JP5943267B2 - Personal model data generation method, generation program, and generation system - Google Patents

Description

  The present invention generates personal model data for generating an internal tissue structure of an individual having different ages, body shapes, and the like from a high-resolution numerical human body model (standard numerical human body model) based on the anatomical structure of the human body. The present invention relates to a method, a generation program, and a generation system.

  A high-resolution numerical human body model having an anatomical structure of the human body has been developed for the purpose of evaluating exposure of radio waves and radiation to the human body. These numerical human models are created based on X-ray CT and MRI image data (hereinafter sometimes referred to as CT / MRI data or personal data) of a human body having an average body shape. Therefore, it is a standard numerical human body model (hereinafter sometimes referred to as a numerical human body model). Such a numerical human body model is used not only for evaluating the influence on a human body in a living body EMC, a car collision, etc., but also for determining an exposure position in radiation therapy in cancer treatment.

  As an example of a numerical human body model, there is a resolution of 2 mm, a number of tissue organs of 51, and a total number of voxels of about 8 million for an adult male and about 60.3 million for an adult female (Non-patent Document 1). These numerical human body models of adult men and women are created by, for example, coloring about 630 to 8 million voxels for each tissue organ by manual editing from MRI image data of the human body. Therefore, an annual period is required to develop such a numerical human body model as a whole.

  However, the body shape and internal organization structure are different for each age group such as infants, children, and adults, and are different for each individual. In other words, there are differences in body shape and internal organization structure due to age differences and individual differences.

  Therefore, when trying to determine the exposure position in radiation therapy among individuals of various ages based on the numerical human body model, the accuracy may be reduced due to the difference between the patient and the numerical human body model. Therefore, it is conceivable to create personal model data (individual numerical human body model) based on the CT / MRI data of the individual patient and determine the exposure position and the like using this personal model data.

  However, it is unrealistic to create individual model data in the same way as a numerical human body model because it takes a great deal of time and enormous costs. It is also impractical to develop a numerical human body model for each age group when evaluating the influence on the human body in biomedical EMC, automobile collision, etc. for each age group.

By the way, there has been proposed a technique for deforming the body shape of a numerical human body model and simulating the internal structure after deformation. For example, Free Form Deformation
(FFD) and the like (Non-Patent Document 2), and is used for living body EMC when the posture is changed. By using such a simulation technique, the body shape of the numerical human body model can be deformed in a short time. As a technique related to the deformation of the numerical model, a lattice generation technique after the deformation of the numerical model (also called a boundary fit method (BFM)) (Non-patent Document 3), and a body surface mesh after the deformation of the body shape are stored in a binary volume database. There is a voxelization technique using graphic hardware for conversion into (Non-Patent Document 4). Further, there is a technique related to an anatomical structure analysis method, an anatomical structure display method, and an anatomical structure display device (Patent Document 1).

  It is conceivable to generate personal model data with different ages, body shapes, etc. by utilizing the body shape transformation technology of the numerical human body model.

JP 2011-138350 A

National Institute of Information and Communications Technology website: http: // emc. nict. go. jp / bio / data / index. html Nagaoka and Watanabe "Human body posture model for electromagnetic wave measurement" Medical Bio 53, pages 7047 to 7061 2008 Co-authored by Thomson, Worge, and Mustin, “Basics and Applications of Lattice Generation,” New York Elsevier, North Holland, 1985 Fang and Chen "High-Speed Hardware Voxelization" Computer Graphic Magazine, No. 24, pages 433 to 442 2000

  However, by transforming the body shape of the numerical human body model into a personal body shape, differences due to age differences and individual differences cannot be reflected as the personal model data. Therefore, the internal structure of the individual is acquired as CT / MRI data, and the internal structure of a numerical human body model that has undergone body deformation is targeted to adjust the difference from the internal structure of the individual. By doing so, it can be considered as personal model data.

  If it does so, the error resulting from the difference in internal organization structure between a numerical human body model and personal model data can be reduced. However, these simulations have the following problems.

  The first problem is related to simulation automation. For example, if it is necessary to set the reference points manually for deforming the body shape, it takes time to set the reference points, and rapid treatment cannot be realized. Here, the reference point is a part to be a reference on the body surface, such as an eye, mouth, ear, joint, etc. (FIG. 9 shows an example in which the reference point SPn is manually set on the body surface. It is a figure.)

  However, if the reference point SPn is manually set, for example, when the mouth is used as the reference point, an error occurs in the simulation depending on where the mouth is used as the reference point SPn. In order to reduce the influence of such an error caused by the reference point setting, some parameter setting may be required. Thus, the quality of the setting accuracy of the reference point SPn affects the simulation accuracy and its reproducibility. In addition, depending on parameter settings, a mathematical solution for simulation may not be obtained.

  Therefore, there is no need to manually set the reference points SPn and parameters, and a simulation that can be automated and has high reproducibility is required.

  The second is the accuracy when the body surface of the numerical human body model is transformed into the body surface of an individual. This is because the internal tissue structure of the numerical human body model must be embedded in the body surface of the deformed numerical human body model to obtain the internal tissue structure after deformation. This is because there is a difference between the internal structure that has been formed and the internal structure that should be embedded.

