JP6299504B2 - Image analysis device - Google Patents

Description

  The present invention relates to an image analysis apparatus that calculates a texture analysis index by performing texture analysis on an image in which a subject is reflected.

  A medical institution is equipped with an image analysis device that performs image analysis on a subject image obtained by radiographing a subject and calculates an evaluation value for evaluating the state of the subject.

  A problem in image analysis is the body thickness of the subject that appears in the image. Radiation has the property that the greater the body thickness, the less likely it is to penetrate the subject. Therefore, as the body thickness of the subject is thicker, the exposure becomes less exposed and the S / N ratio of the obtained tomographic image becomes worse. As the S / N ratio of a tomographic image becomes worse, more noise components appear in the image. Therefore, in order to perform image analysis, it is necessary to correct the body thickness of the subject (for example, see Patent Document 1).

  One of image analysis techniques is texture analysis. In this texture analysis, an image is converted into a co-occurrence matrix and a co-occurrence matrix is calculated to calculate a texture analysis index which is a kind of evaluation value. The co-occurrence matrix represents statistics on how the pixel values constituting the image vary, and is a matrix that varies depending on the state of the subject image on the image (for example, Non-Patent Document 1, Non-Patent Document 1). Patent Document 2).

JP 2001-86409 A Haralick RM. Statistical and structural approaches to texture. Proc IEEE 1979; 67 (5): 786-804. Haralick RM. et al. Textural Features for Image Classification. IEEE Transactions on Systems Man and Cybernetics 1973; 6: 610-621.

However, the image analysis apparatus according to the conventional configuration has the following problems.
That is, according to the image analysis apparatus according to the conventional configuration, the influence of the body thickness of the subject is not sufficiently considered in the image analysis using the texture analysis.

  In order to describe a conventional image analysis apparatus, here, trabecular analysis is taken as an example. A trabecular bone is an elongated structure that forms the sponge within the bone. The subject of trabecular analysis of the conventional configuration is a tomographic image of the subject obtained by performing radiography on the subject.

  In trabecular analysis using texture analysis, it is ideal to calculate a texture analysis index representing only information about trabeculae. In other words, the texture analysis index for trabecular analysis originally varies depending on the state of the trabecular bone and should not vary depending on other conditions.

  However, when a trabecular analysis is performed on an actual tomographic image, the texture analysis index calculated as the body thickness of the subject increases. The reason why such a phenomenon occurs is that the amount of noise components in the tomographic image varies according to the body thickness of the subject. The noise component does not represent the state of the trabecular bone of the subject. Nevertheless, in the texture analysis, the noise component is regarded as a kind of variation in the pixel values constituting the tomographic image, which is taken into account in the calculation of the texture analysis index. Texture analysis captures changes in the content of noise components in tomographic images and reflects them in the texture analysis index.

  If there is a texture analysis method that does not affect the influence of the body thickness on the texture analysis index, a more accurate texture analysis index can be calculated. However, according to the conventional image analysis apparatus, the same texture analysis is performed on subjects having different body thicknesses, and such consideration is not made. How can texture analysis be modified to prevent the effect of body thickness? Note that the conventionally known correction of body thickness is not for texture analysis.

  The present invention has been made in view of such circumstances, and an object thereof is to provide an image analysis apparatus capable of realizing texture analysis that calculates an appropriate texture analysis index without being affected by the body thickness of a subject. There is to do.

The present invention has the following configuration in order to solve the above-described problems.
That is, the image analysis apparatus according to the present invention is an image analysis apparatus that analyzes a radiographic image obtained by radiography of a subject, and divides the gradation of pixel values included in the radiographic image into a plurality of gradation sections. The gradation width setting means for setting the respective gradation widths, and a pair of two pixels whose pixel values belong to a predetermined gradation section, the pixels being separated from each other by a predetermined distance are radiation. A co-occurrence matrix generating means for generating a co-occurrence matrix by counting how many times it appears in the analysis range provided in a part of the image for each combination of gradation intervals, and calculating a texture analysis index based on the co-occurrence matrix And a tone width setting means for setting the tone width so as to increase as the noise component included in the radiographic image increases.

[Operation / Effect] According to the configuration of the present invention, a texture analysis technique in which the difference in body thickness of the subject does not affect the analysis result is specifically shown. That is, the gradation width is set so as to increase as the noise component included in the radiation image increases. This gradation width represents the gradation width to be identified at the time of co-occurrence matrix generation. That is, according to the present invention, when the noise component included in the radiographic image increases, the elements constituting the co-occurrence matrix are reduced. As the noise component of the radiographic image increases, the variation in pixel values in the radiographic image increases accordingly. In texture analysis of such an image, if the elements of the co-occurrence matrix are increased as in the conventional configuration, the variation in pixel values derived from the noise component is also expressed in the co-occurrence matrix. It becomes the value which considered the noise component.
As in the configuration of the present invention, as the noise component of the image increases, the number of elements constituting the co-occurrence matrix is reduced, so that the influence of the variation in pixel values derived from the noise component is absorbed when the co-occurrence matrix is generated during the analysis. Therefore, it does not appear in the texture analysis index. Therefore, according to the image analysis apparatus of the present invention, a more reliable texture analysis index can be calculated.

  In the above-described image analysis apparatus, it is more desirable that the tone width setting unit estimates the noise component based on the body thickness of the subject.

  [Operation / Effect] The above-described configuration more specifically shows the image analysis apparatus of the present invention. The amount of the noise component included in the radiation image varies in correlation with the body thickness of the subject. The amount of the noise component can be estimated from the body thickness of the subject.

