JP2006175057A - Image processing apparatus, and its program and method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To estimate a rib profile by taking the 3D profile change of the rib of an object into consideration. <P>SOLUTION: Standard 2D rib profile obtained using a statistical method from a plurality of rib shapes of 2D chest images acquired by photographing in advance and a plurality of shape deforming vectors A deforming the standard 2D rib profile S2D are stored, and the standard 3D rib shape S3D corresponding to the standard 2D rib shape S2D is stored. The parameter of the shape deforming A deforming the standard 2D rib shape S2D so that it corresponds to the 2D object rib shape is acquired based on the 2D object rib shape photographed in the object chest image P acquired by photographing the object and the standard 2D rib shape S2D. Further, the 3D object rib shape at the time of photographing the object is estimated by deforming the standard 3D rib shape S3D according to the parameter. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to an image processing apparatus, method, and program for estimating a rib shape of a subject imaged in a chest image.

  Conventionally, in the medical field, a digital medical image (hereinafter simply referred to as a medical image) is image-analyzed using a computer for the purpose of reducing the load on an image reader such as a doctor in image diagnosis. (Computer Aided Diagnosis) processing is performed, and as such CAD processing, for example, abnormal shadow detection processing for automatically detecting an abnormal shadow in a medical image is known.

  The abnormal shadow detection process is a process of calculating an index value or the like representing the feature of the abnormal shadow on the image based on the medical image and detecting the abnormal shadow based on the index value or the like. Therefore, an index value similar to an abnormal shadow may be calculated not only in an abnormal shadow but also in an area having image characteristics similar to the abnormal shadow. It may be detected by mistake. For example, in the case of diagnosis in a chest image, there are many cases where a place where a normal structure such as a rib overlaps is erroneously detected as a nodule shadow or a tumor shadow. The presence of such a misdetected normal location, so-called FP (False Positive), imposes a burden on the interpreter and may deteriorate the accuracy of diagnosis. Therefore, reduction of FP is strongly desired.

  For this reason, as described above, when a normal structure constituting a subject in a medical image used for image diagnosis affects the accuracy of abnormal shadow detection, such a structure is detected and the structure is detected. It is necessary to change the abnormal shadow detection process on the object, determine the abnormal shadow candidate detected on the structure with another index, or remove the structure from the medical image. Become.

  Therefore, various methods for detecting the structure of such a structure or generating an image from which the structure is removed have been proposed. For example, in order to observe the temporal change of the affected area in the radiographic image, create a difference image between time-series radiographic images between the images taken of the same subject in the past and the present, and highlight the part that changes with time A time-lapse subtraction (subtraction process) that makes it easy to perform is proposed, and a diagnosis support is proposed by observing the created difference image simultaneously with the time-series radiation image (for example, Non-Patent Document 1).

In addition, when subtraction is performed between images taken in the past and the present to emphasize and detect areas that are suspected of being abnormal in the chest image, the rib shape of the past image is used to improve the accuracy of the subtraction process. A method has been proposed in which warping is performed so as to approximate the rib shape of the current image, and the warped image and the current image are subtracted (for example, Patent Document 1).
A. Kano, K. Doi, H. MacMahon, D. Hassell, MLGinger “Digital image subtraction of temporally sequential chest images for interval charge”, Med. Phys. 21 (3), March 1994, 453-461 [1] JP-A-7-37074

  However, as shown in FIG. 11, when the posture of the subject is different between the past and the current photographing (see FIG. 11A), the photographed chest image is three-dimensional even if the rib shape of the same subject is present. Since the position of the anterior and posterior ribs changes depending on the posture, the positional relationship between the anterior and posterior ribs does not match between the previous and current images (see FIG. (See (b)). For this reason, there is a problem that even if subtraction is performed between the past and current images over time, the ribs cannot be erased and artifacts due to differences occur.

  Therefore, in Patent Document 1, subtraction is performed by warping the two-dimensional rib shape in order to reduce artifacts due to deviation of the rib shape, but since the three-dimensional change of the rib shape is not considered, If the subject's posture at the time of shooting changes significantly between the present and the present, it is expected that an artifact will also occur.

  In view of the above circumstances, an object of the present invention is to provide an image processing apparatus, method and program for estimating the shape of a rib in consideration of a three-dimensional shape change of the rib of a subject. .

