JP2007098028A - Modeling apparatus, modeling method, region extraction apparatus, and program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a technique capable of realizing precise modeling processing. <P>SOLUTION: By deforming a standard model so as to optimize a prescribed evaluation function using a model fitting method, the model of an object is generated. The prescribed evaluation function includes a transcendental knowledge term. The transcendental knowledge term is the term to be optimized when the standard model is deformed so as to coincide with an assumption region RB assumed as the presence region of an object on the basis of transcendental knowledge. The assumption region RB is obtained as a region where the presence probability of the object obtained on the basis of data of the same kind of object is higher than a prescribed value, for example. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention relates to a modeling apparatus for generating a model and a technique related thereto.

  In recent years, in-vivo three-dimensional images taken by an X-ray CT (Computed Tomography) apparatus or an MRI (Magnetic Resonance Imaging) apparatus are widely used in the medical field. As a result, information on internal organs and the like can be visually grasped, and an improvement in diagnostic accuracy can be expected. However, on the other hand, since the three-dimensional image used for diagnosis is composed of tens to hundreds of slice images, only information necessary for diagnosis is obtained from such an enormous amount of information. Is a big burden for doctors.

  Therefore, demands for quantitative or automatic diagnosis with the aid of a computer have increased, and research on a computer-aided diagnosis (CAD: Computer-aided diagnosis) system has been actively conducted (for example, see Non-Patent Documents 1 to 3). .

  In order to perform diagnosis support by a computer, it is an important issue to accurately extract information necessary for diagnosis, that is, an organ region or a shape.

  As one of organ region extraction methods, a model fitting method for finding a target contour by deforming a standard model prepared in advance on the basis of the energy minimization principle has been proposed.

  According to this method, since a corresponding region is extracted by deforming a model having a characteristic in advance, which is a standard model, high-precision extraction processing is possible.

Tsagaan Bigalma, Akinobu Shimizu, Hidefumi Obata, “Development of a method for extracting kidney regions from abdominal CT images using a three-dimensional deformable model”, IEICE Transactions, D-II, January 2002, Vol. D-II, No.1, pp.140-148 Akinobu Shimizu, Hiroki Sakurai, Tomonao Yanagida, Hidefumi Kibata, Shigeru Nino, “Extraction processing of multiple organs from 3D abdominal CT images based on human electronic atlas and its performance evaluation-Comparison with conventional methods”, Shingaku Technical Bulletin, 2005, MI2005-15, pp.7-12 Teruhiko Kitagawa, Kazumasa Okio, Sakae Sakae, Ryujiro Yokoyama, Takeshi Hara, Hiroshi Fujita, Masayuki Kanematsu, Hiroaki Hoshi, "Automatic generation of probabilistic atlas of the liver by deforming the diaphragm in trunk CT images and its automatic extraction Application, IEICE Technical Report, 2005, MI2005-16, pp.13-18

  However, even with the above model fitting method (see Patent Document 1), sufficient accuracy may not be obtained. Specifically, in the above model fitting method, two types of energy, external energy and internal energy, are considered, but sufficient accuracy can be obtained by considering these two types of energy. There may not be.

  Such a problem is a problem that generally occurs not only in the modeling process for the region extraction process as described above but also in the modeling process for creating a model of the object.

  Accordingly, an object of the present invention is to provide a technique capable of realizing a more accurate modeling process.

  In order to solve the above-mentioned problem, the invention of claim 1 is a modeling apparatus for generating a model of an object, an acquisition means for acquiring a standard model related to the object, and a predetermined evaluation function using a model fitting technique. And deforming the standard model so as to optimize the model and generating a model of the object, and the predetermined evaluation function is assumed as an existence region of the object based on a priori knowledge. A priori knowledge term that is optimized when the standard model is deformed so as to coincide with an assumed area that is a predetermined area.

  According to a second aspect of the present invention, in the modeling apparatus according to the first aspect of the invention, the assumed region is a region in which the existence probability of the object is higher than a predetermined value.

  According to a third aspect of the present invention, in the modeling device according to the first or second aspect of the present invention, the a priori knowledge term is in the direction of an approach vector that is a vector that approaches the boundary of the assumed region in the shortest distance. It is optimized when each control point of the standard model moves.

  According to a fourth aspect of the present invention, in the modeling apparatus according to the third aspect of the present invention, when the model deforming means sequentially deforms the standard model, the approach vector is a value of each control point of the standard model after deformation. It is updated according to the position, and the a priori knowledge term is calculated based on the updated approach vector.

  According to a fifth aspect of the present invention, in the modeling apparatus according to the third aspect of the present invention, the approach vector is preset for each control point in accordance with an initial position of each control point, and the model deformation means is the standard. Even when the model is sequentially deformed, the a priori knowledge term is always calculated based on the approach vector preset for each control point according to the initial position of each control point. It is characterized by that.

  According to a sixth aspect of the present invention, in the modeling apparatus according to any one of the first to fifth aspects, the predetermined evaluation function is optimized when the shape of the standard model approaches the contour in the measurement data. It has an external energy term.

  According to a seventh aspect of the present invention, in the modeling apparatus according to any one of the first to sixth aspects, the predetermined evaluation function is obtained when the relationship between the control points of the standard model approaches a certain relationship. It has an internal energy term that is optimized.

  The invention according to claim 8 is the modeling device according to any one of claims 1 to 7, wherein the region corresponding to the generated position of the model of the object is used as the region of the object from the measurement data. And a means for extracting.

