Method and apparatus for determining asymmetry in an image

0Associated
Cases 
0Associated
Defendants 
0Accused
Products 
1Forward
Citation 
0
Petitions 
1
Assignment
First Claim
1. A computerized method for determining asymmetry in a medical image, the method comprising the steps of:
 receiving the medical image;
determining a symmetry plane to divide the image into a first part and a second part, that comprise left and right hemispheres respectively;
determining probability distributions of voxels against intensities for the first part and the second part;
generating a first histogram of intensities representative of the first part and storing the first histogram on a computer memory;
generating a second histogram of intensities representative of the second part and storing the second histogram on the computer memory;
compensating for any relative shift along a predetermined axis between the first histogram and the second histogram;
calculating, using a computer processor, a divergence value based on distance between the first and second histograms; and
determining if the calculated divergence value is greater than a predetermined threshold, a divergence of greater than the predetermined threshold being indicative of asymmetry in the medical image and indicative of one or more abnormalities, whereinthe step of calculating, using the computer processor, the divergence value comprises determining probability distributions of intensities for the first part and the second part and calculating the Jdivergence value according to the following equation;
J(p,q)=I(p,q)+I(q,p)where I(p/q) and I(q/p) are the KullbackLeibler'"'"'s divergences for intensity distributions q and p of the first and second parts respectively.
1 Assignment
0 Petitions
Accused Products
Abstract
A method for determining asymmetry in an image such as an MR image of a brain comprises determining a symmetry plane to divide the image into a first part and a second part representative of, for example, the hemispheres of the brain. The probability distributions of voxels against intensities are determined for the first and second parts and histograms of intensities representative of the parts are generated. Compensation is made for any relative shift along a predetermined axis between the histograms. A divergence value based on a distance between the first and second histograms is then calculated and it is determined if the calculated divergence value is greater than a predetermined threshold. A divergence of greater than the predetermined threshold is indicative of asymmetry in the image that may be considered as suspicious for abnormality. There is also disclosed an apparatus for determining asymmetry in an image.
4 Citations
View as Search Results
Automated target recognition based body scanner using database scans  
Patent #
US 10,445,591 B2
Filed 10/05/2017

Current Assignee
TEK84 INC.

Sponsoring Entity
TEK84 INC.

Optical system and method for invivo transcranial examination of brain tissue of a subject  
Patent #
US 7,610,082 B2
Filed 11/05/2004

Current Assignee
NonInvasive Technology Incorporated

Sponsoring Entity
NonInvasive Technology Incorporated

System and method for threedimensional image rendering and analysis  
Patent #
US 7,274,810 B2
Filed 04/10/2001

Current Assignee
David Yankelevitz

Sponsoring Entity
Cornell Research Foundation Incorporated

Method, system, and computer software product for featurebased correlation of lesions from multiple images  
Patent #
US 7,298,881 B2
Filed 02/14/2005