  In the above-described image analysis apparatus, it is more desirable to include an input unit that allows the operator to input the body thickness.

  [Operation / Effect] The above-described configuration more specifically shows the image analysis apparatus of the present invention. If the configuration is such that the operator inputs the body thickness of the subject used to estimate the amount of the noise component, the body thickness can be acquired more reliably.

  In the above-described image analysis apparatus, it is more desirable to estimate the body thickness based on the variation in pixel values constituting the subject image reflected in the radiation image.

  [Operation / Effect] The above-described configuration more specifically shows the image analysis apparatus of the present invention. If the body thickness is estimated from the intensity of the variation in pixel values seen between the pixels constituting the subject image, the amount of noise component can be calculated more reliably.

  The image analysis apparatus includes a storage unit that stores a relationship between a gradation width that is referred to by the gradation width setting unit and a noise component, and the relationship that the storage unit stores is a copy of a subject having a certain body thickness. It is more desirable if the texture analysis index obtained by texture analysis of the included radiographic image is set to be the same as the texture analysis index for the reference radiographic image.

  [Operation / Effect] The above-described configuration more specifically shows the image analysis apparatus of the present invention. When texture analysis is performed on a radiographic image of a subject with a certain body thickness, the relationship between the tone width and the noise component determined so that the obtained texture analysis index is the same as the texture analysis index for the reference radiographic image If the gradation width setting means operates according to the characteristics, the gradation width setting means operates based on the actual measurement result, so that a more reliable texture analysis index can be acquired.

  It is more desirable to calculate the trabecular index of the subject based on the texture analysis index calculated by the above-described image analysis apparatus.

  [Operation / Effect] As described above, the configuration of the present invention can be specifically applied to trabecular analysis.

  Further, in the trabecular bone analysis apparatus described above, as a texture analysis index calculated by the index calculation means, one of correlation, dissimilarity, contrast, homogeneity, entropy, angler second moment, variance, and inverse differential moment is used. Or it is more desirable if a plurality are selected.

  [Operation / Effect] The above-described configuration more specifically shows the image analysis apparatus of the present invention. The configuration of the present invention can be applied to various texture analysis indices.

  According to the configuration of the present invention, a method of texture analysis in which the difference in body thickness of the subject does not affect the analysis result is specifically shown. That is, the gradation width is set so as to increase as the body thickness of the subject increases. This gradation width represents the gradation width to be identified at the time of co-occurrence matrix generation. In other words, the present invention is devised so that the number of elements constituting the co-occurrence matrix decreases as the body thickness of the subject increases. In this way, the influence of the variation in pixel values derived from the noise component is absorbed when the co-occurrence matrix is generated during the analysis, and a more reliable texture analysis index can be calculated.

1 is a functional block diagram illustrating an overall configuration of an image analysis device according to Embodiment 1. FIG. 6 is a schematic diagram illustrating background processing according to Embodiment 1. FIG. It is a schematic diagram explaining the body thickness value acquisition process which concerns on Example 1. FIG. 3 is a schematic diagram illustrating a table according to Embodiment 1. FIG. FIG. 6 is a schematic diagram illustrating a co-occurrence matrix generation process according to the first embodiment. FIG. 6 is a schematic diagram illustrating a co-occurrence matrix generation process according to the first embodiment. FIG. 6 is a schematic diagram illustrating a co-occurrence matrix generation process according to the first embodiment. FIG. 6 is a schematic diagram illustrating a co-occurrence matrix generation process according to the first embodiment. It is a schematic diagram which shows the relationship between the gradation width which concerns on Example 1, and body thickness value. It is a schematic diagram explaining the principle which the effect of the structure which concerns on Example 1 produces. 6 is a schematic diagram illustrating a method for demonstrating the effect of the configuration according to Embodiment 1. FIG. 6 is a schematic diagram illustrating a method for demonstrating the effect of the configuration according to Embodiment 1. FIG. 6 is a schematic diagram illustrating a method for demonstrating the effect of the configuration according to Embodiment 1. FIG. 6 is a schematic diagram illustrating a method for demonstrating the effect of the configuration according to Embodiment 1. FIG. FIG. 6 is a schematic diagram for explaining the effect of the configuration according to the first embodiment.

  Next, an embodiment of an image analysis apparatus according to the present invention will be described with reference to the drawings. Note that a tomographic image of the subject M acquired by the filter back projection method is taken as an example of an image to be processed by the image analysis processing apparatus of the present invention. The tomographic image corresponds to the radiation image of the present invention. The image analysis apparatus according to the embodiment performs image processing on the input tomographic image to calculate an index value for texture analysis related to trabecular analysis.

  FIG. 1 is a functional block diagram showing the overall configuration of an image analysis apparatus according to the present invention. In the following, details of the operation of each unit shown in FIG. 1 will be described in order.

<Operation of Background Processing Unit 11>
The tomographic image D is first sent to the background processing unit 11. The background processing unit 11 performs background processing such as unsharpness masking processing on the tomographic image D, and generates a tomographic image Da subjected to background processing. By such background processing, it is possible to eliminate the difference in the reflection of the subject image due to the difference in the imaging conditions seen between the different tomographic images D. The background processing performed by the background processing unit 11 is not limited to unsharpness masking processing, and may be other processing.

  FIG. 2 briefly describes the unsharpness masking process. The unsharpness masking process is an image process for subtracting a smoothed image obtained by smoothing the tomographic image D from the tomographic image D. By such image processing, the tomographic image D is converted into a clearer tomographic image Da.