  The invention according to claim 9 is a modeling method for generating a model of an object, wherein a standard model related to the object is arranged at an initial position, and a predetermined evaluation function is optimized using a model fitting technique. And deforming the standard model to generate a model of the object, wherein the predetermined evaluation function is an assumed area which is an area assumed as an existence area of the object based on a priori knowledge. It has a priori knowledge terms that are optimized when the standard model is deformed to match.

  In the invention of claim 10, the standard model relating to the object is arranged at an initial position in the computer, and the standard model is deformed so as to optimize a predetermined evaluation function using a model fitting technique, and the object A program for generating a model of an object, wherein the predetermined evaluation function matches an assumed area that is assumed as an existing area of the object based on a priori knowledge And a priori knowledge term that is optimized when the standard model is deformed.

  The invention according to claim 11 is an area extraction device for extracting an area of an object from measurement data, and optimizes a predetermined evaluation function using an acquisition means for acquiring a standard model related to the object and a model fitting technique. Means for deforming the standard model, and means for extracting, from the measurement data, an area corresponding to the position of the standard model after the deformation as the area of the object, and the predetermined evaluation function includes: It has a priori knowledge term that is optimized when the standard model is deformed so as to coincide with an assumed region where an object is assumed to exist based on a priori knowledge.

  According to the first to eleventh aspects of the invention, since the model can be deformed by reflecting a priori knowledge, a more accurate modeling process can be realized.

  In particular, according to the invention described in claim 4, when the standard model is sequentially deformed, the approach vector is updated according to the position of each control point of the standard model after deformation, and the a priori knowledge term is And calculated based on the updated approach vector. Therefore, by considering the approach vector updated from time to time, model fitting can be performed by reflecting a priori knowledge with higher accuracy.

  According to the fifth aspect of the present invention, it is not necessary to prepare approach vectors in advance for all the pixels in the object existence probability map, and each approach vector at the initial position of each control point of the standard model can be obtained. It is efficient because it can be done.

  Further, according to the invention described in claim 8, by extracting from the measurement data the region corresponding to the position where the generated model exists as the region of the object, the a priori knowledge is reflected, The region can be extracted accurately.

  According to the invention described in claim 11, since the region corresponding to the position of the standard model that is deformed reflecting a priori knowledge is extracted from the measurement data as the region of the object, the a priori knowledge is obtained. Reflecting this, it is possible to accurately extract the region of the object.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings.

<1. Configuration>
FIG. 1 is a diagram showing an outline of a region extraction device 1 according to an embodiment of the present invention. The region extraction device 1 is a device that extracts a desired target region (specifically, a three-dimensional region) from measurement data (specifically, three-dimensional measurement data (also referred to as three-dimensional measurement data)). The region extraction apparatus 1 is also referred to as a “modeling apparatus” because it generates a target object model using a model fitting technique.

  As shown in FIG. 1, the region extraction apparatus 1 includes a personal computer (hereinafter simply referred to as “personal computer”) 2, a monitor 3, a mounting unit 5, and an operation unit 4.

  The personal computer 2 includes a control unit 20, an input / output I / F 21, and a storage unit 22.

  The input / output I / F 21 is an interface (I / F) for transmitting / receiving data between the monitor 3, the operation unit 4, the mounting unit 5, and the personal computer 2. Send and receive.

  The storage unit 22 includes, for example, a hard disk, and stores a software program (hereinafter simply referred to as “program”) PG for executing various processes described below.

  The control unit 20 mainly includes a CPU, a ROM 20a, a RAM 20b, and the like, and is a part that performs overall control of each unit of the personal computer 2.

  The monitor 3 is composed of, for example, a CRT, and visually outputs a display image generated by the control unit 20.

  The operation unit 4 includes a keyboard and a mouse, and transmits various signals to the user (user) according to various operations to the input / output I / F 21.

  The mounting unit 5 can detachably mount a storage medium such as the memory card 51. Various data or programs stored in the memory card 51 attached to the attachment unit 5 can be taken into the control unit 20 or the storage unit 22 via the input / output I / F 21.

  Next, various functions of the region extraction device 1 will be described.

  FIG. 2 is a block diagram illustrating various functions of the region extraction device 1.

  These various functions are realized by executing a predetermined program PG using various hardware such as a CPU in the control unit 20.

  As shown in FIG. 2, the region extraction apparatus 1 includes a model storage unit 11, a measurement data input unit 12, a temporary region extraction unit 13, an initial position determination unit 14, a model fitting unit 15, a region extraction unit 16, and an extraction region output. An object to be extracted by a user (user) or the like from the input three-dimensional image (three-dimensional measurement data) (hereinafter also referred to as “extraction target region (object)”) and output Can do.

  The model storage unit 11 stores a “standard model” used for the model fitting process. The area extracting apparatus 1 can acquire the “standard model” by reading from the model storage unit 11. Here, the “standard model” is a model used in a model fitting method described later, and means a standard model of an extraction target region (for example, an organ).

  The measurement data input unit 12 has a function of inputting three-dimensional measurement data (volume data) acquired by a CT apparatus or an MRI apparatus.

  The provisional region extraction unit 13 has a function of provisionally extracting an extraction target object (region) using a method different from the model fitting method.

  The initial position determination unit 14 has a function of determining an initial position of the standard model using a provisionally extracted extraction target object (region).

  The model fitting unit 15 has a function of generating a model of an object by deforming a standard model using a model fitting method.

  The region extraction unit 16 has a function of extracting a region of an object from measurement data using the model generated by the model fitting unit 15.

  The extraction area output unit 17 has a function of displaying and outputting the extracted target object on the monitor 3.

<2. Operation>
<Overview of operation>
In this embodiment, a model fitting process based on a three-dimensional image (also referred to as volume data) acquired by an X-ray CT apparatus, a region extraction process using the model fitting process, and the like will be described. However, the present invention is not limited to this, and is also applicable to a model fitting process based on other measurement data (for example, a three-dimensional image acquired by an MRI apparatus) and a region extraction process using the model fitting process. can do.

