JP2007105196A - Image processor, image processing method and its program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an image processor capable of improving the detection performance of abnormal shadows, an image processing method and its program. <P>SOLUTION: An out-of-rib area excluding a rib area is extracted from a lung field area on a chest photography image, and a soft part estimated image for which pixel value components contributing to a soft part inside the lung field area on the chest photography image are estimated is generated on the basis of the image of the out-of-rib area. The generated soft part estimated image is removed from the chest photography image, a bone part estimated image composed of the pixel value components contributing to the ribs is generated, and the rib area is detected from the bone part estimated image. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention relates to an image processing device, an image processing method, and a program for detecting a rib region of a subject imaged in a chest image.

  2. Description of the Related Art Conventionally, in the medical field, based on a digital medical image, a diagnostic support device (CAD: Computer Aided Diagnose) that automatically detects an abnormal shadow in the image using a computer has been provided. There is a chest CAD that detects a mass shadow in an image of the chest based on the digital chest X-ray image.

  The chest X-ray image includes a so-called “background image” which is an image in which structures having various anatomical features such as ribs and clavicles appear. This background image is used to detect abnormal shadows. It becomes an obstacle and causes a decrease in detection ability. Thus, a method has been proposed in which such background images are removed by filtering processing and chest CAD processing is performed (for example, Patent Document 1).

Further, the anatomical structure of the chest is complicated, and the chest CAD using the above filtering process has a problem that the background feature image cannot be sufficiently removed and the detection performance of the abnormal shadow is not improved. . Therefore, a method of generating an artificial image and removing an anatomical structure such as a bone as a background image has been proposed (for example, Patent Document 2).
JP-A-6-121792 JP-A-2005-020338

  However, when trying to extract the rib shape of the subject from the chest image, it is difficult to accurately grasp the rib shape because of being affected by structures other than the ribs such as the heart and blood vessels. Therefore, it has been an obstacle to detect abnormal shadows in the chest CAD process.

  In view of the circumstances described above, an object of the present invention is to provide an image processing apparatus, an image processing method, and a program thereof that can further improve the detection performance of abnormal shadows.

An image processing apparatus according to the present invention includes a chest radiograph storage unit that stores a chest radiograph obtained by simple X-ray imaging of a subject's chest,
Rib area estimation means for estimating a rib area from the chest radiograph,
Extra-radial region extraction means for extracting an extra-costal region excluding the rib region from a lung field region on the chest radiograph,
Soft part image estimation means for generating a soft part estimation image that estimates a pixel value component that contributes to a soft part in the lung field area on the chest radiograph based on the image of the extracostal area;
Bone estimated image generating means for removing the soft part estimated image from the chest captured image and generating a bone estimated image composed of pixel value components contributing to the ribs among the pixel values of the chest captured image;
And a rib area detecting means for detecting a rib area from the bone part estimation image.

The image processing method of the present invention includes a chest radiograph storage step for storing a chest radiograph obtained by simple X-ray imaging of a subject's chest in a chest radiograph storage unit;
A rib region estimation step for estimating a rib region from the chest radiograph,
An extra-rib area extraction step for extracting an extra-rib area excluding the rib area from a lung field area on the chest radiograph; and
A soft part image estimation step for generating a soft part estimation image that estimates a pixel value component that contributes to a soft part in the lung field area on the chest radiograph based on the image of the extracostal area;
A bone estimated image generation step for generating a bone estimated image composed of pixel value components contributing to the ribs among the pixel values of the chest captured image by removing the soft portion estimated image from the chest captured image;
A rib region detecting step of detecting a rib region from the bone part estimation image.

Further, the program of the present invention provides a computer,
A chest radiograph storage means for storing a chest radiograph obtained by simple X-ray imaging of the subject's chest;
Rib area estimation means for estimating a rib area from the chest radiograph,
Extra-radial region extraction means for extracting an extra-costal region excluding the rib region from a lung field region on the chest radiograph,
Soft part image estimation means for generating a soft part estimation image that estimates a pixel value component that contributes to a soft part in the lung field area on the chest radiograph based on the image of the extracostal area;
Bone estimated image generating means for removing the soft part estimated image from the chest captured image and generating a bone estimated image composed of pixel value components contributing to the ribs among the pixel values of the chest captured image;
It is made to function as a rib area | region detection means to detect a rib area | region from the said bone part estimation image.

  The “radial area” refers to an area where the ribs are imaged on the chest radiographing area, and the “external rib area” refers to an area obtained by removing the “radial area” from the lung field area on the chest radiographed image.

  The pixel value on the “chest image” is a pixel value that represents the density according to the anatomical structure such as the rib or heart or lung field being photographed, and the soft part such as the heart or lung field overlaps the rib. In other words, the pixel value represents the density influenced by the density due to the soft part and the density due to the rib.

