KR20070091714A - Method and apparatus of diagnosis micro-calcification - Google Patents

Abstract

A method and an apparatus for diagnosing a micro-calcification phenomenon are provided to improve accuracy of a diagnosed result by analyzing images on two interest regions which are taken by a mammography process. An apparatus(100) for diagnosing a micro-calcification phenomenon includes an interface unit(106), a memory unit(104), and a controller(102). The interface unit provides an interface with a camera. The memory unit stores various information. The controller receives upper/lower and inner/outer image information from the camera and determines interest regions for the respective information. The controller generates a characteristic matrix on the interest regions and calculates character vectors on the interest regions from the character matrix. The controller classifies the character vectors whose correlation coefficient is higher than a predetermined value, and selects one of the classified character vectors as an eigenvector. The controller diagnoses the micro-calcification phenomenon by using the eigenvector and the character vectors with small correlation coefficients.

Description

Microcalcification diagnostic system and method {METHOD AND APPARATUS OF DIAGNOSIS MICRO-CALCIFICATION}

1 is a block diagram of a microcalcification diagnostic system according to a preferred embodiment of the present invention.

2 is a flow chart of a microcalcification method according to a preferred embodiment of the present invention.

3 is a diagram illustrating an equation for generating an SGLD matrix;

4 shows SGLD parameters.

5 shows an example of generation of an SGLD matrix.

6 and 7 illustrate definitions and equations for calculating SGLD feature vectors.

8 and 9 are detailed flowcharts of a SGLD feature vector selection method according to a preferred embodiment of the present invention.

10 is a diagram illustrating a change in diagnostic reliability according to an increase in the number of SGLD feature vectors.

The present invention relates to mammography, and more particularly, whether the microcalcifications are benign or malignant with respect to two regions of interest for the same lesion obtained from the upper and lower imaging information obtained by mammography and the internal and external imaging information. A microcalcification diagnostic system and method for discriminating.

Mammography for mammography is an apparatus for capturing a breast image using X-rays, and an image captured by the mammography is called a mammogram. The mammogram can be used to diagnose breast cancer early.

Unlike the above, biopsy and ultrasound may be used for diagnosing breast cancer, but micro-calcification, which is an early stage of breast cancer, can only be diagnosed by mammography using mammography. to be.

The microcalcification is a very small calcium mass and has a size of less than 1 [mm], and has a calcium-like property. Therefore, it is difficult to distinguish the noise from the mammogram. It was hard to make.

Thus, in the related art, a CAD system (Computer Aided Detection System) for microcalcification and mass detection has been used in clinical practice in various countries including the United States in order to assist doctors in diagnosis.

The microcalcified computer aided detection system (CADe) and the microcalcified computer aided diagnosis system (CADx) will be described in more detail.

The CADe extracts and guides a region of interest (ROI) having a possibility of microcalcification based on a characteristic having a high intensity of microcalcification, and a doctor may diagnose the region of interest and diagnose it as malignant or modal. have.

Recently, in order to assist the diagnosis of a doctor, a system for identifying a positive or a positive diagnosis has been proposed, but it provides independent results for each of the upper and lower imaging information and the internal and external imaging information. have.

As the two pieces of imaging information are used independently instead of comprehensively, there is a problem of low specificity and low sensitivity.

The low specificity lowers the diagnostic efficiency of the doctor by judging the malignant lesions positive, and the low sensitivity has a problem of determining the benign as malignant and causing a fatal result.

For this reason, there is an urgent need for the development of a technique for determining whether the microcalcifications are benign or malignant with respect to two regions of interest for the same lesion obtained from two imaging information.

SUMMARY OF THE INVENTION The present invention is to overcome the above-described problems, and whether the microcalcifications are benign or malignant with respect to two regions of interest for the same lesion obtained from the top and bottom imaging information captured by the mammography and the internal and external imaging information. It is an object of the present invention to provide a microcalcifications diagnostic system and method for determining.

Another object of the present invention is to provide a microcalcification diagnostic system and method for selecting some of the SGLD feature vectors for determining whether the microcalcifications are benign or malignant according to correlation coefficients between the SGLD feature vectors.

