JP2002366961A - Method and device for detecting image contour - Google Patents

Abstract

PROBLEM TO BE SOLVED: To detect a cardiothoracic contour in a chest image with high accuracy even in the case cardiothroracic the contour is unclear. SOLUTION: A contour candidate detecting means 30 detects a cardiothoracic contour candidate in a chest radiation image P1, a discontinuous point detecting means 14 detects discontinuous points in the detected cardiothoracic contour candidate, a smoothing processing means 15 performs smoothing processing of the contour candidate when the discontinuous point detecting means 14 detects the discontinuous points, and a redetecting means 16 redetects a contour candidate about the discontinuous points in the contour candidate subjected to smoothing processing. A contour is detected in such a manner that the discontinuous point detecting means 14 detects discontinuous points, the smoothing processing means 15 performs smoothing processing and the redetecting means 16 redetects a contour about the discontinuous points repeatedly in this order until the discontinuous point detecting means 14 can not detect a discontinuous point any more.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method and an apparatus for detecting an outline of an image.

[0002]

2. Description of the Related Art Conventionally, it has been common practice to measure the size, curvature, etc. of anatomical features of bones and organs using medical images such as X-ray images, so as to monitor the presence or absence of disease and follow-up. It is frequently performed at consultations. For example, when diagnosing cardiac hypertrophy, cardiac hypertrophy is measured by measuring each width of the thorax and the heart using a chest radiographic image and calculating the ratio (cardiothoracic ratio = heart width / thoracic width). In the case of judging whether or not the patient has a spinal curvature or performing a diagnosis of spinal curvature, determination of whether or not the patient has spinal curvature is performed by measuring the curvature of the spine.

[0003] For example, when calculating the above-described cardiothoracic ratio, a doctor or a radiological technologist or the like conventionally applies a ruler or the like to a cardiothoracic image in a chest radiographic image to obtain a ribcage (an area in which both left and right lung fields are combined). ) And the width of the heart were measured, and the cardiothoracic ratio was calculated based on the measured values.

[0004] However, in recent years, it has become common to digitize an image and handle it as image data. Since it is easy to perform various kinds of signal processing on this image data, the above-described cardiothoracic ratio is not easily obtained. It is desired that the calculation be performed automatically.

[0005] Therefore, when calculating the cardiothoracic ratio based on the cardiothoracic image in the chest image, the inventor converts the cardiothoracic image in the chest image into polar coordinates to obtain an average cardiothoracic reference. By performing template matching processing on the polar coordinate plane using a template that is substantially similar to the contour of the heart, the contour of the cardiothoracic (cardiothoracic together with the chest) is automatically detected with high accuracy and simplified. We propose a cardiothoracic contour detection method and a cardiothoracic ratio calculation method that automatically measure the width of the thorax and the width of the heart based on the detected contour, and calculate the cardiothoracic ratio from the result of the automatic measurement. (Japanese Patent Application 2000-336859).

[0006]

However, in the method for detecting the contour of the cardiothoracic image as described above, the contour of the cardiothoracic necessary for the automatic measurement may not be clearly visible depending on the patient and imaging conditions. When the accuracy of the detected cardiothoracic contour is reduced, for example, when the cardiothoracic ratio is measured based on the detected cardiothoracic contour, the cardiothoracic ratio is defined as the cardiothoracic contour is clearly defined. The measurement error tended to increase when compared to the automatically measured cardiothoracic ratio when visible.

The present invention has been made in view of the above-described problems. For example, even when the outline of a cardiothoracic image in a chest image is unclear, the outline of the image can be accurately detected. It is an object to provide a detection method and device.

[0008]

According to the image contour detecting method of the present invention, in the image contour detecting method for detecting the contour of an image, a contour candidate of the image is detected, and a discontinuous point in the detected contour candidate is detected. Detecting, detecting a discontinuous point, performing a smoothing process on the contour candidate, performing a contour detection again on the discontinuous point in the contour candidate subjected to the smoothing process, detecting the discontinuous point, performing the smoothing process and It is characterized in that the contour detection is performed by repeating the contour detection for the discontinuous point once or more in this order at least once.

Here, the "image" may be any image as long as it is an image to be subjected to contour detection. However, a medical image such as a radiographic image or an image obtained by a CT apparatus or an MRI apparatus is an image according to the present invention. It is particularly suitable for the contour detection method.

In the medical image, particularly, the image of the cardiothoracic cavity (the portion where the thorax and the heart are combined) is likely to be blurred due to the overlap of shadows such as fat and gas in the abdomen. This is suitable for the image contour detection method using

The above-mentioned "image outline candidate" refers to an outline that may be an image outline.

The above-mentioned "discontinuous point in the contour candidate of the image" means that, when the contour candidate is composed of vertically arranged points, at least one of the vertically adjacent points. When the positional relationship is a point separated by a predetermined distance or more in the left-right direction, or when the outline candidate is composed of points arranged in the left and right directions, the positional relationship with at least one of the points present adjacent to the left and right. Means a point separated by a predetermined distance or more in the vertical direction. That is, for example, when the contour candidate is composed of vertically arranged points as shown in FIG. 14, the black dots shown in the figure are discontinuous points.

The "smoothing process" refers to a process of correcting the position of a discontinuous point in the detected contour candidate so that the discontinuous point has continuity with the contour candidate.

[0014] The above-mentioned "re-detecting the contour of the discontinuous point" means that the discontinuous point subjected to the smoothing process is moved to a position which is more likely to be the contour to be detected. Means

Further, the above-mentioned "repetition of discontinuous point detection, smoothing processing and contour detection for discontinuous point again"
Can be repeated until no discontinuous point is detected.

