JP5640280B2 - Osteoporosis diagnosis support device and osteoporosis diagnosis support program - Google Patents

Description

  The present invention relates to an osteoporosis diagnosis support apparatus and an osteoporosis diagnosis support program.

  With the increase in the elderly population, there is a need for a device that discriminates osteoporosis. Patent Document 1 discloses an osteoporosis diagnosis support apparatus that discriminates osteoporosis from the thickness of the cortical bone of the lower jaw shown in an X-ray image.

International Publication No. 2006/043523

  However, since the osteoporosis diagnosis support apparatus of Patent Document 1 measures the thickness of the cortical bone only at one part of the lower jaw, the measurement accuracy of the cortical bone thickness is not so high. Therefore, in some cases, there is a problem in that it is erroneously determined as osteoporosis based on a measurement result having a large error.

  The present invention has been made in view of the above-described matters, and an object thereof is to provide an osteoporosis diagnosis support apparatus and an osteoporosis diagnosis support program that can measure the thickness of bone such as cortical bone with high accuracy.

In order to achieve the above object, an osteoporosis diagnosis support apparatus according to the first aspect of the present invention provides:
Boundary line specifying means for specifying two boundary lines between cortical bone and other portions reflected in a band shape in an X-ray image;
Based on a plurality of points provided on the boundary line, approximate line calculating means for obtaining an approximate line that is a straight line or a curve that approximates the boundary line;
A plurality of measurement points are set on the approximate line, a distance between the two boundary lines is measured on a measurement auxiliary line extending from the set measurement point in a normal direction of the approximate line, and a plurality of measurement values are measured. Measurement value group acquisition means for acquiring
Cortical bone thickness calculating means for calculating the thickness of cortical bone based on the measured value;
Osteoporosis discriminating means for discriminating osteoporosis based on the cortical bone thickness calculated by the cortical bone thickness calculating means,
It is characterized by that.

The boundary line specifying means includes:
Cortical bone part specifying means for specifying a cortical bone part and other background parts based on the density of the X-ray image;
For each pixel belonging to the cortical bone part, to calculate the shortest distance to the background part, the shortest distance acquisition means for associating the calculated shortest distance and the corresponding pixel,
By tracking the maximum value of the shortest distance, centerline acquisition means for obtaining the centerline of cortical bone,
Boundary line acquisition means for acquiring, as the boundary line, envelopes of a plurality of circles centered on a plurality of points provided on the center line,
The boundary line acquisition means includes
It is desirable that the shortest distance associated with the pixel located at the center point is a radius of each circle.

The cortical bone thickness calculating means includes
Means for obtaining a histogram of the measurement value range and the number of occurrences thereof from the plurality of measurement values;
A plurality of sections in the range of the histogram, and a histogram dividing means for dividing the histogram based on the sections;
Means for selecting a histogram having the maximum total number of appearances among the plurality of divided histograms;
A means for calculating an expected value of a measured value in a histogram in which the sum of the selected number of appearances is maximum, and obtaining the calculated expected value as a thickness of cortical bone, and
The histogram dividing means includes
It is desirable to set the number of divisions so that the difference in the total number of appearances between the histogram having the maximum total number of appearances and the histogram having the second largest total number of appearances after the histogram division is maximized. .

The cortical bone thickness calculating means includes
Means for obtaining an average value of the plurality of measured values as the thickness of the cortical bone,
The osteoporosis discrimination means is
It is desirable to discriminate osteoporosis based on the cortical bone thickness acquired by the cortical bone thickness calculating means and the dispersion value of the measured value.

The histogram dividing means includes
Set acquisition means for randomly associating numerical values of the plurality of measurement values with a predetermined number of sets, and acquiring a plurality of sets;
For each of the plurality of sets, an expected value acquisition means for obtaining a sum within a set of values obtained by multiplying the measured value and the number of times the measured value appears, and an expected value;
One of the numerical values that can be taken by the measured value is selected, compared with the expected value, and the operation of redistributing the selected numerical value to the set having the closest expected value is performed for all possible numerical values of the measured value. Redistribution means,
It is desirable to include a repeating unit that repeats the expected value acquisition unit and the redistribution unit alternately and continues until the redistribution does not occur.

  The approximate line is preferably a straight line or a curve expressed by a polynomial.

  The approximate line may be a curve expressed by a quadratic function.

  The X-ray image is preferably a dental panoramic X-ray image including a mandibular portion of a person.

In order to achieve the above object, an osteoporosis diagnosis support program according to the second aspect of the present invention provides:
A boundary line specifying step for specifying two boundary lines between the cortical bone and the other part appearing in a band shape in the X-ray image;
An approximate line calculation step for obtaining an approximate line that is a straight line or a curve that approximates the boundary line based on a plurality of points provided on the boundary line;
A plurality of measurement points are set on the approximate line, a distance between the two boundary lines is measured on a measurement auxiliary line extending from the set measurement point in a normal direction of the approximate line, and a plurality of measurement values are measured. A measurement value group acquisition step for acquiring
Cortical bone thickness calculating step for calculating the thickness of cortical bone based on the measured value;
It is characterized by causing a computer to execute an osteoporosis determining step of determining osteoporosis based on the cortical bone thickness calculated in the cortical bone thickness calculating step.

  According to the osteoporosis diagnosis support apparatus and the osteoporosis diagnosis support program of the present invention, the thickness of bones such as cortical bone can be measured with high accuracy.

