0002
= ∑HRa (P(Xi)) '=1 (73) where again the inner sum is over the domain of random variable Xi. Like the Shannon entropy, the Renyi entropy of a random vector is a sum of entropies of the independent, component random variables . Now we consider the single contour JRD decision rule function given in equation (43) , here substituted with the random vector X c*= argmaxJR^(jp1,jp2) c
= arg max[HRa0p(X | Z0) + ωπp (X \ Zλ)) c
Figure imgf000036_0003
and using (73) , obtain an expression for the optimal contour on the separate independent variates, c* = argmaxJR θ1- j?.2)
Figure imgf000037_0001
^∑HRa(P(Xi I Z0))-πι∑HRa (p(Xt I Zj))
.=1 .=1 ( 75 )
The JRD applied to M-1 simultaneous contours results when substituting the random vector X into equation (44)
{Cj*,c2*,...,cM_j*} = arg max [jRζ (pl ,..., pM))\
{c l cM-l)
Figure imgf000037_0002
( 76 ) and using (73) we obtain an expression for the multiple simultaneous contours in terms of the independent variates,
{Cj*,c2 *,..-,c _,*}
Figure imgf000037_0003
This is the expression one would use to compute the M-1 contours based on the maximum of the JRD over the d features in each of the M regions .
In summary, it has been shown here the numerical formulas needed to compute contour estimates for the JSD/JRD methods, on image pixel properties represented as several random variables . The method and the accompanying formula is given in the following table.
Figure imgf000037_0004
These are the formulas which, implemented in computer programs, can be used to compute contour estimates using multiple image pixel properties
The above discussion of contour estimation using multiple random variables assumed that the variables were all statistically independent, or that the joint distribution is the product of the individual variate' s unconditioned probabilities (equation (59)). The more general situation is given by equation (57) in which the variates are allowed to be from distributions which may be conditionally dependent. For the two variates Xi and X2, the joint probability p(Xι,X2) is related to the conditional probabilities by definition p(X1,X2) = p(X1 \ X2)p(X2) = p(X2,X,) = p(X2 \ Xl)p(Xl). (?8)
The relative entropy for two distributions of these two random variables, p(Xι,X2) and q(Xι,X2) is given by
D(p(Xl,X2)|| q(Xl,X2)) = ∑∑p(X1,X2)\og * ' 2l
X X <1\Λ\>Λ2)
Figure imgf000038_0001
which is equivalent to the sum of relative entropies,
Figure imgf000038_0002
or more compactly,
D(p(X ,X2) II q(Xλ,X2)) = D(p(Xl | X2) || q(Xx \ X2))D(p(X2) \\ q(X2)) (81) which is equivalent to
D(p(X2,Xl)II q(X2,Xλ)) = D(p(X2 \ Xλ) \\ q(X2 | Xl))D(p(Xl)|| q(Xλ)) { Q2 )
Computing these relative entropies can be done by literally using equation (80) to form the sums of ratios across the rows or columns of the joint distributions p(Xι,X2) and q(Xx,X2) . This is considerably more complicated than computing the sum of relative entropies for the independent variates given in equation (62) , and would have to be repeated for each iteration of the search for the contour's coefficients. And the complications multiply as one adds more random variables to the problem. However, given the speed and memory of current computers, such computations are only lengthy, not impossible. The JSD/JRD image segmentation technique of the present invention is also applicable to non- stationary data. "Non-stationary" applies to pixel data in which random variable means, variances, and distributions vary across the image of a single logical object. Classifiers which depend on a fixed mean and variance (e.g., the multivariate Gaussian model described in Section 2) to sort pixels into one or another objects pictured can produce higher error rates for non-stationary image data.
Since the JSD/JRD methods use functions of the random variables' distributions, it is possible that these methods may be more robust than parametric models since the JSD/JRD functions operate by portioning the data by distribution differences, regardless of any parameters that might be used to describe the distributions. During iterative region contour estimation, non- stationary data would produce small, continuous changes in the region data distributions, and such changes need not increase the error in the JSD/JRD autosegmentation.
Further still, the JSD/JRD image segmentation discussed above was a region-centered approach in which the decision rules operated on distributions of pixel properties . The teachings of the present invention are also applicable to an edge-centered approach in which the decision rules operate on intensity gradients . Edges represented by coheren , high-valued gradient- image pixels should maximize a decision rule that superimposes the contour C* on gradient maxima pixels . Edges contain the highest values in the gradient image so a rule which seeks to maximize the difference between edge-pixel distributions and all the other pixels should force the contour onto the gradient image maxima.
Thus, the present invention can be used in combination to (1) partition an object from a background (as previously described) and (2) separate the edge of the object, as observed in an image of the gray level gradient maxima (which is derived from the original image data set) by maximizing the JSD/JRD divergence between the pixels underneath contour C1 and all other pixels in the gradient image. The center panel of Fig. 1 illustrates an example of a gradient image. Because the set of pixels coincident with an edge will necessarily have a distribution different from that of the gradient image background plus other object edges, maximizing the divergence over these two sets of gradient image pixels should add to the specificity of the boundary estimate produced in the regional divergence estimation. Due to the fact that images are usually never simple, it is highly likely that the regional divergence and the edge divergence would be maximized for different C1 contours . In such cases, the "matching contour" can be defined as the contour that jointly maximizes the combined divergences. Because the JSD/JRD divergences take on arbitrary real values determined by the data at hand, the regional and edge divergences for any given region in any given image might have very different values, thereby requiring a weighting to make them equivalent . Such a function would be expressed as :
F(JSD) = wregionJSDregion (C )+ wedgeJSDedge (C ) ( 83 ) or
F(JRD) = wregionJRDregion {c )+ wedgeJRDedge (σ ) (84) wherein wregion and wege are real number coefficients that can be set to equalize the contributions of the region and edge to the total divergence .
Further, combining separate JSD/JRD methods for the region pixel data and for edge pixel data could result in more accurate estimates of image object boundaries. Also, restrictions on the pixel-classes segmented in the multiple, simultaneous segmentations by the JSD (42) and the JRD (44) might be used to obtain segmentations of like organs along with unlike organs. For instance, the kidneys might share a single distribution, so the rule would differentiate them collectively from the spleen, liver, stomach, or other organs. This is a kind of prior information which could improve the performance any above-described method.
Moreover, the present invention is also applicable to images having more than 2 dimensions . The present invention should work equally well in 3D as 2D, but would require flexible surfaces or volumes instead of a contour.
While the present invention has been described above in relation to its preferred embodiment, various modifications may be made thereto that still fall within the invention's scope, as would be recognized by those of ordinary skill in the art. For example, the invention has been described wherein the contouring program adjusts all coefficients of a contour in finding a maximal divergence value. However, given that the very high order coefficients have a relatively insignificant effect on the resulting contour (they add only very fine details) , a practitioner of the present invention can modify the contouring program to adjust the numeric values of less than all of a contour's coefficients. While some accuracy would be lost, less processing power would be needed. Such modifications to the invention will be recognizable upon review of the teachings herein. As such, the full scope of the present invention is to be defined solely by the appended claims and their legal equivalents .

