JP5641503B2 - Multi-step lattice voxel method - Google Patents

Description

  The present invention relates to a method for forming a three-dimensional model to be photographed by stacking slice data of photographed images including voxels having different sizes, and a computer program. In particular, the present invention relates to a multi-step lattice voxel method capable of speeding up Monte Carlo transport calculation for a human body model.

  Boron Neutron Capture Therapy (BNCT), a next-generation radiotherapy, administers boron drugs that selectively accumulate in cancer cells to the patient and irradiates the lesion with neutrons only to the cancer cells. It is a tumor cell-selective particle radiotherapy that selectively destroys tumor cells. In order to carry out radiation therapy, a treatment plan that can accurately evaluate the dose given to the lesion and its surroundings by radiation irradiation by simulation calculation and determine the optimum irradiation condition is indispensable. In particular, BNCT has a very complex neutron behavior, so it cannot be calculated with simple calculation algorithms such as conventional X-ray therapy and particle beam therapy. is necessary. The present inventor has so far researched and developed this Monte Carlo treatment planning system for BNCT and put it into practical clinical research (Non-patent Document 1: H. Kumada, et al., Development of JCDS, a computational). dosimetry system at JAEA for boron neutron capture therapy, Journal of Physics: Conference Series, Vol. 74, pp.1-7 2007).

  FIG. 1 shows an outline of a treatment planning system JCDS developed by the present inventor. “JCDS” creates a numerical model around the affected area using MRI and CT image data, which are medical diagnostic images, and uses this model, incident beam conditions, nuclear data library, etc. A simulation calculation is performed to obtain a flux distribution of neutrons and photons, convert it into a dose distribution, and provide information necessary for treatment such as displaying the result superimposed on a medical diagnostic image (Non-Patent Document 2: Hiroaki Kumada et al., Development of BNCT dose evaluation system (JCDS) for boron neutron capture therapy, JAERI-Tech 2003-002, (2003)). This is used to create a dose plan before BNCT, and to analyze the relationship with treatment outcome by evaluating the dose distribution after irradiation. Furthermore, it can also be used as a design tool such as an irradiation collimator of a neutron beam facility.

  There are only three treatment planning systems developed for BNCT in the world, including JCDS (the other two are “SERA” jointly developed by MIT's “NCTPlan” and the University of Washington, etc.), JCDS Is one of the few systems in the world that have put a treatment plan based on the Monte Carlo method into practical use. Furthermore, only JCDS is currently being improved and enhanced, and is attracting attention from BNCT researchers in Japan and overseas.

  Dose evaluation by the Monte Carlo method enables high-accuracy dose calculation, and introduction of X-ray therapy and particle beam therapy is also being considered. However, the Monte Carlo method has many problems in that it requires a long calculation time. For example, the evaluation of existing X-ray treatments can be completed within a few minutes, whereas the Monte Carlo method often takes hours. Therefore, in the X-ray therapy and the particle beam therapy, the Monte Carlo method has not been put into practical use in view of the treatment efficiency.

  In dose calculation by the conventional Monte Carlo method of radiotherapy, a three-dimensional model of the human body is created based on the CT data of the patient, and this is converted into a calculation model for dose calculation. At this time, for a computational model with a complex shape such as a human body, as shown in FIG. 2, a rectangular parallelepiped region including the human body is divided into X, Y, and Z directions and divided into small rectangular parallelepipeds (called voxels). A human body calculation model (called a voxel model) is defined by defining appropriate material information for the voxel. At this time, dose evaluation accuracy (calculation accuracy) can be improved by making the size of the voxel as fine as possible (FIG. 2-b). On the other hand, if the number of voxels increases, the number of boundaries between voxels increases, and calculation occurs every time a particle passes through the boundary. That is, there is a correlation between the number of voxels (calculation accuracy resulting from this) and the calculation time.

  In the treatment planning system for BNCT that is currently in practical use, calculation is performed with a voxel model in which the size of the voxel is increased from 1 cm square to 5 mm square due to the limitation of calculation time considering the treatment efficiency (Fig. 2-a). When using this large voxel, the material of each voxel defines information such as air, soft tissue, bone, etc. For example, if the voxel overlaps the boundary between air and skin (soft tissue), the mixed material This mixed material causes a calculation error.