  FIG. 3 is a diagram showing a three-dimensional image TDP1 acquired by the X-ray CT apparatus.

  As shown in FIG. 3, the three-dimensional image TDP1 is composed of a plurality (for example, several tens to several hundreds) of slice images (tomographic images) SD showing a cross section of an object (human body). . Each slice image SD is an image visualized by displaying the amount of X-ray absorption (CT value) at each point (each pixel) in grayscale. In FIG. 3, a slice image PM at a predetermined position among the plurality of slice images SD constituting the three-dimensional image TDP1 is also shown.

  FIG. 4 is a flowchart showing the overall operation of the region extraction apparatus 1.

  The region extraction apparatus 1 extracts the target specified by the user (user) from the input image (three-dimensional image) by executing each process (steps S1 to S4) shown in FIG. The extraction target is designated in advance by the user (user) before the process of step S1 is started. As a designation method, for example, the user can operate the operation unit (for example, mouse) 4 or the like from an item list (for example, an item list in which various organs are listed) stored in the storage unit 22 in advance as a database. A mode in which the (user) selects a desired item can be employed. Hereinafter, assuming that a predetermined organ is designated as the extraction target, each process of FIG. 4 will be described.

  First, in step S1, an object (organ) designated by a user (user) is provisionally extracted from a three-dimensional image input to the region extraction device 1. As will be described later, in step S1, an extraction target region is extracted using a method different from the model fitting method.

  Next, in step S2, the initial position of the standard model (initial position when the standard model is arranged in the measurement data) is determined using the object (organ) temporarily extracted in step S1, and the standard model is determined. Is arranged. According to the provisional extraction process in step S1, this initial position can be obtained accurately.

  Further, in step S3, the standard model placed at the initial position is deformed by the model fitting method, and a model of the object is generated. Then, a region existing at the position of the standard model after deformation is extracted as a region of the object.

  In step S4, the created individual model is output to the monitor 3 or the like as an extraction target object.

  In this way, the extraction target object is finally extracted using the model fitting technique (step S3). Here, a case where a method (region expansion method) different from the model fitting method is used when determining the initial position of the standard model in the model fitting method (steps S1 and S2) is illustrated.

  Further, as will be described later, more accurate model fitting processing can be performed by considering a priori knowledge in the model fitting processing (step S3).

  Hereafter, each process of step S1, S2, S3 is explained in full detail sequentially.

<Temporary region extraction processing (step S1)>
As described above, the provisional area extraction process (step S1) aims to tentatively extract an object (organ) designated by a user (user). In this embodiment, a region expansion method is adopted as an example of the extraction method.

  FIG. 5 is a flowchart showing details of the provisional region extraction process (step S1) using the region expansion method. FIG. 6 is a view showing a slice image PM at a predetermined position when a human body is imaged by an X-ray CT apparatus, and FIG. 7 is a vicinity of adjacent pixels centering on the pixel M1 on the slice image PM shown in FIG. FIG. In FIG. 7, as adjacent pixels of the pixel M1 of the slice image PM, in addition to the adjacent 8 pixels M2 to M9 in the slice image PM of the same hierarchy, the adjacent 9 pixels U1 to U9 in the slice image PU and the adjacent in the slice image PD Nine pixels D1 to D9 are shown. The slice image PU is a slice image adjacent to the slice image PM in the + z direction (in other words, a slice image immediately above the slice image PM), and the slice image PD is −z with respect to the slice image PM. It is a slice image adjacent in the direction (in other words, a slice image immediately below the slice image PM).

  In the following, the provisional region extraction processing using the region expansion method will be described with a specific example of the case where the object (organ) OB1 displayed in the slice image PM of FIG. 6 is provisionally extracted.

  First, in step S11, the expansion start pixel in the extraction target object (organ) is specified by the user (user). Specifically, on an arbitrary slice image (in this case, the slice image PM) on which an object to be extracted (extraction target object) is displayed, any operation in the extraction target object is performed by operating the operation unit (eg, mouse) 4 or the like. Identify points. More specifically, when it is desired to extract the object OB1, an arbitrary point (for example, M1) in the object (organ) OB1 on the slice image PM may be specified with a mouse or the like (see FIG. 6). By this operation, the pixel corresponding to the specific point M1 on the slice image PM becomes the expansion start pixel.

  Next, in step S12, a pixel adjacent to the expansion start pixel is detected as an expansion candidate pixel. Specifically, if the expansion start pixel is specified as M1 in step S11, all the pixels adjacent to the pixel M1 are the expansion candidate pixels. That is, a total of 26 pixels, that is, the pixels M2 to M9 on the slice image PM, the pixels U1 to U9 on the slice image PU, and the pixels D1 to D9 on the slice image PD are extension candidate pixels.

  Next, in step S13, it is determined whether or not an extended candidate pixel is detected in step S12. If an extended candidate pixel is detected in step S12, the process proceeds to step S14.

  In step S14, it is determined whether or not the density value of the detected extension candidate pixel is within a predetermined range. Specifically, the density value range of the object (organ) is determined in advance for each object (organ), and whether or not the density value of the expansion candidate pixel is within the density value range of the extraction target object. judge. Accordingly, when it is determined that the expansion candidate pixel has a density value within a predetermined range, the process proceeds to step S15.

  In step S15, region expansion is performed using the expansion candidate pixel as a pixel in the extraction target object.

  On the other hand, if it is determined in step S14 that the extension candidate pixel does not have a density value within the predetermined range, the area extension of the extension candidate pixel is not performed.