Claims

What is claimed is :
1. A method of identifying the boundary of an object in an image, the image being represented by a data set, the data set comprising a plurality of data elements, each data element having a data value corresponding to a feature of at least part of the image, the method comprising; adjusting a boundary estimate at least partially according to one of the group consisting of a Jensen-Shannon divergence value corresponding to the boundary estimate and a Jensen-Renyi divergence value corresponding to the boundary estimate such that the boundary estimate substantially matches the object boundary.
2. The method of claim 1 wherein the divergence value represents a measure of the difference between the data values of the data elements located in a zone ZΣ and the data values of the data elements located in a zone Z0, Zτ and Z0 being defined by the boundary estimate and representing, respectively, the data elements of the image located inside and outside the boundary estimate.
3. The method of claim 2 wherein each data value is any member of the data value set X' , X' comprising a plurality of possible data values Xi through xn, and wherein each data value x^ is the value of a random variable X drawn from X' according to a probability law p (Xi) =Pr{x=Xi} , each zone having a corresponding probability law such that the probability of observing X with the specific value x± in zone Z0 is the conditional probability p(xi|z0) and the probability of observing X with the specific value Xi in zone Zτ is the conditional probability p (x | Zx) , and wherein the divergence value represents a measure of the difference between a conditional probability distribution (x|Zϊ) and a conditional probability distribution p(x|z0) .
4. The method of claim 3 wherein X' comprises a finite plurality of possible data values.
5. The method of claim 3 wherein the boundary estimate is a contour C.
6. The method of claim 5 further comprising: determining a contour that represents an approximation of the object boundary; calculating one of the group consisting of the Jensen- Shannon divergence value and the Jensen-Renyi divergence for the contour; and wherein the adjusting step comprises adjusting the approximate contour so as to increase its corresponding calculated divergence value.
7. The method of claim 6 further comprising: receiving input from a user or from a programmed method corresponding to a polygon that represents an initial estimate of the object boundary; mapping the received input to the image data set to thereby determine a pixel boundary corresponding to the polygon; and wherein the approximate contour determining step comprises calculating the approximate contour from the determined pixel boundary.
8. The method of claim 6 wherein the divergence value calculating step comprises : identifying which data elements are located in Zτ; identifying which data elements are located in Z0; calculating the conditional probability distribution
calculating the conditional probability distribution p(x|z0) ; and calculating the divergence value between p(x|ZT) and p(x|z0).
9. The method of claim 6 wherein the adjusting step further comprises adjusting the approximate contour so as to substantially maximize its calculated divergence value.
10. The method of claim 9 wherein each contour is a parametric contour C defined by functions of a plurality N of ordered contour coefficients cx through cN, each contour coefficient C having a numeric value, and wherein the adjusting step further comprises adjusting the numeric value of at least one contour coefficient
11. The method of claim 10 wherein the adjusting step further comprises determining a numeric value for Ci of the approximate contour for which the divergence value of the approximate contour is substantially maximum.
12. The method of claim 11 wherein the adjusting step comprises performing the numeric value determining step for Ci before performing a numeric value determining step for Ci+X.
13. The method of claim 10 wherein the adjusting step further comprises individually adjusting the numeric values of at least a plurality of approximate contour coefficients starting from cx and continuing through successively higher order approximate contour coefficients until the approximate contour' s divergence value is substantially maximized.
14. The method of claim 13 wherein the individually adjusting step comprises individually adjusting the numeric values of each approximate contour coefficient starting from Ci and continuing through successively higher order approximate contour coefficients to c until the approximate contour's divergence value is substantially maximized.
15. The method of claim 13 wherein the coefficients correspond to the terms of a finite-length trigonometric series, wherein the number N of the contour coefficients is 4K+2 , and wherein K is a programmably-determined number of harmonics of the trigonometric series.
16. The method of claim 2 wherein the divergence value is a Jensen-Shannon divergence value.
17. The method of claim 2 wherein the divergence value is a Jensen-Renyi divergence value.
18. The method of claim 2 wherein the image is a medical image of an interior portion of a body, wherein the data set comprises a plurality of pixels, each pixel having a pixel value corresponding to image intensity.
19. The method of claim 18 wherein the medical image is an X-ray computed tomography (CT) image.
20. The method of claim 18 wherein the medical image is an X-ray image .
21. The method of claim 18 wherein the medical image is a magnetic resonance image.
22. The method of claim 2 wherein the image comprises a plurality of objects, each object having a boundary, the method further comprising performing the adjusting step simultaneously for at least two of the image objects.
23. The method of claim 2 wherein the data set is a two- dimensional data set.
24. The method of claim 2 wherein each data element has a plurality of data values corresponding to different features of at least part of the image.
25. The method of claim 24 wherein the data values for each data element are conditionally dependent.
26. The method of claim 24 wherein the data values for each data element are mutually independent.
27. An apparatus for identifying the boundary of an object in an image, the image being represented by a data set, the data set comprising a plurality of data elements, each data element having a data value corresponding to a feature of at least part of the image, the apparatus comprising; a processor configured to adjust a boundary estimate at least partially according to one of the group consisting of a Jensen-Shannon divergence value corresponding to the boundary estimate and a Jensen-Renyi divergence value corresponding to the boundary estimate such that the boundary estimate substantially matches the object boundary.
28. The apparatus of claim 27 wherein the divergence value represents a measure of the difference between the data values of the data elements located in a zone Zτ and the data values of the data elements located in a zone Z0, Zi and Z0 being defined by the boundary estimate and representing, respectively, the data elements of the image located inside and outside the boundary estimate.
29. The apparatus of claim 28 wherein each data value is any member of the data value set X' , X' comprising a plurality of possible data values Xi through xn, and wherein each data value Xi is the value of a random variable X drawn from X' according to a probability law p (xi) =Pr{x=Xi} , each zone having a corresponding probability law such that the probability of observing X with the specific value Xi in zone Z0 is the conditional probability p(Xi|z0) and the probability of observing X with the specific value Xi in zone Zτ is the conditional probability p(Xi|zI), and wherein the divergence value represents a measure of the difference between a conditional probability distribution p(x|zt) and a conditional probability distribution p(x|z0) .
30. The apparatus of claim 29 wherein X' comprises a finite plurality of possible data values.
31. The apparatus of claim 29 wherein the boundary estimate is a contour C.
32. The apparatus of claim 31 wherein the processor is further configured to: determine a contour that represents an approximation of the object boundary; calculate one of the group consisting of the Jensen-Shannon divergence value and the Jensen-Renyi divergence for the contour; and adjust the approximate contour so as to increase its corresponding calculated divergence value.
33. The apparatus of claim 32 wherein the processor is further configured to: receive input from a user or from a programmed method corresponding to a polygon that represents an initial estimate of the object boundary; map the received input to the image data set to thereby determine a pixel boundary corresponding to the polygon; and determine the approximate contour by calculating the approximate contour from the determined pixel boundary.
34. The apparatus of claim 32 wherein the processor is configured to calculate the divergence value by: identifying which data elements are located in Zτ; identifying which data elements are located in Z0; calculating the conditional probability distribution p {.x \ zτ) ; calculating the conditional probability distribution p (X I Z0) ; and calculating the divergence value between p(x|zx) and P(X|Z0) .
35. The apparatus of claim 32 wherein the processor is further configured to adjust the approximate contour so as to substantially maximize its calculated divergence value.
36. The apparatus of claim 35 wherein each contour is a parametric contour C defined by functions of a plurality N of ordered contour coefficients Ci through cN, each contour coefficient Ci having a numeric value, and wherein the processor is further configured to adjust the approximate contour by adjusting the numeric value of at least one contour coefficient
37. The apparatus of claim 36 wherein the processor is further configured to adjust the approximate contour by determining a numeric value for cx of the approximate contour for which the divergence value of the approximate contour is substantially maximum.
38. The apparatus of claim 37 wherein the processor is further configured to adjust the approximate contour by performing the numeric value determining step for Ci before performing a numeric value determining step for ci+x.
39. The apparatus of claim 36 wherein the processor is further configured to adjust the approximate contour by individually adjusting the numeric values of at least a plurality of approximate contour coefficients starting from cx and continuing through successively higher order approximate contour coefficients until the approximate contour' s divergence value is substantially maximized.
40. The apparatus of claim 39 wherein the processor is further configured to adjust the approximate contour by individually adjusting the numeric values of each approximate contour coefficient starting from Ci and continuing through successively higher order approximate contour coefficients to cN until the approximate contour's divergence value is substantially maximized.
41. The apparatus of claim 39 wherein the coefficients correspond to the terms of a finite-length trigonometric series, wherein the number N of the contour coefficients is 4K+2 and wherein K is a programmably-determined number of harmonics of the trigonometric series .
42. The apparatus of claim 28 wherein the divergence value is a Jensen-Shannon divergence value.
43. The apparatus of claim 28 wherein the divergence value is a Jensen-Renyi divergence value.
44. The apparatus of claim 28 wherein the image is a medical image of an interior portion of a body, wherein the data set comprises a plurality of pixels, each pixel having a pixel value corresponding to image intensity.
45. The apparatus of claim 44 wherein the medical image is a computed tomography (CT) image.