The image processing apparatus of the present invention transforms a two-dimensional standard rib shape obtained by using a statistical method and a two-dimensional standard rib shape from a rib shape of a plurality of two-dimensional chest images obtained by photographing in advance. Shape deformation data storage means for storing a plurality of shape deformation vectors,
Three-dimensional standard rib shape storage means for storing a three-dimensional standard rib shape corresponding to the two-dimensional standard rib shape;
Based on the two-dimensional subject rib shape captured in the subject chest image obtained by photographing the subject and the two-dimensional standard rib shape, the two-dimensional standard rib shape is changed to the two-dimensional subject rib shape. Parameter acquisition means for acquiring parameters of the respective shape deformation vectors that are deformed so as to coincide;
3D rib shape estimating means for deforming the 3D standard rib shape according to the parameter and estimating the 3D subject rib shape at the time of photographing the subject is provided. .

The program of the present invention is a computer,
Based on the two-dimensional standard rib shape obtained from the rib shapes of a plurality of two-dimensional chest images obtained by photographing in advance and the two-dimensional subject rib shape photographed on the subject chest image obtained by photographing the subject. The two-dimensional standard rib shape obtained by using a statistical method from the rib shapes of the plurality of two-dimensional chest images deforming the two-dimensional standard rib shape to match the two-dimensional subject rib shape. Parameter acquisition means for acquiring parameters of a plurality of shape deformation vectors for deforming
The three-dimensional standard rib shape corresponding to the two-dimensional standard rib shape is deformed in accordance with the parameter, and the three-dimensional rib shape estimating means for estimating the three-dimensional rib shape at the time of photographing the subject is caused to function. It is characterized by this.

The image processing method of the present invention transforms a two-dimensional standard rib shape obtained using a statistical method and a two-dimensional standard rib shape from a rib shape of a plurality of two-dimensional chest images obtained by photographing in advance. A shape deformation data storage step for storing a plurality of shape deformation vectors to be performed;
A three-dimensional standard rib shape storing step for storing a three-dimensional standard rib shape corresponding to the two-dimensional standard rib shape;
Based on the two-dimensional subject rib shape captured in the subject chest image obtained by photographing the subject and the two-dimensional standard rib shape, the two-dimensional standard rib shape is changed to the two-dimensional subject rib shape. A parameter acquisition step of acquiring parameters of each shape deformation vector that is deformed so as to match;
A three-dimensional rib shape estimation step for deforming the three-dimensional standard rib shape in accordance with the parameter and estimating a three-dimensional subject rib shape at the time of photographing the subject. .

  The “two-dimensional chest image” refers to an image obtained by photographing the chest and projecting it two-dimensionally from a simple X imaging device or a CR (Computed Radiography) device.

  “Two-dimensional standard rib shape” refers to a standard two-dimensional rib shape obtained from a two-dimensional rib shape obtained from a number of chest images taken with a simple X imaging device or a CR device. Specifically, for example, an average shape of a two-dimensional rib shape or a shape having a center value (median) can be used.

  The “three-dimensional rib shape” can be obtained from a tomographic image photographed by, for example, a CT (Computed Tomography) apparatus or an MRI (Magnetic Resonance Imaging) apparatus. The “three-dimensional standard rib shape” refers to a standard three-dimensional rib shape. Specifically, for example, an average shape or center value (median value) of a three-dimensional rib shape obtained from a tomographic image is used. ) Can be used.

  The “three-dimensional standard rib shape corresponding to the two-dimensional standard rib shape” includes a two-dimensional standard rib shape obtained from a two-dimensional chest image taken with a simple X-ray imaging device or a CR device, and a CT device. Or a three-dimensional standard rib shape obtained from a tomographic image taken by an MRI apparatus or the like. For example, the “two-dimensional standard rib shape” is an average shape of a two-dimensional rib obtained from a two-dimensional chest image taken in the past, and the “three-dimensional standard rib shape” is obtained from a tomographic image taken in the past. When the average shape of the three-dimensional ribs is used, the two-dimensional average shape and the three-dimensional average shape obtained from a large number of past chest images and tomographic images are statistically three-dimensional averages. Since the projection of the shape in two dimensions is estimated to be approximately the same as the two-dimensional average shape, the past so that it matches the “two-dimensional standard rib shape” obtained from the past chest image. The “three-dimensional standard rib shape corresponding to the two-dimensional standard rib shape” can be obtained by projecting the two-dimensional “three-dimensional standard rib shape” obtained from the tomographic images.

Further, the three-dimensional rib shape estimation means performs the first 3D on the target chest image of two different target chest images obtained by photographing a predetermined subject to be subjected to the difference processing. First subject rib shape acquisition means for acquiring a three-dimensional subject rib shape;
Second subject rib shape acquisition means for acquiring a second three-dimensional subject rib shape by the three-dimensional rib shape estimation means for the other target chest image of the two target chest images;
Based on the first and second three-dimensional subject rib shapes, deform at least one of the two target chest images so that the first and second three-dimensional subject rib shapes match. It may be provided with deformation means.