<Operation of Variation Calculation Unit 12>
The tomographic image Da is sent to the variation calculation unit 12. The variation calculation unit 12 analyzes the tomographic image Da and calculates a standard deviation SD indicating the body thickness of the subject M reflected in the tomographic image Da. FIG. 3 shows a state in which the variation calculation unit 12 analyzes the tomographic image Da. The variation calculation unit 12 first recognizes the analysis region R set by the operator through the input unit 13. The analysis region R is selected so that the portion of the subject image in the image where the bone is not reflected (the portion where the soft tissue of the subject image is reflected) is the entire analysis region R. Therefore, in the analysis region R, the soft tissue of the subject image is reflected. The input unit 13 is configured to allow an operator to input an input value and corresponds to the input unit of the present invention.

  Then, the variation calculation unit 12 performs statistical analysis on the pixels constituting the recognized analysis region R to calculate the standard deviation SD, and sets this as a body thickness value indicating the body thickness of the subject M. It can be said that the variation calculation unit 12 at this time estimates the body thickness based on the size (standard deviation SD) of the pixel values constituting the subject image reflected in the tomographic image Da.

  The body thickness value (standard deviation SD) indicates the variation in the pixel values in the pixels belonging to the analysis region R, and also indicates the variation in the pixel values seen between the pixels constituting the soft tissue image of the subject M. . It can be evaluated that the larger the standard deviation SD, the greater the variation in pixel values. From this standard deviation SD, the amount of noise component (noise amount) included in the tomographic image Da can be known. In the tomographic image Da, a noise component is superimposed on the subject image. Specifically, the noise component is a sandstorm-like pattern. As the amount of noise superimposed on the image increases, a sandstorm-like pattern appears more strongly in the image. That is, the greater the amount of noise, the greater the variation in the pixel value of the pixel.

  That is, the body thickness value (standard deviation SD) is also an index indicating the amount of noise in the image. That is, it can be evaluated that the larger the standard deviation SD, the greater the amount of noise. On the other hand, the soft tissue image of the subject M reflected in the analysis region R does not contribute much to the standard deviation SD. This is because the soft tissue image is composed of similar pixel values.

The reason for setting the analysis region R to the soft tissue of the subject M is as follows. That is, the standard deviation SD calculated by the variation calculation unit 12 is a parameter that varies depending on the body thickness of the subject M. That is, when the standard deviation SD is large, the tomographic image D is captured with insufficient exposure. The underexposed because caused by body thickness of the subject M is thick, standard deviation SD is not so that that indirectly represent the body thickness of the subject M.

  By the way, the bone part of the subject M originally contains a lot of noise components because it hardly transmits X-rays. Therefore, the standard deviation SD of the pixel value in the bone part of the subject M takes a large value regardless of the body thickness of the subject M. Therefore, in order to examine the body thickness value, it is better to calculate the standard deviation of the soft tissue of the subject M.

<Operation of the gradation width setting unit 14>
The variation calculation unit 12 sends the body thickness value (standard deviation SD) in the tomographic image Da to the gradation width setting unit 14. The gradation width setting unit 14 acquires the width value Wa corresponding to the sent standard deviation SD based on a table as shown in FIG. 4 and sets the acquired width value Wa as the gradation width W. Each width value Wa in the table is defined to increase, for example, linearly as the standard deviation SD increases. Therefore, the gradation width setting unit 14 widens each gradation width W when the gradation of the pixel value included in the tomographic image Da is divided into each of a plurality of gradation sections according to an increase in noise components. Will be set. At this time, the gradation width setting unit 14 estimates the amount of noise components included in the radiation image based on the body thickness value (standard deviation SD).

  The gradation width setting unit 14 is configured to add a complementary process to the table to acquire a new width value Wa when the sent standard deviation SD is not in the table. Further, the gradation width setting unit 14 may execute the setting of the gradation width W based on an equation in which the standard deviation SD and the width value Wa are related. The tables and equations described above are stored in the storage unit 20. The gradation width W set by the gradation width setting unit 14 is sent to the matrix generation unit 15. How the table is generated will be described later. The gradation width setting unit 14 corresponds to the gradation width setting unit of the present invention, and the matrix generation unit 15 corresponds to the co-occurrence matrix generation unit of the present invention.

<Operation of Matrix Generation Unit 15>
There is a co-occurrence matrix (GLCM) as a matrix necessary for performing texture analysis. This matrix is generated by the matrix generation unit 15. The tomographic image Da generated by the background processing unit 11 is sent to the matrix generation unit 15 where it is converted into GLCM. FIG. 5 illustrates an operation in which the matrix generation unit 15 generates a GLCM based on the tomographic image Da. The left side of FIG. 5 represents the tomographic image Da as a two-dimensional array of pixel values. For simplicity of explanation, it is assumed that the pixel values of each pixel constituting the tomographic image Da take ten values from 0 to 9.

  As shown on the right side of FIG. 5, the number of rows and the number of columns of GLCM generated from the tomographic image Da both match the number of pixel values that the pixel value of the pixel can take. Since each pixel constituting the tomographic image Da has one of 10 pixel values, the GLCM generated from the tomographic image Da is a two-dimensional matrix of 10 rows and 10 columns. The matrix generating unit 15 completes the GLCM by assigning numerical values to 100 elements constituting the GLCM that is a 10 × 10 matrix. It is determined based on the pixel value of the tomographic image Da what value is to be entered for each element.