An image outline detecting apparatus according to the present invention is an image detecting apparatus for detecting an outline of an image, wherein an outline candidate detecting means for detecting an outline candidate of the image, and a discontinuous point in the detected outline candidate of the image. Discontinuity point detection means, a smoothing processing means for performing smoothing processing on a contour candidate when a discontinuity point is detected by the discontinuity point detection means, and a contour candidate subjected to smoothing processing by the smoothing processing means Re-detecting means for re-detecting the contour at the discontinuous point in step (a), detecting the discontinuous point by the discontinuous point detecting means, smoothing by the smoothing processing means, and re-detecting the discontinuous point by the re-detecting means. Is repeatedly performed once or more in this order to detect a contour.

Further, the detection of the discontinuous point, the smoothing process, and the repetition of the contour detection for the discontinuous point can be repeated until the discontinuous point is not detected by the discontinuous point detecting means. .

[0018]

According to the image contour detecting method and apparatus according to the present invention, a contour candidate of an image is detected, a discontinuous point in the detected contour candidate is detected, and when a discontinuous point is detected, the contour is detected. The candidate is subjected to a smoothing process, the contour is detected again at a discontinuous point in the contour candidate subjected to the smoothing process, and the detection of the discontinuous point, the smoothing process, and the contour detection at the discontinuous point are performed again. Since the contour is repeatedly detected one or more times in order, the contour of an unclear image can be more clearly detected.
Even when the contour of the cardiothoracic in the chest image is not clear due to the influence of the patient and the imaging conditions, etc., it is possible to accurately detect the contour as close as possible to the original contour, and based on the detected cardiothoracic contour, When the ratio is automatically measured, the measurement error can be reduced.

In the case where the detection of the contour is performed by repeating the above-described detection of the discontinuous point, the smoothing process, and the contour detection for the discontinuous point again in this order until the discontinuous point is no longer detected. Thus, a contour that is clearer and as close as possible to the original contour can be detected with high accuracy.

When a radiation image is applied as an image of the image contour detection method and apparatus according to the present invention, the radiation image is an image obtained by projecting a subject in which various structures (such as human body tissues) are complicatedly overlapped. Since the outline of the structure to be detected tends to be unclear, the outline cannot be accurately detected in the unclear portion, and the structure tends to be discontinuous. Therefore, the above-mentioned effect is more remarkable.

[0021]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, a specific embodiment of a cardiothoracic contour detecting apparatus for implementing a cardiothoracic contour detecting method to which the image contour detecting method of the present invention is applied will be described in detail with reference to the drawings.

FIG. 1 is a block diagram showing a schematic configuration of the present embodiment. FIG. 2 is a flowchart showing the processing of the cardiothoracic contour detection method of the present invention. FIG. 3A is a view showing a chest radiographic image P1 to be processed by the cardiothoracic contour detecting method of the present embodiment. (2) is a diagram schematically showing the chest radiation image P1 shown in (1). Usually,
The chest radiographic image is a high-density high pixel value image in which the pixel value increases as the density increases (the direction of blackening) and decreases as the density decreases (the direction of whitening). In various processes of the present embodiment, an image obtained by inverting the pixel value, that is, a pixel value becomes smaller as the density becomes higher (black direction) and becomes smaller as the density becomes lower (white direction). An image with a large low density and high prime value is used. In the schematic diagram shown in FIG. 3B, the outer contours of the left and right lung fields Pa and Pb are represented by solid lines indicated by reference signs PA and PB.
The area surrounded by is the rib cage. In addition, each lung field Pa, Pb
Are represented by solid lines indicated by reference symbols PC and PD. Further, of the contours of the heart Pc, a contour forming a boundary with the left lung field Pb is represented by a solid line indicated by a symbol PE.

The cardiothoracic contour detection device of the present embodiment is shown in FIG.
The rib cage contour P in the chest radiographic image P1 shown in (2)
A and PB, inner contours PC and PD of the lung fields Pa and Pb, and a left contour PE of the heart Pc are detected, respectively. Then, based on the detected contour of the cardiothoracic cage, the cardiothoracic ratio is measured and calculated by the connected measuring means at the subsequent stage.

The cardiothoracic contour detection device of the present embodiment is shown in FIG.
As shown in FIG. 3, the outline contour detection processing is performed on the chest radiation image P1 shown in FIG. 3 (2), and the other side is subjected to the smoothing image processing, so that the outline contour image P of the chest radiation image P1 is obtained.
2 and a smoothed image P3 are obtained, and a process of smoothing the inside of the rib cage by multiplying the outline image P2 and the smoothed image P3 for each corresponding pixel is executed, and the inside of the rib cage shown in FIG. A smoothed image processing means 11 for obtaining a smoothed image P4,
4, a reference center point (xc, yc) which is substantially the center of the substantially arc-shaped contour portion PA of the rib cage contours PA and PB is determined, and the chest radiographic image P1 is subjected to polar coordinate conversion with respect to the reference center point (xc, yc). The first template matching process is performed on the polar coordinate conversion means 12 and the polar coordinate image P5 obtained by the polar coordinate conversion, using the fixed templates T1, T2, and T3, each of which is a reference contour.
A, PB, PC, PD, and PE are detected as candidate points, and further subjected to a second template matching process using predetermined elastic templates T1 ', T2', and T3 ', so that each of the contours PA, PB, PC, PD, and PE is processed. Contour candidate detecting means 30 having a matching processing means 13 for detecting PE candidate points;
A discontinuous point detecting means 14 for detecting whether or not there is a discontinuous point in each of the contours PA, PB, PC, PD, PE composed of candidate points detected by the contour candidate detecting means 30; When a continuous point is detected, smoothing processing means 15 for performing smoothing processing on the contour candidate
Re-detection means 16 for detecting the contour again at the discontinuous point in the contour candidate smoothed by the smoothing processing means 15, and smoothing when the discontinuous point is not detected by the discontinuous point detecting means 14. The contours PA, PB,
Interpolation processing is applied to PC, PD, and PE, and each lung field Pa, Pb
And the interpolation processing means 17 for detecting the contour of the heart Pc. Then, the cardiothoracic contour output from the cardiothoracic contour detection device 10 is output to the measuring means 20.