46. The apparatus of claim 44 wherein the medical image is an X- ray image.
47. The method of claim 44 wherein the medical image is a magnetic resonance image.
48. The apparatus of claim 28 wherein the image comprises a plurality of objects, each object having a boundary, and wherein the processor is further configured to perform the contour adjustment simultaneously for at least two of the image objects.
49. The apparatus of claim 28 wherein the data set is a two- dimensional data set.
50. The apparatus of claim 28 wherein each data element has a plurality of data values corresponding to different features of at least part of the image.
51. The apparatus of claim 50 wherein the data values for each data element are conditionally dependent.
52. The apparatus of claim 50 wherein the data values for each data element are mutually independent.
53. The apparatus of claim 28 wherein the processor is a workstation.
54. The apparatus of claim 28 wherein the processor is a digital signal processor (DSP) .
55. The apparatus of claim 28 wherein the processor is an application-specific integrated circuit (ASIC) .
56. The apparatus of claim 28 wherein the processor is a multiprocessor computer.
57. The apparatus of claim 28 further comprising a medical imaging device in communication with the processor for generating the data set.
58. A method of approximating the boundary of an object in an image, the image being represented by a data set, the data set comprising a plurality of data elements, each data element having a data value corresponding to a feature of at least part of the image, the method comprising: determining which one of a plurality of contours most closely matches the object boundary at least partially according to a divergence value for each contour, the divergence value being selected from the group consisting of Jensen-Shannon divergence and Jensen-Renyi divergence.
59. The method of claim 58 wherein each contour C1 defines a zone Zj1 and a zone ZQ X, Zx representing the data elements inside the contour and Zo1 representing the data elements outside the contour, each zone having a corresponding probability distribution of data values for the data elements therein, and wherein the divergence value for each contour C1 represents a measure of the difference between the probability distributions for the zones x
60. The method of claim 59 wherein each data value is any member of the data value set X' , X' comprising a plurality of possible data values xx through xn, and wherein each data value is the value of a random variable X drawn from X' according to a probability law p (X) =Pr{x=Xi} , each zone having a corresponding probability law such that the probability of observing X with the specific value X£ in zone Z0 is the conditional probability p(xi|z0) and the probability of observing X with the specific value Xi in the zone Zτ is the conditional probability p(xi|zt), the method further comprising: determining a first contour C1; calculating a conditional probability distribution i, \ Z-^) ; calculating a conditional probability distribution p(X|z0 1); calculating the difference between p(x|zI 1) and p(x|z0 1) according one of the group consisting of the Jensen-Shannon divergence and the Jensen-Renyi divergence; determining a second contour C2; calculating a conditional probability distribution p(x|Z! 2); calculating a conditional probability distribution p(x|z0 2); calculating the difference between p(X|zx 2) and p(x|z0 2) according to the one of the group consisting of Jensen-Shannon divergence and Jensen-Renyi divergence that was selected in the first difference calculating step; and wherein the match determining step comprises selecting the contour, C1 or C2, having a larger divergence value as the contour that most closely matches the object boundary.
61. The method of claim 60 wherein each contour C1 is defined by a plurality N of ordered contour coefficients cx through c,,1, each contour coefficient having a numeric value, and wherein the C2 determining step comprises creating C2 from C1 by adjusting the numeric value of a C1 contour coefficient .
62. The method of claim 61 wherein the C2 determining step comprises creating C2 from C1 by adjusting the numeric value of each C1 contour coefficient such that the divergence value of C2 is greater than the divergence value for C1.
63. The method of claim 62 further comprising repetitively performing the method steps until the difference between divergence values for
C1 and C2 is less than a programmably-defined threshold, and wherein C2 becomes C1 on each new iteration of the method steps .
64. The method of claim 63 wherein, on a first iteration of the method steps, the C1 determining step comprises creating C1 from a user-specified pixel boundary.
65. The method of claim 63 wherein, on a first iteration of the method steps, the C1 determining step comprises creating C1 from a programmably-defined pixel boundary.
66. The method of claim 59 wherein the image is a medical image of an interior portion of the body.
67. The method of claim 59 wherein each difference calculating step comprises calculating the Jensen-Shannon divergence value.
68. The method of claim 59 wherein each difference calculating step comprises calculating the Jensen-Renyi divergence value.
69. The method of claim 59 wherein each data element has a plurality of data values corresponding to different features of the image.
70. The method of claim 69 wherein the data values for each data element are conditionally dependent.
71. The method of claim 69 wherein the data values for each data element are mutually independent .
72. A method of segmenting an object from its background in an image, the image being represented by a data set, the data set comprising a plurality of data elements, each data element having a data value corresponding to a feature of at least part of the image, the method comprising: associating the data elements of the data set with either a zone Zτ or a zone Z0, Zx being an estimation of the object in the image, Z0 being an estimation of the background in the image; and adjusting Zτ and Z0 at least partially according to one of the group consisting of a Jensen-Shannon divergence value and a Jensen-Renyi divergence value such that Zx substantially matches the object.
73. The method of claim 72 wherein the divergence value is a measure of the divergence between a probability distribution of the data values of the data elements associated with Zτ and a probability distribution of the data values of the data elements associated with Z0.
74. The method of claim 73 wherein each data value is any member of a data value set X' , X' comprising a plurality of possible data values xx through xn, and wherein each data value i is the value of a random variable X drawn from X' according to a probability law p (Xi) =Pr{X=Xi} , each zone having a corresponding probability law such that the probability of observing X with the specific value X£ in zone Z0 is the conditional probability p(xi|z0) and the probability of observing X with the specific value X£ in zone Zx is the conditional probability p (x \ zx) , and wherein the method further comprises : calculating a conditional probability distribution p {x \ zx) ; calculating a conditional probability distribution p(x|z0); calculating the divergence value as the divergence between
Figure imgf000052_0001
75. The method of claim 74 wherein the adjusting step comprising adjusting Zx and Z0 such that the calculated divergence value is substantially maximum.
76. The method of claim 75 further comprising: determining a boundary estimate, the boundary estimate defining zτ and Z0; and wherein the adjusting step comprises adjusting the boundary estimate.
77. The method of claim 76 wherein the boundary estimate is a parametric contour C comprising a plurality N of ordered contour coefficients c through cN/ each contour coefficient having a numerical value, and wherein the adjusting step comprises adjusting the contour coefficients so as to maximize the divergence value for C.
78. The method of claim 77 wherein the adjusting step comprises first adjusting only contour coefficient cx until the calculated divergence value for the adjusted contour is substantially maximum.
79. The method of claim 78 wherein the adjusting step further comprises next adjusting only contour coefficient c2 until the calculated divergence value for the adjusted contour is substantially maximum.
80. The method of claim 79 wherein the adjusting step further comprises performing the contour coefficient adjusting step for successively higher ordered contour coefficients until the calculated divergence value does not differ from a previous calculated divergence value by more that a predetermined threshold value.
81. The method of claim 73 wherein the divergence value is the Jensen-Shannon divergence value.
82. The method of claim 73 wherein the divergence value is the Jensen-Renyi divergence value.
83. The method of claim 73 wherein the image is a medical image.
84. A method of segmenting an object in an image through optimization of a segmentation estimate, the image being represented by a data set, the data set comprising a plurality of pixels, each pixel having a value corresponding to a feature of at least part of the image, each pixel value being any member of the pixel value set X' , X' comprising a plurality of possible pixel values xx through xn, the segmentation estimate defining a zone Z! representing a set of pixels corresponding to the object and a zone Z0 representing a set of pixels not corresponding to the object, wherein a plurality of pixels are located in Zτ and a plurality of pixels are located in Z0, the method comprising using one of the group consisting of (1) a Jensen-Shannon divergence value representing the difference between a probability distribution p(x|zτ) and a probability distribution p(X|z0) and (2) a Jensen-Renyi divergence value representing the difference between a probability distribution p(x|Zj) and a probability distribution p(x|z0) as a criterion for optimizing the segmentation estimate to substantially match the object boundary.
85. The method of claim 84 wherein the divergence value criterion is the Jensen-Shannon divergence value.
86. The method of claim 84 wherein the divergence value criterion is the Jensen-Renyi divergence value.
87. The method of claim 84 wherein the image is a medical image of an interior portion of the body.
88. The method of claim 84 wherein the segmentation estimate corresponds to a contour C that represents an approximation of the object boundary.
89. A combination region-based and edge-based method of approximating the boundary of an object in an image, the image being represented by a data set, the data set comprising a plurality of pixels, each pixel having a pixel value corresponding to a feature of the image, each pixel value being any member of a pixel value set X' , X' comprising a finite plurality of possible pixel values, the method comprising: maximizing one of the group consisting of a total Jensen- Shannon divergence corresponding to a boundary estimate and a total Jensen-Renyi divergence value corresponding to a boundary estimate, wherein the total divergence value represents both a regional divergence value and an edge divergence value, wherein the total divergence value is a sum of the regional divergence value multiplied by a regional weight and the edge divergence value multiplied by an edge weight, and wherein the boundary estimate for which the total divergence value is maximized represents a matching boundary .
PCT/US2003/023685 2002-08-02 2003-07-29 Image segmentation using jensen-shannon divergence and jensen-renyi divergence WO2004013811A2 (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
AU2003256972A AU2003256972A1 (en) 2002-08-02 2003-07-29 Image segmentation using jensen-shannon divergence and jensen-renyi divergence