  FIG. 5 shows that the matrix generation unit 15 decides the numerical value of the element p (0, 1) located in the row meaning 0 among the rows of GLCM and the column meaning 1 out of each column. Yes. The matrix generation unit 15 counts the number of pairs of pixels in which the pixel value 0 and the pixel value 1 are arranged adjacent to each other in the tomographic image Da, and calculates the count number to the element p (0, 1) of the GLCM. ). In FIG. 5, since there are two pairs of pixels in which the pixel value 0 and the pixel value 1 are arranged adjacent to each other, the value of the element p (0, 1) is 2. Since the arbitrary element p (a, b) in this GLCM is equal to the element p (b, a), the value of the element p (1, 0) in GLCM is also 2.

  The matrix generation unit 15 performs the same operation over the entire area of the GLCM, and determines all the elements of the matrix based on the tomographic image Da. Thus, the matrix generation unit 15 completes the GLCM based on the tomographic image Da.

  FIG. 6 shows how the matrix generation unit 15 generates GLCM based on the tomographic image Da. The generated GLCM increases as the number of pixel values that can be taken by the pixels of the tomographic image Da increases. The GLCM is a matrix having symmetry, and is a matrix in which the values of overlapping elements are the same when folded in half by a diagonal line shown by a dotted line in FIG.

<Change of operation based on gradation width W>
Next, how the operation of the matrix generation unit 15 is changed based on the gradation width W set by the gradation width setting unit 14 which is the most characteristic configuration of the present invention will be described. That is, the matrix generation unit 15 changes the generated GLCM based on the set gradation width W. The elements of the matrix constituting the GLCM according to the present invention are configured to increase or decrease based on the gradation width W.

  FIG. 7 shows a type of GLCM generated by the matrix generation unit 15. The GLCM in FIG. 5 is generated by distinguishing all the gradations from 0 to 9, but the GLCM in FIG. 7 is generated without distinguishing between 0 and 1. Similarly, the GLCM of FIG. 7 does not distinguish between 2, 3, 4, 5, 6, 7, 8, and 9. This GLCM is a two-dimensional matrix of 5 rows and 5 columns. The matrix generation unit 15 completes the GLCM by assigning numerical values to the 25 elements constituting the GLCM that is a 5 × 5 matrix.

  FIG. 7 shows an element p (in a row meaning a gradation interval (gradation interval) from 0 to 1 in each row of GLCM and a column meaning a gradation interval from 0 to 1 in each column. It shows that the matrix generation unit 15 is trying to determine a numerical value of 0 to 1, 0 to 1). The matrix generation unit 15 has a pixel value in which one of the pixel values belongs to the gradation interval (0 to 1) and the other pixel value belongs to the gradation interval (0 to 1) in the tomographic image Da. It is counted whether there is a set, and the count number is set as an element p (0 to 1, 0 to 1) of GLCM. In FIG. 7, two pairs of pixels in which pixel values 0 and 1 are arranged adjacent to each other are two pairs, and one pair of pixels in which pixel values 1 and 1 are arranged next to each other. Therefore, the value of the element p (0 to 1, 0 to 1) is 3.

  FIG. 8 shows a state in which the matrix generation unit 15 generates a GLCM based on the tomographic image Da. The generated GLCM has fewer components than the GLCM described with reference to FIG.

  The gradation width W indicates the width of the gradation section that is identified when the matrix generation unit 15 generates GLCM. For example, since the matrix generation unit 15 in FIG. 5 has distinguished all of the gradations, the gradation width W in this GLCM generation operation is 1. Further, since the matrix generation unit 15 in FIG. 7 views the two gradations as the same, the gradation width W in this GLCM generation operation is 2. The gradation width W is set by the gradation width setting unit 14 before the matrix generation unit 15 generates GLCM.

  The gradation width W when the matrix generation unit 15 generates GLCM is set to increase as the body thickness value (standard deviation SD) of the tomographic image Da increases. Therefore, the GLCM generated by the matrix generation unit 15 changes according to the standard deviation SD of the tomographic image Da as shown in FIG. When the standard deviation SD of the tomographic image Da is small, as shown in the upper part of FIG. 9, since the gradation width W is small, many gradation sections are set, and the generated GLCM is a matrix with many elements. As the standard deviation SD of the tomographic image Da increases, the gradation width W gradually increases, the number of gradation sections to be set decreases, and is generated as shown in the middle and lower stages of FIG. GLCM is a matrix with fewer elements. Therefore, the matrix generation unit 15 operates so that the scale of the GLCM decreases as the standard deviation SD of the tomographic image Da increases.

  The matrix generation unit 15 receives the gradation width W from the gradation width setting unit 14, divides the gradation of the tomographic image Da in units of gradation width W, and sets a plurality of gradation sections. As a result, the matrix generation unit 15 forms a part of the tomographic image D that is a pair of two pixels whose pixel values belong to a predetermined gradation section and are separated from each other by a predetermined distance. A GLCM (co-occurrence matrix) is generated by counting how many times it appears in the set analysis range for each combination of gradation intervals. As the analysis range, the cancellous bone portion of the bone is selected.