  Here, body deformation is a deformation that three-dimensionally deforms the body surface of a numerical human body model and approximates the body surface of an individual (because it is body surface registration (hereinafter sometimes simply referred to as registration)). Improvement in registration accuracy leads to higher accuracy in embedding internal tissue structures.

  Third, the difference between the embedded internal tissue structure and the CT / MRI data must be adjusted.

  For example, even if the body surface of a numerical human body model of an adult woman can be deformed with high accuracy targeting the body surface of a 3-year-old girl, and as a result, the accuracy in embedding the internal tissue structure can be improved, the embedded internal tissue structure Is clearly different from that of a 3-year-old girl. Similarly, when the treatment target is an adult with obesity, the body surface of the numerical human model is transformed into the body surface of the adult with obesity to be treated, and the internal tissue structure is embedded, It is clear that the structure is different from that of obese adults. Therefore, in addition to highly accurate body surface registration, it is necessary to reduce the difference between the embedded internal tissue structure and the individual internal tissue structure.

  Therefore, the present invention has an object to realize a personal model data generation method, a generation program, and a generation system capable of automatically simulating personal model data with high accuracy, in a short time, based on a numerical human body model. It was.

  Means for solving the above problems will be described. The present invention constructs a numerical human body model by transforming a numerical human body model based on personal data (individual CT / MRI data). In general, personal model data is constructed by directly applying non-rigid registration. However, the non-rigid registration process is very expensive and it is difficult to construct a robust algorithm. Thus, as described below, the present invention efficiently constructs a highly accurate model by combining several methods step by step.

  In the present invention, a numerical human body model is modified step by step to approximate personal data. The first stage is a deformation process for the human body region of the whole body (hereinafter sometimes referred to as a body shape deformation process). Here, the whole human body region is an internal region surrounded by the body surface. At this stage, the body shape can be matched with high accuracy. The second stage is an approximation of the internal tissue structure.

  In body deformation processing, shape approximation of the human body region of the whole body, especially approximation of the body surface is mainly performed. Therefore, the body surface is once extracted as a high-definition polygon mesh. Then, approximation processing is performed by using the numerical human body model and the body surface of the personal data as mesh processing. Polygon mesh processing has been extensively studied in various fields (especially in the field of computer graphics), and various methods can be applied. The volume data is three-dimensional data, but the polygon mesh on the body surface is essentially two-dimensional, so there is an advantage that the processing load is light. After transforming the body surface of the numerical human body model to the body surface of the target personal data at the polygon mesh level, the volume data of the numerical human body model is mapped to an internal region surrounded by the body surface. This mapping is obtained by applying a grid generation technique in the numerical calculation field. This technique is called volume refilling in the present invention.

  As a result of the body shape transformation process, the shape of the body surface coincides with high accuracy, the internal tissue structure of the human body is continuously and smoothly mapped, and a human body model having an individual body shape is obtained. However, since the shape of the individual internal tissue is generally different from the shape of the mapped internal tissue, the shape of the internal tissue is matched (or approximated) in order to use the mapped data as the personal model data. ) Processing is required. This second stage process is called a tissue shape deformation process.

  The tissue shape deformation process approximates the internal tissue structure by applying a fluid-based non-rigid registration to the numerical human body model deformed so as to approximate personal data and the personal data. Non-rigid registration is a global optimization process and is highly dependent on the initial value. The numerical human body model having the individual body shape obtained by the body shape deformation has a higher degree of similarity to the target individual data than the original numerical human body model. Therefore, it can be expected that the initial value of the non-rigid registration is better than the original data.

  According to the above-described personal model data generation method, generation program, and generation system according to the present invention, simulation of personal model data based on a numerical human body model can be performed without manually setting a reference point for performing body shape deformation. By performing in stages, it is possible to perform automatically, with high accuracy, in a short time and with high reproducibility.

An example of generating personal model data (c) from the source model (a) and the target model (d) by performing body shape deformation processing on the source model to perform volume refilling (b) and further performing tissue shape deformation processing. Each of the drawings shows a plurality of internal tissues (built-in tissues) of the human body as images. It is a figure which shows the voxel percentage of the muscular tissue (Muscle) of each model shown in FIG. 1, a fat tissue (Fat), and another structure | tissue (Other). It is a flowchart of the production | generation of the personal model data concerning this invention. 4 is a detailed flowchart of internal tissue registration in FIG. 3. It is a figure which shows an example of the system for implementing this invention. It is a figure which shows the example (b) which applied the Laplacian twice to the polygon mesh (a), and the example (c) which applied Laplacian four times. FIGS. 7A to 7C each illustrate an image of the deformation of the muscular tissue in the human body. Similarly to FIG. 7, FIGS. 8A to 8C exemplify the deformation of other tissues in the human body (a tissue other than muscle and fat treated as one tissue) as an image. . It is a figure which shows the example when setting a reference point manually on the body surface.

  Hereinafter, modes for carrying out the present invention will be described. Here, after describing the processing for carrying out the invention, the embodiment will be described.