In the program, a first three-dimensional rib shape estimation unit applies a first three-dimensional rib shape estimation unit to one target chest image of two different target chest images obtained by photographing a predetermined subject to be subjected to a difference process. First subject rib shape acquisition means for acquiring a subject rib shape;
Second subject rib shape acquisition means for acquiring a second three-dimensional subject rib shape by the three-dimensional rib shape estimation means for the other target chest image of the two target chest images;
Based on the first and second three-dimensional subject rib shapes, deform at least one of the two target chest images so that the first and second three-dimensional subject rib shapes match. It may further include a deformation means.

Alternatively, in the image processing method, the first three-dimensional shape is estimated by performing the three-dimensional rib shape estimation step on one target chest image among two different target chest images obtained by photographing a predetermined subject to be subjected to difference processing. A first subject rib shape acquisition step of acquiring the subject rib shape;
A second subject rib shape acquisition step of acquiring a second three-dimensional subject rib shape by the three-dimensional rib shape estimation step with respect to the other target chest image of the two target chest images;
Based on the first and second three-dimensional subject rib shapes, deform at least one of the two target chest images so that the first and second three-dimensional subject rib shapes match. A deformation step may be further provided.

  “Two different chest images taken from the subject” are chest images taken from the same subject, which are two chest images to be subjected to differential processing. For example, 2 taken in the past and the present This is a chest image.

  “Deform at least one of the two target chest images so that the first and second three-dimensional subject rib shapes match” means that the subject captured in the two target chest images For example, if the subject has changed its posture, and one subject's rib shape has been photographed in a forward-bend posture than the other subject's rib shape. This means that the rib shape is deformed so as to have the same degree of forward bending. “Deformation” means that after a three-dimensional rib shape is deformed, a two-dimensional chest image obtained by projecting the three-dimensional rib shape is also changed according to the three-dimensional deformation. This is to transform a two-dimensional image. “Deform at least one” means that one of the first and second three-dimensional subject rib shapes is deformed and matched to the other, and both are deformed to have the same shape. Such modifications are included.

  Further, it is desirable that the deforming means deforms by moving the texture on the rib along with the deformation of the rib shape.

  “Move the texture” means to move the texture of the image in accordance with the three-dimensional deformation, and includes to move the texture on the two-dimensional image corresponding to the three-dimensional deformation.

  Preferably, the three-dimensional rib shape estimation means stores in advance the correspondence between the amount of deformation of the three-dimensional standard rib shape and the parameters of the shape deformation vector.

  Further, it is desirable that the shape deformation vector is a principal component vector obtained by principal component analysis of rib shapes of a plurality of two-dimensional chest images obtained by photographing.

Furthermore, the principal component vector obtained by principal component analysis is
A first principal component vector that is a component that changes the size of the lung field of the three-dimensional standard rib shape;
A second principal component vector that is a component that tilts the three-dimensional standard rib shape back and forth;
A third principal component vector that is a component that tilts the three-dimensional standard rib shape to the left and right;
And a fourth principal component vector that is a component that changes the three-dimensional standard rib shape so as to spread below the shape of the lung field.

  According to the present invention, a two-dimensional standard rib shape obtained by using a statistical method, a shape deformation vector for deforming the two-dimensional standard rib shape, and a three-dimensional standard corresponding to the two-dimensional standard rib shape. Based on the rib shape, a parameter of a shape deformation vector that deforms so that the two-dimensional standard rib shape matches the rib shape captured in the chest image of the subject is obtained, and a three-dimensional standard rib is obtained according to the parameter. By deforming the shape, the three-dimensional rib shape can be estimated from the two-dimensional rib shape captured in the two-dimensional chest image.

  Also, by estimating the 3D rib shape from the two 2D chest images to be subtracted, aligning the 3D rib shape to match, and performing subtraction, artifacts can be obtained. It becomes possible to reduce.

  Further, when deforming the rib shape so that the corresponding positions on the three-dimensional rib coincide with each other, the rib removal image generated by subtraction is also generated by moving and deforming the texture on the rib according to the change. Artifacts generated above can be reduced.

  If the correspondence between the amount of deformation of the three-dimensional standard rib shape and the parameter of the shape deformation vector for deforming the three-dimensional standard rib shape is stored in advance, the three-dimensional standard rib shape is immediately deformed. This makes it possible to estimate the three-dimensional rib shape at high speed.

  In addition, by analyzing the rib shape of a large number of chest images taken in the past, the variation from the standard shape can be expressed by a small number of independent components, and the three-dimensional rib shape can be easily estimated. Is possible.

  Further, the principal component vector obtained by the component analysis includes a component that changes the size of the lung field of the three-dimensional standard rib shape, a component that tilts the three-dimensional standard rib shape back and forth, and a three-dimensional standard. It includes a component that tilts the rib shape to the left and right and a component that changes the three-dimensional standard rib shape so as to spread below the shape of the lung field.