<Effect of changing the scale of GLCM according to the thickness value of the tomographic image Da>
FIG. 10 illustrates the effect of changing the scale of the GLCM generated by the matrix generation unit 15. The upper part of FIG. 10 shows what happens when the GLCM is generated with the gradation width W set to 1 for a part of the tomographic image Da. Since the gradation width W is 1, all the gradations are distinguished and a GLCM is generated. Consider how to treat the vertical 3 × horizontal 3 region shown in the upper part of FIG. 10 in the tomographic image Da when the GLCM is generated. In this region, there are six pairs of two pixels adjacent in the vertical direction and six pairs of two pixels adjacent in the horizontal direction. There are four pairs of two pixels adjacent in the diagonally right direction and four pairs of two pixels adjacent in the diagonally left direction. Since these 20 pairs are all combinations of pixels having a pixel value of 500, they are all treated the same when generating a GLCM. That is, all the 20 pairs are added to the count number of the GLCM element p (500, 500).

  The middle part of FIG. 10 represents a state in which a noise component is superimposed on a part of the tomographic image shown in the upper part of FIG. The pixel values that are all 500 due to the superimposition of the noise component vary depending on the noise component and become 494 or 506. What happens when GLCM is generated with a gradation width W of 1 for a part of a tomographic image on which such noise components are superimposed? Since there is no one combination of pixel values in the 20 pairs on the region, all 20 pairs are treated differently when generating the GLCM. For example, a pixel pair whose pixel value is a combination of 506 and 494 is added to the count number of the GLCM element p (506,494), and the remaining 19 pairs are the elements p (506,494). Not counted in the count.

  From the above, it can be seen that the noise component is superimposed on the tomographic image, thereby affecting the generated GLCM elements. If the tomographic image is disturbed by the noise component, a change appears in the counting operation of the matrix generation unit 15, and the GLCM is also disturbed.

  The lower part of FIG. 10 shows a state where a noise component is superimposed on a part of the tomographic image shown in the upper part of FIG. What happens when a GLCM is generated with a gradation width W of 20 for a part of a tomographic image on which such noise components are superimposed? There is certainly no one combination of pixel values in the 20 pairs on the region. However, since the gradation width W is 20, the matrix generation unit 15 generates GLCM by equating pixel values from 490 to 510, for example. Thus, these 20 pairs will all be treated the same when generating a GLCM. In other words, in this example, all 20 pairs are added to the count number of the GLCM element p (490-510, 490-510).

  From the above, it can be seen that adjusting the gradation width W can prevent the influence of noise components from appearing in the generated GLCM elements. The 20 pairs shown in the lower part of FIG. 10 that should have been treated in the same way are all treated in the same way without being treated differently, and added to the generation of GLCM. is there.

  As the noise component of the tomographic image Da increases, the GLCM is greatly disturbed. In order to prevent such large GLCM disturbance, it is necessary to widen the gradation width W. When the body thickness value (standard deviation SD) of the tomographic image Da is increased, the disturbance of the pixel value is further increased. That is, in the description of the middle stage and the lower stage of FIG. 10, the variation in the pixel value is within the gradation width W of 20, but when the standard deviation SD is increased, it cannot be within the width of the pixel value. Therefore, according to the present invention, the gradation width W at the time of GLCM generation is increased with the increase of the standard deviation SD. In this way, even if the pixel values vary greatly, the range in which the pixel values are identified at the time of GLCM generation is expanded, and the influence of variation is absorbed. As a result, the influence of the noise component of the tomographic image Da does not appear in the GLCM.

<Operation of Texture Analysis Index Calculation Unit 16>
The GLCM is sent to the texture analysis index calculation unit 16. The texture analysis index calculation unit 16 can calculate the texture analysis index by performing various operations on the GLCM. Examples of the texture analysis index that can be calculated by the texture analysis index calculation unit 16 include the following. P (i, j) in the equation is the value of the element in the i-th row and j-th column in GLCM, Σ _i and Σ _j are the total of the elements for the i-th row and j-th column, respectively, and N _g is the tomographic image Da. Is the average value, μ _x and μ _y are the average values in the row direction and the column direction, and σ _x and σ _y are the standard deviations in the row direction and the column direction, respectively. ing. Note that these texture analysis indices ASM (Angular Second Moment), CNT (Contrast), COR (Correlation), VAR (Variance), IDM (Inverse Differential Moment, Inverse Differential Moment). Each of ENT (Entropy) is a part of the 14 types of parameters proposed by Harlick et al. DIS is a texture analysis index called dissimilarity or dissimilarity, and HOM is a texture analysis index called uniformity or homogeneity. The texture analysis index calculation unit 16 corresponds to the index calculation means of the present invention.

  The texture analysis index calculation unit 16 calculates the texture analysis index by performing the above-described various operations on the GLCM. The type and number of texture analysis indices calculated by the texture analysis index calculation unit 16 can be changed as appropriate. As described above, the texture analysis index calculation unit 16 performs texture analysis based on the GLCM (co-occurrence matrix) and calculates the texture analysis index. The texture analysis index is used, for example, to estimate a state change of the trabecular bone. In this case, the texture analysis index is calculated for each of the tomographic images D of the same subject M having different imaging timings.

  Each part 11, 12, 14, 15, 16 which comprises this invention is implement | achieved when CPU runs various programs. Each unit may be realized by an arithmetic device that individually executes the function of each unit.

<Demonstration of effects of the present invention>
Subsequently, the effect of the present invention was actually verified, and the result will be described. In order to understand the effects of the present invention, it is first necessary to know a method for capturing the tomographic image D, which will be described.

  FIG. 11 illustrates the principle of the filter back projection method for generating the tomographic image D. For example, a virtual plane (reference cut section MA) parallel to the top plate 2 (horizontal with respect to the vertical direction) will be described. As shown in FIG. The FPD 4 is synchronized with the opposite direction of the X-ray tube 3 in accordance with the irradiation direction of the cone-shaped X-ray beam B by the X-ray tube 3 so as to be projected onto the fixed points p and q of the X-ray detection surface of the FPD 4. A series of X-ray images are generated while being moved.