Current Assignee
University of Chicago

Sponsoring Entity
University of Chicago

12 Claims
 1. A computerized method for determining asymmetry in a medical image, the method comprising the steps of:
receiving the medical image; determining a symmetry plane to divide the image into a first part and a second part, that comprise left and right hemispheres respectively; determining probability distributions of voxels against intensities for the first part and the second part; generating a first histogram of intensities representative of the first part and storing the first histogram on a computer memory; generating a second histogram of intensities representative of the second part and storing the second histogram on the computer memory; compensating for any relative shift along a predetermined axis between the first histogram and the second histogram; calculating, using a computer processor, a divergence value based on distance between the first and second histograms; and determining if the calculated divergence value is greater than a predetermined threshold, a divergence of greater than the predetermined threshold being indicative of asymmetry in the medical image and indicative of one or more abnormalities, wherein the step of calculating, using the computer processor, the divergence value comprises determining probability distributions of intensities for the first part and the second part and calculating the Jdivergence value according to the following equation;
J(p,q)=I(p,q)+I(q,p)where I(p/q) and I(q/p) are the KullbackLeibler'"'"'s divergences for intensity distributions q and p of the first and second parts respectively.  View Dependent Claims (2, 3, 4, 5, 6, 7, 8, 9, 10, 12)
 11. A computer program product comprising:
a nontransitory computer readable medium having a computer readable code and computer readable system code embodied on said medium for determining a symmetry plane to divide the image into a first part and a second part;
that comprise left and right hemispheres respectively, said computer program product further comprising;computer readable code within said computer usable medium for; determining probability distributions of voxels against intensities for the first part and the second part; generating a first histogram of intensities representative of the first part; generating a second histogram of intensities representative of the second part; compensating for any relative shift along a predetermined axis between the first histogram and the second histogram; calculating a divergence value based on distance between the first and second histograms; and determining if the calculated divergence value is greater than a predetermined threshold, a divergence of greater than the predetermined threshold being indicative of asymmetry in the image and indicative of one or more abnormalities, wherein the step of calculating the divergence value comprises determining probability distributions of intensities for the first part and the second part and calculating the Jdivergence value according to the following equation;
J(p,q)=I(p,q)+I(q,p)where I(p/q) and I(q/p) are the KullbackLeibler'"'"'s divergences for intensity distributions q and p of the first and second parts respectively.
1 Specification
The present invention relates to a method and apparatus for determining asymmetry in an image. In particular, the present invention relates to an apparatus and method for determining asymmetry in an image such as a magnetic resonance imaging (MRI) image of, for example, a brain.
There are a number of conventional methods for detecting abnormalities in a brain from a study of neuroimages, such as those obtained using magnetic resonance imaging (MRI) techniques. One conventional method is based on the analysis of tissue classes and another conventional method is based on the analysis of the symmetry between the two hemispheres of the brain by, for example, extracting structures, or finding pointtopoint interhemispheric correspondence and crosscorrelation.
A number of papers have been published describing the analysis of tissue classes, for example, the paper entitled Validity Of Guided Clustering For Brain Tumour Segmentation, by Velthuizen, R. P., published by the Engineering in Medicine and Biology Society, 1995, IEEE 17th Annual Conference, V. 1, 1995, 413414. In this paper, a method is described which detects abnormalities in the brain by allowing the validity of small classes, like tumours, to have a noticeable effect on the validity measure. However, only three tumour data sets, and no normal data sets were tested and reported in this publication.
Unsupervised Segmentation For Automatic Detection Of Brain Tumours In MRI, by Capelle A. S., Alata O., Fernandez C., Lefevre S., Ferrie J. C., published as Proceedings of International Conference on Image Processing IPMI 2000, V. 1, 2000, 613616 describes a multiple resolution segmentation method in which the brain is divided into homogeneous Gaussian distributed classes. In this method, the maximum a posteriori method is used to estimate the parameters of each class to detect the existence of tumours in a twodimensional MR image. This method is based on the segmentation of tumours using knowledge of the intensity distribution. However, due to the overlapping intensity of tumour(s) with other normal tissues, this method may segment the tissues incorrectly. This iterative method may also be time consuming although no indication is given in this paper of the run time.