  Next, in step S16, it is determined whether or not there is an extension candidate pixel that has not passed through the extension candidate pixel density determination step (step S14).

  If there is an extension candidate pixel that has not undergone the density determination step (step S14), the processes of steps S14 and S16 are performed on the unfinished extension candidate pixel.

  On the other hand, if the density determination step (step S14) has been completed for all the expansion candidate pixels, the process returns to step S12, and the pixels expanded in step S15 (in other words, the pixels in the region). A pixel that is further adjacent to the specified pixel) and has not yet been determined whether or not it exists in the region is detected as a new expansion candidate pixel. Thereafter, the above-described steps S13 to S16 are performed on the newly detected extended candidate pixel. Steps S12 to S16 are repeated as much as possible to detect a new extension candidate pixel. When a new extension candidate pixel cannot be detected, the provisional region extraction process ends (step S13).

  As described above, by repeatedly executing the steps S12 to S16, the region is gradually expanded from the expansion start pixel, and the object (organ) designated by the user (user) is provisionally extracted.

<Initial position determination process (step S2)>
Next, the initial position determination process (step S2) for determining the initial position of the standard model in the model fitting method will be described with reference to FIGS. FIG. 8 is a flowchart showing details of the initial position determination process (step S2) using the barycentric point. FIG. 9 is a diagram showing the object OBa included in the three-dimensional image (measurement data), and FIG. 10 is a diagram conceptually showing how the standard model SO is moved to the initial position. FIG. 10 also shows the positional relationship between the extraction target object and the standard model when the standard model is initially arranged.

  In the following, for simplification of illustration, it is assumed that an object OBa existing in the three-dimensional image TDP1 as shown in FIG. 9 is an extraction target object.

  First, in step S21, the center of gravity GZ of the object OBa temporarily extracted in step S1 is calculated.

  Next, in step S22, the standard model is initially placed so that the center point GH of the standard model used in model fitting described later matches the center point GZ calculated in step S21. For example, as shown in FIG. 10, the standard model SO is moved so that the center of gravity GH of the standard model SO matches the center of gravity GZ of the extraction target object OBa, and the standard model SO is initially arranged in the measurement data.

  When the initial placement of the standard model using the barycentric point is completed, the process proceeds to model fitting processing (step S3).

<Model fitting process (step S3)>
In the model fitting process in step S3, information (shape, etc.) obtained from a “standard model”, which is a model of a general (standard) extraction target object (also simply referred to as a target object) prepared in advance, is obtained. ). In the present application, a standard model after deformation by model fitting processing (in other words, a standard model reflecting information of an extraction target object) is also referred to as an “individual model”.

  In this model fitting process, a process for extracting an extraction target object is performed by deforming a standard model arranged at an initial position by a model fitting method.

  FIG. 11 is a flowchart showing details of the model fitting process (step S3). FIG. 12 is a schematic diagram in which the vicinity of a region RQ (see FIG. 10) centered on the boundary point Q1 of the extraction target object OBa is enlarged, and FIG. 13 is a diagram showing some slice images PQ1, PQ2, and PQ4. It is. FIG. 14 is a schematic diagram in which control points are connected by virtual springs.

  The standard model of the extraction target object is composed of, for example, a minute polygon (for example, a triangle) and is stored in the storage unit 22 or the like. The standard model composed of polygons can express the surface shape of the model by the three-dimensional coordinates of the vertices of each polygon. In addition, of the vertices of the polygon in the standard model, all vertices or some representative vertices are also referred to as “control points”.

  Hereinafter, the model fitting process will be described in detail with reference to FIGS.

  First, in step S31, a process of determining a point corresponding to the control point Cj of the standard model SO (hereinafter also referred to as “corresponding point”) is performed (see FIG. 11). Here, a method is used in which a pixel existing nearest to the control point Cj among the pixels (boundary points) indicating the outline (boundary) of the extraction target object OBa is a corresponding point of the control point Cj. Note that the outline (boundary) of the extraction target object OBa is acquired by performing an edge extraction process on each slice image SD.

  For example, in FIG. 12 (elevation), the corresponding point of the control point C1 is a boundary point (pixel) Q1 that is present nearest to the control point C1. Here, in FIG. 12, only boundary points in a predetermined plane (two-dimensional space) parallel to the xz plane are shown, but actually, the control point C1 in the three-dimensional space as shown in FIG. The boundary point existing closest to is a corresponding point. In FIG. 13, the projection points in the slice images PQ1, PQ2, and PQ4 of the control point C1 are indicated by crosses.

  Next, in step S32, an arbitrary one of the control points Cj of the standard model SO (hereinafter also referred to as “movement target point”) (for example, the control point C1) is moved by a minute amount L in one direction (for example, the A1 direction). (See FIG. 12).

  Further, in step S33, the total energy Ue of the model (hereinafter also referred to as “temporary deformation model”) in a state where the movement target point is moved and temporarily deformed in step S32 is calculated.

  The total energy Ue takes into account the external energy term Fe regarding the distance between the control point Cj and the corresponding point, the internal energy term Ge to avoid excessive deformation, and a priori knowledge as shown in the equation (1). It is expressed by the sum of the a priori knowledge term He for The external energy term (also simply referred to as external energy) Fe, the internal energy term (also simply referred to as internal energy) Ge, and the a priori knowledge term He will be described later.

  Next, in step S34, it is determined whether or not the movement target point has been moved in all directions. For example, if the movement direction of the movement target point in all directions in the three-dimensional space is 26 directions (direction toward 26 pixels adjacent to the movement target point (pixel)), the control point C1 is moved in all 26 directions. It is determined whether or not.