  Hereinafter, a first embodiment 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.

  The first image processing apparatus 1 shown in FIG. 1 includes a two-dimensional standard rib shape S2D obtained by using a statistical method from the rib shapes of a plurality of two-dimensional chest images obtained by photographing in advance, and the 2 Shape deformation data storage means 10 for storing a shape deformation vector A for deforming a three-dimensional standard rib shape, and a three-dimensional standard rib shape storage means 20 for storing a three-dimensional standard rib shape S3D corresponding to the two-dimensional rib shape. And the two-dimensional standard rib shape S2D based on the two-dimensional subject rib shape captured in the chest image P obtained by photographing the subject and the two-dimensional standard rib shape S2D. Parameter acquisition means 40 for acquiring parameters of each shape deformation vector A that is deformed so as to match the shape, and deforming the three-dimensional standard rib shape S3D according to the parameters to obtain a three-dimensional subject at the time of photographing the subject Rib shape 3D rib shape estimation means 50 for estimating PS3D.

  First, a method for obtaining a two-dimensional standard rib shape and a shape deformation vector for deforming the two-dimensional standard rib shape will be described in advance.

As shown in FIG. 2, the rib shape (FIG. 2B) is extracted from the chest image S representing many normal breasts. 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). The shape vector X representing the entire shape of the rib is described in, for example, “Peter de Souza,“ Automatic Rib Detection in Chest Radiographs ”, Computer Vision, Graphics and image Processing 23, 129-161 (1983)”. 100 points on the ribs are extracted by using a method of detecting the rib shape by an edge extraction filter or a Hough transform for detecting a parabola, and the coordinates (x, y) of the 100 points are sequentially arranged. As a vector, it can be expressed as in equation (1).

  Therefore, a large number of chest images S representing normal chest images taken in the past are collected, the shape vector represented by the above equation (1) is extracted, and the principal component analysis (specifically, the literature [ The principal component analysis method described in Matthew Turk, Alex Pentland, Eigenfaces for Recognition, Journal of Cognitive Neuroscience, vol. 3, num1, 1991] can be used.) The rib shape of the accumulated chest image S can be represented by a small number of independent vector components.

Here, as proposed by the present applicant in Japanese Patent Application No. 2004-009363, the average shape vector Xb representing the average shape of the ribs, and the difference between the shape vector X of the chest image and the average shape vector Xb (that is, standard A method of obtaining a principal component vector by obtaining a difference vector X-Xb representing a change from the shape and performing principal component analysis on the difference vector will be described. The rib model shape M includes first to n-th principal component vectors Ai (i = 1, 2,..., N) obtained by principal component analysis of the difference vector X-Xb, and an average shape vector Xb. Are used to obtain a plurality of rib model shapes obtained by deforming the average shape vector Xb. Each rib model shape can be expressed as shown in Equation (2).

  Here, αi is a parameter for each principal component vector. The parameter for this principal component vector is hereinafter referred to as a rib generation parameter. The average shape of the ribs will be described below as a two-dimensional standard rib shape S2D.

  For example, as shown in FIG. 3, the average shape (standard rib shape S2D) is obtained from the rib shape obtained from N two-dimensional chest images ((a) in the figure), and the difference between the average shape and the rib shape is obtained. Each principal component vector Ai (i = 1, 2,..., N) obtained as a result of the principal component analysis of the vector is represented by the first principal component as shown in FIGS. The vector appears as a component that expands the two-dimensional standard rib shape S2D in the direction of the arrow in (b). The second principal component vector expands the two-dimensional standard rib shape S2D in the direction of the arrow in (c). Appears as an ingredient. This principal component vector is obtained as a shape deformation vector A for deforming the two-dimensional standard rib shape S2D, and is stored in the shape deformation data storage means 10 together with the two-dimensional standard rib shape S2D.