It is a block diagram which shows the structure of the osteoporosis diagnosis assistance apparatus which concerns on this Embodiment. It is a figure of a dental panoramic X-ray image. It is a flowchart which shows the outline | summary of the procedure of the diagnostic assistance process which the osteoporosis diagnostic assistance apparatus which concerns on this Embodiment performs. It is a flowchart which shows the procedure of a mandible area extraction process. It is a figure which shows the image which the mandible area | region was selected and binarized. It is a flowchart which shows the procedure of a cortical bone outline clarification process. FIG. 6 is a diagram illustrating an image obtained by cutting out an original image using the binarized image of FIG. 5 as a mask and applying edge enhancement by applying high-pass filtering. It is the figure which displayed the distance from the pixel (black) used as the starting point based on the distance of 8 vicinity. It is a figure which shows the image which performed distance conversion, noise removal, binarization, morphological closing, and opening to the image of FIG. It is a flowchart which shows the procedure of a cortical bone boundary line specific process. It is a figure which shows the image which performed distance conversion to the image of FIG. It is a figure which shows the image which displayed the cortical bone centerline. It is the figure which displayed some of the circles arranged centering on the pixel on the cortical bone center line. When more circles are placed, the circle and the area occupied by it are shown in gray. It is a flowchart which shows the procedure of a cortical bone thickness measurement process. It is a figure which shows the image which displayed the cortical bone boundary line and the quadratic function which applied to this. It is a figure which shows the image which showed the normal line of the approximate line. It is a figure showing the histogram of a thickness measurement value group. It is a flowchart which shows the procedure of an osteoporosis discrimination | determination process. It is the figure which displayed the largest cluster selected in step S505. FIG. 18 is a diagram obtained by removing maximum and minimum 10% data from the histogram of FIG. 17. It is a figure which shows the ROC curve about the mandible right side. It is a figure which shows the boundary line determined using the support vector machine and the plot of the average value and dispersion value of a thickness measurement value group.

  Hereinafter, the osteoporosis diagnosis support apparatus according to the present embodiment will be described with reference to the drawings. The osteoporosis diagnosis support apparatus 1 according to the present embodiment is a diagnosis support apparatus that generates information for supporting diagnosis of whether or not osteoporosis is based on the thickness of the cortical bone of the mandible captured in an X-ray image. As shown in FIG. 1, the osteoporosis diagnosis support apparatus 1 includes an X-ray image acquisition unit 11, a control unit 12, a storage unit 13, an output unit 14, and an operation unit 15.

  The X-ray image acquisition unit 11 is an input device for acquiring an X-ray image captured by an X-ray device (not shown). For example, a USB (for reading information from a semiconductor memory or the like in which an X-ray image is stored) (Universal Serial Bus) connector. When the X-ray image acquisition unit 11 is connected to the control unit 12, the control unit 12 reads an X-ray image from the semiconductor memory and stores it in the storage unit 13.

  In the following description, it is assumed that the X-ray image stored in the storage unit 13 is a dental panoramic X-ray image. The dental panoramic X-ray image is an X-ray image obtained by rotating the X-ray apparatus around the face and photographing the entire jaw and teeth. In this embodiment, it is assumed that the dental panoramic X-ray image is a gray scale image as shown in FIG.

  Further, in the present embodiment, diagnosis support information is generated from the thickness of the cortical bone portion of the jaw bone. Therefore, in the dental panoramic X-ray image, as in the portion surrounded by the white frame in FIG. It is assumed that the cortical bone (band-like part) of the bone is shown.

  Furthermore, in order to avoid erroneous recognition of cortical bone in image processing to be described later, in principle, structures other than cortical bone are not reflected in the dental panoramic X-ray image, as shown by the white frame in FIG. Shall. However, a structure having a width that is clearly smaller than the width of the cortical bone may be reflected in the image. For example, the trabecular bone of the cancellous bone is substantially smaller than the width of the cortical bone, and may be shown in the image. “Cortical bone” refers to the outer peripheral edge of bone with high bone density, and “cancellous bone” refers to a sponge-like portion within bone that is surrounded by cortical bone and has low bone density. .

  The control unit 12 includes a processor such as a CPU (Central Processing Unit) and controls each unit of the osteoporosis diagnosis support apparatus 1. Further, the control unit 12 starts a diagnosis support process for generating diagnosis support information in accordance with a program stored in a work memory (not shown). In addition, the control unit 12 determines whether or not the patient to be examined has osteoporosis by comparing verification data stored in a storage unit 13 described later and diagnosis support information.

  The storage unit 13 is composed of a storage device such as a hard disk and is a dental panoramic X-ray image or collation data that is a basis for determining osteoporosis (for example, data obtained by measuring the thickness of cortical bone of a plurality of persons, Various data such as data associated with classification between osteoporosis patients and those who do not are stored.

  The output unit 14 includes a display device such as a display, and displays osteoporosis diagnosis support information, osteoporosis discrimination results, and the like.

  The operation unit 15 is composed of an input device such as a keyboard and a mouse, and the user operates the operation window displayed on the output unit 14 using the mouse or the like, so that the control unit 12 performs diagnosis support processing and the like. Command execution.

  The configuration of the osteoporosis diagnosis support apparatus 1 has been described above. Next, the operation of the osteoporosis diagnosis support apparatus 1 will be described. When the execution of the process is instructed from the operation unit 15, the control unit 12 starts a diagnosis support process for displaying osteoporosis diagnosis support information. Hereinafter, the diagnosis support processing will be described with reference to the flowchart of FIG.