<Body shape processing>
The body deformation process extracts the numerical human body model and the body surface of the personal data as a polygon mesh, deforms the body surface mesh of the numerical human body model to approximate the body surface mesh of the personal data, and finally deforms the numerical value The numerical human body model is mapped using volume refilling inside the body surface of the human body model.

  Here, volume refilling will be described first, and then polygon mesh deformation will be described as application processing for applying it to body shape deformation processing.

<Volume refilling>
Volume refilling was developed to realize the transformation of numerical human body model data by deformation of the body surface. By this method, posture deformation and body shape deformation can be controlled by deformation of the body surface. The contents of this method are as follows.

  A point on the body surface can naturally give a position coordinate and an index (called a texture coordinate) of a voxel including the point. When the body surface is deformed, the position coordinates of the points change, but the texture coordinates do not change. That means, for points on the body surface, this texture coordinate can be used to sample the voxels of a numerical human model. That is, a numerical human body model can be mapped to the deformed body surface. Therefore, if the texture coordinates can be generated for the inside of the body surface, the numerical human model can be mapped.

  The above method is mathematically formulated as follows.

The source volume data set and the target volume data, which are numerical human models, are assumed to be V0 and V1, respectively. These are a subset of the three-dimensional Euclidean space E3. The human body regions included in the respective volume data are D0 and D1. It is assumed that D0 and D1 have the same topology, that is, D0 can be continuously transformed into D1. Let ∂D0, ∂D1 be the boundary between D0 and D1. (∂D0 assumes the body surface before deformation of the numerical human body model, and ∂D1 assumes the body surface after deformation.)
Now, given a continuous mapping g (Equation 1) from ∂D1 to ∂D0, find a continuous mapping G (Equation 2) from D1 to D0 that matches g on ∂D1 (Equation 3). .

Since D0, D1, etc. are a subset of the three-dimensional Euclidean space E3, the mappings g and G can be expressed as a set of three component functions (Equations 4 and 5).

と表せる。これは、Dirichletの境界条件になる。そこで、次の条件

を（式６）、（式７）、（式８）に追加して、Giを求めることができる。このような解Giは、必ずユニークに存在する。これにより、Gを決定することができる。 From (Equation 1) to (Equation 3), i = 0,1,2

It can be expressed. This becomes the boundary condition of Dirichlet. Therefore, the following conditions

Can be added to (Equation 6), (Equation 7), and (Equation 8) to obtain Gi. Such a solution Gi always exists uniquely. Thereby, G can be determined.

In practice, (Equation 8) and (Equation 9) are discretized to obtain G. Since (Equation 9) is an elliptic partial differential equation and linear, this discretized problem is to solve a large-scale simultaneous linear equation. Here, for example, the SOR method which is an iterative solution method that is simple to implement can be used.

<< Approximate deformation processing of polygon mesh >>
<Extracting polygon mesh on body surface>
A polygon mesh representing the body surface is extracted from the volume data of the numerical human body model. In the present invention, the voxel boundary of the human body region is extracted as a polygon. The vertex coordinates are the voxel vertex coordinates, and coordinates normalized by the volume resolution are added as texture coordinates.

このように抽出したメッシュは階段状の形状を持つが、これはメッシュの変形の際に自己交差をしたり、メッシュ間の最短距離を計算するときに好ましくない状況を引き起こすので、それらの問題を防ぐためにメッシュのラプラシアンに基づくフェアリング処理を行い滑らかにしておく。

The mesh extracted in this way has a stepped shape, but this causes self-intersection when the mesh is deformed and causes undesirable situations when calculating the shortest distance between meshes. In order to prevent this, a fairing process based on the Laplacian of the mesh is performed and smoothed.

  Next, the target personal data mesh is extracted. Again, the mesh is extracted by the same method as for the numerical human body model. An approximate deformation process is performed using the mesh thus obtained.

<Local approximation processing>
At this time, if the two meshes are sufficiently similar and the distance between the meshes is close, a highly accurate approximation process can be performed by a local approximation process. Here, the local approximating process refers to a process of searching for the closest point among the mesh vertices of the personal data for all the mesh vertices of the numerical human body model and moving to that point.

しかしながら、この局所的な近似処理では、適切に近似できない場合は、その前処理として大域的な近似処理が必要である。大域的な近似処理は、メッシュの点ごとではなく、ある程度メッシュの頂点が連動しながら変形して、局所的な近似処理がうまく動作する程度になるまでの変形を行う。大域的な近似処理を自動的に行う手法として、一度メッシュをボクセル化して低解像度でレジストレーションを実行してその変形情報である変位ベクトルを求め、それを利用してメッシュを変形する。

However, if this local approximation process cannot be approximated appropriately, a global approximation process is required as a pre-process. The global approximation processing is not performed for each point of the mesh, but is performed until the vertexes of the mesh are deformed while interlocking to some extent, and the local approximation processing is performed well. As a method of automatically performing global approximation processing, a mesh is once voxeled, registration is performed at a low resolution to obtain a displacement vector as deformation information, and the mesh is deformed by using it.