  Next, a method for obtaining a three-dimensional standard rib shape and a method for associating the three-dimensional standard rib shape with a two-dimensional standard rib shape will be described. First, a rib shape is extracted from a tomographic image group acquired by CT, MRI, or the like, and a three-dimensional rib shape is obtained based on the interval at which the tomographic images are taken. In this way, a three-dimensional rib shape is obtained from tomographic images obtained by photographing many subjects, and the average shape of the three-dimensional rib shape is used as the three-dimensional standard rib shape S3D. From the statistical point of view, the average shape of the two-dimensional ribs and the average shape of the three-dimensional ribs obtained from the two-dimensional chest images taken in the past are the projection of the three-dimensional standard rib shape S3D in two dimensions. It is considered that the object is substantially identical to the two-dimensional standard rib shape S2D. Accordingly, the corresponding positions are obtained based on the shapes obtained by projecting the two-dimensional standard rib shape S2D and the three-dimensional standard rib shape S3D on the two-dimensional plane. For example, as shown in FIG. 4, the feature points on the ribs as shown in FIG. 2B correspond to the standard rib shape S3D of the three-dimensional rib, and the XY plane of the two-dimensional standard rib shape S2D. The depth direction values (Z direction, z1, z2, z3, z4,...) Of the three-dimensional standard rib shape S3D corresponding to the above feature points (q1, q2, q3, q4,...) The three-dimensional standard rib shape storage means is obtained by expressing the three-dimensional standard rib shape S3D using the position of the feature point on the two-dimensional plane (the feature point of the two-dimensional standard rib shape S2D) and the value in the depth direction thereof. 20 stored.

  The parameter acquisition means 40 extracts the subject rib shape from the two-dimensional subject chest image P obtained by photographing the subject and deforms the two-dimensional standard rib shape S2D (hereinafter referred to as a rib generation parameter). To get.

  In the present embodiment, the parameter acquisition means 40 generates and stores a large number of rib model images M in advance, and the rib model image having the rib shape most similar to the subject rib shape extracted from the chest image P from among them. A method of searching for M and obtaining a rib generation parameter according to the searched rib model image M will be described.

  As shown in FIG. 5, the parameter acquisition unit 4 includes a rib database 41 that stores the rib model image M, a rib shape extraction unit 42 that extracts a subject rib shape from the chest image P, and a subject rib shape from the rib model image M. And a rib shape estimating means 43 for searching the rib model image M having the rib shape most similar to the above and estimating the rib shape.

  First, the rib generation parameter αi in the equation (2) is varied in various ways, and a large number of rib model shapes are generated (see FIG. 2D) and stored in the rib database 41 as shown in FIG. Specifically, for example, the rib generation parameter αi may be allocated at equal intervals between the maximum value and the minimum value of the rib generation parameter αi obtained from the normal rib shape of the accumulated image.

  In this way, a rib model shape that appears in a normal rib shape, for example, gradually changing features such as the size of the lung field of the rib shape and the width between ribs, is comprehensively generated, and this rib model shape is A large number of images are prepared as rib model images M. The multiple rib model images M generated in this way store the rib generation parameter αi data used when generating the rib model shape together in the rib database 41. The rib model image M may be an image in which the bone portion and the lung field soft portion have different textures. For example, the bone portion may have a white texture and the lung field soft portion may have a gray texture. It is desirable to prepare a large number of rib model images M (for example, 1000 or more) so that a shape approximate to the rib shape of the subject in the chest image P obtained by photographing the subject can be searched.

  Further, the rib database 41 is configured to include database management software having a structure for processing the search reliably or at high speed, and includes information such as gender, age, race, and the like targeted for the rib model image M. It is desirable to manage them in association with each other and to store and manage them so that a number of similar rib shapes can be easily searched from among the rib model images M.

  From now on, the flow of the parameter acquisition means 40 in this embodiment is demonstrated according to the flowchart of FIG.

  First, the rib shape extraction means 42 creates a rib-weighted image using a filter that emphasizes the ribs on the chest image P obtained by photographing the subject, and extracts the rib shape. The filter that emphasizes the rib is an image processing filter in which the filter coefficient is adjusted so as to emphasize a rod-shaped object having a thickness and direction suitable for the rib. For example, a filter having a filter coefficient as shown in FIG. It is. In the case of this filter, the value is large for the horizontal bar-shaped structure, has no sensitivity to the vertical bar-shaped structure, and the value increases as the structure in the horizontal direction is longer. However, the filter is rotated to create filters in various directions, and the maximum value of the filter output values is taken to create a rib-enhanced image (S100).

  Next, the rib model image M stored in the rib database 41 is also subjected to rib emphasis processing using the filter for emphasizing the ribs to create a rib emphasis image (S101).

  Therefore, the rib shape estimation means 43 obtains a correlation between the rib emphasized image in which the rib shape of the chest image P is emphasized and the rib emphasized image in which the rib shape of the rib model image M stored in the rib database 41 is emphasized. Is searched from the rib model image M with the maximum correlation, and the rib shape of the rib model image M with the maximum correlation is estimated as the subject rib shape captured in the chest image P.

Specifically, the correlation between the rib-enhanced image of the chest image P and the rib-enhanced image of the rib model image M is obtained using normalized cross-correlation (S102). Normalized cross-correlation, chest image pixel values of pixels included in the rib enhanced image of P I _{1 (x,} y) and the pixel values of pixels included in the rib enhanced image ribs model image M I _{2 (x,} y ), It is expressed as in equation (3).