  First, the control unit 12 acquires an X-ray image from the storage unit 13 (step S100). The X-ray image acquired here is an image of a part of the jaw, such as a part surrounded by a white frame in FIG. Since many dental panoramic X-ray images have low contrast, a contrast enhancement process is appropriately performed on the acquired X-ray image.

  When the acquisition of the X-ray image is completed, the control unit 12 then starts a mandible region extraction process for extracting the mandible region (step S200). Here, the “mandible” is a mandibular bone composed of cortical bone and cancellous bone. This process is intended to remove a background portion that is not an image processing target from the X-ray image. Here, the “background portion” is a portion other than the mandible in the X-ray image. By this processing, only the mandible part is cut out from the X-ray image. Hereinafter, the mandible region extraction processing will be described with reference to the flowchart of FIG.

  The control unit 12 generates a “mask image” as shown in FIG. 5 that is binarized by setting the pixel of the mandible part to 1 (white) and the pixel of the background part to 0 (black) (step S201). The mask image is used to cut out the mandibular portion from the X-ray image in a process (step S202) described later. There are various methods of binarization, and the method is not limited to one method, but in the present embodiment, a method based on pixel clustering is used. Hereinafter, the binarization method using clustering will be described in detail.

A cluster refers to a set of pixels classified according to a predetermined definition. In this method, first, all the pixels in the measurement region are classified based on the pixel values, and a cluster for each pixel value is created. That is, all the pixels in the image whose pixel value is z belong to the cluster C _z . Since the number of clusters by this classification is equal to the number of pixel values in the image, the upper limit of the number of clusters is 256. All the obtained clusters are arranged based on the pixel value characterizing each cluster. Thereafter, cluster integration is repeatedly performed by the method described below, and finally all pixels are integrated into two clusters.

  First, in two clusters, the concepts of intra-cluster dispersion and inter-cluster dispersion are introduced. The inter-cluster variance is a variance of pixel values calculated in the union of clusters A and B when they exist. The intra-cluster dispersion is such that when the average value of pixel values belonging to cluster A is mA and the average value of pixel values belonging to cluster B is mB, all pixel values of cluster A are mA and all pixel values of cluster B are mB. Is the variance of the pixel values in the union of cluster A and cluster B. Also, the product of intra-cluster variance and inter-cluster variance is defined as “inter-cluster distance”.

  Next, in the array of clusters based on the pixel values defined above, the inter-cluster distance is calculated for all the combinations of two adjacent clusters. A combination of two clusters that minimizes the obtained inter-cluster distance is searched, and the two clusters are integrated. Thereafter, the inter-cluster distance calculation and integration operations are repeated until the number of clusters reaches 2. Of the obtained two clusters, a pixel value 1 is given to a pixel belonging to a cluster having a large pixel value, and a pixel value 0 is given to a pixel belonging to the other cluster. By this processing, the image can be binarized. The data generated as a result is a mask image in which the pixel value of the mandible is 1 and the pixel value of the background is 0.

  When the generation of the mask image is completed, the control unit 12 then applies the mask image to the original X-ray image (hereinafter referred to as “original image”) to remove the background portion (step S202). Specifically, for each pixel of the original image, the product of the pixel value and the pixel value of the mask image is taken as a new pixel value. By this process, a grayscale image in which the background portion is completely black is generated. When the gray scale image with the background portion being black is acquired, the control unit 12 ends the mandible region extraction process.

  When the mandible region extraction process is completed, the control unit 12 returns to the flow of FIG. 3 and starts the cortical bone outline clarification process (step S300). This process aims to clarify the contour of the cortical bone and to smooth the contour. Hereinafter, the cortical bone clarification process will be described with reference to the flowchart of FIG.

  The control unit 12 performs edge enhancement by applying high-pass filtering to the grayscale image generated in step S202 (step S301). By this processing, an image as shown in FIG. 7 is obtained.

  Next, the control unit 12 applies the same clustering method as in step S201 to the image processed in step S301, and sets the pixel value of the structure centered on the belt-like cortical bone portion to 1 A binary image is generated with the pixel values of (white) and other background portions set to 0 (black) (step S302). The binarization method may be a method different from the method described in step S201.

  Next, the control unit 12 performs distance conversion processing on all the pixels in the region displayed in white of the binarized image by a method described later (step S303). As a result of the distance conversion process, the pixel value is equal to the distance value for the image area that was white before the process, and the pixel value remains 0 (black) for the image area that was black before the process. An image is generated.

  The distance conversion is performed as follows. The distance to the pixel of the nearest background part (black part) is calculated in all the pixels of the structure part (white part) centering on the belt-like cortical bone part. At this time, the control unit 12 calculates the distance based on the distance conversion in the vicinity of 8. Here, the distance conversion in the vicinity of 8 is, as shown in FIG. 8, the distance to the 8 pixels adjacent to the starting pixel (center black pixel) is 1, and the number of pixels to the end pixel is This is a method for determining the distance between two pixels. The control unit 12 replaces the obtained distance value with each pixel value of the mandible part, and generates a new grayscale image (step S303).

  Next, in order to remove the trabecular image around the cortical bone (the fine bone tissue forming the cancellous bone) and the small island-like structure, the control unit 12 removes a pixel value smaller than a predetermined threshold value from the grayscale image. Are replaced with black (pixel value 0) (step S304). Note that the threshold value is desirably about 10% of the maximum pixel value in the image. Since a pixel having a small value means a pixel that is close to the background portion, a small island-like structure in the gray scale image, a trabecular image around the cortical bone, and the like are removed by this processing.