  In the series of X-ray images, the projected image of the subject M is reflected while changing the position. Then, when this series of X-ray images is reconstructed, images (for example, fixed points p and q) positioned on the reference cut surface MA are accumulated and imaged as X-ray tomographic images.

  On the other hand, the point I not located on the reference cut surface MA is reflected as a point i in a series of subject M images while changing the projection position on the FPD 4. Unlike the fixed points p and q, such a point i is blurred without forming an image when the X-ray projection images are superimposed. In this way, by superimposing a series of projection images, an X-ray tomographic image in which only an image positioned on the reference cut surface MA of the subject M is reflected is obtained. In this way, when the projected images are simply superimposed, a tomographic image at the reference cut surface MA is obtained.

  Furthermore, according to the filter back projection method, a similar tomographic image can be obtained even at an arbitrary cut surface horizontal to the reference cut surface MA. During shooting, the projection position of the point i moves in the FPD 4, but this moving speed increases as the separation distance between the point I before projection and the reference cut surface MA increases. By utilizing this, reconstruction is performed while shifting the acquired series of subject M images in the body axis direction A at a predetermined pitch, and a tomographic image at a cutting plane parallel to the reference cutting plane MA is obtained. That is why.

  That is, the tomographic image D is an image obtained by superimposing X-ray images with different imaging directions. If each of the X-ray images is underexposed, the tomographic image D is also underexposed. The reason why each of such X-ray images is underexposed is that the body thickness of the subject M is thicker than expected at the time of imaging. When the body thickness of the subject M to be imaged is thicker, X-rays are less likely to pass therethrough, so that the X-ray dose that is detected by the FPD 4 through the subject M is reduced. This is the reason why each X-ray image is underexposed and the tomographic image D is underexposed. Such an underexposed tomographic image contains more noise components than a tomographic image when exposure is appropriate.

<Generate table>
In order to confirm the effect of the present invention, it is necessary to generate the table described in FIG. This table relates the body thickness value (standard deviation SD) included in the tomographic image and the gradation width W. Therefore, to generate the table, it is necessary to prepare a plurality of tomographic images having different body thickness conditions of the subject M. This tomographic image can be prepared by the following method. First, as shown on the left side of FIG. 12, a phantom Ph that is regarded as the subject M is placed on the top 2 and a tomographic image D0 is taken based on the principle explained in FIG.

  Subsequently, as shown on the right side of FIG. 12, the acrylic plate Ac is placed on the upper side of the phantom Ph, and the tomographic image D1 is taken again in this state. Since this acrylic plate Ac does not transmit part of the X-rays to be transmitted, it can produce the same effect as when the body thickness of the subject M is increased. Thereafter, as shown on the left side of FIG. 13, if the tomographic images D2, D3, and D4 are taken while increasing the number of acrylic plates Ac placed on the upper side of the phantom Ph, the subjects M having different body thicknesses are taken respectively. A tomographic image like that can be obtained. The tomographic images D0, D1, D2, D3, and D4 obtained in this way have the same phantom Ph but have different body thickness values (standard deviation SD).

  When a conventional texture analysis is performed using the tomographic image D0 taken without placing the acrylic plate Ac on the phantom Ph, some texture analysis index is calculated. On the other hand, the texture analysis index can be calculated even if the conventional texture analysis is performed using the tomographic image D1 taken with the acrylic plate Ac placed on the phantom Ph. However, the texture analysis index obtained from the tomographic image D0 does not match the texture analysis index obtained from the tomographic image D1. This is because the body thickness values (standard deviation SD) calculated by the tomographic images D0 and D1 are different, and this influence causes a discrepancy in the value of the texture analysis index. Conventional texture analysis refers to analysis that generates a co-occurrence matrix based on a certain gradation width W.

  Here, when the texture analysis is performed on the tomographic image D1, the gradation width W used when generating the co-occurrence matrix is increased, and the texture analysis index is generated again. Then, the texture analysis index approaches the value of the texture analysis index for the tomographic image D0 by increasing the gradation width W. This is because the texture analysis index value is not affected by the noise component of the tomographic image D1 when the gradation width W is wide.

  When the texture analysis is repeated for the tomographic image D1 with the same acrylic plate Ac while gradually increasing the gradation width W, the texture analysis index becomes the same as the texture analysis index for the tomographic image D0 without the acrylic plate Ac. . The gradation width W at this time is an appropriate gradation width W for texture analysis of the tomographic image D1. If the gradation width W is further increased, the texture analysis method is unnecessarily separated from that of the tomographic image D0. Such a situation may cause a problem when comparing texture analysis indices. The value of the texture analysis index being the same means that the numerical value obtained by dividing the texture analysis index related to the tomographic image D0 from the texture analysis index related to the tomographic image D1 is sufficiently close to 1.

  If the above-described adjustment of the gradation width W is performed for each of the tomographic images D2, D3, and D4 with different numbers of the acrylic plates Ac, the appropriate gradation width W is calculated for each of the tomographic images D2, D3, and D4. .

  By the way, the standard deviation SD can be calculated individually for the tomographic images D0, D1, D2, D3, and D4. Therefore, as shown on the right side of FIG. 13, each tomographic image D0, D1, D2, D3, and D4 can be plotted in a coordinate system having a standard deviation on one axis and an appropriate gradation width W on the other axis. It should be possible. These plots can be considered to be aligned on an approximate expression that forms a straight line. The relationship between the width value Wa and the standard deviation SD shown in the table in FIG. 4 is actually obtained by such a calculation method.