H. Kumada, et al., Development of JCDS, a computational dosimetry system at JAEA for boron neutron capture therapy, Journal of Physics: Conference Series, Vol. 74, pp.1-7 2007 Hiroaki Kumada et al .: Development of BNCT dose evaluation system (JCDS) for boron neutron capture therapy, JAERI-Tech 2003-002, (2003)

  An object of the present invention is to provide a method with high calculation accuracy and a short processing time and a computer program thereof when creating a three-dimensional model of a captured image.

The present inventor has intensively studied to solve the above problems.
If one pixel (pixel: about 0.5x0.5x2mm) of CT data that is the basis of the calculation model can be converted into voxels as it is, the calculation accuracy can be maximized, but at present it cannot be realized due to the limitation of calculation time. In response to this problem, the present inventor forms a voxel model while maintaining the pixel level of CT data to improve calculation accuracy, and is a modeling technique that can improve calculation efficiency, “multi-step lattice voxel method” (MLV method) was devised and the present invention was completed.

That is, the present invention is a method of stacking slice data of captured images including voxels having different sizes to form a three-dimensional model to be captured, and includes the following steps:
(a) dividing the sliced image data into a plurality of areas so that a predetermined number of pixels are included in one area;
(b) When the pixels included in one divided region all represent tissues or materials having the same characteristics, the pixels are connected to each other and replaced with one voxel data;
(c) when a pixel representing a tissue or a substance having different characteristics is included in one divided region, defining pixel data in the one region as mixed voxel data;
(d) repeating the steps (b) and (c) to obtain slice data including the substituted voxel data and the mixed voxel data; and
(e) a step of stacking a plurality of slice data acquired in the step (d) in multiple stages,
A method comprising:

Further, the present invention is a program for stacking slice data of photographed images including voxels of different sizes on a computer to form a three-dimensional model to be photographed, and the following procedure:
(a) a procedure for dividing slice data of a captured image into a plurality of regions so that a predetermined number of pixels are included in one region;
(b) When pixels included in one divided region all represent tissues or substances having the same characteristics, a procedure for combining the pixels and replacing them with one voxel data;
(c) When a pixel representing a tissue or a substance having different characteristics is included in one divided area, a procedure for defining pixel data in the one area as mixed voxel data,
(d) repeating the steps (b) and (c) to obtain slice data including replaced voxel data and mixed voxel data; and
(e) a procedure of stacking a plurality of slice data acquired in the procedure (d) in multiple stages;
It is the program for executing.

In the method and program of the present invention, examples of the captured image include medical images of the human body. The slice data of the captured image is based on at least one selected from the group consisting of CT, MRI, Bitmap, and JPEG.
In one embodiment of the present invention, the number of regions divided in the step (a) or the procedure (a) is, for example, 1024, and the number of pixels included in one divided region is, for example, There are 256.
Furthermore, the present invention provides a computer-readable recording medium that records the program.

  The present invention provides a multi-step lattice voxel method. The method of the present invention can reduce the number of voxels forming the human body model (the number of boundary surfaces), and provides a more efficient calculation model by utilizing a Lattice function capable of efficient calculation. be able to. Thus, the method of the present invention is extremely useful in planning radiation therapy, for example.

It is a figure which shows the outline | summary of the Monte Carlo treatment planning system (JCDS) for BNCT. It is a figure which shows the voxel model from which the magnitude | size of a voxel differs. It is a figure which shows the preparation procedure of the human body three-dimensional model used as the basis of a calculation model. It is a figure which shows the outline of a multistep lattice voxel method. It is a figure which shows the process flowchart of voxel model creation by the multistep lattice voxel method. It is a figure which shows the cylindrical water phantom model by each modeling method. It is a figure which shows the full pixel model and nonuniform voxel model with respect to the arbitrary cross sections of a head. It is a figure which shows the full pixel model with respect to the arbitrary cross sections of a head, a 2 mm square mixed material model, and a non-uniform voxel model. It is a figure which shows the thermal neutron flux distribution on the beam central axis in a phantom calculated | required with each calculation model. It is a detailed block diagram of a system for executing the program of the present invention.

Hereinafter, the present invention will be described in detail.
The present invention relates to a method of stacking slice data of photographed images including voxels having different sizes to form a three-dimensional model to be photographed.