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
US10/211,779 US7187800B2 (en) 2002-08-02 2002-08-02 Method and apparatus for image segmentation using Jensen-Shannon divergence and Jensen-Renyi divergence
US10/211,779 2002-08-02

Publications (2)

Publication Number Publication Date
WO2004013811A2 true WO2004013811A2 (en) 2004-02-12
WO2004013811A3 WO2004013811A3 (en) 2004-09-16

Family

ID=31187654

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/US2003/023685 WO2004013811A2 (en) 2002-08-02 2003-07-29 Image segmentation using jensen-shannon divergence and jensen-renyi divergence

Country Status (3)

Country Link
US (1) US7187800B2 (en)
AU (1) AU2003256972A1 (en)
WO (1) WO2004013811A2 (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109643485A (en) * 2016-12-30 2019-04-16 同济大学 A kind of urban highway traffic method for detecting abnormality

Families Citing this family (62)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7466848B2 (en) * 2002-12-13 2008-12-16 Rutgers, The State University Of New Jersey Method and apparatus for automatically detecting breast lesions and tumors in images
GB2398379A (en) * 2003-02-11 2004-08-18 Qinetiq Ltd Automated digital image analysis
WO2004114063A2 (en) * 2003-06-13 2004-12-29 Georgia Tech Research Corporation Data reconstruction using directional interpolation techniques
US20060056701A1 (en) * 2004-03-02 2006-03-16 Gozde Unal Joint segmentation and registration of images for object detection
SE0400731D0 (en) * 2004-03-22 2004-03-22 Contextvision Ab Method, computer program product and apparatus for enhancing a computerized tomography image
US7369645B2 (en) * 2004-06-21 2008-05-06 Derek Graham Lane Information theoretic inverse planning technique for radiation treatment
US7519209B2 (en) * 2004-06-23 2009-04-14 Vanderbilt University System and methods of organ segmentation and applications of same
US7421415B2 (en) * 2004-09-07 2008-09-02 Siemens Corporate Research, Inc. Methods and systems for 3D object detection using learning
US7430321B2 (en) * 2004-09-09 2008-09-30 Siemens Medical Solutions Usa, Inc. System and method for volumetric tumor segmentation using joint space-intensity likelihood ratio test
US7697755B2 (en) * 2004-09-29 2010-04-13 Drvision Technologies Llc Method for robust analysis of biological activity in microscopy images
US7711172B2 (en) * 2005-04-08 2010-05-04 Siemens Medical Solutions Usa, Inc. Automatic nodule measurements
US7590273B2 (en) * 2005-04-08 2009-09-15 Siemens Corporate Research, Inc. Estimation of solitary pulmonary nodule diameters with a hybrid segmentation approach
US7720271B2 (en) * 2005-04-08 2010-05-18 Siemens Medical Solutions Usa, Inc. Estimation of solitary pulmonary nodule diameters with reaction-diffusion segmentation
US7860344B1 (en) 2005-05-06 2010-12-28 Stochastech Corporation Tracking apparatus and methods using image processing noise reduction
US7813581B1 (en) 2005-05-06 2010-10-12 Fitzpatrick Ben G Bayesian methods for noise reduction in image processing
US7400755B2 (en) * 2005-06-02 2008-07-15 Accuray Incorporated Inverse planning using optimization constraints derived from image intensity
US7880154B2 (en) 2005-07-25 2011-02-01 Karl Otto Methods and apparatus for the planning and delivery of radiation treatments
US20080314395A1 (en) 2005-08-31 2008-12-25 Theuniversity Of Virginia Patent Foundation Accuracy of Continuous Glucose Sensors
US7603000B2 (en) * 2005-08-31 2009-10-13 Siemens Medical Solutions Usa, Inc. System and method for learning relative distance in a shape space using image based features
US8176414B1 (en) * 2005-09-30 2012-05-08 Google Inc. Document division method and system
KR100698845B1 (en) * 2005-12-28 2007-03-22 삼성전자주식회사 Method and apparatus for editing image using algorithm of extracting personal shape
US20070153091A1 (en) * 2005-12-29 2007-07-05 John Watlington Methods and apparatus for providing privacy in a communication system
US20070160274A1 (en) * 2006-01-10 2007-07-12 Adi Mashiach System and method for segmenting structures in a series of images
US20070263915A1 (en) * 2006-01-10 2007-11-15 Adi Mashiach System and method for segmenting structures in a series of images
US7907772B2 (en) * 2006-03-30 2011-03-15 Accuray Incorporated Delineation on three-dimensional medical image
US7653425B2 (en) 2006-08-09 2010-01-26 Abbott Diabetes Care Inc. Method and system for providing calibration of an analyte sensor in an analyte monitoring system
US7737972B2 (en) * 2006-04-13 2010-06-15 Varian Medical Systems, Inc. Systems and methods for digital volumetric laminar tomography
US20080260229A1 (en) * 2006-05-25 2008-10-23 Adi Mashiach System and method for segmenting structures in a series of images using non-iodine based contrast material
US7773807B2 (en) * 2006-08-29 2010-08-10 Siemens Medical Solutions Usa, Inc. Seed segmentation using l∞ minimization
US7929762B2 (en) * 2007-03-12 2011-04-19 Jeffrey Kimball Tidd Determining edgeless areas in a digital image
USRE46953E1 (en) 2007-04-20 2018-07-17 University Of Maryland, Baltimore Single-arc dose painting for precision radiation therapy
US20090097704A1 (en) * 2007-10-10 2009-04-16 Micron Technology, Inc. On-chip camera system for multiple object tracking and identification
US7904530B2 (en) * 2008-01-29 2011-03-08 Palo Alto Research Center Incorporated Method and apparatus for automatically incorporating hypothetical context information into recommendation queries
US20090213984A1 (en) * 2008-02-26 2009-08-27 United Technologies Corp. Computed Tomography Systems and Related Methods Involving Post-Target Collimation
US7639777B2 (en) * 2008-02-26 2009-12-29 United Technologies Corp. Computed tomography systems and related methods involving forward collimation
US20090225954A1 (en) * 2008-03-06 2009-09-10 United Technologies Corp. X-Ray Collimators, and Related Systems and Methods Involving Such Collimators
US8238521B2 (en) * 2008-03-06 2012-08-07 United Technologies Corp. X-ray collimators, and related systems and methods involving such collimators
US7876875B2 (en) * 2008-04-09 2011-01-25 United Technologies Corp. Computed tomography systems and related methods involving multi-target inspection
US8718743B2 (en) * 2008-04-24 2014-05-06 Duke University Methods for single-pass volumetric bidirectional blood flow imaging spectral domain optical coherence tomography using a modified hilbert transform
US7888647B2 (en) * 2008-04-30 2011-02-15 United Technologies Corp. X-ray detector assemblies and related computed tomography systems
US20090274264A1 (en) * 2008-04-30 2009-11-05 United Technologies Corp. Computed Tomography Systems and Related Methods Involving Localized Bias
US8133127B1 (en) * 2008-07-21 2012-03-13 Synder Terrance W Sports training device and methods of use
WO2010062638A2 (en) * 2008-10-28 2010-06-03 Washington University In St. Louis Applying renyi entropy to detect changes in scattering architecture
US8693745B2 (en) * 2009-05-04 2014-04-08 Duke University Methods and computer program products for quantitative three-dimensional image correction and clinical parameter computation in optical coherence tomography
US8663210B2 (en) 2009-05-13 2014-03-04 Novian Health, Inc. Methods and apparatus for performing interstitial laser therapy and interstitial brachytherapy
WO2011072259A1 (en) * 2009-12-10 2011-06-16 Indiana University Research & Technology Corporation System and method for segmentation of three dimensional image data
IL202788A (en) 2009-12-17 2016-08-31 Elta Systems Ltd Method and system for enhancing a sar image
WO2011160235A1 (en) 2010-06-22 2011-12-29 Karl Otto System and method for estimating and manipulating estimated radiation dose
EP2617009A1 (en) * 2010-09-17 2013-07-24 Koninklijke Philips Electronics N.V. Contour delineation for radiation therapy planning with real-time contour segment impact rendering
JP2012115658A (en) * 2010-11-12 2012-06-21 Toshiba Corp Diagnostic imaging apparatus
KR101439412B1 (en) * 2010-11-22 2014-09-11 한국전자통신연구원 Method of segmenting lesions in images
JP5565810B2 (en) * 2010-11-29 2014-08-06 国立大学法人浜松医科大学 Mass spectrometry data processing method and apparatus
WO2012100213A2 (en) 2011-01-21 2012-07-26 Duke University Systems and methods for complex conjugate artifact resolved optical coherence tomography
KR101886333B1 (en) 2012-06-15 2018-08-09 삼성전자 주식회사 Apparatus and method for region growing with multiple cores
EP2901153A4 (en) 2012-09-26 2016-04-27 Abbott Diabetes Care Inc Method and apparatus for improving lag correction during in vivo measurement of analyte concentration with analyte concentration variability and range data
US9950194B2 (en) 2014-09-09 2018-04-24 Mevion Medical Systems, Inc. Patient positioning system
US9964499B2 (en) * 2014-11-04 2018-05-08 Toshiba Medical Systems Corporation Method of, and apparatus for, material classification in multi-energy image data
CN105046712B (en) * 2015-08-07 2017-06-30 江西理工大学 Based on the circle detection method that adaptability difference of Gaussian develops
US11281993B2 (en) * 2016-12-05 2022-03-22 Apple Inc. Model and ensemble compression for metric learning
CN107180442A (en) * 2017-04-13 2017-09-19 太原理工大学 A kind of photoacoustic image based on Renyi entropys rebuilds prefilter
CN107180433A (en) * 2017-06-06 2017-09-19 衢州学院 A kind of local cross-entropy measures the level set image segmentation algorithm of fuzzy C-mean algorithm
US10930386B2 (en) 2018-12-11 2021-02-23 International Business Machines Corporation Automated normality scoring of echocardiograms