The measuring means 20 includes the cardiothoracic contour detecting device 10
Of heart and thorax PA, PB, PC, P
Based on D and PE, as shown in FIG. 19, the width L1 of the heart Pc and the width L2 of the thorax (right lung field Pa and left lung field Pb), as shown in FIG.
Are respectively detected, and the ratio L1 / L2 of the obtained heart width L1 and thorax width L2 is calculated to be a cardiothoracic ratio.

Next, the operation of the cardiothoracic contour detecting apparatus according to the present embodiment will be described with reference to the flowchart shown in FIG.

First, the outline detection process (# 1) and the smoothing process (#) are performed on the input chest radiation image P1.
2) is performed. The outline detection process (# 1) is performed for each (a) of FIGS. 4A to 4C on the chest radiation image P1.
6 directions (0 °, 3 ° in each (b) of FIG.
(6 directions of 0 °, 60 °, 90 °, 120 °, 150 °)
5 (1) through 5 (1)-
Six images (first to sixth outline contour images P2a to P2f) in which the contours extending in the directions corresponding to the respective contour detection masks are emphasized using the edge detection masks for each extension direction in (3)
(Not shown)) and three images (seventh to ninth outline contour images P2g to P2i (not shown)) in which edges extending in directions corresponding to the edge detection masks in the extending directions are enhanced. In this process, a pixel having a maximum (lowest density) pixel value is selected by associating the pixels of these nine images P2a to P2i with each other to synthesize a single outline image P2.

That is, each (b) of FIGS.
In the contour detection mask shown in FIG. 2 (a) for detecting a straight line extending in a specific angle direction as shown in (a), the positive (+ sign) portion of the mask is a straight line so as to easily respond to a straight line in each direction. It has an elongated elliptical shape to fit the direction,
In addition, the negative part is chosen to be distributed on both sides of the positive part. Such a negative portion is indispensable for giving sufficient orientation selectivity to these masks. on the other hand,
The edge detection mask shown in FIG. 4A for detecting an edge extending in a specific angle direction and having a specific density gradient direction as shown in FIG. 4B to FIG. In order to easily respond to the straight line in each direction and the gradient direction of the density in a specific direction (in particular, the inner contours PC and PD of the lung fields Pa and Pb and the left contour PE of the heart Pc), the positive ( The (+ sign) portion has an elongated elliptical shape so as to fit in the linear direction, and the negative portion is selected to be distributed on one side of the positive portion. Such a negative portion is indispensable for providing these masks with sufficient azimuth selectivity and edge density gradient.

Then, using a contour detection mask whose longitudinal direction of the elongated ellipse is six specific directions and an edge detection mask whose longitudinal direction of the elongated ellipse is the specific direction and the density gradient direction is the specific direction, Convolve the radiation image P1. Here, this mask corresponds to a simple cell in the visual cortex of the cerebrum, and is created by a Gabor function. This Gabor function is represented by the following equation.

[0030]

(Equation 1) Here, it is a real part in the equation (1).

(Equation 2) Is used to create a contour detection mask in a specific direction. Further, according to the initial values of kx and ky in the equation (2), the longitudinal direction of the elongated ellipse is 0 °, 30 °, 60 °,
Each mask of 90 °, 120 °, and 150 ° can be created.

On the other hand, the edge detection mask uses the imaginary part (detects the first derivative) of the Gabor function.

The receptive field size of the contour detection mask in a specific direction is determined so that it does not easily react to a thin contour component other than the required contour component in the chest radiographic image P1. That is, the mask is the chest radiation image P1
Of these, it is easy to react to structures considered to be the rib cage and the outline PA of the rib cage. By determining the size of the receptive field of the mask in this manner, the above-described contours PA and PB can be detected satisfactorily irrespective of the presence of the background. Thus, a contour component (FIG. 4B) in a specific direction suitable for each mask is extracted from the chest radiation image P1.

Similarly, the contours PC, PD, and PE can be favorably detected by the edge detection mask.
By convolving the chest radiation image P1 with an edge detection mask in a specific direction, the chest radiation image P1 is obtained.
From the edge component in a specific direction suitable for each mask (Fig. 5
(B)) is extracted.

Here, the convolution and non-linear processing of each contour component of the chest radiographic image P1 by the contour detection mask in the specific direction are performed by the following equation (3).

[0035]

(Equation 3) By the above-described outline detection process (# 1), the outline outline image P2 is obtained as shown in FIG.

The smoothed image processing (# 2) for the chest radiographic image P1 is a general smoothed image processing using, for example, a mask created by a Gaussian function. A smoothed image P3 in which ribs and collarbones are not conspicuous as shown in FIG. Note that the Gaussian coefficient and size are set to such an extent that ribs and collarbones are inconspicuous.

Next, the inside of the rib cage is smoothed by multiplying the outline image P2 and the smoothed image P3 obtained in this way for each corresponding pixel (# 3). The reconstructed chest ribs smoothed image P4 is shown in FIG.
As shown in (1), since the inner region of the rib cage in the smoothed image P3 has a higher density than the outer region of the rib cage, the inside of the rib cage has a relatively lower pixel value (high density) than the outer region of the rib cage. The inner ribs are the contours of the cardiothoracic PA, PB, PC,
It is relatively smoother than PD and PE. Hereinafter, in order to simplify the display of the image, description will be given using the simplified image P4 shown in FIG. 7 (2).