The method of the present invention includes the following steps.
(a) dividing the sliced image data into a plurality of areas so that a predetermined number of pixels are included in one area;
(b) When the pixels included in one divided region all represent tissues or materials having the same characteristics, the pixels are connected to each other and replaced with one voxel data;
(c) when a pixel representing a tissue or a substance having different characteristics is included in one divided region, defining pixel data in the one region as mixed voxel data;
(d) repeating the steps (b) and (c) to obtain slice data including the substituted voxel data and the mixed voxel data; and
(e) A step of stacking a plurality of slice data acquired in the step (d) in multiple stages.
The method of the present invention makes effective use of the Lattice function incorporated in the calculation code, rather than creating and arranging individual voxels individually, and using this Lattice function over and over again to produce a single calculation model. It is a technique to form. Compared to a calculation model formed by arranging voxels of the same size, the calculation time can be reduced to about half or less by using this method, and the calculation result with the same accuracy can be obtained because exactly the same geometry can be defined. Can do.

In the present invention, the target for forming the three-dimensional model is not particularly limited. For example, a human body, a non-human animal body, a precision instrument, food, an aircraft, a passenger car, etc. Any image data may be used as long as the image data is classified for each slice by three-dimensional data created by software. Among these, it is preferable that the subject to be imaged is a human body for use in a human body model forming method for dose calculation for a human body shape. The imaging range may be partial such as the head, chest, abdomen, and extremities, or the entire human body.
Hereinafter, a human body model will be described as an example.

1. Procedure for creating a human body model using JCDS
In dose evaluation by JCDS, first, medical image data such as CT and MRI of a subject (for example, a patient) is read and superimposed to create a three-dimensional model. With this 3D model as it is, dose calculation using the Monte Carlo calculation code cannot be performed, so it is converted to a calculation model that can be calculated (called a voxel model). FIG. 3 shows an outline of the data creation procedure of the three-dimensional human body model that is the basis of this calculation model.

  First, a 3D model base is created based on CT data. At this time, since the behavior of neutrons greatly affects the hydrogen density, the captured image of the human body is in a material, that is, a region representing air, bone, or soft tissue with significantly different hydrogen densities (in this case, pixels representing three tissues or materials). Sort. In this specification, “material” may be described as meaning data derived from a photographed tissue (bone or soft tissue).

  This material classification is performed automatically using the CT values recorded in the CT data. The CT value is information recorded for each pixel at the time of CT imaging, and is determined from the source weak coefficient of X-ray irradiation at the time of CT imaging. Accordingly, the material of the pixel can be obtained by analyzing the CT value. For example, the CT value of air is defined as -1000, and water is defined as 0. Bone CT value is high (CT value: 1000), followed by soft tissue / muscle (CT value: 30-60) → blood (CT value: 50) → water (CT value: 0) → fat (CT value: -100) → Air (CT value: -1000). Using this CT value, regions with different hydrogen densities are automatically separated.

Subsequently, cancer lesions, risky organs, and regions of interest (referred to as ROI) are difficult to automatically sort by CT value, so MRI is used for these regions. These regions of interest are manually extracted using MRI. By superimposing this MRI information on the CT side that is the base of the model, a three-dimensional model of one human body is formed. Here, there are basically only three types of material definitions classified by CT value, but in the case of BNCT, a large amount of boron 10 accumulates in the tumor, so only the tumor area should be set as another material. Can do. Also, if there is an area that you want to calculate transport separately from the soft tissue and bone, extract the area separately as ROI on MRI, overlay it on the CT side, and manually input the material of that area Can be defined.
These operations complete a three-dimensional model (image data) that is the basis of the voxel model. At this time, material information is defined for each pixel constituting the three-dimensional model.

2. Multi-step lattice voxel method The present invention is characterized in that image data of a three-dimensional human body model created by the above-described processing is converted into a voxel model that can be transported and calculated by a Monte Carlo calculation code.

  By the way, if the number of voxels forming the voxel model (the number of boundary surfaces) can be reduced, the particle transport calculation of the Monte Carlo method can be made efficient. In the conventional voxel model modeling method, the same material is adjacent to each other inside the air or soft tissue region, but the calculation model is formed by dividing the voxel with the same size. This is because the Lattice function built into the calculation code (MCNP or PHITS) is used. Here, the Lattice function is a function that performs efficient transportation calculation by defining a material for each voxel by equally dividing a rectangular parallelepiped in the X, Y, and Z directions.