  If the creation of the temporary deformation model in which the movement target point (control point C1) is moved in all 26 directions has not been completed, the movement direction of the movement target point is changed and moved again by a small amount L, and different movement direction patterns are obtained. A temporary deformation model is created and the total energy Ue of each temporary deformation model is calculated (steps S32 and S33).

  If it is determined in step S34 that the movement in all directions has been completed, the process proceeds to step S35.

  In step S35, a temporary deformation model that minimizes the total energy Ue is selected from the generated temporary deformation models of all patterns. In other words, a temporary deformation model that minimizes the total energy Ue is selected from the temporary deformation models generated by moving a certain control point Cj in 26 directions. For example, when only the external energy Fe is considered among the three terms Fe, Ge, and He described later, a temporary deformation model in which the control point moves in a direction approaching the corresponding point is selected. When the internal energy Ge is taken into consideration, a temporary deformation model in which the control point moves in a direction different from the direction approaching the corresponding point may be selected. Further, when the a priori knowledge term He is taken into account, the a priori knowledge is reflected, and therefore, a temporary deformation model that moves to a position where the existence probability of the same type of object is high may be selected.

  Next, in step S36, it is determined whether or not all the control points Cj have been moved. Specifically, it is determined whether or not the movement of the minute amount L has been completed for all the control points Cj of the standard model SO. If there is a control point that has not been completed (also referred to as an incomplete point), the object to be moved is determined. The point is changed to the control point (unfinished point), and the operations in steps S32 to S35 are repeated to create a temporary deformation model in which the movement of all the control points Cj is completed. On the other hand, if the movement of all the control points has been completed, the process proceeds to step S37.

  In step S37, it is determined whether or not to end the model fitting process. Specifically, when the condition is satisfied on condition that the average value of the distances between the plurality of control points and the corresponding points in the temporary deformation model in which the movement of all the control points Cj is completed is equal to or less than a predetermined value In addition, the model fitting process may be terminated. According to this, when each control point approaches the corresponding point to a predetermined extent, the same process can be ended. Alternatively, or in addition to this, the amount of change in the total energy Ue between the previous temporary deformation model (which has completed the movement of all control points Cj) and the current temporary deformation model is equal to or less than a predetermined amount. Whether or not it may be used as a criterion for end determination. According to this, a process can be complete | finished when the total energy does not change so much even if a control point is moved.

  If it is determined in step S37 that the model fitting process is not finished, the process proceeds to step S38.

  In step S38, it is determined whether the unit processing loop has been executed a predetermined number of times W (for example, 10 times) using the processing executed in steps S32 to S37 described above as a unit processing loop. If the predetermined number of times W is not executed, the processing loop of steps S32 to S37 is repeated again until the predetermined number of times W is executed. If the predetermined number of times W is executed, the process proceeds to step S39. That is, the unit processing loop is repeated until a temporary deformation model that completes the movement of all control points is created W a predetermined number of times.

  In step S39, the corresponding points determined in step S31 are updated. Specifically, the corresponding point is updated with the pixel (boundary point) closest to each control point having moved W a predetermined number of times by the above processing loop as a new corresponding point for each control point, and the step is again performed. The processes of S32 to S39 are repeated. According to such “update of corresponding point”, it is possible to optimize the corresponding point even when the nearest boundary point of the control point changes with the movement of the control point.

  On the other hand, when it is determined in step S37 that the model fitting process is to be ended, the finally obtained temporary deformation model is determined as an individual model corresponding to the extraction target object. Thereby, a model of the extraction target object is generated. Thereafter, an area existing at the position of the finally obtained individual model (that is, the generated object model) is extracted from the measurement data as the object area.

  With the above processing, the processing in step S3 is completed.

  In such model fitting processing (step S3), the individual model corresponding to the extraction target region is created by gradually deforming the standard model SO by a minute amount L for each control point.

  Here, the external energy Fe and the internal energy Ge constituting the total energy Ue will be described.

  The external energy Fe is expressed as in Expression (2) using the square of the distance between each control point Cj and the corresponding point Qj corresponding to each control point Cj.

  Here, α is a constant, Nt is the number of control points, and | Cj−Qj | is the distance between the control point Cj and the corresponding point Qj.

  Since the temporary deformation model in which the external energy Fe becomes very large, that is, the temporary deformation model in which the distance between the control point Cj and the corresponding point Qj is larger than that before the movement, the total energy Ue becomes large. It becomes difficult to be employed in the above-described step S35 (step of employing movement of the control point that minimizes the total energy Ue). In other words, by considering the external energy Fe, a temporary deformation model is selected in which the distance between each control point Cj and the corresponding point Qj becomes smaller than before the movement (that is, each control point Cj approaches the corresponding point Qj). It becomes easy to be done.

  The external energy Fe is also expressed as an energy term reflecting the relationship between each control point Cj of the standard model SO and each corresponding point Qj in the measurement data. Alternatively, the external energy Fe is also expressed as an energy term that is optimized when the distance between each control point Cj of the standard model and each corresponding point Qj in the measurement data approaches. In short, the external energy Fe plays a role of bringing the shape of the standard model SO closer to the contour in the measurement data.

  Assuming that the internal energy Ge is connected to the control points Cj by virtual springs SPR (SPR1, SPR2, SPR3,...) As shown in FIG. It is expressed as follows.

  Where β is a constant, K is the spring coefficient of the virtual spring, Nh is the number of virtual springs, and w is the amount of displacement from the natural length of each virtual spring.

  According to Equation (3), excessive movement of each control point Cj is expressed as an increase in energy stored in the virtual spring SPR. For example, if one control point Cz moves to a certain point Vz and the relative displacement with other control points increases, energy from the extension of each virtual spring is stored in the virtual springs SPR1, SPR2, and SPR3. The internal energy Ge and thus the total energy Ue increases.