  As described above, the relationship between the standard deviation SD and the gradation width W indicated by the table is that the texture analysis index obtained when a tomographic image D having a predetermined standard deviation SD is subjected to texture analysis by copying a subject having a certain body thickness. It is determined to be the same as the texture analysis index for the reference tomographic image D0.

In this demonstration, an approximate expression of W = 0.0777SD−2.5915 (R ² = 0.9998) was obtained based on the plots of the respective tomographic images. Of the texture analysis indices, the target of calculation is entropy. Note that the standard deviation SD and the appropriate gradation width W calculated when acquiring the approximate expression were calculated from the entire tomographic image.

<Use for actual texture analysis>
Since the texture analysis was actually performed based on the relationship between the width value Wa and the standard deviation SD thus obtained, the result will be described. Texture analysis was performed on three different parts of the femur shown in FIG. Specifically, the three parts are a distribution area of the main compression trabecular bone, a distribution area of the main tensile trabecular bone, and a ward triangle. The main compressive trabecular bone is a trabecular bone to which a strong pressure is applied, and the main tensile trabecular bone is a trabecular bone to which a strong tension is applied. The ward triangle is a specific part of the femur where the trabecular bone is vacant.

  The left side of FIG. 15 shows the result of texture analysis performed on the tomographic image D by a conventional method that does not depend on the present invention. The conventional method is a method for setting a constant gradation width when generating a co-occurrence matrix. The left side of FIG. 15 compares texture analysis indices for the main compressive trabecular bone, main tensile trabecular bone, and ward triangle among six different tomographic images D. Of the texture analysis indices, the target of calculation is entropy. The calculation target is desirably the same texture analysis index as that used in the creation of the table. The six tomographic images D are one image taken without the acrylic plate Ac, and the acrylic plate Ac placed on the upper side of the phantom Ph has a thickness of 1 cm, 2 cm, 3 cm, 4 cm, and 5 cm. There are 5 photos taken by each.

  Since all of the phantoms Ph appearing in the six tomographic images D with different conditions of the acrylic plate Ac at the time of photographing are the same, it is ideal that the texture analysis index has the same value in any tomographic image D. is there. In this regard, the main pressure trabecular bone on the left side of FIG. 15 is almost ideal. Therefore, it can be said that the reliability of the texture analysis index obtained by the conventional analysis method is high for the main compression trabecular bone.

  However, for the main tensile trabeculae and the ward triangle, the texture analysis index varies depending on the number of acrylic plates Ac that appear in the tomographic image. Therefore, it cannot be said that the reliability of the texture analysis index obtained by the conventional analysis method is high for the main tensile trabecular bone and the ward triangle. This is because the texture analysis index has fluctuated due to the difference in body thickness of the subject M. It is considered that the fluctuation of the texture analysis index is caused by the difference in body thickness value (standard deviation SD) among the six tomographic images D.

  The right side of FIG. 15 shows the result of texture analysis performed on the tomographic image D according to the method of the present invention. The method of the present invention is a method of changing the gradation width in accordance with the body thickness value (standard deviation SD) of the tomographic image D. The tomographic images D used for the analysis are the same six images as the analysis on the left side of FIG. 15, and the target of calculation among the texture analysis indices is entropy. In the demonstration on the left side of FIG. 15, the gradation width is determined so that the standard deviation SD obtained by analyzing the soft tissue image of the subject M in the tomographic image D has a linear relationship.

  According to the right side of FIG. 15, it can be seen that the variation of the texture analysis index depending on the acrylic conditions at the time of photographing is suppressed for the main tensile trabeculae and the ward triangle by applying the method of the present invention. Therefore, it can be said that the reliability of the texture analysis index obtained by the analysis method according to the present invention is high for the main tensile trabeculae and the ward triangle. This is because the texture analysis index does not vary due to the difference in body thickness of the subject M.

  On the other hand, as described above, a sufficiently reliable texture analysis index is obtained with the conventional analysis method for the main compression trabecular bone. In this regard, when the texture analysis was performed on the main compression trabecular bone according to the method of the present invention, the obtained texture analysis index did not vary greatly between the tomographic images D. That is, the reliability of the texture analysis index does not decrease even when the texture analysis is performed on the main compression trabecula using the analysis method of the present invention instead of the conventional analysis method.

  From these facts, it can be said that a more reliable texture analysis index can be calculated by using the analysis method according to the present invention than the conventional analysis method.

  As described above, according to the configuration of the present invention, the texture analysis technique in which the difference in body thickness of the subject M does not affect the analysis result is specifically shown. That is, the gradation width W is set so as to increase as the body thickness of the subject M increases and the noise component increases. This gradation width W represents the width of the gradation to be identified at the time of GLCM generation. That is, according to the present invention, when the body thickness of the subject M is increased and the noise component included in the tomographic image Da is increased, the elements constituting the GLCM are reduced. As the noise component of the tomographic image D increases, the variation in pixel values in the tomographic image D increases accordingly. In texture analysis of such images, if the number of GLCM elements is increased as in the conventional configuration, pixel value variations derived from noise components are also expressed in GLCM, so the texture analysis index takes noise components into account. It will become the value.

  When the number of elements constituting the GLCM is reduced as the noise component of the image increases as in the configuration of the present invention, the influence of the variation in the pixel value derived from the noise component is absorbed when the GLCM is generated during the analysis, and the texture is It does not appear in the analysis index. Therefore, according to the image analysis apparatus of the present invention, a more reliable texture analysis index can be calculated.