  “A pixel value component that contributes to the soft part in the lung region on the chest radiograph” refers to the anatomical structure of the soft part excluding the influence of the ribs among the pixel values in the lung field on the chest radiograph The pixel value component that contributes to

  “The pixel value component that contributes to the ribs among the pixel values of the chest radiograph image” refers to “the pixel value of the pixels constituting the chest radiograph image” that contributes to the rib excluding the influence of anatomical structures other than the ribs This is a pixel value component.

  Further, it is desirable that the soft part image estimation means estimates based on the result of analyzing the soft part pixel value of the chest radiograph obtained by photographing a large number of subjects using a statistical analysis means.

The analysis means is principal component analysis,
If the main component image of the soft part is obtained as a result,
The pre-soft part image estimation means may generate the soft part estimation image in which the pixel value of the soft part of the predetermined subject is estimated by weighted addition of the obtained principal component images. .

  A “principal component image” refers to an image representing a principal component obtained as a result of principal component analysis of pixel values of a soft part.

  According to the present invention, the soft part image is estimated from the extracostal area excluding the rib area on the chest radiographed image, the soft part image estimated from the chest radiograph image is removed, and the bone part estimation image is generated, and the influence of the soft part is removed. By detecting the rib region from the estimated bone part estimation image, the rib region can be detected with high accuracy.

  In addition, by estimating the soft part image from the extracostal area excluding the rib area by principal component analysis, it becomes possible to estimate the soft part image that was taken over the rib, and the bone part excluding the influence of the soft part It is possible to accurately estimate the rib area from the image.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings. FIG. 1 is a diagram showing a schematic configuration of an image processing apparatus according to the present invention.

  As shown in FIG. 1, the image processing apparatus 1 includes a chest photographed image storage unit 10 that stores a chest photographed image 100 obtained by performing simple X-ray photography of a subject's chest, and a rib that estimates a rib region from the chest photographed image. Based on the region estimation means 20, the extracostal region extraction means 30 for extracting the extracostal region from the lung field region on the chest radiograph image, and the extracostal region image, A soft part image estimating means 40 for generating a soft part estimated image in which a pixel value component contributing to a soft part in the lung field region is estimated; and removing the soft part estimated image from the chest photographed image; Of these, a bone portion estimated image generating means 50 for generating a bone portion estimated image composed of pixel value components contributing to the ribs and a rib region detecting means 60 for detecting a rib region from the bone portion estimated image are provided.

  The chest image 100 is an image obtained by performing simple X-ray imaging of a subject using a CR device or the like. In an image obtained by simple X-ray imaging, each tissue appears with a pixel value having a density corresponding to the X-ray transmittance (or absorption rate) of the anatomical tissue in the subject's chest. Since ribs and the like have a high X-ray absorption rate, the site where the ribs are present appears white on the chest radiograph image 100, but the concentration appearing on the chest radiograph image 100 is the absorption rate of all organs through which X-rays have passed. Therefore, even at the location where the rib is imaged, the density is affected by other organs such as the lung field and heart other than the rib imaged over the rib. Therefore, even if the ribs have the same thickness, they appear on the chest radiograph 100 at different concentrations depending on whether the organ under the ribs is the lung field or the heart.

  Here, the flow of processing for detecting a rib region by removing the influence of an organ under the rib from the chest image 100 of the subject to be inspected by the image processing apparatus 1 will be described with reference to the flowchart of FIG.

  First, the bone region estimation means 20 generates an edge image from the chest radiograph image using the edge extraction filter from the chest radiograph image 100 of the subject stored in the chest radiograph image storage means 10, and detects a parabola from the edge image. A method of detecting rib shape by detecting a parabola that seems to be ribs using Hough transform (eg Peter de Souza, “Automatic Rib Detection in Chest Radiographs”, Computer Vision, Graphics and image Processing 23, 129-161 (1983 )), The rib area is estimated from the rib shape (S100). The rib region estimated from the chest image 100 in this manner is affected by the structure of soft parts such as the heart and blood vessels on the chest image 100, and thus has a low accuracy.

  Next, a soft part estimation image in which the soft part is estimated is generated from the extracostal area obtained by removing the rib area from the lung field area on the chest photographed image 100, and the soft part estimation image is deleted from the chest photographed image 100. An accurate rib area is detected from the image.

  Further, the extra-radial region extraction means 30 detects the lung field region from the chest image 100 (specifically, for example, as disclosed in Japanese Patent Laid-Open No. 2003-6661 proposed by the present applicant). Then, polar coordinates are converted with reference to the approximate center point of the chest image of the chest image, and template matching is performed on the polar coordinate plane using a template that is approximately similar to the average cardiothoracic contour. A method of automatically extracting the outline of the ribs can be used.) An extra-radial region excluding the rib region estimated by the bone region estimating means 20 is extracted from the detected lung field region (S101). When the rib region is cut out from the lung field of the chest radiograph 100 as shown in FIG. 4A, the image 110 in the extra-rib region becomes a soft part image as shown in FIG. 4B.