  From the obtained normalized cross-correlation value, the most similar rib shape of the rib model image M is estimated as the rib shape of the chest image P (S103). Therefore, the rib generation parameter αi of the most similar rib model image M is acquired from the rib database 41 as the rib generation parameter of the subject rib shape (S104).

  By the way, the two-dimensional standard rib shape S2D is deformed using the shape deformation vector A, that is, the main component vector, and the size and inclination (front and rear, left and right) of the three-dimensional standard rib shape S3D are changed to a two-dimensional plane. When compared with the projected shape, (1) the first principal component vector is a component that changes the size of the lung field of the three-dimensional standard rib shape, and (2) the second principal component vector is the three-dimensional (3) The third principal component vector is a component that tilts the three-dimensional standard rib shape to the left and right. (4) The fourth principal component vector is a three-dimensional component. It can be seen that it is a component that changes the standard rib shape so as to spread below the shape of the lung field. These principal component vectors coincide with changes due to the physique and posture of the subject, which are the main factors that determine the rib shape of the chest image when the chest image is taken. The shape deformation vector parameter (rib generation parameter) increases the overall size of the three-dimensional standard rib shape, changes the forward / backward inclination, changes the left / right inclination, and extends to the lower part of the lung field shape. It is possible to cope with deformation caused by this. For example, although the second principal component vector is a component representing the front-rear inclination, if the value of the bone generation parameter of the second principal component vector is “100”, the inclination is stored in association so that the inclination is 3 degrees. To do.

  As shown in FIG. 4, the three-dimensional rib shape is on the two-dimensional plane (XY plane) of each feature point so as to correspond to the feature points on the two-dimensional rib shape (see FIG. 2 (b)). The three-dimensional rib shape estimation means 50 is expressed by using the coordinate value and the component in the depth direction of the feature point (value in the Z direction) according to the rib generation parameter αi obtained as described above. The size of the three-dimensional standard rib shape S3D is changed, or the three-dimensional rib shape of the subject is obtained by tilting back and forth and right and left.

  As described above in detail, the three-dimensional rib shape of the subject can be estimated from the rib shape of the two-dimensional chest image by using a statistical method.

  Next, in the second embodiment, a method of performing alignment when performing subtraction will be described using the above-described method of estimating a three-dimensional rib shape from a two-dimensional chest image. The same reference numerals are given to the same components as those in the above embodiment, and the detailed description is omitted.

  The second image processing device 2 shown in FIG. 8 is the same as the first image processing device 1 and one target chest image of two different target chest images obtained by photographing a subject to be subjected to difference processing. The first subject rib shape acquisition unit 60 that acquires the first three-dimensional subject rib shape using the three-dimensional rib shape estimation unit 50, and the other target chest image among the target chest images, the three-dimensional rib Based on the first and second three-dimensional subject rib shapes, the second subject rib shape obtaining means 70 for obtaining the second three-dimensional subject rib shape using the shape estimation means, and the first and second Deformation means 80 for deforming one of the two target chest images so that the three-dimensional subject rib shapes match, and difference image acquisition means 90 for generating a difference image from the two target chest images after the deformation. Prepare.

  The first image processing apparatus 1 includes a shape deformation data storage unit 10, a three-dimensional standard rib shape storage unit 20, a parameter acquisition unit 40, and a three-dimensional rib shape estimation unit 50.

  In the present embodiment, a case where the temporal subtraction process is performed to remove the ribs will be specifically described. When observing a change in a lesion part of a subject having a disease such as lung cancer, a temporal subtraction process is performed on the past image and the current image to remove the ribs and observe only the change in the soft part. However, it is difficult to photograph the subject with the same posture at the past and the current photographing, and in many cases, the posture is changed between the past and the current photographing, and the rib shape is photographed differently. Therefore, if time-dependent differential traction is performed on the past image and the current image as they are, artifacts due to changes in the rib shape occur. Therefore, in order to reduce artifacts, subtraction processing is performed after aligning the rib shape of the subject of the chest image captured in the past with the rib shape of the subject of the currently captured chest image.

  First, the first subject rib shape acquisition means 60 acquires the first three-dimensional subject rib shape using the past chest image P1 as the target chest image. First, the subject rib shape is extracted from the chest image P1 by the parameter acquisition means 40 of the first image processing apparatus, the rib model image M most similar to the extracted subject rib shape is searched from the rib database 41, and this rib is searched. The rib generation parameter αi corresponding to the model image M is acquired. Next, the first three-dimensional subject rib shape corresponding to the acquired rib generation parameter αi is obtained using the three-dimensional rib shape estimating means 50 of the first image processing apparatus.