  In such a temporary deformation model with excessive deformation, the internal energy Ge increases and the total energy Ue also increases. Therefore, the above-described step S35 (step of adopting the movement of the control point that minimizes the total energy Ue). It becomes difficult to be adopted.

  In other words, the temporary deformation model that reduces the internal energy Ge and decreases the total energy Ue is selected in step S35, thereby obtaining an effect of preventing excessive deformation due to movement of each control point Cj. it can.

  That is, by introducing such internal energy Ge, the control point Cj can be moved without impairing the shape of the standard model SO, that is, the shape of each polygon constituting the standard model SO. When considering the internal energy Ge, a temporary deformation model in which each control point Cj moves in a direction different from the direction approaching the corresponding point Qj may be selected.

  In addition, there is a case where the corresponding point Qj is not correct because the boundary extraction process for obtaining the corresponding point Qj in step S31 is not accurate. In such a case, when only the external energy Fe is considered, a certain control point moves toward an inaccurate corresponding point, so that inaccurate model fitting is performed. On the other hand, considering the internal energy Ge, even in such a case, a deformation operation that does not impair the shape of the standard model SO is performed, so a more accurate modeling process is possible. become.

  This internal energy Ge is also expressed as an energy term that is optimized when the relationship between the control points Cj of the standard model SO approaches a certain relationship (that is, a state in which all virtual springs have a natural length). . In short, the internal energy Ge plays a role of maintaining the standard model SO in a certain shape.

  Next, the a priori knowledge term He will be described.

  The a priori knowledge term He is based on data (a priori knowledge) obtained in advance for an object of the same type as the measurement object. For example, the a priori knowledge term He is based on the existence probability of the object as follows. Determined.

  FIG. 15 is a diagram in which the existence positions of a plurality of (here, five) objects (organs) B1 to B5 of the same type obtained in advance are superimposed and FIG. 16 is an existence probability of the object of that type. It is a figure which shows map MP. The existence probability map MP indicates the existence probability of the object at each position. In FIG. 16, the existence probability is expressed by the density of points, and the existence probability of a relatively black portion is higher than that of a relatively white portion.

  The existence probability map MP (FIG. 16) relating to a certain object is obtained based on a priori data relating to the existence positions (see FIG. 15) of a plurality of objects of the same type. In the existence probability map MP of FIG. 16, as can be seen from comparison with FIG. 15, the existence probability is high at a position where a large number of objects exist among a plurality of objects B1 to B5 of the same type. The existence probability is low at a position where only a small number of objects are present among the objects B1 to B5. Here, for simplification, five objects are illustrated in FIG. 15, but in reality, it is possible to generate the existence probability map MP related to the object of that type based on a larger number of objects. preferable. Further, the existence probability map MP is shown in a planar manner in FIG. 15 and the like, but actually, a three-dimensional map (stereoscopic map) showing the existence probability for each point (pixel) in the three-dimensional space. As obtained.

  Then, an area where the object is assumed to exist based on a priori knowledge of the existence probability (in other words, the existence probability distribution of the object) at each position of the object (in other words, the existence area of the object) RB (referred to as “assumed region” hereinafter) RB (see FIG. 17) is obtained using the existence probability map MP of FIG. Here, in the existence probability map MP, an area composed of pixels whose existence probability is higher than a predetermined threshold value TH1 (for example, about 80% to 90%) is obtained as the assumed area RB. This threshold value TH1 may be determined so that, for example, the volume of the assumed region RB is the same as the average volume of a plurality of objects. In FIG. 17, the boundary of the assumed region RB is indicated by a solid line.

  FIG. 18 is a diagram showing a vector p (hereinafter also referred to as “approach vector”) p that approaches the boundary of the assumed region RB from each position at the shortest distance. Specifically, the approach vector p is a vector that starts from each pixel in the existence probability map MP and connects each pixel to the boundary of the assumed region RB with the shortest distance. In FIG. 18, only the approach vectors p for some representative points are shown for simplification of illustration, but in reality, the approach vectors p for all pixels are required. And A map that displays the approach vector p at each position as shown in FIG. 18 is also referred to as an “approach vector map VM”.

  FIG. 19 is a diagram showing an approach vector pj at each control point Cj (indicated by a white circle in FIG. 19) of the standard model SO arranged in the initial state. Each approach vector pj is obtained by directly adopting the approach vector p at the position of each control point Cj in the approach vector map VM of FIG.

  The a priori knowledge term He is expressed by, for example, Expression (4) using such an approach vector pj.

  However, γ represents a constant, u represents a vector representing the movement state (direction and amount of movement) of the control point Cj in step S32, and θj represents the angle between the vector u and the approach vector pj. (See FIG. 20). In addition, the symbol (·) means an inner product of vectors.

  As shown in Expression (4), when the direction of movement of each control point Cj and the direction of the approach vector pj are greatly different, the value (1-cos θj) becomes large (for example, when θj = 180 (deg)) (1-cos θj) = 2). For this reason, the a priori knowledge term He becomes a relatively large value, and as a result, the total energy Ue becomes large. In the temporary deformation model when each control point Cj moves in a direction significantly different from the approach vector pj in this way, the a priori knowledge term He increases and the total energy Ue also increases. It becomes difficult to be adopted.