Next, a process (# 4) for determining a reference center point is performed on the intrathoracic smoothed image P4. The process of determining the reference center point (xc, yc) (# 4) is described in FIG.
This is a process of determining a pole (center point) at the time of polar coordinate conversion of the intrathoracic smoothed image P4 shown in (2). This pole is located at substantially the same distance from the substantially arc-shaped contour part PA of the thorax contour. Adopt a point. Specifically, first, the physical center point (x0, y0) of the intrathoracic smoothed image P4 is set as a temporary reference center point (xc, yc) (FIG. 2B). That is, xc = x0 and yc = y0.

Here, while the y-coordinate is fixed and the x-coordinate is moved in the left-right direction by x pixels from xc, at each of the moving positions, an image of the inside of the rib cage P4 centered on the moving position (xc + j, yc) is obtained. Left and right profiles ((xc +
The movement position j at which the correlation between the pixel value of (j + i, yc) and the pixel value of (xc + ji-y, yc) is maximized. That is, the correlation value cor (j) is

(Equation 4) And the moving position j at which the correlation value becomes maximum is jmax.
Then, when (xc + jmax, yc) is the center,
Since the left and right profiles of the intrathoracic-smoothed image P4 have the highest correlation, x = xc + jmax is equal to the chest radiation image P1.
It turns out that it becomes a left-right symmetric axis. Therefore, the normal reference center point is (xc + jmax, yc). The eligibility of the y-coordinate of the reference center point will be described later with reference to a process (#
Consider in 6).

Next, as shown in FIG. 8, the intrathoracic smoothed image P4 is subjected to polar coordinate transformation with respect to the reference center point (xc + jmax, yc) (# 5). In other words, the smoothed image inside the rib cage P4 represented as shown in FIG. 7 (2) on the real image plane is directed downward from the distance r from the reference center point (xc + jmax, yc) and from the reference center point (xc + jmax, yc). Polar coordinate transformed image P expressed by angle θ with vector
Convert to 5.

Next, the polar coordinate transformed image P5 is converted into a fixed template T1 representing the reference rib cage contours PA 'and PB', a fixed template T2 representing the reference rib cage inner contours PC 'and PD', And a fixed template T3 representing a left-sided heart contour PE 'of the thorax as a reference, and the first and second cardiothoracic contours PA, P are fixedly obtained by a first template matching process (# 6).
B, PC, PD, and PE candidate points are detected. Here, the reference cardiothoracic contours PA ′, PB ′, PC ′,
PD ′ and PE ′ are averages of a large number of clinically obtained cardiothoracic contours, and on the real image plane, cardiothoracic contours PA, PB, PC, PD, PE Although it has a substantially similar shape, on the polar coordinate plane, as shown in FIG. 9, for example, it has substantially the same shape as the contours PA, PB, PC, PD, and PE of the cardiothoracic chest, but it is assumed that it has been translated mainly in the r direction. Is represented.

The fixed template T1 representing the ribcage outlines PA 'and PB' serving as a reference includes a plurality of pixels (the ribcage coordinates (ri, θ
i), which is created as a set and serves as a reference
The fixed template T2 representing C ′, PD ′ is composed of a plurality of pixels (thoracic coordinates (r
i, θi), and a fixed template T3 representing a contour PE ′ on the left side of the heart as a reference is a plurality of pixels (thoracic coordinates (r
i, θi).

Then, the outline of the rib cage PA ′ serving as the reference is
T1 based on PB and PB '
5, and the total value d (r, θ) of the pixels constituting the template T1 (formula (5) below)
The position of the template T1 at which is maximized is determined. Similarly, the template T2 based on the lung field inner contours PC ′ and PD ′ serving as references is moved up, down, left, and right on the polar coordinate image P5, and the sum value d (r, θ) of the values of the pixels constituting the template T2 is obtained. Is determined, the fixed template T3 based on the contour PE 'on the left side of the heart is moved up, down, left, and right on the polar coordinate image P5, and the sum value d (r) of the values of the pixels constituting the template T3 is obtained. ,
The position of the template T3 at which θ) is maximized is determined.

[0044]

(Equation 5) The moving range of each of the templates T1 to T3 is about ± 30 pixels in the r direction and about ± 10 ° in the θ direction, but is not limited thereto. In addition, when calculating the pixel value sum value, the detected density value may be weighted and calculated for each pixel constituting the template T. This is similar to the target image of the present embodiment, in which the lower end portion PB ′ of the lung field in the real image plane (θ ranges from approximately 0 ° to approximately 30 ° in the polar coordinate plane in FIG.
(A range of about 360 ° to about 360 °), there is a large individual difference. This makes it possible to perform matching with priority given to. Further, the maximum pixel value in the range A of several pixels adjacent to each pixel constituting each of the templates T1 to T3 is applied as max {g (ri + r, θi + θ)}, and the pixel value sum value according to the following equation (6) It is more preferable to determine the position of the template T at which d (r, θ) is maximized. Each contour PA, PB, P in the polar coordinate image P5
Even if C, PD and PE are slightly different from the shapes of the templates T1 to T3, the contours PA, PB, PC, PD,
This is because PE can be detected.

[0045]

(Equation 6) Note that the pixel value sum value d (r, θ) represented by the equation (6)
Of the reference center point (xc + jmax, y
By moving the y coordinate of c) k pixels at a time, new reference center points (xc + jmax, yc ± k) are respectively set, and polar coordinate transformation centered on the new reference center points (xc + jmax, yc ± k) is performed. , Rib cage internal smoothed image P4
Are respectively subjected to polar coordinate conversion, and a new polar coordinate transformed image P
5 ′, and the y coordinate when the maximum value dmax of the sum of the pixel values of the obtained new polar coordinate transformed image P5 ′ becomes maximum is nyc, and finally, the reference center point is (nxc ( = Xc + jmax), nyc). When obtaining nyc, the template T1 may be used. As a result, the eligibility of the y coordinate of the reference center point obtained first is determined. Then, if the polar coordinate transformed image P5 'obtained by performing polar coordinate transformation on the rib cage internal smoothed image P4 around the reference center point (nxc, nyc) is displayed in polar coordinates, (nri (= ri + rmax), nθ)
i (= θi + θmax)). Where rmax and θ
max indicates the amount of movement in the r direction and the θ direction obtained by template matching when the maximum value dmax of the sum of pixel values is the largest.