Family Cites Families (138)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
GB1574823A (en) 1976-03-27 1980-09-10 Emi Ltd Video display arrangements
EP0068961A3 (en) * 1981-06-26 1983-02-02 Thomson-Csf Apparatus for the local heating of biological tissue
US4567896A (en) * 1984-01-20 1986-02-04 Elscint, Inc. Method and apparatus for calibrating a biopsy attachment for ultrasonic imaging apparatus
US4764971A (en) 1985-11-25 1988-08-16 Eastman Kodak Company Image processing method including image segmentation
US4751643A (en) * 1986-08-04 1988-06-14 General Electric Company Method and apparatus for determining connected substructures within a body
US4791567A (en) 1986-09-15 1988-12-13 General Electric Company Three dimensional connectivity system employing an equivalence schema for determining connected substructures within a body
JPH0738681B2 (en) 1987-03-20 1995-04-26 富士ゼロックス株式会社 Area recognition device
US5185809A (en) * 1987-08-14 1993-02-09 The General Hospital Corporation Morphometric analysis of anatomical tomographic data
US4961425A (en) 1987-08-14 1990-10-09 Massachusetts Institute Of Technology Morphometric analysis of anatomical tomographic data
US4991579A (en) * 1987-11-10 1991-02-12 Allen George S Method and apparatus for providing related images over time of a portion of the anatomy using fiducial implants
US4896673A (en) * 1988-07-15 1990-01-30 Medstone International, Inc. Method and apparatus for stone localization using ultrasound imaging
US4994013A (en) * 1988-07-28 1991-02-19 Best Industries, Inc. Pellet for a radioactive seed
US5227969A (en) * 1988-08-01 1993-07-13 W. L. Systems, Inc. Manipulable three-dimensional projection imaging method
FR2639238B1 (en) 1988-11-21 1991-02-22 Technomed Int Sa APPARATUS FOR SURGICAL TREATMENT OF TISSUES BY HYPERTHERMIA, PREFERABLY THE PROSTATE, COMPRISING MEANS OF THERMAL PROTECTION COMPRISING PREFERABLY RADIOREFLECTIVE SCREEN MEANS
US5072384A (en) 1988-11-23 1991-12-10 Arch Development Corp. Method and system for automated computerized analysis of sizes of hearts and lungs in digital chest radiographs
US5205289A (en) 1988-12-23 1993-04-27 Medical Instrumentation And Diagnostics Corporation Three-dimensional computer graphics simulation and computerized numerical optimization for dose delivery and treatment planning
US5133020A (en) * 1989-07-21 1992-07-21 Arch Development Corporation Automated method and system for the detection and classification of abnormal lesions and parenchymal distortions in digital medical images
JP2845995B2 (en) * 1989-10-27 1999-01-13 株式会社日立製作所 Region extraction method
US5222499A (en) * 1989-11-15 1993-06-29 Allen George S Method and apparatus for imaging the anatomy
US5187658A (en) * 1990-01-17 1993-02-16 General Electric Company System and method for segmenting internal structures contained within the interior region of a solid object
JPH04364829A (en) * 1990-02-15 1992-12-17 Toshiba Corp Magnetic resonance image processing method and apparatus therefor
FR2660543B1 (en) * 1990-04-06 1998-02-13 Technomed Int Sa METHOD FOR AUTOMATICALLY MEASURING THE VOLUME OF A TUMOR, IN PARTICULAR A PROSTATE TUMOR, MEASURING DEVICE, METHOD AND APPARATUS COMPRISING THE SAME.
FR2660732B1 (en) 1990-04-06 1992-09-04 Technomed Int Sa TRANSLATABLE END ARM AND THERAPEUTIC TREATMENT APPARATUS, INCLUDING APPLICATION.
US5457754A (en) 1990-08-02 1995-10-10 University Of Cincinnati Method for automatic contour extraction of a cardiac image
DE69132412T2 (en) 1990-10-19 2001-03-01 Univ St Louis LOCALIZATION SYSTEM FOR A SURGICAL PROBE FOR USE ON THE HEAD
US5204625A (en) * 1990-12-20 1993-04-20 General Electric Company Segmentation of stationary and vascular surfaces in magnetic resonance imaging
US5166876A (en) 1991-01-16 1992-11-24 General Electric Company System and method for detecting internal structures contained within the interior region of a solid object
US5662111A (en) 1991-01-28 1997-09-02 Cosman; Eric R. Process of stereotactic optical navigation
US5640496A (en) * 1991-02-04 1997-06-17 Medical Instrumentation And Diagnostics Corp. (Midco) Method and apparatus for management of image data by linked lists of pixel values
US5210799A (en) 1991-03-28 1993-05-11 Texas Instruments Incorporated System and method for ranking and extracting salient contours for target recognition
DE69230440T2 (en) * 1991-04-25 2000-04-13 Unisys Corp METHOD AND DEVICE FOR ADAPTIVE THRESHOLD PROCESSING OF GRAY SCALES
EP0516047B1 (en) 1991-05-27 1998-03-11 Hitachi, Ltd. Method of and apparatus for processing multi-dimensional data
US5417210A (en) * 1992-05-27 1995-05-23 International Business Machines Corporation System and method for augmentation of endoscopic surgery
US5239591A (en) 1991-07-03 1993-08-24 U.S. Philips Corp. Contour extraction in multi-phase, multi-slice cardiac mri studies by propagation of seed contours between images
US5260871A (en) 1991-07-31 1993-11-09 Mayo Foundation For Medical Education And Research Method and apparatus for diagnosis of breast tumors
US5242373A (en) 1991-09-17 1993-09-07 Scott Walter P Medical seed implantation instrument
US5371810A (en) 1991-09-27 1994-12-06 E. I. Du Pont De Nemours And Company Method of determining the interior points of an object in a background
US5289374A (en) * 1992-02-28 1994-02-22 Arch Development Corporation Method and system for analysis of false positives produced by an automated scheme for the detection of lung nodules in digital chest radiographs
DE4207463C2 (en) * 1992-03-10 1996-03-28 Siemens Ag Arrangement for the therapy of tissue with ultrasound
US5299253A (en) * 1992-04-10 1994-03-29 Akzo N.V. Alignment system to overlay abdominal computer aided tomography and magnetic resonance anatomy with single photon emission tomography
US5603318A (en) 1992-04-21 1997-02-18 University Of Utah Research Foundation Apparatus and method for photogrammetric surgical localization
US5389101A (en) * 1992-04-21 1995-02-14 University Of Utah Apparatus and method for photogrammetric surgical localization
US5574799A (en) 1992-06-12 1996-11-12 The Johns Hopkins University Method and system for automated detection of microcalcification clusters in mammograms
US5537485A (en) * 1992-07-21 1996-07-16 Arch Development Corporation Method for computer-aided detection of clustered microcalcifications from digital mammograms
FR2694881B1 (en) 1992-07-31 1996-09-06 Univ Joseph Fourier METHOD FOR DETERMINING THE POSITION OF AN ORGAN.
US5391139A (en) * 1992-09-03 1995-02-21 William Beaumont Hospital Real time radiation treatment planning system
DE4233978C1 (en) * 1992-10-08 1994-04-21 Leibinger Gmbh Body marking device for medical examinations
US5319549A (en) * 1992-11-25 1994-06-07 Arch Development Corporation Method and system for determining geometric pattern features of interstitial infiltrates in chest images
DE4240722C2 (en) * 1992-12-03 1996-08-29 Siemens Ag Device for the treatment of pathological tissue
US5517602A (en) * 1992-12-03 1996-05-14 Hewlett-Packard Company Method and apparatus for generating a topologically consistent visual representation of a three dimensional surface
CA2110148C (en) 1992-12-24 1999-10-05 Aaron Fenster Three-dimensional ultrasound imaging system
DE4304571A1 (en) * 1993-02-16 1994-08-18 Mdc Med Diagnostic Computing Procedures for planning and controlling a surgical procedure
US5787886A (en) 1993-03-19 1998-08-04 Compass International Incorporated Magnetic field digitizer for stereotatic surgery
IL109385A (en) * 1993-04-22 1998-03-10 Pixsys System for locating the relative positions of objects in three dimensional space
CA2161430C (en) * 1993-04-26 2001-07-03 Richard D. Bucholz System and method for indicating the position of a surgical probe
US5491627A (en) * 1993-05-13 1996-02-13 Arch Development Corporation Method and system for the detection of microcalcifications in digital mammograms
US5526812A (en) * 1993-06-21 1996-06-18 General Electric Company Display system for enhancing visualization of body structures during medical procedures
US5494039A (en) * 1993-07-16 1996-02-27 Cryomedical Sciences, Inc. Biopsy needle insertion guide and method of use in prostate cryosurgery
US5391199A (en) * 1993-07-20 1995-02-21 Biosense, Inc. Apparatus and method for treating cardiac arrhythmias
US5412563A (en) * 1993-09-16 1995-05-02 General Electric Company Gradient image segmentation method
US5411026A (en) * 1993-10-08 1995-05-02 Nomos Corporation Method and apparatus for lesion position verification
EP0649117A3 (en) 1993-10-15 1996-01-31 George S Allen Method for providing medical images.
US5452367A (en) 1993-11-29 1995-09-19 Arch Development Corporation Automated method and system for the segmentation of medical images
US5460592A (en) 1994-01-24 1995-10-24 Amersham Holdings, Inc. Apparatus and method for making carrier assembly for radioactive seed carrier
US5531227A (en) * 1994-01-28 1996-07-02 Schneider Medical Technologies, Inc. Imaging device and method
US5458126A (en) 1994-02-24 1995-10-17 General Electric Company Cardiac functional analysis system employing gradient image segmentation
US5433199A (en) * 1994-02-24 1995-07-18 General Electric Company Cardiac functional analysis method using gradient image segmentation
US5734739A (en) * 1994-05-31 1998-03-31 University Of Washington Method for determining the contour of an in vivo organ using multiple image frames of the organ