  Similarly, the second subject rib shape acquisition unit 70 uses the current chest image P2 as the target chest image, acquires the rib generation parameter αi by the parameter acquisition unit 40, and acquires it using the three-dimensional rib shape estimation unit 50. A second three-dimensional subject rib shape corresponding to the calculated rib generation parameter αi is obtained.

  Since the first and second three-dimensional rib shape feature points obtained by the first and second subject rib shape acquisition means 60 and 70 correspond to each other, the deforming means 80 is obtained from the past chest image P1. The first three-dimensional subject rib shape feature point matches the position of the second three-dimensional subject rib shape feature point obtained from the current chest image P2. Transform.

  In the present embodiment, since a difference image is acquired from the past and current chest images, deformation is performed on the two-dimensional plane so that the positions of the feature points of the chest image P2 coincide with the positions of the feature points of the chest image P1. .

  As shown in FIG. 9, when a feature point (black circle) on the first three-dimensional subject rib shape is projected on a two-dimensional plane (FIG. 9A), it substantially matches the rib shape of the chest image P1. Therefore, the feature point obtained by projecting the three-dimensional subject rib shape obtained by the first subject rib shape acquisition means 60 onto the two-dimensional plane is superimposed on the chest image P1, and the position of the feature point on the chest image P1 is obtained. . Similarly, the position of the feature point on the chest image P2 is obtained by projecting the second three-dimensional subject rib shape obtained by the seventh subject rib shape obtaining means 70 on the two-dimensional plane, as shown in FIG. In addition, the image of the position of the feature point of the past chest image P1 is the position of the feature point of the current chest image P2 so that the position of the feature point on the chest image P1 matches the position of the feature point on the chest image P2. Warping to match. At this time, the texture on the ribs of the chest image P1 is also moved along with warping to generate an image obtained by deforming the past chest image P1.

  However, when the three-dimensional rib shape differs between the past and the present depending on the posture of the subject, the positional relationship between the anterior rib and the posterior rib changes as shown in FIG. Therefore, the density of the texture where the ribs overlap is not the same as the warped image texture. Therefore, the texture of the overlapping part of the ribs is estimated from the density of the texture of the surrounding ribs. For example, as shown in FIG. 11 (b), when QL = 100 of the overlapping portion and the surrounding bone A = 20 and bone B = 60, the bone density in the overlapping is defined as the contribution of bones A and B. Ask for the same ratio. Therefore, the density of the overlapped portion is determined by setting the contribution ratio of the overlapped portion even after warping as bone A: bone B = 25: 75.

  The difference image acquisition means 90 performs a subtraction process on the image obtained by deforming the previous chest image P2 and the current chest image P2, and acquires a difference image.

  In the above description, the past chest image is matched with the current chest image. However, the current chest image may be matched with the past chest image. Alternatively, both chest images may be deformed so that both chest images have the same shape.

  As described above in detail, when acquiring a three-dimensional rib shape from a two-dimensional chest image and performing alignment, the artifacts are deformed so that the points on the three-dimensional rib shape coincide with each other. It is possible to generate a difference image without any difference.

1 is a diagram illustrating a schematic configuration of a first image processing apparatus according to a first embodiment. The figure for demonstrating the method of producing a rib model image by principal component analysis The figure which shows the average shape and principal component vector of the rib Diagram showing 3D rib shape The figure which shows the structure of the parameter acquisition means Flowchart showing the flow of processing of parameter acquisition means An example of a filter to emphasize the ribs The figure which shows schematic structure of the 2nd image processing apparatus in 2nd Embodiment. The figure for demonstrating the method of projecting a three-dimensional rib shape on a two-dimensional plane The figure for demonstrating the method of warping in consideration of the change of a three-dimensional rib shape The figure which shows a mode that the position where a rib overlaps changes on a two-dimensional image when a three-dimensional rib shape changes.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 1st image processing apparatus 2 2nd image processing apparatus 10 Shape deformation data memory | storage means 20 3D standard rib shape memory | storage means 40 Parameter acquisition means 41 Radius database 42 Radius shape extraction means 43 Radius shape estimation means 50 3D rib shape Estimation means 60 First subject rib shape acquisition means 70 Second subject rib shape acquisition means 80 Deformation means

Claims (10)