On the other hand, the present inventor has conceived that the number of voxels (number of boundaries) can be greatly reduced by joining adjacent regions of the same material and replacing them with large voxels while keeping the material arrangement of the model the same. However, when adjacent materials are combined, it is possible in principle to create an overall model by defining each combined region using the Cell function, but the number of cell definitions It is assumed that the number of definitions exceeds the upper limit of the number of cells in the calculation code and cannot be modeled. Furthermore, if each Cell is defined one by one, there will be an event that results in a longer calculation time than modeling with Lattice, which allows efficient calculations. Therefore, the present invention proposes a method for defining a more efficient calculation model by utilizing a Lattice function capable of efficient calculation.
FIG. 4 shows an outline of the method of the present invention.

One slice of CT data is divided into a plurality of regions so that a predetermined number of pixels are included in one region. “Slice” means a photographed slice when a subject to be photographed (for example, a human body) is photographed in a circular shape, and the thickness of one slice can be arbitrarily set. Further, the number of pixels included in one region is not particularly limited, and can be arbitrarily set. For example, assuming that 512 × 512 = 262,144 pixels are included in one slice, this is divided into n × m areas. n and m are values of powers of 2 (2 ^x ), which are smaller than 512 (2 ⁹ ) (0 <m <9), and m and n are the horizontal direction (x-axis direction) and the vertical direction (y-axis), respectively. Direction), which are 1, 2, 4, 8, 16, 32, 64, 128, 256, 512, preferably 8, 16, 32, 64, and more preferably 32, respectively. m and n may be the same number or different numbers, but are preferably the same number. Therefore, for example, when image data of 512 × 512 pixels is first divided into 32 × 32 = 1024 pieces, 16 × 16 = 256 pixels are included in one divided region (in one section). As described above, an area (in the above example, 32 × 32 areas) obtained by dividing one slice into n × m is referred to as a “Large Voxel area”.

At this time, if all the materials in one region (16 × 16 pixel region) are the same material, the pixels are combined and replaced with one large voxel (that is, 1 × 1 voxel) to reduce the number of voxels. The cell when it is replaced with one large voxel is called “Single Large Voxel Cell”.
If two or more materials are mixed in the Large Voxel area, the voxel is defined by setting a small pixel level voxel (Small Voxel Cell). The cell defined in this way is called “Mixed Large Voxel Cell”.
A human body model (non-uniform voxel model) in which voxels are stacked in multiple stages by repeatedly using the process of replacing Single Large Voxel Cell and the process of defining Mixed Large Voxel Cell, and finally voxels of different sizes are mixed. Form.
FIG. 5 shows a flowchart of these processes.
For convenience of explanation, it is assumed that the Large Voxel region includes 16 × 16 = 256 pixels.

In order to separate the 16 × 16 pixel Large Voxel region into Single Large Voxel Cell or Mixed Large Voxel Cell, first, the material information of 16 × 16 = 256 pixels in the Large Voxel region is checked by full scan. When all the pixels are made of the same material (air, soft tissue, bone, or tumor), 256 pixels are replaced with one cell (Single Large Voxel cell). If any other material data is mixed in 256 pixels, replace the mixed large voxel cell with the Lattice function. Mixed Large Voxel cell is created by arranging 256 Small Voxel Cells of pixel size in order using Lattice function.

“Lattice function is used” means that when particles are transported to a complicated calculation system such as a human body model, the particles pass through the model and the positional relationship (where in the divided Voxel the particles passed) ) Is immediately discriminated and processing for improving transportation efficiency is performed.
The above Lattice function processing is repeated for all Large Voxel regions (32x32 = 1024 regions), and one slice is converted into a slice composed of two types of Single Large Voxel and Mixed Large Voxel. Finally, the two large Voxel cells that have been replaced are arranged in order using the Lattice function to create one Voxel Slice.

  This series of operations is further performed for each slice, and finally each slice is arranged in turn using the Lattice function to form one three-dimensional voxel model.

In the method of the present invention, as described above, the large voxel area is set to 16x16 pixels, but the large voxel area having a size of 8x8 = 64 or 32x32 = 1024 depending on the complexity of the material shape of the original human body. Can be set to As a result, the number of single voxel cells that can be replaced increases, and the total number of voxels can be reduced. The number of pixels in the Large Voxel area may be used for each slice.
Further, the creation of the Voxel Slice is processed for each slice, but a plurality of slices (for example, two, three,...) Can be processed together.