  Next, the control unit 12 generates a binarized image again by applying the same method as in step S201 to the image generated in step S304 (step S305).

  Furthermore, the controller 12 removes cavities and chips in the cortical bone by applying morphological opening and morphological closing to the binarized image generated in step S305, as shown in FIG. A new binarized image is generated (step S306). Here, the morphological opening means that a small component (figure) selected appropriately is arranged so as to fit inside the portion of pixel value 1 (white), and the portion where this component is not arranged It is a method of smoothing the boundary line by removing it. An example of a figure used as a small component is a circle. Morphological closing is a method of performing the same operation as morphological opening on a pixel value 0 (black) portion. By this process, cavities and chips in the cortical bone are removed, and the contour becomes smooth. When the process of step S306 ends, the control unit 12 ends the cortical bone contour clarification process.

  When the cortical bone contour clarification process is completed, the control unit 12 returns to the flow of FIG. 3 and starts the cortical bone boundary line specifying process (step S400). The purpose of this processing is to specify two boundary lines that are boundaries between the cortical bone and the other parts and are substantially parallel to each other. Hereinafter, the cortical bone boundary line specifying process will be described with reference to the flowchart of FIG.

  The control unit 12 again applies the 8-neighbor distance transform described in step S303 to the image generated in step S306 to generate a grayscale image as shown in FIG. 11 (step S401).

  Based on the grayscale image obtained in step S401, the control unit 12 specifies a center line that penetrates the center of the belt-like cortical bone in the longitudinal direction (step S402). By this processing, as shown in FIG. 12, a center line that penetrates the center of the cortical bone is specified. The center line is specified by tracking the maximum value of the distance value (the pixel value of the grayscale image). Hereinafter, a method for specifying the center line will be described in detail.

The center line is specified by generating a number of paths crossing the cortical bone portion from side to side in a predetermined method, and selecting an optimal path from among the paths. The route is generated by selecting the starting point and moving the starting point. Here, the case where the starting point is the right end of the screen and the movement is performed from the right end to the left end will be described, but the left and right may be reversed. As the starting point, all the pixels at the right end among the pixels constituting the cortical bone portion are selected. The movement of the starting point when focusing on one starting point is performed as follows.
(1) The pixel values of the three pixels on the left side, the upper left side, and the lower left side of the starting pixel are compared.
(2) A pixel having the largest pixel value is selected from the three pixels and set as a new starting pixel.
(3) Repeat (1) and (2) until the starting pixel reaches the left edge of the screen.
By applying this movement to all selected origins, multiple paths are obtained. Next, for all paths, the sum of pixel values on the path is taken. A route that maximizes the sum of pixel values is specified as a center line.

  When the specification of the center line is completed, the control unit 12 arranges a plurality of points on the center line, and draws a plurality of circles with the points as the center points as shown in FIG. The radius of the circle to be drawn is the pixel value of the center point, that is, the shortest distance from the pixel serving as the center point to the background portion (black portion). The control unit 12 specifies the upper and lower envelopes (envelope) formed by a plurality of circles as the boundary line of the cortical bone (step S403). Thereby, the control part 12 can specify a smoother line as a boundary line of a cortical bone. When the boundary line of the cortical bone is specified, the control unit 12 ends the cortical bone boundary line specifying process.

  When the cortical bone boundary line specifying process is completed, the control unit 12 returns to the flow of FIG. 3 and starts the cortical bone thickness measurement process (step S500). Hereinafter, the cortical bone thickness measurement process will be described with reference to the flowchart of FIG.

  Since the two boundary lines have irregularities, the direction perpendicular to the cortical bone cannot be determined appropriately. Therefore, the control unit 12 applies the least square method to a plurality of points provided on the boundary line for any one of the two boundary lines, and approximates the boundary line as shown in FIG. Is obtained (step S501). The approximate line is an analytical function and may be any function such as a polynomial, but is preferably a function that well reproduces the shape of the cortical bone, particularly the curvature. One suitable example of a function is a quadratic function. The approximate line is used to specify a straight line (hereinafter referred to as “measurement auxiliary line”) extending in a direction perpendicular to the longitudinal direction of the cortical bone in the process (step S502) described later.

  The control unit 12 sets a plurality of measurement points on the approximate line, and, as shown in FIG. 15, a straight line extending from the set measurement point in the normal line direction of the approximate line (FIG. 16A), respectively, as a measurement auxiliary line ( It determines as FIG. 16, c) (step S502). In order to improve the measurement accuracy, it is desirable to take as many measurement points as possible on the approximate line.

  The control unit 12 acquires data obtained by measuring the distance between the two boundary lines on the determined plurality of measurement auxiliary lines as measurement data (measurement value group) of cortical bone thickness (step S503).

  This measurement data is likely to contain noise. Because, when the cortical bone part is separated by binarization, if there is a region with a low pixel value in the cortical bone, a cavity is generated by the binarization and cannot be removed by subsequent processing. Because there is. If the cortical bone thickness is calculated based on the measurement data including noise, an incorrect value is calculated as the cortical bone thickness. Therefore, the control unit 12 executes the following processing (steps S504 to S505) to remove noise from the measurement data.

  The control unit 12 creates a histogram as shown in FIG. 17 from the value of the cortical bone thickness expressed in pixel units and the number of times the value has been measured (step S504).