Next, using the candidate points of the rib cage contour obtained from the fixed template T as initial values, predetermined elastic templates T1 'to T1
3 ′ using a second template matching process (#
Perform 7). These elastic templates T1 'to T3'
Does not mean that all of the constituent pixels move integrally as in the case of the fixed templates T1 to T3, and that each pixel is constrained between adjacent pixels by a constraining force corresponding to the moving amount r of each pixel. This is a template which is set to be movable independently in the r direction while being subjected to a virtual spring constraint, and the entire template is configured to be elastically deformed.

Here, in the polar coordinate plane, the elastic template T1 'has a contour portion PB' at the lower end of the lung field (in the polar coordinate plane of FIG. 9, θ ranges from approximately 0 ° to approximately 30 ° and approximately 330 ° to approximately 360 °). Is set in a range excluding the range (FIG. 11). This is because it is easier to search for the contour portion PB at the lower end of the lung field on the real image plane.

The elastic template T1 'is arranged on the polar image plane of the thorax with the candidate points of the ribcage contour obtained by the fixed template T1 as its initial values (state without elastic deformation) (FIG. 12 (1)). , This elastic template T1 '
Are independently moved in the r direction (the vertical direction in FIG. 11). At this time, the movement amount of each pixel of the elastic template T 'is obtained as follows. First, the difference between each pixel value g (nri ± r, nθi) and the pixel value g (nri, nθi) at the initial position is determined in the peripheral range (initial position ± r) of each pixel. At this time, the pixel value at the position where r is small is subtracted from the pixel value at the position where r is large. Then, the sum of the differences is obtained by the following equation (7).

[0049]

(Equation 7) This pixel value difference sum takes a positive value if there is a pixel brighter (high pixel value (low density)) in the direction where r is larger than the initial position, and is brighter (high pixel value) in the direction r smaller than the initial position. (Low density)) If a pixel exists, a negative value is taken. Further, by dividing the difference by r, the difference between pixels close to the initial value is weighted. That is, since the contour PA is brighter than the periphery, the pixel value sum takes a positive value if the contour PA is in a direction where r is larger than the initial position, and the pixel value sum is negative if the contour PA is in a direction where r is smaller than the initial position. The positive or negative sign gives a pointer for the direction of movement from the initial position, and its absolute value provides a pointer for the amount of movement.

Therefore, the moving amount (including the direction) r of each pixel constituting the template T 'is defined as in the following equation (8) using a predetermined coefficient b.

[0051]

(Equation 8) The movement amount rn of each pixel n obtained in this way is a movement amount for moving each pixel independently, but as described above, this elastic template T1 ' Since the pixel is constrained to the pixel of the template T1 ', the pixel does not move as it is by the above-described moving amount r, but instead moves to an adjacent pixel (for example, a pixel (n−n
1) and the pixel (n + 1)) and the amount of movement of each pixel (pixel (n−2), pixel (n−1), pixel (n + 1) and pixel (n + 2)) including the pixel adjacent thereto, etc. rk
(K = n, n ± 1,...) Is determined by the following equation (9).

[0052]

(Equation 9) Here, it is preferable that the spring constant ak is set to be large for the pixel of interest n itself, and to be sequentially reduced for the pixels n ± 1,. That is, in equation (9),

(Equation 10) Therefore, the movement amount rn of the target pixel n is a movement amount corresponding to the difference between the movement amount rn of the pixel of interest n and the movement amount rk of the adjacent pixel, and is constrained by the virtual elastic force. Become.

As described above, the elastic template T '
By repeating the operation of moving each pixel little by little, a contour portion of the contour PA of the rib cage pa excluding the lower end portion of the lung field can be accurately detected. The end of the repetition is determined based on whether the total value of the movement amounts is equal to or less than a predetermined threshold value or whether a predetermined number of repetitions has been reached.

Similarly to the above, the inner contour of the lung field detected by the fixed template T2 is set as the initial position, the fixed template T2 is replaced with the elastic template T2 ', and the left contour of the heart detected by the fixed template T3 is set as the initial position. Fixed template T3 is replaced with elastic template T
Instead of 3 ′, candidate points of the lung inner contours PC and PD and the cardiac left contour PE can be detected respectively.

On the other hand, as for the contour PB of the lower end of the lung field, as shown in FIG. 13, the elastic template T1 ″ corresponding to the contour PB ′ of the lower end of the lung field as a reference in the real image plane.
, A second template matching process using the intrathoracic smoothed image P4 as a target image may be performed to detect a contour portion PB corresponding to the lower end of the lung field of the thorax. As described above, since the shape of the contour PB at the lower end of the lung field has not only a large individual difference as described above, but also a large fluctuation in the signal value difference, instead of using the intrathoracic smoothed image P4 as the target image, FIG. It is preferable to perform the second template matching process using the smoothed image P3 as a target image as shown in FIG. This is because the degree of change in the signal value difference can be reduced, and the following by the elastic template T1 ″ becomes relatively easy. In this case, the moving amount is calculated by the following equation (10).
As shown in (11), the setting is made so as to move to the pixel where the change in the pixel value adjacent in the vertical direction (y direction) in FIG. 13 is the largest. Specifically, it is set so as to move toward an edge that changes from a high density lung field to a portion below a low density lung field.