  Conversely, when the direction of movement of each control point Cj is close to the direction of the approach vector pj, the value (1-cos θj) is small (for example, the value (1-cos θj) = 0 when θj = 0 (deg)). . Therefore, the a priori knowledge term He becomes small, and the total energy Ue becomes small. In particular, the a priori knowledge term He is optimized (minimized) when the direction of movement of each control point Cj matches the direction of the approach vector pj. In the temporary deformation model when each control point Cj moves in the direction closer to the approach vector pj in this way, the a priori knowledge term He is reduced and the total energy Ue is also reduced. It becomes easy to be adopted in the process of adopting the movement of the control point that minimizes the total energy Ue.

  In other words, the temporary deformation model that reduces the a priori knowledge term He and decreases the total energy Ue is selected in step S35, thereby causing the standard model SO to be deformed so as to approach the assumed region RB. Obtainable.

  Further, the a priori knowledge term He is also expressed as a term that is optimized when the standard model is deformed so as to coincide with the assumed region RB. This a priori knowledge term He is a term that is optimized when each control point Cj of the standard model SO moves in the direction approaching the boundary of the assumed region RB with the shortest distance (in short, the direction of the approach vector pj). It is also expressed as. By introducing such a priori knowledge term He, it becomes possible to move each control point Cj more stably using a priori knowledge.

  Further, in this embodiment, when the standard model SO is sequentially deformed, the approach vector pj in the a priori knowledge term He in the equation (4) is at the position of each control point Cj of the standard model after deformation. It shall be updated as needed. Specifically, after the operations of steps S32 to S36 (FIG. 11) are repeated to complete the movement of all control points, the approach vector pj is updated. Here, in step S39, the approach vector update operation is performed together with the corresponding point update operation. Since the approach vector p of each pixel is obtained in the approach vector map VM (see FIG. 18), each approach vector pj in the a priori knowledge term He in the equation (4) is each control point in the standard model after deformation. By using the approach vector p at the position of Cj (that is, the updated position after movement), it is updated as needed (see FIG. 21). Then, the a priori knowledge term He is calculated by the equation (4) using the updated approach vector pj. According to this, it becomes possible to prompt the deformation so as to approach the assumed region RB more accurately. That is, by considering the approach vector updated from time to time, model fitting can be performed by reflecting a priori knowledge with higher accuracy. FIG. 21 is a diagram showing the updated approach vector pj at each control point Cj after movement. In FIG. 21, the position of the standard model SO before movement is indicated by a broken line, and the position of the standard model SO after movement is indicated by a solid line.

  In the process of step S3 described above, the temporary deformation model that minimizes the total energy Ue expressed by the sum of the external energy Fe, the internal energy Ge, and the a priori knowledge term He optimizes the standard model SO. Is determined as a temporary deformation model (also referred to as “optimal deformation model”), that is, an individual model. In other words, a temporary deformation model that optimizes the total energy Ue as the evaluation function is determined as an individual model.

  According to the above processing, the standard model SO is deformed so as to minimize (optimize) the total energy Ue (evaluation function) including the a priori knowledge term He, and a model of the object is generated. The model can be transformed to reflect a priori knowledge. Therefore, a more accurate modeling process is possible. In addition, since an area existing at the corresponding position of the individual model obtained by the modeling process is extracted from the measurement data as an extraction target area, an appropriate area extraction process is realized.

  In addition, as described above, there is a case where the corresponding point Qj is not correct because the boundary extraction process at the time of obtaining the corresponding point Qj in step S31 is not accurate. In such a case, when only the external energy Fe is considered, a certain control point moves toward an inaccurate corresponding point, so that inaccurate model fitting is performed. On the other hand, by considering not only the external energy Fe but also the a priori knowledge term He, the deformation operation reflecting the a priori knowledge is performed even in such a case. A more accurate modeling process is possible.

<3. Modification>
Although the embodiments of the present invention have been described above, the present invention is not limited to the contents described above.

  In the above embodiment, when the standard model is sequentially transformed, each approach vector pj in the a priori knowledge term He in the equation (4) is updated as needed according to the movement of each control point Cj. However, it is not limited to this. For example, when the standard model is sequentially deformed, each approach vector pj with respect to the initial position of each control point Cj of the standard model SO is used as each control point as each approach vector pj in the a priori knowledge term He of Equation (4). A priori knowledge term He may be calculated without updating each approach vector pj in equation (4) even when the standard model is sequentially deformed for each Cj. Specifically, as each approach vector pj for each control point Cj, each approach vector pj for the initial position as shown in FIG. 19 may always be used (fixed) even after the control point Cj is updated. According to this, it is not necessary to prepare the approach vector p in advance for all the pixels in the existence probability map MP, and it is only necessary to obtain each approach vector pj at the initial position of each control point Cj of the standard model SO. Can be achieved.

  In the embodiment described above, the a priori knowledge term He that makes it easy for each control point Cj to move in the direction of the approach vector pj using Equation (4) is used, but the present invention is not limited to this. For example, a priori knowledge term He based not only on the direction of the approach vector pj but also on the magnitude of the approach vector pj may be used using the following equation (5).

  FIG. 22 is a flowchart showing an operation according to such a modification.

  This modification is different from the flowchart (FIG. 11) of the above embodiment in that each process of steps S51 to S54 is included between step S35 and step S36.

  Specifically, while incrementing the value k (for example, k = 2, 3,..., Nk), the control is performed in the same direction as the direction of movement corresponding to the temporary deformation model selected in step S35. A temporary deformation model obtained by moving the point by an integral multiple (k times) of the original movement amount | u | (= L) is obtained (step S51), and each total energy Ue of the temporary deformation model after each movement is calculated (step S51). The operation of S52) is repeated. Then, after it is determined (step S53) that this operation has been performed a predetermined number of times (for example, (Nk-1) times), a temporary deformation model that optimizes the total energy Ue is selected, and the selected temporary The control point is moved to a position corresponding to the deformation model (step S54). According to this, as shown in the equation (5), the a priori knowledge term He is minimized when a deformation having a size close to the size of the approach vector pj is performed, and the assumed region RB is stored in the assumed region RB. It can make it easier to approach quickly. That is, the processing speed can be increased.