  The soft part image estimation means 40 estimates an image of the entire soft part from the extracted image 110 of the extracostal region, and artificially generates a soft part estimated image (S102). Specifically, the soft part image estimation unit 40 generates a soft part estimation image using a result of statistical analysis of a large number of soft part images obtained by energy subtraction. Specifically, a principal component analysis (PCA: Principal Component Analysis) process is performed on a large number of soft part images obtained by energy subtraction to obtain a principal part (vector component) of the soft part image, and the principal component is used. Thus, the soft part estimation image can be artificially reproduced.

  Therefore, first, the soft part image as shown in FIG. 2 is transformed into a standardized shape such as a rectangle.

When deforming, when the coordinates before deformation are B (x _B , y _B ) and the coordinates of the corresponding positions after deformation are A (x _A , y _A ), the highest and lowest parts of the lung field _Are converted into y coordinates of A corresponding to the y coordinates of B as shown in equation (1), respectively _{, as yB, up} , yB _{, down} , yA _{, up} , yA _{, down} .

Further, when the position of the lung left in _{y B x B, left, x} B, and _right, to the position of the lung left in _{y A x A, left, x} A, and _right, lung right in _{y B} Is deformed as shown in Equation (2) so that the position of A coincides with the left and right positions of the lung field in _A.

By performing principal component analysis on the average image (average soft part density image) X _{ave of} the soft part image transformed into a rectangle in this way (see FIG. 2 (a)) and the difference image between the soft part image and the average soft part density image The soft part principal component density image X _i (i = 1, 2, 3,..., N), which is the obtained first to n-th principal components (principal component images), is obtained (see FIG. 5B). In this case, the seventh principal component is obtained), and the soft part estimated image X is obtained by the weighted sum of the average soft part density image X _ave and the soft part principal component density image X _i (i = 1, 2,..., N). It can be expressed as equation (3).
X = X _ave + Σ _i a _i · X _i (3)
X: Vector with the pixel values of the pixels on the soft part image as components
X _ave : A vector whose component is the pixel value of the pixel on the average soft part density image
X _i : principal component vector representing the i-th soft principal component density image
a _i : Weight coefficient for the i-th principal component vector

  Therefore, when estimating the soft part image of the chest image 100 of the subject to be inspected, the weighting coefficient is determined so as to match the pixel value of the region outside the rib, using the relationship of the above equation (3), A soft part estimation image of the subject to be inspected as shown in FIG. 5 is generated.

  Alternatively, as shown in FIG. 3, the average shape of the soft part is normalized to generate an average soft part density image as shown in FIG. 3 (a) and a main component density image as shown in FIG. 3 (b). Then, the weight coefficient is determined from the chest image 100 of the subject so as to match the density of the soft part other than the ribs, and the soft part estimation image X (for example, an image as shown in FIG. 5) of the subject is generated. Also good.

  Next, the bone estimated image generating means 50 removes the density contributing to the soft part from the chest photographed image 100 by subtracting the soft part estimated image from the chest photographed image 100, and the rib of the pixel value of the chest photographed image 100 is removed. A bone estimation image as shown in FIG. 6 composed of pixel value components contributing to the above is generated (S103).

  Therefore, the rib area detection unit 60 recognizes the rib shape from the bone part estimation image again by the edge extraction filter or the Hough transform for detecting the parabola as in the bone region estimation unit 20 and detects the rib area from the rib shape. (S104).

  Alternatively, the rib region detection means 60 extracts the rib shape from the chest image obtained by photographing the subject, and performs the principal component analysis on the rib shapes of the plurality of chest images obtained by photographing. A number of rib model shapes M are generated, and a rib model shape M most similar to the extracted subject rib shape is searched from among the generated plurality of rib model shapes M, and the searched rib model shapes M are searched. It is also possible to estimate the rib shape of the subject and detect the rib area based on the rib shape.

  Specifically, the rib shape is subjected to principal component analysis as follows to generate a rib model shape.

First, as shown in FIG. 7A, the rib shape (FIG. 7B) is extracted from a chest image S representing many normal breasts in advance. The overall shape of the rib is expressed using a number of points on the rib that form the outline of the rib (feature points, black circles in FIG. 5B). For the shape vector X representing the overall shape of the rib, for example, 100 points on the rib are extracted using a technique as described in Non-Patent Document 1, and the coordinates (x , Y) can be expressed as the equation (4) with a vector in which the frames are sequentially arranged as a rib shape vector.