  Furthermore, it is possible to add one or two stages of Lattice processing. For example, after replacing the first Large Voxel area (16x16 pixels) with Single large Voxel Cell and Mixed large Voxel Cell, check the adjacent 2, 4, 6, or 8 areas of this Large Voxel area. If all cells are single large voxel cells, the number of voxels can be reduced by replacing them with a single larger voxel cell (for example, 1024 pixels when replacing four areas with one cell). Is possible. This process can be repeated further. This method can also be used to improve efficiency of mixed large voxel cells. For example, if the Large Voxel area is set to 8x8 or 4x4, and the material is the same, replace it with a Single Large Voxel Cell. By repeating this, the number of voxels can be significantly reduced.

3. Computer Program The program of the present invention is a program for stacking slice data of photographed images including voxels having different sizes on a computer to form a three-dimensional model to be photographed, and includes the following procedure.
(a) a procedure for dividing slice data of a captured image into a plurality of regions so that a predetermined number of pixels are included in one region;
(b) When pixels included in one divided region all represent tissues or substances having the same characteristics, a procedure for combining the pixels and replacing them with one voxel data;
(c) When a pixel representing a tissue or a substance having different characteristics is included in one divided area, a procedure for defining pixel data in the one area as mixed voxel data,
(d) repeating the steps (b) and (c) to obtain slice data including replaced voxel data and mixed voxel data; and
(e) A procedure of stacking a plurality of slice data acquired in the procedure (d) in multiple stages.
FIG. 10 shows a configuration example showing a system for executing a computer in the program of the present invention.

  In FIG. 10, a system 100 includes a calculation unit 110 for converting into a slice composed of two types of Single Large Voxel and Mixed Large Voxel, a database (hereinafter referred to as “DB”) 109, a control unit 101, a transmission / reception unit 102, The input unit 103, output unit 104, ROM 105, RAM 106, hard disk drive (HDD) 107, CD-ROM drive 108, and the like.

  The control unit 101 is a central processing unit such as a CPU or MPU and controls the operation of the entire system 100. In particular, the communication control of the transmission / reception unit 102 is performed, or by using the shooting data stored in the DB 109, a large Voxel cell or a mixed large Voxel cell is created, and a display data reading process such as a result display is performed. Do.

The transmission / reception unit 102 can perform imaging data transmission and reception processing with the user terminal based on an instruction from the control unit 101. The user terminal may be connected via the Internet line 111. The transmission / reception unit 102 transmits information necessary for MLV processing and the like to the calculation unit 110.
The input unit 103 is a keyboard, a mouse, a touch panel, or the like, and is operated when inputting processing conditions or updating the contents of the DB 109. The output unit 104 is an LCD (liquid crystal display) or the like, and converts code data from the control unit 101 into display data each time the DB 109 is updated, and performs display processing. The ROM 105 stores a processing program for the system 100. The RAM 106 temporarily stores data necessary for the processing of the system 100. The HDD 107 stores a program and the like, reads out the stored program or data based on an instruction from the control unit 101, and stores it in the RAM 106, for example. Based on an instruction from the control unit 101, the CD-ROM drive 108 reads out a program stored in the CD-ROM 120 and writes it in the RAM 106 or the like. Instead of the CD-ROM 120, a rewritable CD-R, CD-RW or the like can be used as a recording medium. In that case, a CD-R or CD-RW drive is provided instead of the CD-ROM drive 108. In addition to the above medium, a medium such as a DVD, an MO, or a flash memory stick may be used, and a drive corresponding to the medium may be provided.

4). Computer-readable recording medium The program of the present invention can be written in, for example, C language, Java (registered trademark), Perl, Fortran, Pascal, and the like, and is designed to be compatible with a cross platform. Therefore, this software can be operated on Windows (registered trademark) 95/98/2000 / XP, Linux, UNIX (registered trademark), and Macintosh.

The program of the present invention can be stored in a computer-readable recording medium or a storage means that can be connected to a computer. A computer recording medium or storage means containing the program of the present invention is also included in the present invention. Recording media or storage means include, but are not limited to, magnetic media (flexible disks, hard disks, etc.), optical media (CD, DVD, etc.), magneto-optical media (MO, MD), flash memory, etc. Is not to be done.
Hereinafter, the present invention will be described more specifically with reference to examples. However, the present invention is not limited to these examples.