US5570430A (en) 1994-05-31 1996-10-29 University Of Washington Method for determining the contour of an in vivo organ using multiple image frames of the organ
US5672172A (en) 1994-06-23 1997-09-30 Vros Corporation Surgical instrument with ultrasound pulse generator
US5398690A (en) * 1994-08-03 1995-03-21 Batten; Bobby G. Slaved biopsy device, analysis apparatus, and process
US6025128A (en) * 1994-09-29 2000-02-15 The University Of Tulsa Prediction of prostate cancer progression by analysis of selected predictive parameters
CA2201877C (en) * 1994-10-07 2004-06-08 Richard D. Bucholz Surgical navigation systems including reference and localization frames
US5765561A (en) * 1994-10-07 1998-06-16 Medical Media Systems Video-based surgical targeting system
WO1996012187A1 (en) * 1994-10-13 1996-04-25 Horus Therapeutics, Inc. Computer assisted methods for diagnosing diseases
US5583659A (en) 1994-11-10 1996-12-10 Eastman Kodak Company Multi-windowing technique for thresholding an image using local image properties
US5626829A (en) * 1994-11-16 1997-05-06 Pgk, Enterprises, Inc. Method and apparatus for interstitial radiation of the prostate gland
US5817022A (en) 1995-03-28 1998-10-06 Sonometrics Corporation System for displaying a 2-D ultrasound image within a 3-D viewing environment
US5868673A (en) * 1995-03-28 1999-02-09 Sonometrics Corporation System for carrying out surgery, biopsy and ablation of a tumor or other physical anomaly
US5797849A (en) 1995-03-28 1998-08-25 Sonometrics Corporation Method for carrying out a medical procedure using a three-dimensional tracking and imaging system
US5833627A (en) 1995-04-13 1998-11-10 United States Surgical Corporation Image-guided biopsy apparatus and methods of use
US5660185A (en) 1995-04-13 1997-08-26 Neovision Corporation Image-guided biopsy apparatus with enhanced imaging and methods
US5820623A (en) 1995-06-20 1998-10-13 Ng; Wan Sing Articulated arm for medical procedures
US5810007A (en) 1995-07-26 1998-09-22 Associates Of The Joint Center For Radiation Therapy, Inc. Ultrasound localization and image fusion for the treatment of prostate cancer
US6256529B1 (en) * 1995-07-26 2001-07-03 Burdette Medical Systems, Inc. Virtual reality 3D visualization for surgical procedures
US5638819A (en) * 1995-08-29 1997-06-17 Manwaring; Kim H. Method and apparatus for guiding an instrument to a target
US5906574A (en) * 1995-10-06 1999-05-25 Kan; William C. Apparatus for vacuum-assisted handling and loading of radioactive seeds and spacers into implant needles within an enclosed visible radiation shield for use in therapeutic radioactive seed implantation
US5709206A (en) * 1995-11-27 1998-01-20 Teboul; Michel Imaging system for breast sonography
JPH09154961A (en) * 1995-12-07 1997-06-17 Toshiba Medical Eng Co Ltd Radiation therapy program method
US5742263A (en) * 1995-12-18 1998-04-21 Telxon Corporation Head tracking system for a head mounted display system
US5682886A (en) 1995-12-26 1997-11-04 Musculographics Inc Computer-assisted surgical system
US5727538A (en) * 1996-04-05 1998-03-17 Shawn Ellis Electronically actuated marking pellet projector
NL1003528C2 (en) * 1996-07-05 1998-01-07 Optische Ind Oede Oude Delftoe Assembly of a capsule for brachytherapy and a guide.
NL1003543C2 (en) * 1996-07-08 1998-01-12 Optische Ind Oede Oude Delftoe Brachytherapy capsule and brachytherapy capsule assembly and guide.
US5778043A (en) * 1996-09-20 1998-07-07 Cosman; Eric R. Radiation beam control system
US5830145A (en) 1996-09-20 1998-11-03 Cardiovascular Imaging Systems, Inc. Enhanced accuracy of three-dimensional intraluminal ultrasound (ILUS) image reconstruction
US5776063A (en) * 1996-09-30 1998-07-07 Molecular Biosystems, Inc. Analysis of ultrasound images in the presence of contrast agent
US5860909A (en) * 1996-10-18 1999-01-19 Mick Radio Nuclear Instruments, Inc. Seed applicator for use in radiation therapy
US5669382A (en) 1996-11-19 1997-09-23 General Electric Company System for measuring myocardium in cardiac images
US6423009B1 (en) * 1996-11-29 2002-07-23 Life Imaging Systems, Inc. System, employing three-dimensional ultrasonographic imaging, for assisting in guiding and placing medical instruments
KR20000069165A (en) * 1996-11-29 2000-11-25 라이프 이미징 시스템즈 인코퍼레이티드 Apparatus for guiding medical instruments during ultrasonographic imaging
WO1998032388A2 (en) * 1997-01-24 1998-07-30 Koninklijke Philips Electronics N.V. Image display system
US5859891A (en) * 1997-03-07 1999-01-12 Hibbard; Lyn Autosegmentation/autocontouring system and method for use with three-dimensional radiation therapy treatment planning
US6033357A (en) * 1997-03-28 2000-03-07 Navius Corporation Intravascular radiation delivery device
US6309339B1 (en) * 1997-03-28 2001-10-30 Endosonics Corporation Intravascular radiation delivery device
US5931786A (en) 1997-06-13 1999-08-03 Barzell Whitmore Maroon Bells, Inc. Ultrasound probe support and stepping device
JP4212128B2 (en) * 1997-07-02 2009-01-21 株式会社東芝 Radiation therapy equipment
US5871448A (en) * 1997-10-14 1999-02-16 Real World Design And Development Co. Stepper apparatus for use in the imaging/treatment of internal organs using an ultrasound probe
US6129670A (en) * 1997-11-24 2000-10-10 Burdette Medical Systems Real time brachytherapy spatial registration and visualization system
US6083166A (en) * 1997-12-02 2000-07-04 Situs Corporation Method and apparatus for determining a measure of tissue manipulation
US6213932B1 (en) * 1997-12-12 2001-04-10 Bruno Schmidt Interstitial brachytherapy device and method
US5938583A (en) 1997-12-29 1999-08-17 Grimm; Peter D. Precision implant needle and method of using same in seed implant treatment of prostate cancer
US6027446A (en) * 1998-01-12 2000-02-22 Washington Univ. Of Office Of Technology Transfer Pubic arch detection and interference assessment in transrectal ultrasound guided prostate cancer therapy
US6083167A (en) * 1998-02-10 2000-07-04 Emory University Systems and methods for providing radiation therapy and catheter guides
US5928130A (en) * 1998-03-16 1999-07-27 Schmidt; Bruno Apparatus and method for implanting radioactive seeds in tissue
US6048312A (en) * 1998-04-23 2000-04-11 Ishrak; Syed Omar Method and apparatus for three-dimensional ultrasound imaging of biopsy needle
KR20010020580A (en) * 1998-05-04 2001-03-15 노보스트 코포레이션 Intraluminal radiation treatment system
US6238342B1 (en) * 1998-05-26 2001-05-29 Riverside Research Institute Ultrasonic tissue-type classification and imaging methods and apparatus
US6036632A (en) * 1998-05-28 2000-03-14 Barzell-Whitmore Maroon Bells, Inc. Sterile disposable template grid system
US6425865B1 (en) * 1998-06-12 2002-07-30 The University Of British Columbia Robotically assisted medical ultrasound
CA2337333A1 (en) * 1998-07-20 2000-02-03 Marvin O. Andrews Brachytherapy device including an anti-static handle
US6387034B1 (en) * 1998-08-17 2002-05-14 Georia Tech Research Corporation Brachytherapy treatment planning method and apparatus
US20030074011A1 (en) * 1998-09-24 2003-04-17 Super Dimension Ltd. System and method of recording and displaying in context of an image a location of at least one point-of-interest in a body during an intra-body medical procedure
IL126333A0 (en) * 1998-09-24 1999-05-09 Super Dimension Ltd System and method of recording and displaying in context of an image a location of at least one point-of-interest in body during an intra-body medical procedure
ATE324144T1 (en) * 1998-10-14 2006-05-15 Terumo Corp WIRE-SHAPED RADIATION SOURCE AND CATHETER ARRANGEMENT FOR RADIATION THERAPY
US6366796B1 (en) * 1998-10-23 2002-04-02 Philips Medical Systems (Cleveland), Inc. Method and apparatus for planning brachytherapy surgical procedures
US6196963B1 (en) * 1999-03-02 2001-03-06 Medtronic Ave, Inc. Brachytherapy device assembly and method of use
US6766036B1 (en) * 1999-07-08 2004-07-20 Timothy R. Pryor Camera based man machine interfaces
US6266453B1 (en) * 1999-07-26 2001-07-24 Computerized Medical Systems, Inc. Automated image fusion/alignment system and method
US6379302B1 (en) * 1999-10-28 2002-04-30 Surgical Navigation Technologies Inc. Navigation information overlay onto ultrasound imagery
US6213110B1 (en) * 1999-12-16 2001-04-10 Odyssey Paintball Products, Inc. Rapid feed paintball loader
US6361487B1 (en) * 2000-03-09 2002-03-26 Neoseed Technology Llc Method and apparatus for brachytherapy treatment of prostate disease
US6358195B1 (en) * 2000-03-09 2002-03-19 Neoseed Technology Llc Method and apparatus for loading radioactive seeds into brachytherapy needles
US6572525B1 (en) * 2000-05-26 2003-06-03 Lisa Yoshizumi Needle having an aperture for detecting seeds or spacers loaded therein and colored seeds or spacers
US6416492B1 (en) * 2000-09-28 2002-07-09 Scimed Life Systems, Inc. Radiation delivery system utilizing intravascular ultrasound
DE10051244A1 (en) * 2000-10-17 2002-05-16 Philips Corp Intellectual Pty X-ray free intravascular localization and imaging procedure
US6540679B2 (en) * 2000-12-28 2003-04-01 Guided Therapy Systems, Inc. Visual imaging system for ultrasonic probe
US7388988B2 (en) * 2002-01-14 2008-06-17 Hewlett-Packard Development Company, L.P. Systems and methods for processing boundary information of a graphical object