A number of papers have been published describing the analysis of the symmetry between the two hemispheres of the brain. Human brains exhibit an approximate bilateral symmetry with respect to the interhemispheric (longitudinal) fissure bisecting the brain, known as the midsaggital plane (MSP). These methods are based on the assumption that a healthy human brain is roughly symmetrical and an abnormality may be detected from brain asymmetry. A common way of detecting abnormalities using such a method is to consider local geometrical asymmetries, such as changes in the relative shape and structure of the hemispheres. Such a method is described in the paper entitled Cortical Variability and Asymmetry in Normal Ageing and Alzheimer'"'"'s Disease, by Thompson P M, Moussai J, Zohoori S, Goldkorn A, Khan A A, Mega M S, Small G W, Cummings J L, Toga A W which was published in Cereb. Cortex. 1998 September; 8(6):492509].
In a paper entitled Knowledgebased Classification And Tissue Labelling Of MR Images Of Human Brain Medical Imaging, by Chunlin Li; Goldgof, D. B.; Hall, L. O. which was published in IEEE Transactions, 1993, Vol. 12(4), 740750, it is suggested that, using a knowledge based approach involving an estimation of the symmetry of cerebrospinal fluid (CSF), a tumour can be detected only in the slices containing CSF. The measures used are based strictly on predefined intensity thresholds which can vary from one data set to another. It was assumed that the tumors appear to have intensity higher than that of grey matter on T2weighted images.
A development of the knowledgebased approach described in the abovementioned paper was published in a paper entitled MR Brain Image Segmentation Using Fuzzy Clustering, by OckKyung Yoon; DongMin Kwak, DongWhee Kim, KilHoum Park and published as IEEE International Fuzzy Systems Conference Proceedings, 1999, FUZZIEEE '"'"'99, Volume: 2, 853857. In this paper a method is described where fuzzy cmeans are used in slices containing CSF to separate grey matter, white matter, and CSF in the cerebrum. In this method, a symmetrical measure based on the number of pixels, moment invariants, and Fourier descriptors is described as being used to quantify the normality of image slices of the brain being studied. However, this algorithm has a number of disadvantages, for example, the quantification of normality is based only on 40 slices in 1 normal and 2 abnormal T2weighted studies, also, as the symmetrical measure uses a large number of user defined parameters, which are difficult to estimate.
In the publication by Joshi S, Lorenzen P, Gerig G, Bullitt E. entitled Structural and radiometric asymmetry in brain images, Med Image Anal. 2003 June; 7(2):155170, structural and radiometric asymmetry was analysed through large deformation image warping in three dimensions. Nine tumour and four normal cases were tested, however, there is no information given on the running time. The second stage of the algorithm described in this paper is based on a Christensen warping algorithm which has an extremely long run time and is described in the paper by Christensen G E, R D Rabbit, M I Miller entitled Deformable Templates Using Large Deformation Kinematics, IEEE Transactions on Image Processing, 5(10), 1996, pp. 14351447.
All the conventional methods mentioned above, with the exception of that described in the publication by Joshi S, Lorenzen P, Gerig G, Bullitt E. entitled Structural and radiometric asymmetry in brain images, Med Image Anal. 2003 June; 7(2):155170, work in two dimensions.
In general terms, the present invention proposes the determination of asymmetry in an image by using an informational divergence measure to calculate the similarity between two parts of the image.
Preferably, embodiments of the invention provide a fast and accurate technique for detecting and locating abnormalities in images, such as MR images on the basis of an analysis of symmetry, and for providing the number of pathological sites in the volumetric image being studied.
The present invention has many applications, for example in the field of neuroimaging, and in the study of pathologies displayed in MR images.
According to a first aspect of the present invention there is provided a method for determining asymmetry in a volumetric image the method comprising the steps of:
 determining a symmetry plane to divide the image into a first part and a second part;
 determining probability distributions of voxels against intensity for the first part and the second part;
 generating a first histogram of intensities representative of the first part;
 generating a second histogram of intensities representative of the second part;
 compensating for any relative shift along a predetermined axis between the first histogram and the second histogram;
 calculating a divergence value based on distance between the first and second histograms; and
 determining if the calculated divergence value is greater than a predetermined threshold, a divergence of greater than the predetermined threshold being indicative of asymmetry in the image and indicative of one or more abnormalities.
For the purposes of the present specification, the term volumetric image is hereinafter referred to as ‘image’.
According to a second aspect of the invention there is provided an apparatus arranged to perform a method for determining asymmetry in an image according to the method defined above.
According to a third aspect of the invention there is provided a computer program product comprising:
 a computer usable medium having a computer readable code and computer readable system code embodied on said medium for determining a symmetry plane to divide the image into a first part and a second part; said computer program product further comprising:
 computer readable code within said computer usable medium for:
 determining probability distributions of voxels against intensity for the first part and the second part;
 generating a first histogram of intensities representative of the first part;