<Registration-based global approximation>
We are also considering an automatic method that transforms global approximation processing by applying full algorithm processing by applying 3D registration. This generates volume data having a human body region and two other regions, and performs registration. Using the displacement vector field of volume data deformation obtained as a by-product of the result, the polygon mesh is deformed to obtain a global approximation. The registration method will be described in detail in the tissue shape deformation process described later.

  Since the three-dimensional registration at this stage is intended to approximate the shape of the human body region, the internal region information is not necessary. Moreover, since global approximation is intended, high-precision approximation may be performed by local approximation of a polygon mesh. Therefore, registration is performed with a binary volume representing a human body shape having a low resolution.

<Binary voxel with boundary condition of deformed numerical human model mesh>
In order to apply volume refilling, it is necessary to voxel the polygon mesh. The volume data has two types of information. (A) a binary volume defining a deformed human body region;
(A) A volume obtained by voxelizing a boundary mesh that defines boundary conditions. This voxel value is a three-dimensional texture coordinate.

  Using these data, the deformed numerical human body model is obtained by sampling the numerical human body model using the three-dimensional texture coordinates for sampling generated from (a) for the voxels inside the human body region determined in (a). Is generated.

<Summary of body deformation processing flow>
The processing flow of body deformation is summarized.
(A) A body surface is extracted as a polygon mesh from a numerical human body model and volume data of personal data.
(B) The two meshes are brought closer to each other by global approximation processing of the polygon mesh. At this time, a landmark base or a registration base is selected.
(C) A local approximation process is performed to approximate the mesh of the numerical human body model to the mesh of personal data with high accuracy.
(D) After transforming the deformed numerical human mesh into a binary volume, volume refilling is applied to fill the interior.

<< Tissue shape deformation process >>
For the internal structure of the deformed numerical human body model (hereinafter sometimes referred to as a deformed human body model) obtained by body deformation, a process for approximating the internal structure of personal data with higher accuracy is performed. Therefore, a flow-based three-dimensional registration process is applied. The registration process is premised on the same organization having the same voxel value. However, it is common that the numerical human body model and personal data have different voxel values even in the same organization. Therefore, it is necessary to easily identify the organization of personal data.

<Simple organization identification of personal data>
In order to apply registration, the same organization needs to have the same voxel values for the deformed human body model and the personal data. It is none other than identifying all tissues, but this goes against the goal. Therefore, the identification is limited to only the tissues important for simulation in the human body. Regarding these tissues, the values in the CT / MRI data are relatively easily separated, and thus are identified by simple threshold processing. The important tissues in the simulation are fat, muscle, and skeleton. Other organizations are considered as a virtual organization.

<Flow-based registration algorithm>
In the present invention, a flow-based algorithm is adopted as a non-rigid registration method.

  Flow-based registration was developed for the purpose of achieving large deformations based on a hydrodynamic model of the registration process. Flow-based registration interprets the differential energy of the image as the pressure potential of the fluid. Then develop it in time steps. First, registration based on the image difference for one hour step will be described. Next, flow-based registration will be described.

<Registration based on differential energy of image>
Let I0 and I1 be source and target images, the mapping of deformation is f, and the displacement vector field generated by the mapping is v. The relationship between f and v is given by

レジストレーションの目的は、ｆでI0を変形した画像I0・ｆについて、変形した画像I0・ｆとI1の差分を小さいこと、ｆを定めるｖが滑らかでなるべく小さいこととなるようなｆを求めることである。

The purpose of the registration is to find a difference f between the deformed images I0 · f and I1 with respect to the image I0 · f obtained by transforming I0 with f, and to determine f that makes f as smooth as possible. It is.

によってｆで変形した画像I0を定義する。 Here, when the image I0 is volume data,

Defines an image I0 transformed by f.

とする。E1が小さいときに、画像の差分も小さくなる。したがって、画像が一致するためには、E1を最小にすることが必要である。一方、変化が小さいｖの中で滑らかで最小のものを選ぶために、ベクトル場のエネルギー

を加える。ここでLは、ベクトル場を滑らかにする作用のある微分作用素とする。 At this time, the difference energy between the images I0 · f and I1 obtained by transforming I0 with f is

And When E1 is small, the image difference is also small. Therefore, it is necessary to minimize E1 in order for the images to match. On the other hand, the energy of the vector field is used to select the smoothest and smallest of the v with small changes.

Add Here, L is a differential operator having an effect of smoothing the vector field.

と記述する。
（１３）を実現するｖは、

が成り立つことが必要である。（１４）の左辺は、

であり、（１４）の右辺は

である。ここで

である。
（１４）（１５）（１６）から

ここで、（１７）を離散化すると、連立一次方程式となり、数値的に解であるｖを求めることができる。 E1 and E2 are combined into an energy function E. At this time, the registration is to obtain f that minimizes E. Considering that f is now determined by v, obtaining f is also obtaining v. v minimizes E

Is described.
V realizing (13) is

It is necessary to hold. The left side of (14) is

And the right side of (14) is

It is. here

It is.
(14) From (15) (16)

Here, when (17) is discretized, it becomes a simultaneous linear equation, and v, which is a numerical solution, can be obtained.