The microcalcifications diagnostic method according to the present invention for achieving the above object and to solve the problems of the prior art comprises the steps of selecting a region of interest for each of the upper and lower and internal and external imaging information of the breast; Generating an SGLD matrix for the regions of interest; Calculating SGLD feature vectors for the regions of interest from the SGLD matrix for the regions of interest; Calculating a correlation coefficient of SGLD feature vectors for the ROIs; Grouping the SGLD feature vectors whose correlation coefficient is greater than or equal to a predetermined threshold and selecting one of the SGLD feature vectors in the group as a representative feature vector; And performing a diagnosis by using the representative feature vector and the SGLD feature vectors below the predetermined threshold.

In addition, the microcalcifications diagnostic method according to the present invention comprises the steps of selecting a region of interest for each of the upper and lower and internal and external imaging information of the breast; Generating an SGLD matrix for the regions of interest; Calculating SGLD feature vectors for the regions of interest from the SGLD matrix for the regions of interest; Calculating a correlation coefficient of SGLD feature vectors for the ROIs; Grouping the SGLD feature vectors whose correlation coefficient is greater than or equal to a predetermined threshold and selecting one of the SGLD feature vectors in the group as a representative feature vector; Generating an SGLD matrix for each of the regions of interest; Calculating SGLD feature vectors for each of the regions of interest from the SGLD matrix for each of the regions of interest; For each of the grouped SGLD feature vectors, a correlation coefficient between SGLD feature vectors for each of the regions of interest corresponding to each of the grouped SGLD features is calculated, and a feature vector having the largest value among the calculated correlation coefficients is represented. Selecting to; And performing a diagnosis by using the representative feature vector and the SGLD feature vectors below the predetermined threshold.

The configuration of a microcalcifications diagnostic system according to a preferred embodiment of the present invention is shown.

The diagnostic apparatus 100 is a computer system or the like and includes a control unit 102, a memory unit 104, a user interface unit 106, a display control unit 108, a display 110, and a communication module 112.

The control unit 102 controls the diagnosis apparatus 100 as a whole, and receives upper and lower imaging information and internal and external imaging information from the mammography 114 according to a preferred embodiment of the present invention. A plurality of SGLD feature vectors are calculated for the four ROIs, a portion of the plurality of SGLD feature vectors is selected according to a correlation coefficient between the SGLD feature vectors, and a positive or malignant is determined from the selected SGLD feature vectors.

The memory unit 104 stores various information including a processing program of the control unit 102.

The user interface unit 106 receives various commands from the user and provides them to the controller 102.

The display controller 108 displays various types of information according to the control of the controller 102 on the display 110.

The communication module 112 provides an interface between the mammography 114 and the controller 102.

The mammography 114 performs image capturing up and down and internal and external sides of the breast, converts the scratch signal into digital imaging information, and then transmits the image to the control unit 102.

A microcalcification method according to a preferred embodiment of the present invention applicable to the microcalcification diagnosis system will be described with reference to the flowchart of FIG. 2.

The mammography 114 captures the upper and lower and inner and outer sides of the breast, and digitally converts the captured image information and transmits the captured image information to the control unit 102 of the diagnostic apparatus 100 (steps 200 and 202).

The controller 102 receives the up-and-down and internal / external-side imaging information and selects a region of interest ROI of the same region among the imaging information (step 204).

After selecting a region of interest for the up-and-down and internal / external-side imaging information, the controller 102 generates a SGLD (Sptial Gray Level Dependence) matrix for the region of interest for the up-down and internal-side imaging information.

The SGLD matrix represents a frequency of occurrence of combinations of gray levels of pixels in an image. The generation of the SGLD matrix is according to the equation of FIG. 3, and if the gray intensity is Ng, It becomes Ng * Ng size. Since the generation process of the SGLD matrix is already known, a detailed description thereof will be omitted.

For example, suppose that an inter-pixel distance D is 1 and produces an SGLD matrix having a parameter of 0 degrees. As shown in FIG. 4, the zero degree targets pixels in which the target pixel and the consideration pixel have a horizontal relation of zero degree. And since D is 1, a relationship where D is 1 and 0 degrees considers two pixels that are one pixel difference horizontally.

If the image having a size of 5 * 4 is assumed as the target image as shown in FIG. 5A, the image coordinates are (1,1), (1,2), (1,2), (1,3) Pixels such as are targeted. Here, the coordinates of (L, M) represent vertical coordinates and horizontal coordinates.