 generating a second histogram of intensities representative of the second part;
 compensating for any relative shift along a predetermined axis between the first histogram and the second histogram;
 calculating a divergence value based on distance between the first and second histograms; and
 determining if the calculated divergence value is greater than a predetermined threshold, a divergence of greater than the predetermined threshold being indicative of asymmetry in the image and indicative of one or more abnormalities.
According to a fourth aspect of the invention there is provided an apparatus for determining asymmetry in an image, the apparatus comprising:
 a device for determining a symmetry plane to divide the image into a first part and a second part, that comprise left and right hemispheres respectively;
 a device for determining probability distributions of voxels against intensities for the first part and the second part;
 a generator for generating a first histogram of intensities representative of the first part and a second histogram of intensities representative of the second part;
 a compensator for compensating for any relative shift along a predetermined axis between the first histogram and the second histogram;
 a calculator for calculating a divergence value based on distance between the first and second histograms; and
a device for determining if the calculated divergence value is greater than a predetermined threshold, a divergence of greater than the predetermined threshold being indicative of asymmetry in the image and indicative of one or more abnormalities.
According to a fifth aspect of the invention there is provided a scanner comprising the apparatus defined above.
Preferred embodiments of the invention will now be described by way of example and with reference to the accompanying drawings in which:
In a preferred embodiment, the symmetry of the brain is analysed from the point of view of information theory. If the brain were absolutely symmetrical, the amount of information in both hemispheres would be the same. The brain is usually regarded as being roughly symmetrical. However, brain abnormalities may cause changes in images obtained by MRI techniques. For example, a brain tumors may belong to one or both of the following categories: masseffect, in which the diseased tissue displaces healthy tissue; and infiltrating, in which healthy tissue has become diseased. Masseffect brain tumors cause structural asymmetry by displacing healthy tissue, and may cause radiometric asymmetry in adjacent normal structures due to edema. Infiltrating tumors have a different radiometric response from healthy tissue. Both of the categories increase interhemisphere asymmetry.
It has been appreciated that by comparing the probability distributions, that is, histograms, of the brain it is possible to determine the value of asymmetry and detect abnormalities.
The brain tissue volume is different from one person to another and the manifestation of brain tissues in MRI also changes from one scanner to another. There are numerous protocols which also influence the nature of an MR image. In view of these factors, normalisation of images with respect to a standard brain structure or structures, or with respect to a particular scanner is difficult. In such situations, selfnormalisation of an MR image on the basis of intensity distributions of two hemispheres provides, in some sense, a uniformed representation.
The next stage 3 is to divide the MRI volume into the two hemispheres of the brain, the right and left hemispheres being separated by the MSP. The probability distributions of voxels against intensities are then obtained for each hemisphere to create two histograms, one representing the left hemisphere and the other representing the right hemisphere. It has been observed that sometimes these histograms have some shift in the intensity values as shown at
The next stage 5 in the method embodying the invention, as shown in
The value of the Jdivergence is then compared in a further stage 6 with a threshold value. If the value of the Jdivergence does not exceed the threshold, the scan is considered in a further stage 7 to be normal and no further action is required. Alternatively, if the value of the Jdivergence exceeds the threshold, this is considered in a further stage 8 to be an indication that an abnormality has been found in the scan and the results are then analysed in a still further stage 9 to determine the maximum (principal) summands in the Jdivergence and the tissues to which they correspond. These tissues are considered to be the main contributors to the asymmetry. The process then ends in a final stage 10 and the results may then be studied further to determine the nature of the detected abnormality.
As mentioned above, after shift reduction has been applied to the calculated histograms, at every relative position of the histograms, Jdivergence is calculated. The minimum of Jdivergence corresponds to a maximum reduction in the influence of the RF inhomogeneity and provides histograms for the left and right parts of the MRI which reflect the real interdependence of the intensity distributions.
The Jdivergence may be calculated as follows:
A threedimensional volumetric image may be considered to be a union of two parts, one part representing the left hemisphere and the other part representing the right hemisphere, the two parts being divided by the MSP. The distributions of values that voxel intensities take on these parts may be considered to be the probability distributions of a discrete random value. These probability distributions may be denoted as p={p_{i}} and q={q_{i}}, where p_{i }and q_{i}, are the probabilities of occurrence of the voxel with intensity i in the left and right parts respectively, a voxel being a volume which is the smallest distinguishable boxshaped part of a threedimensional image.