[0056]

[Equation 11] By the above processing, each of the elastic templates T1 'to T3'
And T1 ″, the candidate points of the cardiothoracic contours PA, PB, PC, PD, PE are respectively detected. Then, the elastic templates T1 ′ to T1 ″ are detected.
Each pixel constituting each of 3 ′ and T1 ″ is returned to the real image P1, and in order to further improve the accuracy of each candidate point, each pixel value Pi is set in the peripheral range (initial value ± x) of each candidate point. = ((Px (i ± x), py (i)), i = 0
To N) and the pixel value Pi = ((px (i), p
y (i)), i = 0 to N) (first differential value). Then, a position where the difference is maximum is found and each candidate point is moved (this moved candidate point is referred to as Pni =
((Pxn (i), pin (i)), i = 0 to N).

Here, the candidate points of each contour of the cardiothoracic cage detected as described above do not indicate an appropriate position when the contour of the chest radiation image P1 to be detected is not clear due to the patient or imaging conditions. There are cases. For example, when the contour PE of the heart shadow of the cardiothoracic cavity is detected by the above-described processing, if the contour is not clear, as shown in FIG. In some cases, discontinuous points are detected, and an accurate contour cannot be detected.

Therefore, in the cardiothoracic contour detection method according to the present invention, a more accurate and accurate cardiothoracic contour is detected by performing the following processing on the contour candidate points including these discontinuous points. .

First, regarding the contour candidate points detected by the second template matching process (# 7),
A discontinuous point is detected for each contour (# 8). Here, only the detection processing of the contour PE of the heart shadow will be described as a representative of the detection processing of each contour. In the discontinuous point detection process (# 8), first, the Pni is set as an initial value, and each point of the Pni is connected by a virtual spring (spring coefficient a). Then, the spring force Fx (i) (i = 0 to N) is calculated based on the difference of the x values between the upper and lower points according to the following equation (12).

[0060]

(Equation 12) Then, a point at which the spring force Fx (i) is equal to or less than a predetermined threshold value is detected as a discontinuous point. In FIG. 13A, a black dot is detected as a discontinuous point. When this discontinuous point is detected, each candidate point is subjected to a smoothing process (#
9) is performed. The smoothing process (# 9) corresponds to the above equation (1)
In this process, the x value of each candidate point is gradually corrected based on the value of the spring force Fx (i) obtained in 2). Specifically, the x value pxn of each candidate point is calculated according to the following equation (13).
(I) is calculated, and each candidate point is moved according to this value. Note that this process requires that the x value be modified little by little.
Repeat several times. Alternatively, the processing may be repeated until the sum of the correction amounts becomes equal to or less than a predetermined value.

[0061]

(Equation 13) Here, α is a learning coefficient, and is set to a value between 0 and 1. For example, it is preferably set to 0.1. By the smoothing process (# 9), the candidate points around the discontinuous point move by a moving amount according to each spring force as shown in FIG.

Next, after the smoothing process (# 9) is performed, the re-detection process (# 10) is performed only on the discontinuous points detected by the discontinuous point detection process (# 8). .
The re-detection process (# 10) means that only the discontinuous point and its vicinity (search range shown in FIG.
Is a process of calculating the average primary differential value Dx in accordance with the formula (1), and moving the discontinuous point to the position Pm where Dx becomes the maximum.

[0063]

[Equation 14] The search range (−70 ≦ x ≦
70, −10 ≦ y ≦ 10) indicates that the chest radiographic image P1 is
Pixel density 5 pixels / mm, pixel size 1760 pixels x 17
This is a search range for 60 pixels. In addition, the above Dx
Is smaller than a predetermined threshold, the discontinuous point is not moved. As shown in FIG. 13 (3) by the re-detection process (# 10), the hatched circle point that has deviated from the original contour due to the smoothing process (# 9) is a black circle point that is the position of the original contour again. Will return to. In addition, the discontinuous point (black dot) whose movement is not indicated by the arrow does not move because the contour does not exist in the search range, the Dx becomes equal to or less than a predetermined value.

Then, the discontinuous point detection processing (# 8) is performed again,
The smoothing process (# 9) and the re-detection process (# 10) are repeated, and each candidate point moves as shown in (1) to (3) of FIG. Further, the above processing is repeated until the discontinuous point is not detected in the discontinuous point detection processing (# 8), but in the present embodiment, the discontinuous point is detected in the fifth discontinuous point detection processing (# 8). If no points are found,
(1) to (3) of FIG. 15 and (1) to (3) of FIG.
Are the third discontinuous point detection process (# 8), the smoothing process (# 9) and the re-detection process (# 10), the fourth discontinuous point detection process (# 8), and the smoothing process (# 8). # 9) and the movement of each candidate point by the re-detection process (# 10).

The fifth discontinuous point detection process (#
If no discontinuous point is detected in 8), the same smoothing processing (# 11) as described above is performed on each candidate point, and then interpolation processing between adjacent pixels (such as linear interpolation or spline interpolation) is performed. And are connected by a closed curve as shown in FIG. 18 (# 1
2) By doing so, the contours PA, PB, PC, P
D and PE can be extracted as a contour curve. Also, when a discontinuous point is not detected in the first discontinuous point detection processing (# 8), the same smoothing processing (# 1)
After 1) is performed, an interpolation process (# 12) is performed, and each contour is detected. Note that the above-described discontinuous point detection processing, smoothing processing, and re-detection processing are contour detection processing in which continuity between upper and lower points is taken into consideration. Is also good. In this case, in the discontinuous point detection processing, the spring force is calculated based on the difference between the y values between the left and right points, the smoothing processing is performed in the y direction, and the re-detection processing is performed. The average primary differential value Dy in the y direction may be calculated (in this case, the search range needs to be set to a range different from the above search range).