Since the pair of pixels having the image brightness (1,1) among the target pixels is a total of one, the SGLD matrix generated based on the target image shown in (A) of FIG. 5 is 1 at the position of (1,1). Is recorded, and 2 is recorded at the position of (1, 2) since the pairs of pixels having the image brightness of (1, 2) are two times in total. This SGLD matrix is expressed as P (I, J, D = 0,0 °). The conversion of the uppercase letter to the relative frequency of the SGLD matrix is called a normalized SGLD matrix, and is expressed as p (i, j, d = 0,0 °). Here, the leveling refers to dividing each element of the matrix by the sum of all elements.

When generation of the leveled SGLD matrix of the ROI of the up-and-down and internal / external side imaging information is completed, the control unit 102 determines 12 SGLD feature vectors from the leveled SGLD matrices of the ROI of the top-down and internal / external side imaging information. To calculate.

Definitions and equations for calculating the twelve SGLD feature vectors are as shown in FIGS. 6 and 7, and the twelve SGLD feature vectors are already known, and thus a detailed description thereof will be omitted.

When the 12 SGLD feature vectors for the ROI of the up and down and the internal and external imaging information are calculated, the control unit 102 converts some of the 12 SGLD feature vectors for the ROI of the up and down and internal and external imaging information into the final feature vector. Select (step 208).

When the final feature vector is selected, the controller 102 determines whether the target region of interest is benign or malignant through the SVM classifier and guides it to a doctor or the like (step 210).

The SGLD feature vector selection process (step 208) will be described in more detail.

The present invention measures a correlation coefficient between twelve SGLD feature vectors for the ROI of up-and-down and internal / external-side imaging information, stores only feature vectors whose correlation coefficient is greater than or equal to a predetermined threshold, and stores the feature vectors stored in the cluster. Only one of them is selected as the representative feature vector.

The selection method includes measuring a correlation coefficient between all SGLD feature vectors in a corresponding cluster and a label value of data, and selecting an SGLD feature vector having the largest correlation coefficient value, or a cluster. For all SGLD feature vectors inside, measure the correlation coefficient of the corresponding SGLD feature vector between the top and bottom image and the internal / external side image information in the malicious data, and similarly the corresponding SGLD between the top and bottom image information and the internal and external side image information in the positive data. After measuring the correlation coefficient of the feature vector, selecting the SGLD feature vector having the largest difference as the representative feature vector may be employed.

Such SGLD feature vector selection methods will be described in more detail with reference to FIGS. 8 and 9.

First, the SGLD feature vector selection method according to FIG. 8 will be described.

The control unit 102 of the diagnostic apparatus 100 selects a region of interest from the upper and lower and internal and external imaging information, and calculates 12 SGLD feature vectors f1, f2, .. f12 for the selected region of interest (step 300). ). When the SGLD feature vectors are calculated, the control unit 102 calculates a correlation coefficient between 12 SGLD feature vectors according to Equation 1 below (step 302).

The number of correlation coefficients generated according to Equation 1 is 132 (12 * 11).

The controller 102 stores the SGLD feature vectors having a value equal to or greater than a predetermined threshold value in the same cluster (step 304). For example, if the correlation coefficient of f1 and f2 is equal to or greater than the threshold, f1, f2 is stored in the first cluster C1. If the correlation coefficient of f3, f5, f7 is equal to or greater than the threshold, the second cluster ( F3, f4, f5, etc. can be stored in C2).

Thereafter, the controller 102 selects a representative feature vector from each cluster (step 306). In more detail, the correlation coefficient between all SGLD feature vectors and data label values in each cluster is measured, and the SGLD feature vector having the largest value among the correlation coefficient values is selected as the representative feature vector. For example, after calculating R (f1, label) and R (f2, label) in the first cluster C1, the SGLD feature vector having the larger of the two values is selected as the representative feature vector.

The label of the data means a true value indicating whether the data is substantially malicious or substantially positive.

For example, if the label is malicious but the system determines to be positive, this is a discrimination error, ie false positive (FALSE NAGATIVE), and if the label is malicious and the system determines to be malicious, then it is a normal discrimination, that is, a correct malicious (TRUE POSITIVE).

In addition, if the label is positive and the system determines to be malignant, this is a discrimination error, ie false malicious (FALSE POSITIVE), and if the label is positive and the system is positive, then it is normal discrimination, i.e. true positive (TRUE). NEGATIVE).

Accordingly, the present invention improves the accuracy (ACCURACY), sensitivity, and specificity of the discrimination through the correlation coefficient of the label value.