The quantity log(1/p_{i}) is termed ‘surprise’ or ‘unexpectedness’. If p_{i}=1 then an event is certain to happen, and zero surprise is expected if the event does occur. If the event is nearly impossible (p_{i }0), this means that there is infinite surprise if the event does occur. The difference in unexpectedness for these events may be denoted as (log(1/p_{i}}−log {1/q_{i})). Averaging over all the intensities gives a divergence of unexpectedness which may be expressed as:
The above function I(p/q) is known as the KullbackLeibler'"'"' divergence or the crossentropy. It gives an information divergence measure between the two probability distributions p and q. In other words, it is a measure of the distance between the distributions. The function I(p/q) is nonnegative, and additive but not symmetric.
In the preferred embodiment of the invention, a symmetric measure is used which is termed the Jdivergence which may be expressed as:
J(p,q)=I(p,q)+I(q,p)
This measure provides a comparison of informational contents of intensity distributions in the left and right parts of an MR image of a brain and substantially provides the distance between the distributions for both of the hemispheres.
Jdivergence is a sum of I(p,q) and I(q,p). These values depend on the ratio of p_{i }and q_{i }but not the values themselves; so Jdivergence has a selfnormalisation feature, Jdivergence may be used in the consideration of MR data with different pulse sequences. The similarity of roughly symmetrical structures may be estimated using the Jdivergence as an indication of the measure of similarity of the two distributions.
As mentioned above, abnormalities change the radiometric response of tissues, and this affects the probability distribution, of intensities for the hemispheres so, using Jdivergence, it is possible to measure the dissimilarity. By virtue of the roughly symmetrical nature of the brain, the cases with a value of Jdivergence measure greater than a predetermined threshold value may be considered as suspicious for abnormality. In a preferred embodiment, the empiric threshold T_{o }which has been determined to minimise the sum of false negative and false positive (wrong) results is approximately 0.008125.
A study has been conducted comprising the analysis of only the part of brain above eyes, as the inferior part of the head starting from the level of the paranasal sinuses is highly asymmetrical. Detection of the start slice has been carried out manually in the study.
The method embodying the invention was tested on phantom and real data from different sources and comprised 165 studies of 97 patients. In real tumour cases, Jdivergence may sometimes detect a tumour with a diameter of as small as about 10 mm.
In tumour cases a complex of tumour tissue, edema and distortion of neighborhood structures all contribute to the asymmetry of the hemispheres and increase Jdivergence.
Out of the 165 studies considered, 55 were healthy patients and 42 were patients with different brain pathologies manifested in MRI. The results of the investigation are shown in Table 1 below.
The values of Jdivergence for the 165 studies of different pulse sequences were calculated. All the information on the normality of the data was known a priori. Abnormalities in the brains studied included different types of tumours, neuroma, multiple metastasis, hematoma, meningioma, and hydrocephalus.
Threshold for the separation of normal and abnormal cases is defined empirically. In the study, the empiric threshold T_{o }which was found to minimise the sum of false negative and false positive (wrong) results is approximately 0.008125.
In Table 2 below, information on the detection of abnormalities using the method embodying the invention and different pulse sequences in the scanner is provided. Table 2 is based on cases, not patients; so if patient has several cases of different pulse sequences and abnormality was detected only for some of the patient'"'"'s cases, other failed cases have been included in the FN column, even if abnormality were not manifested in the particular case.
The contribution of different intensities to the Jdivergence was analysed and the maximum (principal) summands determined. These summands correspond to the tissues which contribute mainly to asymmetry. All nonprincipal intensities were removed from the image (that is, they were set to black colour).
The scans from the brain of
In summary, embodiments of the invention use an informational divergence measure to calculate the similarity of two distributions corresponding to both hemispheres of a brain to estimate brain asymmetry. The embodiments of the invention provide a rapid method for identification and localisation of abnormalities in the cerebrum. Furthermore, the methods embodying the invention are pulse sequence independent
Preferred methods embodying the invention provide a technique for detecting and locating abnormalities in the MR images on the basis of an analysis of symmetry.
The present invention has many applications, for example in the field of neuroimaging, and in the study of pathologies displayed in MR images.
Using a Pentium IV, 2.4 GHz processor, RAM 512M, it has been found that it takes less than 3 seconds using a method according to an embodiment of the invention to judge the normality of the data being studied.
It will be appreciated that the scope of the present invention is not restricted to the described embodiments. Numerous other modifications, changes, variations, substitutions and equivalents will therefore occur to those skilled in the art without departing from the spirit and scope of the present invention.