  Here, the classification based on the thickness value is defined in the histogram. Each set generated by classification is called a cluster. A certain cluster is a set of measurement values belonging to a range defining the cluster, and the size of the cluster is defined by the number of components of the cluster. The cluster with the largest size (hereinafter referred to as “maximum cluster”), that is, the cluster with the largest number of measurements, is the measurement value group that most accurately reflects the thickness of the cortical bone.

  Therefore, the control unit 12 selects the maximum cluster by clustering the histogram while changing the number of clusters from 2 to 10 (step S505).

Note that the k-means method is preferable as a clustering method. Specifically, the following processes (1) to (5) are sequentially performed on the above-described histogram. If clustering is performed properly, the difference in the number of components between the largest cluster and the second largest cluster is maximized.
(1) Set the number of clusters to be classified. Let N be the number of clusters set.
(2) Randomly associate one of the clusters with one measurement value in all the numerical values that can be taken by the measurement value expressed in pixel units.
(3) For each of the N clusters, the expected value in that cluster is obtained by multiplying the measured value belonging to that cluster by the number of occurrences of the corresponding measured value in the histogram.
(4) One is selected from the measured values, compared with the expected values of N clusters, and the measured values are redistributed (correction of association) to the cluster having the closest expected value. This operation is performed for all measured values.
(5) The processes of (3) and (4) are repeated until no reallocation occurs.

  The control unit 12 obtains an average value of the measured values in the selected maximum cluster, and uses it as an estimated value of cortical bone thickness. In addition, the control unit 12 calculates an average value and a variance value of the measurement values acquired in step S503 in order to provide a determination material to a doctor or the like. The control unit 12 stores the calculated three values (estimated value, average value, and variance value) in the storage unit 13 as diagnosis support information (step S506). When the storage of the diagnosis support information is completed, the control unit 12 ends the cortical bone thickness measurement process.

  When the cortical bone thickness measurement process is completed, the control unit 12 returns to the flow of FIG. 3 and starts the osteoporosis determination process (step S600). The thickness of the cortical bone has a strong relationship with the progression of osteoporosis, and the control unit 12 can determine whether the target patient has osteoporosis from the diagnosis support information. Hereinafter, the osteoporosis discrimination process will be described with reference to the flowchart of FIG.

  The control unit 12 acquires collation data stored in advance in the storage unit 13 from the storage unit 13 (step S601). The collation data includes a number of results of diagnosis of osteoporosis and diagnostic support information (estimated cortical bone thickness, average thickness value, variance value) by a reference method (for example, DXA, dual energy X-ray absorption measurement method). It is obtained from the subject and is made into a database by associating them.

  Based on this collation data, the control unit 12 determines a threshold value used for discrimination of osteoporosis in the processing described later (step S602). Specifically, the threshold value is determined as follows. First, an arbitrary threshold value is provisionally determined. Then, a subject whose cortical bone thickness estimate is smaller than the provisionally determined threshold is determined as “osteoporosis positive”, and a subject who is not so is determined as “negative”. This provisional discrimination is performed for all subjects. Next, the sensitivity and specificity are calculated by collating with the diagnosis result of osteoporosis by the reference method. Here, the sensitivity is the proportion of people diagnosed as positive for osteoporosis by the standard method, and determined to be positive based on the estimated cortical bone thickness, and the specificity is negative by the standard method. This is the percentage of people diagnosed as negative based on the estimated cortical bone thickness. Finally, the threshold value is raised and lowered, and the threshold value is determined so as to match the diagnosis result by the reference method as much as possible while taking into consideration both sensitivity and specificity.

  The control unit 12 acquires the diagnosis support information generated in step S500 from the storage unit 13 (step S603).

  The control unit 12 extracts an estimated value of cortical bone thickness from the diagnosis support information and collates it with the threshold value generated in step S602. Then, the control unit 12 determines “osteoporosis positive” when the estimated value of cortical bone thickness is smaller than the threshold, and “negative” otherwise, and stores the determination result in a work memory (not shown) or the like (step) S604). When the storage of the determination result is completed, the control unit 12 ends the osteoporosis determination process.

  When the osteoporosis discrimination process is completed, the control unit 12 returns to the flow of FIG. 3, acquires the estimated value of the cortical bone thickness and the discrimination result from the work memory or the like, and outputs them to the output unit 14 (step S700). . A doctor or the like diagnoses whether or not the patient has osteoporosis based on the determination result displayed on the output unit 14. When the output of the estimated cortical bone thickness and the discrimination result to the output unit 14 is completed, the control unit 12 ends the diagnosis support process.

  According to the present embodiment, it is possible to accurately predict the thickness of cortical bone. This is because the thickness of the cortical bone is measured at a plurality of measurement points, and the estimated thickness value is calculated by statistical processing from the plurality of measurement values.

  Further, according to the present embodiment, the cortical bone centerline is acquired, and the envelopes of a plurality of circles centered on a plurality of points provided on the centerline are acquired as the boundary lines of the cortical bone. The boundary line is smooth. For this reason, it is hard to be influenced by noise when measuring the thickness of cortical bone.

  In the present embodiment, a measurement auxiliary line is used when measuring the thickness of the cortical bone. What is used as the measurement auxiliary line is a normal line of an approximate curve that suitably reproduces the boundary line of the cortical bone, and preferably reflects the direction orthogonal to the longitudinal direction of the cortical bone. Taking such a method is suitable as a method for measuring the thickness of a curved belt-like structure such as cortical bone.

  The osteoporosis diagnosis support apparatus according to the present embodiment is entirely automated by a program, and gives the same result regardless of who performs the diagnosis support process.