また、（１２）における微分作用素LをNavier-Stokes方程式における作用素とする。すなわち、

とする。これを（１７）に代入し、

となる。 <Flow-based registration>
In the flow-based algorithm, the image difference in the registration based on energy is regarded as the potential of pressure, and the gradient is used as the pressure. When the pressure vector is p,

Further, the differential operator L in (12) is an operator in the Navier-Stokes equation. That is,

And Substituting this into (17)

It becomes.

  The above process is repeated as one time step process, and the process is repeated.

The algorithm is as follows.

<Multi-channel registration and integration>
At this time, the volume data is not registered as it is, but a binary volume is generated for each organization and is registered. As a result, a result of registration for each tissue is obtained.

しかしながら、実用上はレジストレーション誤差が発生してしまう。レジストレーションの誤差発生の原因は、おもに組織のトポロジーが異なっていることに起因する。レジストレーションは連続変形であるが、連続変形によって組織を一致させるためには、組織のトポロジーが一致していることが条件になる。しかし、実データ上の組織形状およびトポロジーは複雑であり、一致していることは期待できない。したがって、レジストレーションで得られる結果は、ベストエフォートである。レジストレーション誤差のため、各組織のレジストレーション結果は、（式２１、式２２）を満たさない。これは、組織間の間に隙間が生じたい、組織が重なることを意味する。 In principle, the organization of personal data is a set that does not intersect, so the union of each organization matches the entire volume data. Therefore, if there is no registration error in the registration results for each tissue, volume data can be obtained by simply taking the set sum of the tissue regions. That is, the following set of equations holds for both personal data and the result of registering a numerical human body model.

However, in practice, a registration error occurs. The cause of registration error is mainly due to the difference in the topology of the tissue. Registration is a continuous deformation, but in order for the tissues to be matched by the continuous deformation, the topology of the tissues must be matched. However, the structure and topology of the actual data are complex and cannot be expected to match. Therefore, the result obtained with registration is best effort. Due to the registration error, the registration result of each tissue does not satisfy (Equation 21 and Equation 22). This means that there is a gap between the tissues, and the tissues overlap.

  Therefore, the integration corrects gaps and overlaps based on some systematic rules. For example, the gaps are filled with adipose tissue, and for overlap, internal tissue data is overwritten so that the skeleton takes precedence over muscle, the muscle takes precedence over other tissues, and the other tissue takes precedence over fat.

  Each of the above corrections is an exemplification, and does not exclude filling the gap with a tissue other than the fat tissue. Further, the priority other than the above is not excluded for the correction of the overlap.

  An embodiment of a method for generating personal model data according to the present invention will be described. At the same time, an embodiment of the personal model data generation program and generation system will be described.

  In the description of the embodiment, a Japanese adult male model (hereinafter sometimes referred to as a source model) is used as a numerical human body model, and a Japanese adult female model (hereinafter referred to as a target model) as an individual to be deformed. May be used). The reason for using a Japanese adult male model as the source model and a Japanese adult female model as the target model is that the amount of muscle tissue and adipose tissue differ greatly, and the necessity of tissue shape deformation processing is necessary. It is because it can show notably.

  Prior to the description of the embodiment, FIGS. 1 to 4 will be described.

  FIG. 1 shows that the source model (a) and the target model (d) are subjected to body shape deformation processing and volume refilling (b), and further subjected to tissue shape deformation processing to obtain personal model data (c). A generated example is shown.

  The voxel percentages of muscle tissue (Muscle), adipose tissue (Fat), and other tissues (Other) in each model shown in FIGS. 1 (a) to (d) are the four bar graphs VP1 to VP4 in FIG. Is shown in In the bar graphs VP1 to VP4, the lower area indicates the voxel percentage of the muscle tissue, the middle area indicates the voxel percentage of the adipose tissue, and the upper area indicates the voxel percentage of the other tissues. For example, in the bar graph VP1 showing the voxel percentage of the source model, the muscle tissue is 39.76%, the fat tissue is 29.63%, and the other tissues are 30.61%.

  FIG. 3 is a flowchart of generating personal model data according to the present invention, and FIG. 4 is a flowchart of internal organization registration. Hereinafter, the present invention will be described focusing on these flowcharts.

<Polygon mesh extraction>
The first stage of processing is performed from polygon mesh extraction (STEP 10). Here, the body surface is extracted as a polygon mesh (numerical human model polygon mesh) from the volume data of the source model (STEP 11). The body surface is extracted from the volume data of the target model as a polygon mesh (individual polygon mesh) (STEP 12).

  Such polygon mesh extraction is performed in a process of performing a calculation based on a known algorithm or the vertex coordinate addition algorithm shown in the above-mentioned “polygon mesh extraction of body surface”. These operations can be executed by a computer program.

  FIG. 5 shows an example of a system for implementing the present invention (hereinafter sometimes referred to as the present system). The polygon mesh extracting means in this system can be realized by the computer 10 that stores the computer program. Here, the storage of the computer program may be temporary storage as well as permanent storage, and further the computer program may be transmitted via a communication line for temporary storage. Good (the same applies to the storage of computer programs).