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
BEN HAMZA A ET AL: "An information divergence measure for ISAR image registration" PROCEEDINGS OF THE 11TH IEEE SIGNAL PROCESSING WORKSHOP ON STATISTICAL SIGNAL PROCESSING (CAT. NO.01TH8563), PROCEEDINGS OF SSP2001. 11TH IEEE WORKSHOP ON STATISTICAL SIGNAL PROCESSING, SINGAPORE, 6-8 AUG. 2001, 6 August 2001 (2001-08-06), pages 130-133, XP010561093 2001, Piscataway, NJ, USA, IEEE, USA ISBN: 0-7803-7011-2 cited in the application *
GOMEZ-LOPERA J F ET AL: "An analysis of edge detection by using the Jensen-Shannon divergence" JOURNAL OF MATHEMATICAL IMAGING AND VISION, vol. 13, no. 1, August 2000 (2000-08), pages 35-56, XP008033153 KLUWER ACADEMIC PUBLISHERS, NETHERLANDS ISSN: 0924-9907 *
HIBBARD L S ED - VEMURI B: "Maximum a posteriori segmentation for medical visualization" WORKSHOP ON BIOMEDICAL IMAGE ANALYSIS, 1998. PROCEEDINGS, SANTA BARBARA, CA, USA 26-27 JUNE 1998, 26 June 1998 (1998-06-26), pages 93-102, XP010291414 LOS ALAMITOS, CA, USA, IEEE, US ISBN: 0-8186-8460-7 cited in the application *
UNAL G ET AL.: "Active polygons for object tracking" PROCEEDINGS FIRST INTERNATIONAL SYMPOSIUM ON 3D DATA PROCESSING VISUALIZATION AND TRANSMISSION, PADOVA, ITALY, 19-21 JUNE 2002, 19 June 2002 (2002-06-19), pages 696-699, XP002290194 2002, Los Alamitos, CA, USA, IEEE Comput. Soc, USA ISBN: 0-7695-1521-5 *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109643485A (en) * 2016-12-30 2019-04-16 同济大学 A kind of urban highway traffic method for detecting abnormality
CN109643485B (en) * 2016-12-30 2021-04-30 同济大学 Urban road traffic abnormity detection method