  As described above, the variation calculation unit 12 calculates the standard deviation SD more reliably by calculating the standard deviation SD based on the intensity of the variation in the pixel values seen between the pixels constituting the soft tissue image of the subject M. Can do. Since the bone image that is not a soft tissue image of the subject M is an image of a bone that hardly transmits X-rays originally, it includes a lot of noise components regardless of the body thickness.

  As described above, when a tomographic image D having a predetermined standard deviation SD is imaged by imprinting a subject having a certain thickness, the obtained texture analysis index is the same as the texture analysis index for the reference tomographic image D. If the gradation width setting unit 14 operates according to the relationship between the determined gradation width W and the standard deviation SD, the gradation width setting unit 14 operates based on the actual measurement result. High texture analysis index can be acquired.

  The configuration of the present invention is not limited to the above-described configuration, and can be modified as follows.

  (1) In the configuration of the first embodiment, the body thickness value of the image is recognized based on the standard deviation SD, but the present invention is not limited to this configuration. A contrast value may be used instead of the standard deviation SD. In addition, the variation calculation unit 12 can change the configuration so as to calculate an index other than the standard deviation SD as an index of pixel value variation.

  (2) In the configuration of the first embodiment, the body thickness value (standard deviation SD) of the image is recognized by analyzing the tomographic image D, but the present invention is not limited to this configuration. The actual body thickness or standard deviation SD of the subject M may be recognized based on information input by the operator through the input unit 13 or information embedded in an image file constituting the tomographic image D. As information referred to at this time, for example, the measurement result of the body thickness of the subject M may be indicated, or the amount of the operator's eyes may be indicated. Further, for example, it may be the BMI value of the subject M, or may indicate the driving status of the photo timer at the time of imaging. According to this modification, a body thickness calculation unit is provided instead of the variation calculation unit 12. The body thickness calculation unit calculates the body thickness of the subject M based on the various information described above. According to this modification, the body thickness calculation unit calculates the body thickness of the subject M based on the operator's input value. The gradation width setting unit 14 operates while referring to a table in which the body thickness of the subject M and the gradation width W are related.

  (3) The image analysis apparatus of the present invention is not dedicated to trabecular analysis. The present invention can also be applied to other analyzes such as lung bronchial analysis, lung structural analysis, and tumor shape analysis.

  (4) The image analysis apparatus of the present invention always uses the same table, but the present invention is not limited to this configuration. A unique table may be used depending on the purpose of the analysis and the target part. According to this modification, for example, when performing the analysis of a warped triangular trabecular bone, the texture analysis index is calculated based on the relationship between the body thickness value (standard deviation SD) and the width value Wa defined by the dedicated table. Will be made.

  (5) In the above description, the table of FIG. 4 is calculated by analyzing the entire tomographic image, but the present invention is not limited to this configuration. You may make it calculate a table by performing image analysis about the soft tissue of a tomographic image, the background area | region where the phantom, etc. are not reflected.

  (6) The image processing apparatus of the present invention may be configured to calculate a single texture analysis index for an image to be analyzed, or may be configured to calculate a plurality of texture analysis indexes simultaneously. Also good.

SD standard deviation (body thickness value)
13 Input section (input means)
14 Gradation width setting part (gradation width setting means)
15 Matrix generator (co-occurrence matrix generator)
16 Texture analysis index calculation unit (index calculation means)

Claims (7)

An image analysis apparatus for analyzing a radiographic image obtained by radiography on a subject,
A gradation width setting means for setting each gradation width when dividing the gradation of the pixel value included in the radiation image into a plurality of gradation sections;
A pair of two pixels whose pixel values belong to each other in a predetermined gradation section and whose pixels are separated from each other by a predetermined distance appear in the analysis range provided in a part of the radiation image how many times. A co-occurrence matrix generating means for generating a co-occurrence matrix by counting each combination of gradation intervals;
Index calculation means for calculating a texture analysis index based on the co-occurrence matrix,
The gradation width setting means sets the gradation width so as to increase according to an increase in a noise component included in the radiation image ;
An image analyzing apparatus according to claim 1, wherein a gradation interval for generating the co-occurrence matrix is determined by a gradation width set by the gradation width setting means . The image analysis apparatus according to claim 1,
The image analysis apparatus, wherein the gradation width setting means estimates the noise component based on a body thickness of the subject. The image analysis apparatus according to claim 2,
An image analysis apparatus comprising input means for allowing an operator to input the body thickness. The image analysis apparatus according to claim 2,
An image analysis apparatus, wherein the body thickness is estimated based on a variation in pixel values constituting a subject image reflected in the radiation image. The image analysis apparatus according to any one of claims 2 to 4,
Storage means for storing the relationship between the gradation width referred to by the gradation width setting means and the noise component;
The relevance stored by the storage means is determined so that the texture analysis index obtained by texture analysis of a radiographic image in which an object having a certain body thickness is copied is the same as the texture analysis index for the reference radiographic image. An image analysis apparatus characterized by   6. A trabecular bone analysis apparatus that calculates a trabecular index of the subject based on the texture analysis index calculated by the image analysis apparatus according to claim 1. The image analysis apparatus according to any one of claims 1 to 6,
One or more of correlation, dissimilarity, contrast, homogeneity, entropy, angler second moment, variance, and inverse differential moment are selected as the texture analysis index calculated by the index calculation means An image analysis apparatus characterized by the above.

Images