  The computer 10 transmits a source database 11 storing the volume data SD of the source model, a target database 12 storing the volume data data TD of the target model, and generated personal model data via a communication line such as a LAN as necessary. It is connected to a personal model database 13 for storing PD. The system may be connected to an image monitor 14, a medical device such as a printer (not shown), or an X-ray CT apparatus as necessary.

  Both extracted polygon meshes are smoothed by applying a fairing process (STEP 13 and STEP 14) applying Laplacian. FIG. 6 is a diagram showing an example in which the body surface is smoothed by applying a fairing process to the polygon mesh (a), and FIG. 6B is an example in which the Laplacian is applied twice, and FIG. Is an example of applying Laplacian four times.

<Body surface registration>
In the body surface registration (STEP 20), one or both of global approximation (STEP 21) and local registration (STEP 22) is selected and performed.

  The global approximation is a non-rigid registration performed by minimizing the differential energy between the source model mesh shape image and the target model mesh shape image. The global approximation is performed in a step of performing an operation based on the algorithms (Equation 10 to Equation 20, etc.) shown in the aforementioned “registration-based global approximation process” and “tissue shape deformation process”. These operations can be executed by a computer program.

  Local registration searches all the mesh vertices of the source model for the closest point in the mesh vertices of the target model, and moves the mesh vertices of the numerical human model to the closest point Registration. The local registration is performed in a step of performing an operation based on the algorithm shown in the above-mentioned “local approximation process”. These operations can be executed by a computer program.

  Here, the local registration is a registration that moves the mesh vertices, so even if it is automated, it always converges, and it is not necessary to set parameters manually, so that the reproducibility of computation is high. The global approximation minimizes the energy of the difference between the images in the shape of both meshes, so it always converges even if it is automated, and there is no need to set parameters manually, and the reproducibility of calculations is high.

  Therefore, means for automatically executing global approximation and local registration can be realized by the computer 10 of the present system shown in FIG. 5 (hereinafter, the mesh of the source model deformed by registration is transformed into a deformed human body). Sometimes referred to as model mesh.)

  The mesh of the source model is shown as an outline of its body shape (having male characteristics) in FIG. Similarly, the deformed human body model (FIG. 1B) shows the contour of the body shape (having female characteristics) on the same figure. The body surface shape of the generated deformed human body model thus approximates the body surface shape of the target model with high accuracy.

<Volume refilling>
In volume refilling (STEP 30), the mesh of the deformed human body model is binarized (STEP 31), and then volume refilling is applied to embed the internal tissue structure in the mesh of the deformed human body model (STEP 32). The internal tissue thus embedded may be referred to as the internal structure of the deformed human model, and the structure may be referred to as the internal structure of the deformed human model.

  The volume refilling is performed in a step of performing an operation based on the algorithm (Equation 10 to Equation 20 etc.) shown in the above-mentioned “volume refilling”. These operations can be executed by a computer program.

  Here, as described above, since volume refilling always has a unique solution as described above, it automatically converges even if it is automated, and there is no need to manually set parameters, and the reproducibility of computation is high. Therefore, the means for automatically executing volume refilling can be realized by the computer 10 of the present system shown in FIG.

  In this way, as shown in FIG. 1B, the male internal tissue structure as the source model is embedded in the mesh of the deformed human body model (the female body surface mesh as the target model).

  However, as shown in the bar graph VP2 in FIG. 2 showing the internal tissue voxel percentage of the deformed human model, the deformed human model has 40.91% muscle tissue, 30.50% fat tissue, and other tissues. 28.59%.

  Comparing these with the bar graph VP4 of FIG. 2 showing the voxel percentage in the target model (30.45% muscle tissue, 41.44% adipose tissue, and 28.11% other tissue), the deformed human body model and the target model are: While there is a difference in their voxel percent, the voxel percent of the deformed human body model is that of the source model (39.76% for muscle tissue, 29.63% for adipose tissue and 29.63% for other tissues shown in bar graph VP1). Close to 30.61%). That is, the internal organization structure of the deformed human body model is closer to the source model than the target model.

  For example, FIG. 7 is a diagram illustrating an example of tissue shape deformation processing in a muscle tissue, which will be described later. The model in FIG. 7A is a male mesh on a deformed human body model (a model deformed into a female model body shape). Only the muscle tissue of the model is embedded. Here, the muscle tissue structure of the deformed human body model (for example, the thigh is strong and masculine) is more than the muscle tissue structure of the target model FIG. Close to the muscle tissue structure of the source model (male).

  This is the “first stage processing”.

  The internal tissue registration that approximates the internal tissue structure of the deformed human body model that is still approximated to the source model as described above is the “second stage process” described below.

<Internal organization registration>
In the second stage process (which is the above-mentioned “tissue shape deformation process”), simple tissue identification limited to tissues (for example, fat, muscle, and skeleton) that are important for simulation in the human body is performed (STEP 41). Regarding these tissues, the values in the CT / MRI data are relatively easily separated, and can be identified by simple threshold processing (there is no factor that hinders automation). Here, the other organizations are collectively identified as other organizations and identified as virtual organizations (also STEP 41).