  In addition, you may perform osteoporosis discrimination as follows. Here, osteoporosis is determined using two values of the average value and the variance value of the cortical bone thickness in the diagnosis support information. First, an average value and a variance value of cortical bone thicknesses of a large number of subjects associated with the presence or absence of osteoporosis are acquired from the collation data. Next, the subject's data is plotted with the variance value and the average value as the vertical axis and the horizontal axis, respectively. In this plane, a boundary line for separating an osteoporosis patient from a person who does not is determined using the method described below.

The boundaries for separating osteoporotic patients from those who are not are determined using a support vector machine. Hereinafter, a case where the support vector machine is applied to a two-dimensional plane will be described. The support vector machine is a technique for obtaining a boundary line that separates this plane into two based on a large number of vectors (sample vectors) classified into binary on the plane. The boundary line is defined by setting a weight vector and a threshold value, and is defined as a set of vectors whose inner product with the weight vector is equal to a predetermined threshold value. However, in the support vector machine, the distance defined by the kernel function may be used instead of the Euclidean distance when calculating the distance. By using the distance defined by the kernel function, a non-linear boundary line can be obtained. The weight vector and the threshold are determined so as to satisfy the following conditions.
(1) The degree of coincidence between the classification of the sample vectors based on whether or not the inner product with the weight vector exceeds a predetermined threshold and the above binary classification.
(2) The distance between the generated boundary line and the sample vector (support vector) closest to the boundary line is the largest.
When this method is applied to diagnosis support information, each sample vector is a vector having an average value and a variance value of cortical bone thickness as elements, and a binary classification is a classification based on the presence or absence of osteoporosis. In this way, a boundary line for separating osteoporosis patients from those who are not is required.

  When categorizing such an area and discriminating for a new patient, the average and variance values of the patient's measurements are applied to this figure, and if it is in the osteoporosis area, it is positive for osteoporosis. If not, it is determined to be negative.

  In this discrimination method, discrimination is performed in consideration of not only the average value of cortical bone thickness but also the variance value. The variance of cortical bone is also likely to give information correlating with osteoporosis, and this method further increases the accuracy of osteoporosis discrimination.

  In specifying the center line in step S402, an algorithm using dynamic programming may be used for tracking the maximum value of the distance value.

  When generating a plurality of circles on the center line in step S402, the circumference of the circle may be configured as a set of pixels. When obtaining the circumference as a set of pixels, for example, points are generated on the circumference for each angle of 0.001 radians, and the X and Y coordinates of these points are rounded, rounded up, rounded down, etc. to be converted into integers. A set of pixels whose positions are specified in this way may be a circumference.

  The upper and lower envelopes formed by a plurality of circles in step S403 may also be configured as pixel columns. For example, out of the circumferences of a plurality of circles configured as a pixel set, a pixel column including pixels that do not overlap with the inside of another circle may be used as an envelope.

  It is desirable that the number of points provided on the boundary line when determining the approximate line in step S501 is as many as possible. When the boundary line is composed of pixel columns, for example, it is desirable to use all the pixels constituting the boundary line.

  In step S502, it is desirable that as many measurement auxiliary lines as possible be provided on the approximate line. For example, it is desirable to select all points where the X coordinate of the approximate line is an integral multiple of the distance between the two pixels as the measurement point and determine the measurement auxiliary line.

  In step S503, the distance between the two boundary lines on the measurement auxiliary line may be calculated as follows. When the boundary line is a pixel column, the pixel closest to the measurement auxiliary line in the upper boundary line and the pixel closest to the measurement auxiliary line in the lower boundary line are two of the measurement auxiliary line and the boundary line. It may be considered to represent one intersection. Therefore, the Euclidean distance between these two pixels may be the cortical bone thickness on the measurement auxiliary line. Cortical bone thickness measurement data may be converted into integers for pixel processing for later processing.

  In step S602, the reference method used for collation as the diagnosis result of osteoporosis may be bone density measurement of the lumbar spine and the femoral neck by DXA (dual energy X-ray absorption measurement method).

  The osteoporosis diagnosis support apparatus 1 described above can be realized not only by a dedicated system but also by using a normal computer system. For example, the apparatus may be configured by storing and distributing a program for executing the above-described operation in a computer-readable recording medium, installing the program in a computer, and executing the above-described processing. Further, it may be stored in a disk device provided in a server device on a network such as the Internet so that it can be downloaded to a computer, for example. Further, the above-described functions may be realized by the collaboration of the OS and application software. In this case, only the part other than the OS may be stored and distributed in a medium, or may be downloaded to a computer.

  As a recording medium for recording the above program, USB memory, flexible disk, CD, DVD, Blu-ray Disc (registered trademark), MO, SD card, MS (memory stick) (registered trademark), magnetic disk, optical disk A computer-readable recording medium such as a magneto-optical disk, a semiconductor memory, or a magnetic tape can be used. In addition, it is also possible to use a recording medium that is usually fixed to a system or apparatus, such as an HDD (hard disk) or an SSD (solid state drive).

  As shown below, cortical bone thickness was measured using the dental panoramic X-ray image shown in FIG. The panoramic X-ray image in FIG. 2 is obtained by scanning a film-type device (Asahi X-ray AZ-3000) at a resolution of 300 dpi and converting it into a digital image. From this image, a region with a width of 300 pixels around the pit hole surrounded by a rectangle was extracted at two locations on the left and right sides, and subjected to a diagnosis support process.