Also Published As

Publication number Publication date
WO2004013811A3 (en) 2004-09-16
AU2003256972A1 (en) 2004-02-23
US20040022438A1 (en) 2004-02-05
AU2003256972A8 (en) 2004-02-23
US7187800B2 (en) 2007-03-06

Similar Documents

Publication Publication Date Title
US7187800B2 (en) Method and apparatus for image segmentation using Jensen-Shannon divergence and Jensen-Renyi divergence
US9262827B2 (en) Lung, lobe, and fissure imaging systems and methods
El-Dahshan et al. Computer-aided diagnosis of human brain tumor through MRI: A survey and a new algorithm
US10121243B2 (en) Advanced computer-aided diagnosis of lung nodules
Mansoor et al. Deep learning guided partitioned shape model for anterior visual pathway segmentation
CN111105424A (en) Lymph node automatic delineation method and device
Kuang et al. Segmenting hemorrhagic and ischemic infarct simultaneously from follow-up non-contrast CT images in patients with acute ischemic stroke
El-Sayed et al. Computer-aided diagnosis of human brain tumor through MRI: A survey and a new algorithm
Aranguren et al. Improving the segmentation of magnetic resonance brain images using the LSHADE optimization algorithm
CN111798462A (en) Automatic delineation method for nasopharyngeal carcinoma radiotherapy target area based on CT image
US20070223815A1 (en) Feature Weighted Medical Object Contouring Using Distance Coordinates
Priyadarsini et al. Survey on segmentation of liver from CT images
CN111008984A (en) Method and system for automatically drawing contour line of normal organ in medical image
CN116097302A (en) Connected machine learning model with joint training for lesion detection
Dutande et al. Deep residual separable convolutional neural network for lung tumor segmentation
Koundal et al. Challenges and future directions in neutrosophic set-based medical image analysis
Peng et al. H-ProSeg: Hybrid ultrasound prostate segmentation based on explainability-guided mathematical model
Peng et al. H-SegMed: a hybrid method for prostate segmentation in TRUS images via improved closed principal curve and improved enhanced machine learning
Alves et al. Extracting lungs from ct images using fully convolutional networks
CN115830163A (en) Progressive medical image cross-mode generation method and device based on deterministic guidance of deep learning
Shrestha et al. A novel solution of using deep learning for prostate cancer segmentation: enhanced batch normalization
Stough et al. Regional appearance in deformable model segmentation
Srilatha et al. Performance analysis of ultrasound ovarian tumour segmentation using GrabCut and FL-SNNM
Ganasala et al. Semiautomatic and automatic brain tumor segmentation methods: performance comparison
Leonardi et al. 3D reconstruction from CT-scan volume dataset application to kidney modeling

Legal Events

Date Code Title Description
AK Designated states

Kind code of ref document: A2

Designated state(s): AE AG AL AM AT AU AZ BA BB BG BR BY BZ CA CH CN CO CR CU CZ DE DK DM DZ EC EE ES FI GB GD GE GH GM HR HU ID IL IN IS JP KE KG KP KR KZ LC LK LR LS LT LU LV MA MD MG MK MN MW MX MZ NI NO NZ OM PG PH PL PT RO RU SC SD SE SG SK SL SY TJ TM TN TR TT TZ UA UG US UZ VC VN YU ZA ZM ZW

AL Designated countries for regional patents

Kind code of ref document: A2

Designated state(s): GH GM KE LS MW MZ SD SL SZ TZ UG ZM ZW AM AZ BY KG KZ MD RU TJ TM AT BE BG CH CY CZ DE DK EE ES FI FR GB GR HU IE IT LU MC NL PT RO SE SI SK TR BF BJ CF CG CI CM GA GN GQ GW ML MR NE SN TD TG

121 Ep: the epo has been informed by wipo that ep was designated in this application
122 Ep: pct application non-entry in european phase
NENP Non-entry into the national phase

Ref country code: JP

WWW Wipo information: withdrawn in national office

Country of ref document: JP