Next, based on the contours PA, PB, PC, PD, PE (FIG. 18) of the heart and thorax detected as described above, as shown in FIG. 19, the width L1 of the heart Pc and the thorax ( The width L2 of the right lung field Pa and the left lung field Pb) are respectively detected (# 13), and the obtained heart width L1 and thorax width L are obtained.
The ratio L1 / L2 to the ratio 2 is calculated (# 14) and used as the cardiothoracic ratio.

Note that, as shown in FIG.
Is divided into left and right regions on the basis of, the maximum heart widths a and b and the maximum thoracic widths La and Lb are detected in each region, and the maximum heart widths a and b are added, and the heart width L1 and the maximum thoracic width L are added.
a and Lb may be added to obtain the rib cage width L2.

It should be noted that the detected heart-thoracic contours PA, P
The processing (# 13) for obtaining the width L1 of the heart Pc and the width L2 of the thorax based on B, PC, PD, and PE is the maximum width of the heart in the horizontal direction of the image representing the detected outline of the heart and thorax. This is performed by obtaining L1 and the width L2 of the rib cage, respectively.

According to the cardiothoracic contour detecting method and apparatus of the present invention, a cardiothoracic contour candidate is detected, a discontinuous point in the detected cardiothoracic contour candidate is detected, and a discontinuous point is detected. The contour candidate is subjected to a smoothing process, the contour is detected again at the discontinuous point in the smoothed contour candidate, and the detection of the discontinuous point, the smoothing process, and the discontinuity are performed until the discontinuous point is not detected. Since the contour detection is repeated by repeating the contour detection for the point in this order, even if the contour of the cardiothoracic region in the chest image is not clear due to the influence of the patient or imaging conditions, the contour can be restored to the original contour as much as possible. Close contours can be detected with high accuracy.For example, when the cardiothoracic ratio is automatically measured based on the detected cardiothoracic contours, the measurement error can be reduced. Kill.

Further, measurement points for calculating the cardiothoracic ratio using the detected contours, the measured values of the cardiothoracic ratio, and the like may be displayed on a display device. FIG. 20 shows an example of the display. The measurement point of the cardiothoracic ratio is the point K at both ends where the image of the heart width L1 is maximum in the horizontal direction.
2, K3 and the points K1 and K4 at both ends where the image of the rib cage width L2 is maximum in the horizontal direction. The vertical lines at the x-coordinate position of each measurement point are referred to as measurement lines Y1 to Y4. In the display example shown in the figure, the measurement points K1 to K4 and the measurement lines Y1 to Y4, and the measurement values (A of the cardiothoracic ratio calculated based on the measurement points K1 to K4 or the measurement lines Y1 to Y4) are shown.
% = L1 / L2) is displayed. Note that the colors, sizes and line types of the measurement points and measurement lines to be displayed, the colors and fonts of the measurement values, the number of digits, and the like can be set according to the user's preference.

In this display device, the user can move the displayed measurement points and measurement lines to desired positions. For example, when fat is attached to the heart due to cardiac hypertrophy or the like, the detection accuracy of the contour of the heart may be low, so the positions of the measurement points and the measurement lines automatically set based on the detected contour. May slightly deviate from the position of the actual contour on the image. In such a case, the measurement point or the measurement line automatically set and displayed can be moved to a desired position by the user and finely adjusted, so that the measured value of the cardiothoracic ratio is made more accurate. be able to. When the measurement points and the measurement lines are moved, the measurement points and the measurement lines may be dragged using, for example, a mouse. The measurement lines can be moved only in the horizontal direction of the image, and the measurement points can be moved in all directions. Further, the cardiothoracic ratio at the position of the measurement point or the measurement line after the user has moved can be instantly recalculated, and the measured value can be displayed.

When the size and resolution of the image to be processed are different from those of the image which is assumed to be the image to be processed in advance, the setting of the measurement points fails and the order of the four measurement lines is completely reversed. And so on. In such a case, a message indicating that the setting of the measurement point has failed may be displayed, and a measurement line may be displayed at a predetermined position set in advance so that fine adjustment can be easily performed later.
Specifically, when the order of the four measurement lines is completely reversed, or when the maximum diameter of the heart width L1 is larger than the maximum diameter of the thorax width L2, the cardiothoracic ratio Since the measured value of is a negative value or 1 (100%) or more, it is used as an index to determine whether the setting of the measuring point has failed,
If the setting fails, the measurement line Y1 is placed at a position 1/5 × X (the horizontal size of the image is X) from the left end of the image, and the measurement line Y2 is placed at 2/5 × from the left end of the image. At the position of X, the measurement line Y3 is placed at the position of 3/5 × X from the left end of the image,
The measurement line Y4 is displayed at a position of 4/5 × X from the left end of the image. In this way, even if the setting of the measurement point fails, the user can finely adjust the measurement line on the display device, and thus can obtain a substantially accurate cardiothoracic ratio by a simple operation.

Further, a storage device is provided for storing the measured values of the calculated cardiothoracic ratio in association with the image, and when displaying the image on the display device, the measured values stored in the storage device are called and displayed. It can also be done. At this time,
The final measurement value obtained by the user fine-adjusting the measurement points and measurement lines on the image displayed on the display device, together with the initial value of the measurement value automatically calculated based on the detected contour. Also save. When you call the image,
The measurement point, the measurement line and the measurement value after the fine adjustment are displayed, and when the reset button is pressed, the initial value of the measurement value automatically calculated first and the measurement point and the measurement line at that time are displayed. By doing so, it is not necessary to recalculate the image for which the measurement value has been calculated once each time it is displayed, so that the processing time can be reduced.

The method and apparatus for detecting a cardiothoracic contour according to the present invention can detect not only the contour of the cardiothoracic contour but also the contour detection for measuring the size and curvature of anatomical features in other medical images. It can also be applied to the detection of the contour of the trachea. Further, the measurement values that can be calculated by the display device are not limited to the cardiothoracic ratio, and the image to be processed is not limited to the chest image.