  FIG. 17 shows a histogram of cortical bone thickness measurement values for the left side. In the histogram of FIG. 17, there is a set of values in which a major peak is seen around 41 pixels and noise is seen around 20 pixels.

  Clustering was performed on the histogram of FIG. 17 to extract main clusters. FIG. 19 shows a histogram of the extracted main cluster. Here, it can be seen that the noise is removed and the main cluster is suitably selected.

  In order to verify the effectiveness of the above-described noise removal method, noise removal was also performed by other methods and compared. Here, a method of removing the measured values of the maximum 10% and the minimum 10% was used. A histogram obtained by this noise removal is shown in FIG. It can be seen that a small cluster remains in the vicinity of the thickness measurement value of 20 pixels, and noise has not been completely removed. Thus, it can be seen that noise removal can be effectively performed by using the method based on clustering of the present embodiment.

  Osteoporosis was determined using the estimated cortical bone thickness obtained by the above clustering method.

  First, subjects were selected for the creation of collation data as follows. From 531 women who underwent bone density measurements at the lumbar and femoral neck by dental panoramic X-ray images and DXA between 1996 and 2001, from postmenopausal women over 50 years old with no history of osteoporosis A limited number of 100 people were extracted as subjects. In this extraction, in order to make the conditions uniform, the diagnosis of menstrual / metabolic bone diseases and cancer bone metastases within the past year, medications that may affect bone metabolism such as female hormones, uterine and ovarian Limited to subjects with no excision, no history of smoking, and no bone destruction in the jawbone.

  Next, the threshold value for determining osteoporosis was set as follows. When the threshold is set so that the sensitivity is 92.0% with respect to the above collation data, the threshold on the right side of the mandible is 34.3 pixels when the lumbar bone density is used as a reference, and the femoral neck bone density Was 36.6 pixels. As for the left side of the mandible, it was 37.4 pixels when the lumbar bone density was used as a reference, and 33.6 pixels when the femoral neck bone density was used as a reference.

  Table 1 shows the specificity, sensitivity, positive predictive value, negative predictive value, and accuracy of the diagnostic results of the above 100 subjects. Here, the positive predictive value is the ratio of those diagnosed as positive by DXA among the persons identified as positive for osteoporosis by the dental panoramic X-ray image, and the negative predictive value is the dental panoramic X-ray. This is the ratio of people diagnosed as negative by DXA among the people judged negative by the image. The accuracy is the percentage of people who have a diagnosis result by DXA and a diagnosis result by a dental panoramic X-ray image corresponding to the total number of diagnoses.

  As shown in Table 1, when sensitivity is 92.0%, the specificity (the proportion of those who were not detected falsely among those who are not osteoporotic patients) is based on the lumbar bone density for the right side of the mandible 67.0% and 55.0% based on the femoral neck bone density, and 64.0% and 75.0% on the left side, respectively.

  The above results were compared with a diagnostic method based on one-point measurement described in Patent Document 1 (Agus Zainal Arifin, A. Asano, A. Taguchi, T. Nakamoto, M. Ohtsuka, M. Tsuda, Y. Kudo, and K. Tanimoto, "Computer-aided system for measuring the mandibular cortical width on dental panoramic radiographs in idenifying postmenopausal women with low bone mineral density," Osteoporosis International, Vol. 17, No. 5, pp 753-759, 2006). In a one-point diagnosis, the sensitivity and specificity are 88.0% and 58.7%, respectively, based on lumbar bone density, and 87.5% and 56, respectively, based on femoral neck bone density. 3%. In addition, in the diagnosis results obtained by manual measurement by skilled dental radiologists described in the above-mentioned literature, the sensitivity and specificity are 92.0% and 60.0%, respectively, when the lumbar bone density is a reference, and the femur They were 87.5% and 64.8%, respectively, based on the neck bone density. From the above, it can be seen that the osteoporosis diagnosis support apparatus of the present invention exceeds the method described in Patent Document 1 by one-point measurement in most cases, and has the same or better diagnostic ability than the manual measurement by skilled persons. It was.

  In order to verify the accuracy of discrimination of osteoporosis in the present example, an operation characteristic analysis of the examinee was performed. Receiver operating characteristic (ROC) analysis is a method for comprehensively handling sensitivity and specificity for various thresholds and evaluating the ability of a diagnostic method. In ROC analysis, (1-specificity) is plotted on the horizontal axis and sensitivity is plotted on the vertical axis, and the relationship between the two when the threshold value is changed is represented in a plot (ROC curve). Ideally, higher sensitivity and lower (1-specificity) are desirable. Therefore, when the area of the lower part of the ROC curve is larger, the sensitivity is higher and the ROC curve passes through a place where (1-specificity) is lower. Can be obtained.

  FIG. 21 shows an ROC curve (example on the right side of the mandible) obtained by this example. FIG. 21A corresponds to the case where the lumbar bone density is used as a reference, and FIG. 21B corresponds to the case where the femoral neck bone density is used as a reference. The area under the ROC curve is 0.851 for the right side of the mandible when based on the lumbar bone density, 0.830 when the base for the femoral neck bone density, and 0.841 for the left side, respectively. It was 0.863. In the method based on the single point measurement, 0.777 was obtained when the lumbar bone density was used as a reference, and 0.803 when the femoral neck bone density was used as a reference. From this point, the method according to the present invention can provide a better diagnosis. You can see that you have the ability.

  Further, osteoporosis was also determined by a method using both the average value and the dispersion value of the thickness measurement value group. In FIG. 22, the plot of the average value and dispersion value of a thickness measurement value group is shown.