  In the above embodiment, the total energy Ue (see equation (1)) represented by the sum of the three terms of the external energy term Fe, the internal energy term Ge, and the a priori knowledge term He is used as the evaluation function. However, it is not limited to this. For example, the total energy Ue represented by the sum of the external energy term Fe and the a priori knowledge term He may be used as the evaluation function without considering the internal energy term Ge.

  In the above-described embodiment, the case where the standard model is acquired from the model storage unit 11 in the region extraction device 1 is exemplified. However, the present invention is not limited to this, and other devices connected to the region extraction device 1 can be connected to the network. The “standard model” may be acquired via

It is a figure which shows the outline | summary of an area | region extraction apparatus. It is a block diagram which shows the various functions of an area | region extraction apparatus. It is a figure which shows the three-dimensional image acquired by X-ray CT apparatus, and the slice image of a predetermined position. It is a flowchart which shows the whole operation | movement of an area | region extraction apparatus. It is a flowchart which shows the temporary area | region extraction process using the area | region expansion method. It is a figure which shows the slice image in a predetermined position. FIG. 7 is an enlarged view near a pixel M1 on the slice image shown in FIG. 6. It is a flowchart which shows the initial position determination process using a gravity center point. It is a figure which shows the object contained in a three-dimensional image. It is a figure which shows a mode that a standard model is moved to an initial position. It is a flowchart which shows a model fitting process. It is a figure which shows the relationship between the standard model initially arranged, and an extraction object. It is a figure which shows a some slice image. It is the schematic diagram which connected between control points with the virtual spring. It is a figure which overlaps and shows the existing position of the same kind of several target object. It is a figure which shows an existence probability map. It is a figure which shows the assumption area | region assumed that a target object exists. It is a figure which shows an approach vector map. It is a figure which shows the approach vector pj in each control point of the standard model arrange | positioned in the initial state. It is a figure which shows the relationship between the vector u and the approach vector pj. It is a figure which shows the approach vector pj updated in each control point Cj after a movement. It is a flowchart which shows the operation | movement which concerns on a modification.

Explanation of symbols

1 region extraction device Cj control point Fe external energy term Ge internal energy term He a priori knowledge term Ue total energy VM approach vector map RB assumption region SO standard model p, pj approach vector

Claims (11)

A modeling device for generating a model of an object,
Obtaining means for obtaining a standard model for the object;
Model deformation means for deforming the standard model so as to optimize a predetermined evaluation function using a model fitting method, and generating a model of the object;
With
The predetermined evaluation function is an a priori knowledge term that is optimized when the standard model is deformed so as to coincide with an assumed area that is assumed as an existing area of the object based on a priori knowledge. A modeling apparatus comprising: The modeling device according to claim 1,
The modeling apparatus, wherein the assumed area is an area in which the existence probability of the object is higher than a predetermined value. The modeling device according to claim 1 or 2,
The a priori knowledge term is optimized when each control point of the standard model moves in the direction of an approach vector that is a vector that approaches the boundary of the assumed region at the shortest distance. Modeling equipment. The modeling device according to claim 3, wherein
When the model deforming means sequentially deforms the standard model, the approach vector is updated according to the position of each control point of the standard model after the deformation, and the a priori knowledge term is updated. A modeling apparatus, wherein the modeling apparatus is calculated based on the approach vector. The modeling device according to claim 3, wherein
The approach vector is preset for each control point according to the initial position of each control point,
Even when the model deforming means sequentially deforms the standard model, the a priori knowledge term is always the preset approach for each control point according to the initial position of each control point. A modeling apparatus characterized by being calculated based on a vector. The modeling device according to any one of claims 1 to 5,
The modeling apparatus according to claim 1, wherein the predetermined evaluation function has an external energy term that is optimized when the shape of the standard model approaches a contour in measurement data. The modeling device according to any one of claims 1 to 6,
The modeling apparatus, wherein the predetermined evaluation function has an internal energy term that is optimized when the relationship between the control points of the standard model approaches a certain relationship. The modeling device according to any one of claims 1 to 7,
Means for extracting from the measurement data the region corresponding to the position of the generated model of the object as the region of the object;
A modeling apparatus further comprising: A modeling method for generating a model of an object,
Placing a standard model for the object in an initial position;
Deforming the standard model to optimize a predetermined evaluation function using a model fitting technique, and generating a model of the object;
With
The predetermined evaluation function is an a priori knowledge term that is optimized when the standard model is deformed so as to coincide with an assumed area that is assumed as an existing area of the object based on a priori knowledge. The modeling method characterized by having. On the computer,
Placing a standard model for the object in an initial position;
Deforming the standard model to optimize a predetermined evaluation function using a model fitting technique, and generating a model of the object;
A program for executing
The predetermined evaluation function is an a priori knowledge term that is optimized when the standard model is deformed so as to coincide with an assumed area that is assumed as an existing area of the object based on a priori knowledge. The program characterized by having. An area extraction device for extracting an area of an object from measurement data,
Obtaining means for obtaining a standard model for the object;
Means for transforming the standard model to optimize a predetermined evaluation function using a model fitting technique;
Means for extracting from the measurement data a region corresponding to the position of the standard model after deformation as a region of the object;
With
The predetermined evaluation function has an a priori knowledge term that is optimized when the standard model is deformed so as to coincide with an assumed region in which an object is assumed to exist based on a priori knowledge. An area extraction device.