Verification of calculation performance by the method of the present invention (1) Verification under cylindrical water phantom conditions In order to verify the calculation performance of this method, JRR-4 (research reactor of Japan Atomic Energy Agency), Japan Research Reactor The experiment using the cylindrical water phantom (diameter: about 20cm, length: 20cm) conducted in No.4) was reproduced, and the calculation by the following various modeling methods was executed to compare the calculation accuracy and calculation time. The calculation models are (i) a mixed material voxel model composed of conventional 5x5x5mm square (125mm ³ ) voxels, (ii) a mixed material voxel model composed of conventional 2x2x2mm square (8mm ³ ) voxels, and (iii) original Four pixel models are created: a full pixel model that defines voxels with the same size as the CT pixels, and (iv) a non-uniform voxel model created by the multi-step lattice voxel method (MLV method) of the present invention. Then, using the respective calculation models, we performed transport calculations using the Monte Carlo code PHITS and compared the calculation times. FIG. 6 shows a phantom model by each modeling method.

(2) Verification with a human body model Following the phantom model, a 3D head model was created based on CT data of the human head, and for dose calculation, (i) a calculation model in which all pixels were converted directly into voxels ( Full pixel model) and (ii) non-uniform voxel models using multi-step lattice voxel method, and transport calculation using Monte Carlo code PHITS for each to compare the calculation time. 7 and 8 show voxel models created by two methods. FIG. 6 also shows a sample produced with a 2 × 2 × 2 mm square mixed material model.

<Results and discussion>
(1) Calculation results of the cylindrical water phantom model First, the number of voxels in each model is about 64,000 for the 5mm square mixed material model, about 1,000,000 for the 2mm square mixed material model, and 15,728,640 for the full pixel model. Whereas the number of voxels of the non-uniform voxel model created by the MLV method was about 39,000 (about 650 / slice × 60 slices).

FIG. 9 shows the results calculated by three methods for the thermal neutron flux distribution on the central axis of the beam in the cylindrical water phantom. FIG. 9B is an enlarged view of the boundary surface between the air on which the beam is incident and the phantom surface. From this result, in the region deeper than 1 cm from the surface of the phantom, the calculation results of the three methods matched within the statistical error. On the other hand, in the boundary region between the air region and the phantom surface, the full pixel model and the non-uniform voxel model of the MLV method matched within the statistical error. However, the calculation result of the mixed material model of 5 mm square and 2 mm square produced an error of 18% at maximum with respect to the above two results. Here, it was confirmed that the results of the full pixel model and the non-uniform voxel model of the MLV method agree with the experimental values obtained in the experiment within the range of experimental error and statistical error.
Next, the calculation time by each model was compared. Table 1 shows the calculation time for each model, as well as the calculation time ratio for the mixed voxel model and the calculation time ratio for the full pixel model.

  As shown in Table 1, the calculation with the non-uniform voxel model by the MLV method completed the dose calculation in about half the time of the full pixel model. Also, compared to the conventional mixed material model, the calculation time increment was about 17% for the mixed material model using 2x2x2mm square voxels. For a 5x5x5mm square mixed material model with a significantly smaller number of voxels, approximately twice as much calculation time was required.

(2) Calculation results of the human model The number of voxels in the corner model is 262,144 per slice plane (512 x 5125) in the case of the full pixel model, whereas in the non-uniform voxel model of the MLV method, for each slice The number of voxels is different, and the number of voxels on the slice plane shown in FIG. 7 is about 66,000, and the number of voxels can be reduced to about 25% with respect to the full pixel model. In addition, the number of voxels on the slice plane shown in FIG. 8 was about 77,800, which was also reduced to about 30% compared to the full pixel model.
The calculation results by each model were consistent within the range of statistical accuracy in all regions, and it was confirmed that the two models had equivalent calculation accuracy.

  Next, Table 2 shows the results of comparison of calculation time per 10 million particle transport using two calculation models. The non-uniform voxel model calculation by the MLV method was able to reduce the calculation time by about 50% compared with the calculation by the full pixel model.

  As shown in Table 2, the calculation of the non-uniform voxel model by the MLV method was able to complete the dose calculation by reducing the time and time by about half while maintaining the same calculation accuracy as the calculation by the full pixel model.