A two-dimensional standard rib shape obtained by using a statistical method and a plurality of shape deformation vectors for deforming the two-dimensional standard rib shape are stored from rib shapes of a plurality of two-dimensional chest images obtained by photographing in advance. Shape deformation data storage means,
Three-dimensional standard rib shape storage means for storing a three-dimensional standard rib shape corresponding to the two-dimensional standard rib shape;
Based on the two-dimensional subject rib shape captured in the subject chest image obtained by photographing the subject and the two-dimensional standard rib shape, the two-dimensional standard rib shape is changed to the two-dimensional subject rib shape. Parameter acquisition means for acquiring parameters of the respective shape deformation vectors that are deformed so as to coincide;
An image processing apparatus comprising: a three-dimensional rib shape estimating unit that deforms the three-dimensional standard rib shape according to the parameter and estimates a three-dimensional subject rib shape at the time of photographing the subject. . The first three-dimensional subject rib shape is acquired by the three-dimensional rib shape estimation means for one target chest image of two different target chest images obtained by photographing a predetermined subject to be subjected to the difference processing. First subject rib shape acquisition means;
Second subject rib shape acquisition means for acquiring a second three-dimensional subject rib shape by the three-dimensional rib shape estimation means for the other target chest image of the two target chest images;
Based on the first and second three-dimensional subject rib shapes, deform at least one of the two target chest images so that the first and second three-dimensional subject rib shapes match. The image processing apparatus according to claim 1, further comprising a deformation unit.   The image processing apparatus according to claim 2, wherein the deformation unit is configured to move and deform a texture on the rib together with deformation of the shape of the rib.   4. The three-dimensional rib shape estimation means stores in advance a correspondence between an amount of deformation of the three-dimensional standard rib shape and a parameter of the shape deformation vector. Image processing apparatus.   5. The image processing according to claim 1, wherein the shape deformation vector is a principal component vector obtained by principal component analysis of rib shapes of a plurality of two-dimensional chest images obtained by photographing. apparatus. The principal component vector obtained by the principal component analysis is
A first principal component vector that is a component that changes the size of the lung field of the three-dimensional standard rib shape;
A second principal component vector that is a component that tilts the three-dimensional standard rib shape back and forth;
A third principal component vector that is a component that tilts the three-dimensional standard rib shape to the left and right;
6. The image processing apparatus according to claim 5, further comprising a fourth principal component vector, which is a component that changes the three-dimensional standard rib shape so as to spread below the shape of the lung field. Computer
Based on the two-dimensional standard rib shape obtained from the rib shapes of a plurality of two-dimensional chest images obtained by photographing in advance and the two-dimensional subject rib shape photographed on the subject chest image obtained by photographing the subject. The two-dimensional standard rib shape obtained by using a statistical method from the rib shapes of the plurality of two-dimensional chest images deforming the two-dimensional standard rib shape to match the two-dimensional subject rib shape. Parameter acquisition means for acquiring parameters of a plurality of shape deformation vectors for deforming
The three-dimensional standard rib shape corresponding to the two-dimensional standard rib shape is deformed in accordance with the parameter, and the three-dimensional rib shape estimation means for estimating the three-dimensional subject rib shape at the time of photographing the subject is caused to function. program. The first three-dimensional subject rib shape is acquired by the three-dimensional rib shape estimation means for one target chest image of two different target chest images obtained by photographing a predetermined subject to be subjected to the difference processing. First subject rib shape acquisition means;
Second subject rib shape acquisition means for acquiring a second three-dimensional subject rib shape by the three-dimensional rib shape estimation means for the other target chest image of the two target chest images;
Based on the first and second three-dimensional subject rib shapes, deform at least one of the two target chest images so that the first and second three-dimensional subject rib shapes match. 8. The program according to claim 7, further comprising deformation means. A two-dimensional standard rib shape obtained by using a statistical method and a plurality of shape deformation vectors for deforming the two-dimensional standard rib shape are stored from rib shapes of a plurality of two-dimensional chest images obtained by photographing in advance. A shape deformation data storing step,
A three-dimensional standard rib shape storing step for storing a three-dimensional standard rib shape corresponding to the two-dimensional standard rib shape;
Based on the two-dimensional subject rib shape captured in the subject chest image obtained by photographing the subject and the two-dimensional standard rib shape, the two-dimensional standard rib shape is changed to the two-dimensional subject rib shape. A parameter acquisition step of acquiring parameters of each shape deformation vector that is deformed so as to match;
An image processing method comprising: a three-dimensional rib shape estimation step for deforming the three-dimensional standard rib shape according to the parameter to estimate a three-dimensional subject rib shape at the time of photographing the subject. . A first three-dimensional subject rib shape is acquired by the three-dimensional rib shape estimation step with respect to one target chest image of two different target chest images obtained by photographing a predetermined subject to be subjected to difference processing. A first subject rib shape acquisition step;
A second subject rib shape acquisition step of acquiring a second three-dimensional subject rib shape by the three-dimensional rib shape estimation step with respect to the other target chest image of the two target chest images;
Based on the first and second three-dimensional subject rib shapes, deform at least one of the two target chest images so that the first and second three-dimensional subject rib shapes match. The image processing method according to claim 9, further comprising a deformation step.

Images