  In the numerical human body model (Non-Patent Document 1) having 51 tissue organs, 51 tissue organs are assigned IDs, and when the tissue identification is performed, the IDs of the identified tissues are attached. (For other organizations, a new ID is assigned.)

  Internal tissue registration can be performed by the same algorithm as the global approximation in body surface registration (STEP 42). That is, it is performed in a step of performing an operation based on the algorithms shown in the above-mentioned “registration-based global approximation process” and “tissue shape deformation process”. These operations can be executed by a computer program.

  Since such global approximation always converges even if it is automated as described above, it is not necessary to set parameters manually, and the reproducibility of computation is high. Therefore, means for automatically executing internal tissue registration can be realized by the computer 10 of the present system shown in FIG.

  For example, when the tissue shape deformation process is performed on the muscle tissue, the internal tissue resist is further added to the muscle tissue structure of the model of FIG. 7A generated by embedding only the muscle tissue of the source model in the mesh of the deformed human body model. The target model is approximated to the muscle tissue structure of FIG. 7C (see the model of FIG. 7B).

  This muscle tissue registration changes the muscle tissue voxel percentage 40.91% in the deformed human body model (see the bar graph VP2 in FIG. 2) to 30.85% (see the bar graph VP3 in FIG. 2), and the target model Approximate the voxel percentage 30.45% (see bar graph VP4 in FIG. 2)) of the muscle tissue structure of (female model).

  Similarly, a tissue shape deformation process for a fat tissue is performed. This process changes the adipose tissue voxel percentage 30.50% (see the bar graph VP2 in FIG. 2)) to 42.12% (see the bar graph VP3 in FIG. 2) in the deformed human model, and the target model (female model). ) Fat tissue structure percent voxel percent 41.44% (see bar graph VP4 in FIG. 2)).

  FIG. 8 is a diagram illustrating an example of a tissue shape deformation process in another tissue. The model shown in FIG. 8A is obtained by embedding only the other organization of the source model in the mesh of the deformed human body model, and the model shown in FIG. 8C shows the other organization structure of the target model. The model in FIG. 8B is a model obtained by applying internal tissue registration to the other tissues in the model in FIG. 8A to approximate the other tissues in the target model.

  By this process, the voxel percentage 28.59% of other tissues in the deformed human body model (see the bar graph VP2 in FIG. 2) is changed to 27.03% (see the bar graph VP3 in FIG. 2), and the other in the target model The voxel percentage of the tissue structure is 28.11% (see the bar graph VP4 in FIG. 2)).

<Multi-channel registration and integration>
In the internal organization registration (STEP 40), multi-channel registration and integration processing are further performed. This is because, as described above in “Multi-channel registration and integration processing”, a gap or an overlap between internal tissues occurs between internal tissues.

  Therefore, as shown in “Multi-channel registration and integration processing”, the gap between the internal tissues is filled (STEP 43), and the internal tissues are overwritten in the same priority order as illustrated to correct the overlap between the internal tissues (STEP 44). .

  Since the gap between the internal tissues can always be automatically detected as an area where there is no data of a plurality of internal tissues to be processed, adipose tissue can be embedded with high reproducibility. In addition, since an overlap between internal tissues can always be automatically detected as a region where data of a plurality of internal tissues to be processed overlap, it is possible to perform a process of overwriting with a preferred internal structure with high reproducibility. Personal model data is generated through such multi-channel registration and integration processing.

  Such embedding processing and overwriting processing can be executed by a computer program, and means for automatically executing the embedding processing and overwriting processing can be realized by the computer 10 of this system shown in FIG.

  In addition, this invention is not limited to the Example mentioned above, It can deform | transform and implement suitably, without changing the meaning.

  The personal model data generation system and generation program according to the present invention is a target for manufacturing and sales in the medical device industry, and can be used in medical research, medical practice, and the like. The present invention has economic value and industrial applicability because it can be used in medical research and medical practice.

STEP 10 Polygon mesh extraction step STEP 11 Numerical human body model polygon mesh extraction step STEP 12 Personal polygon mesh extraction step STEP 13 Numerical human body model polygon mesh fairing processing step STEP 14 Personal polygon mesh fairing processing step STEP 20 Body surface registration step STEP 21 Global approximation ( (Non-rigid registration) step STEP22 Local registration step STEP30 Volume refilling step STEP31 Binary volume creation step STEP32 of deformed human body model Internal tissue structure embedding step STEP40 of deformed human body model Internal tissue registration step STEP41 Simple Tissue identification process STEP42 Internal tissue registration process STE 43 internal tissue between the gap filling step STEP44 overlap correction process between internal tissue

Claims (6)