When the representative feature vectors of each cluster are selected in the above-described manner, the control unit 102 selects the SGLD feature vectors that are not stored in the cluster as the final feature vector according to the correlation coefficient below the threshold with the selected representative feature vectors. Step 308). The selected final feature vector is input to an SVM determiner to determine whether it is benign or malignant for the region of interest.

Now, the SGLD feature vector selection method shown in FIG. 9 will be described.

The control unit 102 of the diagnostic apparatus 100 selects a region of interest from the upper and lower and internal and external imaging information, and calculates 12 SGLD feature vectors f1, f2, .. f12 for the selected region of interest (step 400). ). When the feature vectors of the SGLD are calculated, the control unit 102 calculates a correlation coefficient between 12 SGLD feature vectors according to Equation 1 (step 404).

The control unit 102 stores the SGLD feature vectors having a value equal to or greater than a predetermined threshold value in the same cluster (step 406). For example, if the correlation coefficient of f1 and f2 is equal to or greater than the threshold, f1, f2 is stored in the first cluster C1. If the correlation coefficient of f3, f5, f7 is equal to or greater than the threshold, the second cluster ( F3, f4, f5, etc. can be stored in C2).

The control unit 102 also calculates twelve SGLD feature vectors f1cc, f1mlo, f2cc, f2mlo ... f12cc, and f12mlo for each of the regions of interest of the up-and-down and internal and external imaging information. More specifically, the feature vectors (f1cc, f2cc, ... f12cc) are extracted from the top and bottom images, and the feature vectors (f1mlo, f2mlo… f12mlo) are extracted from the internal and external images, and 24 SGLD feature vectors (f1cc, f1mlo, f2cc, f2mlo ...) are extracted. f12cc, f12mlo) are generated (step 402).

Thereafter, the controller 102 selects a representative feature vector from each cluster (step 408). In more detail, the control unit 102 calculates a correlation coefficient between all feature vectors in each cluster and feature vectors in up and down and internal and external imaging data of malicious data, and also captures up and down and internal and external imaging information of positive data. The correlation coefficient between the feature vectors is calculated, and the feature vector having the largest difference is selected as the representative feature vector. For example, if the feature vector f1 and the feature vector f2 are selected as shown in step 406 of FIG. 9 in the first cluster C1, after measuring R (f1cc, f1mlo) and R (f2cc, f2mlo), It is to select a feature vector with a large value.

In this way, the correlation coefficient of the corresponding SGLD feature vector between the upper and lower imaging information and the internal and external imaging information in the malicious data is measured, and the correlation coefficient of the corresponding SGLD feature vector between the upper and lower imaging information and the internal and external imaging information is also measured in the positive data. After that, the SGLD feature vector having the largest difference is selected as the representative feature vector.

When the representative feature vectors of each cluster are selected in the above-described manner, the controller 102 selects feature vectors that are not stored in the cluster as the final feature vector according to the correlation coefficient below the threshold with the selected representative feature vectors (410). step).

The selected final feature vector is input to the SVM discriminator to determine whether it is benign or malignant for the region of interest.

As described above, the present invention does not use the 12 SGLD feature vectors obtained from the two ROIs selected from the upper and lower images and the external / external side image pickup information, but rather the correlation between the 12 SGLD feature vectors for the two ROIs. Therefore, only a part of the selective use of microcalcification improves the reliability of discriminating whether it is positive or malignant.

As shown in FIG. 10, the diagnostic reliability increases with an increase in the number of SGLD feature vectors, but decreases again as the number of SGLD feature vectors with a high correlation coefficient increases.

Accordingly, the present invention has the advantage of improving the diagnostic reliability by enabling the diagnosis to be performed by selecting only one of the SGLD feature vectors having a high correlation coefficient.

As described above, the present invention has the advantage of being able to discriminate positively or malignantly the microcalcification of two regions of interest for the same lesion obtained from the top and bottom imaging information captured by mammography and the internal and external imaging information. There is this.

In addition, the present invention has the advantage of improving the diagnostic reliability by selecting some of the SGLD feature vectors for determining whether the microcalcification is positive or malignant according to the correlation coefficient between the label or the SGLD feature vectors.

As described above, the present invention has been described by way of limited embodiments and drawings, but the present invention is not limited to the above-described embodiments, which can be variously modified and modified by those skilled in the art to which the present invention pertains. Modifications are possible.

Accordingly, the spirit of the present invention should be understood only by the claims set forth below, and all equivalent or equivalent modifications thereof will belong to the scope of the present invention.