Therefore, the shape vector represented by the above equation (4) is extracted from a collection of a large number of chest photographed images S representing normal breasts photographed in the past, and the principal component analysis is performed on the shape vector. By obtaining the component vector, the rib shape of the stored chest image S can be represented by a small number of independent vector components. Specifically, the first to n-th principal component vectors Ai (i = 1, 2,..., N) are analyzed by principal component analysis of the difference vector between the average shape and the rib shape of the stored chest radiograph S. ), The rib model shape M is expressed as in equation (5) using the average shape vector of the ribs and the principal component vector Ai (see FIG. 7C).

  A plurality of rib model shapes M are generated by changing the coefficient of equation (5), and the rib shape closest to the photographed rib shape is selected from among the plurality of rib model shapes M, and this rib shape is Based on this, the rib area is detected.

  The rib area detected from the chest image obtained by removing the soft part image in this way is more accurate than the rib area estimated by the rib area estimation means 20.

  As described above in detail, the rib area can be accurately detected by detecting the rib area by removing the background image of the soft tissue imaged. If the rib image is removed from the original image using the rib image obtained in this way, the soft part image can be accurately extracted, and an abnormal shadow due to cancer or the like can be accurately detected. become.

  Also, the rib area estimation means 20 may estimate the rib area using the principal component analysis method described in the rib area detection means 60.

  In addition, by installing a program including the above-described units in a computer, the computer can be operated as an image processing apparatus.

The figure which shows schematic structure of the image processing apparatus of this invention Example of results of principal component analysis of soft part images (Part 1) Example of results of principal component analysis of soft part image (Part 2) Chest image and example of extracostal region Example of soft part estimation image Example of bone estimation image Diagram for explaining principal component analysis of rib shape Flowchart for explaining the flow of processing of the image processing apparatus

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Image processing apparatus 10 Chest radiograph image storage means 20 Radius area estimation means 30 Extra rib area extraction means 40 Soft part image estimation means 50 Bone estimation image generation means 60 Radius area detection means 100 Chest radiograph M M rib model shape

Claims (5)

A chest radiograph storage means for storing a chest radiograph obtained by simple X-ray imaging of the subject's chest;
Rib area estimation means for estimating a rib area from the chest radiograph,
Extra-radial region extraction means for extracting an extra-costal region excluding the rib region from a lung field region on the chest radiograph,
Soft part image estimation means for generating a soft part estimation image that estimates a pixel value component that contributes to a soft part in the lung field area on the chest radiograph based on the image of the extracostal area;
Bone estimated image generating means for removing the soft part estimated image from the chest captured image and generating a bone estimated image composed of pixel value components contributing to the ribs among the pixel values of the chest captured image;
An image processing apparatus comprising: a rib area detecting means for detecting a rib area from the bone part estimation image.   The soft part image estimation means estimates the pixel value of the soft part of the chest image obtained by photographing a large number of subjects based on the result of analysis using a statistical analysis means. Item 6. The image processing apparatus according to Item 1. The analysis means is principal component analysis;
As a result, a principal component image of the soft part is obtained,
3. The image according to claim 2, wherein the pre-soft part image estimation unit generates a soft part estimation image in which a pixel value of a soft part of the predetermined subject is estimated by weighted addition of the obtained principal component images. Processing equipment. A chest-captured image storage step for storing a chest-captured image obtained by simple X-ray imaging of the subject's chest in a chest-captured image storage means;
A rib region estimation step for estimating a rib region from the chest radiograph,
An extra-rib area extraction step for extracting an extra-rib area excluding the rib area from a lung field area on the chest radiograph; and
A soft part image estimation step for generating a soft part estimation image that estimates a pixel value component that contributes to a soft part in the lung field area on the chest radiograph based on the image of the extracostal area;
A bone estimated image generation step for generating a bone estimated image composed of pixel value components contributing to the ribs among the pixel values of the chest captured image by removing the soft portion estimated image from the chest captured image;
An image processing method comprising: a rib area detecting step of detecting a rib area from the bone portion estimation image. Computer
A chest radiograph storage means for storing a chest radiograph obtained by simple X-ray imaging of the subject's chest;
Rib area estimation means for estimating a rib area from the chest radiograph,
Extra-radial region extraction means for extracting an extra-costal region excluding the rib region from a lung field region on the chest radiograph,
Soft part image estimation means for generating a soft part estimation image that estimates a pixel value component that contributes to a soft part in the lung field area on the chest radiograph based on the image of the extracostal area;
Bone estimated image generating means for removing the soft part estimated image from the chest photographed image and generating a bone estimated image composed of pixel value components contributing to the ribs among the pixel values of the chest photographed image;
A program that functions as a rib area detecting unit that detects a rib area from the bone part estimation image.