<Discussion>
From comparison verification of dose calculation by each calculation method for the cylindrical water phantom system and the human head system, it was confirmed that the calculation modeled by the MLV method has the same calculation accuracy as the full pixel model. This is because both the full pixel model and the non-uniform voxel model by the MLV method have exactly the same geometric structure in terms of material definition. However, since the number of voxels differs greatly, it is possible to reduce the calculation time to about half by using the MLV method without changing the calculation accuracy.

  Compared to the mixed material model currently used in treatment, the calculation time increment is about 17% for the 2mm square mixed material model for the simple shape phantom model, and the MLV method It is considered that the dose calculation by the method is sufficiently practical. On the other hand, the 5mm square mixed material model with significantly fewer voxels takes about twice as much calculation time. However, the mixed material model causes a large error at the interface between different materials, so there are still problems in practical use. However, using the MLV method can improve the calculation accuracy to the same level as the actual measurement value. Is possible.

  The MLV method can further optimize and enhance the algorithm, and currently defines a Large Voxel for each 16x16 pixel per slice. For example, the MLV method combines two slices into a 16x16x2 slice (512 By replacing (pixel) with a single Large Voxel, the number of voxels can be further reduced. Table 3 shows the calculation result by the non-uniform voxel model when the above head model is combined with the 16 × 16 × 2 pixels.

By using this combination, the calculation time could be further reduced by about 34%, and for the full pixel model, the calculation time could be reduced to about 1/3.
As described above, this algorithm can be further improved in calculation efficiency by optimizing according to the calculation system. In addition, this method can be used in general-purpose Monte Carlo codes having a Lattice function such as MCNP and PHITS that are generally widely used (with almost no code improvement).

  This method was developed in advance in the field of BNCT where the Monte Carlo method is used for treatment, but it can also be applied to the fields of conventional X-ray therapy and particle beam therapy. By using this method, there is a possibility of realizing a treatment plan by the Monte Carlo method in X-ray therapy and particle beam therapy. In addition, although this method has been verified for a human body model for radiation therapy, it can actually be applied to all objects with complex shapes.

Claims (11)

A computer, a stack of slice data of the captured image that includes the different voxels sizes, a method of causing formation of 3-dimensional model of the imaging target, the following steps:
(a) dividing the sliced image data into a plurality of areas so that a predetermined number of pixels are included in one area;
(b) When the pixels included in one divided region all represent tissues or materials having the same characteristics, the pixels are connected to each other and replaced with one voxel data;
(c) when a pixel representing a tissue or a substance having different characteristics is included in one divided region, defining pixel data in the one region as mixed voxel data;
(d) repeating the steps (b) and (c) to obtain slice data including the substituted voxel data and the mixed voxel data; and
(e) a step of stacking a plurality of slice data acquired in the step (d) in multiple stages,
Including said method.   The method according to claim 1, wherein the captured image is a medical image of a human body.   The method according to claim 1, wherein the slice data of the photographed image is based on at least one selected from the group consisting of CT, MRI, Bitmap, and JPEG.   The method according to claim 1, wherein the number of regions to be divided in step (a) is 1024.   The method according to claim 1, wherein the number of pixels included in one divided region is 256. A program for causing a computer to stack slice data of photographed images including voxels having different sizes to form a three-dimensional model to be photographed. The procedure is as follows:
(a) a procedure for dividing slice data of a captured image into a plurality of regions so that a predetermined number of pixels are included in one region;
(b) When pixels included in one divided region all represent tissues or substances having the same characteristics, a procedure for combining the pixels and replacing them with one voxel data;
(c) When a pixel representing a tissue or a substance having different characteristics is included in one divided area, a procedure for defining pixel data in the one area as mixed voxel data,
(d) repeating the steps (b) and (c) to obtain slice data including replaced voxel data and mixed voxel data; and
(e) a procedure of stacking a plurality of slice data acquired in the procedure (d) in multiple stages;
The said program for performing. The program according to claim 6, wherein the photographed image is a medical image of a human body.   The program according to claim 6, wherein the slice data of the photographed image is at least one selected from the group consisting of CT, MRI, Bitmap, and JPEG.   The program according to claim 6, wherein the number of areas to be divided in step (a) is 1024.   The program according to claim 6, wherein the number of pixels included in one divided area is 256.   The computer-readable recording medium which recorded the program of any one of Claims 6-10.