Claims (8)

In the microcalcification diagnostic method, Selecting a region of interest for each of the upper and lower sides of the breast and the internal and external imaging information; Generating an SGLD matrix for the regions of interest; Calculating SGLD feature vectors for the regions of interest from the SGLD matrix for the regions of interest; Calculating a correlation coefficient of SGLD feature vectors for the ROIs; Grouping the SGLD feature vectors whose correlation coefficient is greater than or equal to a predetermined threshold and selecting one of the SGLD feature vectors in the group as a representative feature vector; Performing a diagnosis using the representative feature vector and the SGLD feature vectors below the predetermined threshold; Microcalcifications diagnostic method comprising a. The method of claim 1, SGLD feature vectors whose correlation coefficient is greater than or equal to a predetermined threshold are stored in differently assigned clusters, Calculating a correlation coefficient between each SGLD feature vector stored in each cluster and the data label, And the SGLD feature vector having the largest correlation coefficient is selected as the representative feature vector. In the microcalcification diagnostic method, Selecting a region of interest for each of the upper and lower sides of the breast and the internal and external imaging information; Generating an SGLD matrix for the regions of interest; Calculating SGLD feature vectors for the regions of interest from the SGLD matrix for the regions of interest; Calculating a correlation coefficient of SGLD feature vectors for the ROIs; Grouping the SGLD feature vectors whose correlation coefficient is greater than or equal to a predetermined threshold and selecting one of the SGLD feature vectors in the group as a representative feature vector; Generating an SGLD matrix for each of the regions of interest; Calculating SGLD feature vectors for each of the regions of interest from the SGLD matrix for each of the regions of interest; For each of the grouped SGLD feature vectors, a correlation coefficient between SGLD feature vectors for each of the regions of interest corresponding to each of the grouped SGLD features is calculated, and a feature vector having the largest value among the calculated correlation coefficients is represented. Selecting to; Performing a diagnosis using the representative feature vector and the SGLD feature vectors below the predetermined threshold; Microcalcifications diagnostic method comprising a. The method according to claim 1 or 3, The correlation coefficient is a microcalcifications diagnostic method, characterized in that calculated according to Equation 2.

In microcalcifications diagnostic system, An interface unit for providing an interface with the imaging device; A memory unit for storing various kinds of information; Receiving the upper and lower and inner and outer side imaging information of the breast from the imaging apparatus, and selecting a region of interest for each, Generate an SGLD matrix for the regions of interest, Calculating SGLD feature vectors for the regions of interest from the SGLD matrix for the regions of interest; Calculating a correlation coefficient of SGLD feature vectors for the ROIs, Grouping the SGLD feature vectors whose correlation coefficient is greater than or equal to a predetermined threshold, selecting one of the SGLD feature vectors in the group as a representative feature vector, A control unit for performing a diagnosis using the representative feature vector and the SGLD feature vectors below the predetermined threshold Microcalcifications diagnostic system comprising a. The method of claim 5, The control unit, SGLD feature vectors whose correlation coefficient is greater than or equal to a predetermined threshold are stored in clusters that are given differently, Calculating a correlation coefficient between each SGLD feature vector stored in each cluster and the data label, And a SGLD feature vector having the largest correlation coefficient as a representative feature vector. In microcalcifications diagnostic system, An interface unit for providing an interface with the imaging device; A memory unit for storing various kinds of information; Receiving the upper and lower and inner and outer side imaging information of the breast from the imaging apparatus, and selecting a region of interest for each, Generate an SGLD matrix for the regions of interest, Calculating SGLD feature vectors for the regions of interest from the SGLD matrix for the regions of interest; Calculating a correlation coefficient of SGLD feature vectors for the ROIs, Grouping SGLD feature vectors whose correlation coefficient is greater than or equal to a predetermined threshold, Generating an SGLD matrix for each of the regions of interest, Calculating SGLD feature vectors for each of the regions of interest from the SGLD matrix for each of the regions of interest; For each of the grouped SGLD feature vectors, a correlation coefficient between SGLD feature vectors for each of the regions of interest corresponding to each of the grouped SGLD features is calculated, and a feature vector having the largest value among the calculated correlation coefficients is represented. , Select A control unit for performing a diagnosis using the representative feature vector and the SGLD feature vectors below the predetermined threshold Microcalcifications diagnostic system comprising a. The method according to claim 5 or 7, The correlation coefficient is a microcalcifications diagnostic system, characterized in that calculated according to the equation (3).

Images