  In FIG. 22, 0 is a patient with osteoporosis diagnosed by DXA, 1 is a person who is not, FIG. 22 (a) is based on the lumbar bone density, and (b) is based on the femoral neck bone density. It corresponds to the case. In the plot, there are basically areas of osteoporosis patients and those who are not, and there are also areas where both are mixed. The optimal boundary between the two regions determined using the support vector machine is also shown in FIG. “training” is learning data that has already been diagnosed, and “classified” is data that has been diagnosed based on the obtained boundary line. As described above, if the boundary line is determined in advance by a large number of learning data, it is possible to determine osteoporosis.

  The osteoporosis diagnosis support apparatus of the present invention is premised on the use of a panoramic X-ray image taken in a dental examination, and it is not necessary to newly take an image. For this reason, the cost required for osteoporosis diagnosis can be significantly reduced.

1 Osteoporosis diagnosis support device 11 X-ray image acquisition unit 12 Control unit 13 Storage unit 14 Output unit 15 Operation unit

Claims (9)

Boundary line specifying means for specifying two boundary lines between cortical bone and other portions reflected in a band shape in an X-ray image;
Based on a plurality of points provided on the boundary line, approximate line calculating means for obtaining an approximate line that is a straight line or a curve that approximates the boundary line;
A plurality of measurement points are set on the approximate line, a distance between the two boundary lines is measured on a measurement auxiliary line extending from the set measurement point in a normal direction of the approximate line, and a plurality of measurement values are measured. Measurement value group acquisition means for acquiring
Cortical bone thickness calculating means for calculating the thickness of cortical bone based on the measured value;
Osteoporosis discriminating means for discriminating osteoporosis based on the cortical bone thickness calculated by the cortical bone thickness calculating means,
An osteoporosis diagnosis support apparatus characterized by the above. The boundary line specifying means includes:
Cortical bone part specifying means for specifying a cortical bone part and other background parts based on the density of the X-ray image;
For each pixel belonging to the cortical bone part, to calculate the shortest distance to the background part, the shortest distance acquisition means for associating the calculated shortest distance and the corresponding pixel,
By tracking the maximum value of the shortest distance, centerline acquisition means for obtaining the centerline of cortical bone,
Boundary line acquisition means for acquiring, as the boundary line, envelopes of a plurality of circles centered on a plurality of points provided on the center line,
The boundary line acquisition means includes
The shortest distance associated with the pixel located at the center point is the radius of each circle,
The osteoporosis diagnosis support apparatus according to claim 1. The cortical bone thickness calculating means includes
Means for obtaining a histogram of the measurement value range and the number of occurrences thereof from the plurality of measurement values;
A plurality of sections in the range of the histogram, and a histogram dividing means for dividing the histogram based on the sections;
Means for selecting a histogram having the maximum total number of appearances among the plurality of divided histograms;
A means for calculating an expected value of a measured value in a histogram in which the sum of the selected number of appearances is maximum, and obtaining the calculated expected value as a thickness of cortical bone, and
The histogram dividing means includes
After dividing the histogram, set the number of categories so that the difference between the total number of appearances of the histogram with the maximum number of appearances and the histogram with the second largest total number of appearances is maximized.
The osteoporosis diagnosis support apparatus according to claim 1 or 2. The cortical bone thickness calculating means includes
Means for obtaining an average value of the plurality of measured values as the thickness of the cortical bone,
The osteoporosis discrimination means is
Discriminate osteoporosis based on the thickness of the cortical bone obtained by the cortical bone thickness calculating means and the variance of the measured value,
The osteoporosis diagnosis support apparatus according to claim 1 or 2. The histogram dividing means includes
Set acquisition means for randomly associating numerical values of the plurality of measurement values with a predetermined number of sets, and acquiring a plurality of sets;
For each of the plurality of sets, an expected value acquisition means for obtaining a sum within a set of values obtained by multiplying the measured value and the number of times the measured value appears, and an expected value;
One of the numerical values that can be taken by the measured value is selected, compared with the expected value, and the operation of redistributing the selected numerical value to the set having the closest expected value is performed for all possible numerical values of the measured value. Redistribution means,
Repeating the expected value acquisition means and the redistribution means alternately, and continuing until the redistribution does not occur;
The osteoporosis diagnosis support apparatus according to claim 3, further comprising: The approximate line is a straight line or a curve expressed by a polynomial.
The osteoporosis diagnosis support apparatus according to any one of claims 1 to 5. The approximate line is a curve expressed by a quadratic function.
The osteoporosis diagnosis support apparatus according to any one of claims 1 to 6. The X-ray image is a dental panoramic X-ray image including a mandibular portion of a person.
The osteoporosis diagnosis support apparatus according to any one of claims 1 to 7. A boundary line specifying step for specifying two boundary lines between the cortical bone and the other part appearing in a band shape in the X-ray image;
An approximate line calculation step for obtaining an approximate line that is a straight line or a curve that approximates the boundary line based on a plurality of points provided on the boundary line;
A plurality of measurement points are set on the approximate line, a distance between the two boundary lines is measured on a measurement auxiliary line extending from the set measurement point in a normal direction of the approximate line, and a plurality of measurement values are measured. A measurement value group acquisition step for acquiring
Cortical bone thickness calculating step for calculating the thickness of cortical bone based on the measured value;
Causing the computer to execute an osteoporosis determining step of determining osteoporosis based on the thickness of the cortical bone calculated in the cortical bone thickness calculating step.
An osteoporosis diagnosis support program characterized by that.