The processing speed may be increased by reducing the size of the image used for contour detection and then detecting the contour within the allowable range of the measurement accuracy.

[Brief description of the drawings]

FIG. 1 is a schematic configuration diagram of an embodiment of a cardiothoracic contour detection device that performs a cardiothoracic contour detection method of the present invention.

FIG. 2 is a flowchart showing processing of an embodiment of a cardiothoracic contour detection method of the present invention.

FIG. 3 is a diagram showing a chest radiographic image to be processed by the cardiothoracic contour detection method of the embodiment shown in FIG. 2;

FIG. 4 is a diagram showing an example of a contour detection mask for each extension direction (a) and an extension direction (b) of a detectable contour.

FIG. 5 is a diagram illustrating an example of an edge detection mask for each extension direction (a) and an extension direction of a detectable edge and a density gradient direction (b).

FIG. 6 is a diagram showing a rough outline image P2 obtained by a rough outline detection process (# 1) and a smoothed image P3 obtained by a smoothed image process (# 2).

FIG. 7 is a diagram showing a rib cage internal smoothed image P4 obtained by rib cage internal smoothing processing (# 3).

FIG. 8 is a diagram showing a polar coordinate conversion image P5 obtained by polar coordinate conversion (# 5).

FIG. 9 is a view showing an outline PB of a rib cage as a reference on a polar coordinate plane;

FIG. 10 shows the contours of the rib cage PA ′, PB ′, P serving as a reference.
Fixed template T based on C ', PD', PE '
Diagram showing 1, T2 and T3 respectively

FIG. 11 shows an elastic template T on a polar coordinate plane.
Figure showing 1 '

FIG. 12 is a diagram illustrating a process in which an elastic template T1 ′ follows a detailed shape of a contour PA.

FIG. 13 shows an elastic template T on a real image plane.
Figure showing 1 "

FIG. 14 is a diagram illustrating a first discontinuous point detection process (# 8), a smoothing process (# 9), and a re-detection process (#) according to the present embodiment.
Figure showing 10)

FIG. 15 shows the second discontinuous point detection processing (# 8), smoothing processing (# 9), and re-detection processing (#) in the embodiment.
Figure showing 10)

FIG. 16 shows a third discontinuous point detection process (# 8), a smoothing process (# 9), and a re-detection process (#) in the embodiment.
Figure showing 10)

FIG. 17 shows the fourth discontinuous point detection processing (# 8), smoothing processing (# 9), and re-detection processing (#) in this embodiment.
Figure showing 10)

FIG. 18 shows a contour P after interpolation processing on a real image plane.
Diagram showing A, PB, PC, PD, PE

FIG. 19 is a diagram showing a heart width L1 and a rib cage width L2.

FIG. 20 is a diagram illustrating a display example of measurement points, measurement lines, and measurement values set based on a detected contour.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 10 Cardiothoracic contour detection device 11 Smoothed image processing means 12 Polar coordinate conversion means 13 Matching processing means 14 Discontinuity point detection means 15 Smoothing processing means 16 Redetection means 17 Interpolation processing means 20 Measurement means 30 Contour candidate detection means

──────────────────────────────────────────────────続 き Continued on the front page (51) Int.Cl. ⁷ Identification symbol FI Theme coat ゛ (Reference) G01R 33/32 G01N 24/02 520Y F term (Reference) 4C093 AA26 CA50 DA02 DA03 FD04 FD07 FD13 FF06 FF12 FF16 FF22 FF37 FF50 4C096 AB38 AB44 AC04 AD14 DC09 DC19 5B057 AA07 BA03 CE05 CE06 DA06 DB02 DB09 DC03 DC05 DC09 DC16 5L096 AA06 BA06 BA13 DA01 EA06 EA28 EA33 FA06 FA34 FA54 FA62 FA64 FA67 FA69 GA10 JA09

Claims (8)

[Claims] 1. An image outline detection method for detecting an outline of an image, comprising detecting an outline candidate of the image, detecting a discontinuous point in the outline candidate of the detected image, and detecting the discontinuous point. Performing a smoothing process on the contour candidate, performing the contour detection again on the discontinuous point in the contour candidate subjected to the smoothing process, detecting the discontinuous point, the smoothing process and the discontinuity An image contour detection method, wherein the contour detection is performed by repeating the point contour detection once or more in this order. 2. The image contour detecting method according to claim 1, wherein the contour is detected by performing the repetition until the discontinuous point is no longer detected. 3. The method according to claim 1, wherein the image is a radiation image. 4. The cardiothoracic image according to claim 1, wherein the image is a cardiothoracic image.
4. The image contour detection method according to any one of claims 1 to 3. 5. An image contour detection device for detecting a contour of an image, wherein: a contour candidate detecting means for detecting a contour candidate of the image; and a discontinuous point detection for detecting a discontinuous point in the contour candidate of the detected image. Means, a smoothing processing means for performing a smoothing process on the contour candidate when the discontinuous point is detected by the discontinuous point detecting means, and the contour subjected to the smoothing process by the smoothing processing means. Re-detection means for detecting the contour again with respect to the discontinuous point in the candidate, detection of the discontinuous point by the discontinuous point detection means, the smoothing processing by the smoothing processing means and the re-detection The contour detection is performed by repeating the contour detection of the discontinuous point by means at least once in this order. 6. An image contour detecting apparatus according to claim 5, wherein said contour is detected by performing said repetition until said discontinuous point is no longer detected by said discontinuous point detecting means. 7. The image contour detection device according to claim 5, wherein the image is a radiation image. 8. The cardiothoracic image, wherein the image is a cardiothoracic image, and the contour is the cardiothoracic image.
8. The image contour detection device according to any one of claims 1 to 7.