A personal model data generating method for generating a personal internal organization structure by computer simulation based on a numerical human body model having an internal organizational structure,
The computer extracts the body surface of the numerical human body model as a numerical human body model polygon mesh based on the volume data of the numerical human body model stored in the source database, and the personal volume data stored in the target database. A polygon mesh extracting step of extracting the individual's body surface as a personal polygon mesh based on
A body surface registration step in which the computer transforms the mesh of the numerical human body model into a deformed human body model mesh that approximates the mesh of personal data;
A volume refilling step in which the computer voxels the mesh of the deformed human body model and then embeds an internal tissue structure based on the volume data of the numerical human body model to generate an internal tissue structure of the deformed human body model;
The computer approximates the shape of one or more internal tissues constituting the internal tissue structure of the deformed human body model to the shape of the one or more individual internal tissues of the individual corresponding to the internal tissue of the deformed human body model. by, by modifying the internal organizational structure of the deformable body model, the method of generating the personal model data and having an internal tissue registration process of the internal tissue structure of the individual. The deformation of the numerical human body mesh in the body surface registration process is as follows :
Non-rigid registration performed by minimizing energy difference between the image of the mesh shape of the numerical human body model and the image of the mesh shape of the personal data, and all the mesh vertices of the numerical human body model On the other hand, one or both of local registration is performed by searching for the nearest point among the vertexes of the mesh of the personal data and moving the mesh vertex of the numerical human body model to the nearest point. Is,
The approximate deformation of the shape of the internal tissue of the deformed human body model to the shape of the internal tissue of the individual in the internal tissue registration step includes an image of the shape of the internal tissue of the deformed human body model and the shape of the internal tissue of the individual. The method of generating personal model data according to claim 1, wherein non-rigid registration is performed to minimize the energy of the difference between the image and the image. A personal model data generation program for causing a computer to execute a procedure for generating an internal organization structure of an individual by simulation based on a numerical human body model having an internal organization structure,
Based on the volume data of the numerical human body model stored in the source database, the body surface of the numerical human body model is extracted as a numerical human body model polygon mesh, and based on the volume data of the individual stored in the target database, A polygon mesh extraction procedure for extracting the individual's body surface as a personal polygon mesh;
A body surface registration procedure that transforms the mesh of the numerical human body model into a deformed human body model mesh that approximates the mesh of personal data;
Volume refilling procedure for generating an internal tissue structure of the deformed human body model by embedding an internal tissue structure based on the volume data of the numerical human body model after voxelizing the mesh of the deformed human body model,
By approximating the shape of one or more internal tissues constituting the internal tissue structure of the deformed human body model to the shape of one or more individual internal tissues of the individual corresponding to the internal tissues of the deformed human body model, the deformation by modifying the internal tissue structures of the human body model, individual model data generation program characterized that you have an internal tissue registration procedure to internal tissue structure of the individual. The deformation of the numerical human body model mesh in the body surface registration procedure is as follows :
Non-rigid registration performed by minimizing energy difference between the image of the mesh shape of the numerical human body model and the image of the mesh shape of the personal data, and all the mesh vertices of the numerical human body model On the other hand, one or both of local registration is performed by searching for the nearest point among the vertexes of the mesh of the personal data and moving the mesh vertex of the numerical human body model to the nearest point. Is,
The shape of the internal tissue of the deformable body model in the internal tissue registration procedure, approximates the deformation in the shape of the inner tissue of the individual,
The non-rigid registration that minimizes a difference energy between an image of an internal tissue shape of the deformed human body model and an image of an internal tissue shape of the individual. Personal model data generation program. A personal model data generating system for generating a personal internal organizational structure by computer simulation based on a numerical human body model having an internal organizational structure,
Based on the volume data of the numerical human body model stored in the source database, the body surface of the numerical human body model is extracted as a numerical human body model polygon mesh, and based on the volume data of the individual stored in the target database, Extracting the person's body surface as a personal polygon mesh, the polygon mesh extracting means of the computer ,
A body surface registration means included in the computer, wherein the mesh of the numerical human body model is transformed into a deformed human body model mesh approximated to a personal data mesh;
Volume refilling means possessed by the computer for generating the internal tissue structure of the deformed human body model by embedding the internal tissue structure based on the volume data of the numerical human body model after voxelizing the mesh of the deformed human body model,
By approximating the shape of one or more internal tissues constituting the internal tissue structure of the deformed human body model to the shape of one or more individual internal tissues of the individual corresponding to the internal tissues of the deformed human body model, the deformation by modifying the internal tissue structures of the human body model, the internal tissue structure of the individual product system individual model data and having an internal tissue registration means the computer has. The deformation of the mesh of the numerical human body model in the body surface registration means is as follows :
Non-rigid registration performed by minimizing energy difference between the image of the mesh shape of the numerical human body model and the image of the mesh shape of the personal data, and all the mesh vertices of the numerical human body model On the other hand, one or both of local registration is performed by searching for the nearest point among the vertexes of the mesh of the personal data and moving the mesh vertex of the numerical human body model to the nearest point. Is,
The approximate deformation of the shape of the internal tissue of the deformed human body model in the internal tissue registration means to the shape of the internal tissue of the individual is :
The non-rigid registration that minimizes energy difference between an image of an internal tissue shape of the deformed human body model and an image of an internal tissue shape of the individual according to claim 5. A system for generating personal model data.

Images