Programs, information processing devices, radiation therapy planning devices, servers, methods

Abstract

We propose a new method for obtaining dose distribution information related to organs. [Solution] The program to be executed by the computer causes the computer to calculate dose distribution information for organs based on first information that can identify areas in organs that are at risk of radiation exposure.

Description

[Technical Field]

[0001] This invention relates to programs, etc. [Background technology]

[0002] Conventionally, there are techniques (radiation therapy) for treating tumors such as lung cancer (e.g., non-small cell lung cancer: NSCLC) using radiotherapy (e.g., Patent Document 1). However, it is known that adverse events (side effects) such as radiation pneumonitis (RP) can occur in such cases. [Prior art documents] [Patent Documents]

[0003] [Patent Document 1] Japanese Patent Publication No. 2023-049895 [Overview of the project] [Problems that the invention aims to solve]

[0004] For example, in the above example, radiation pneumonitis is highly dangerous and frequent, and often fatal to patients, making conventional preventive methods insufficient, and overcoming it is still a long way off. Furthermore, while dose distribution simulations (radiation simulations) are commonly performed during radiation therapy, the simulation results differ even for the same patient depending on the specialist (such as a doctor), making this alone insufficient. The same can be said for cases where adverse events may occur due to radiation exposure.

[0005] This invention has been made in view of these problems, and one of its objectives is to propose a new method for obtaining dose distribution information related to organs. [Means for solving the problem]

[0006] According to a first aspect of the present invention, a program to be executed by a computer causes the computer to calculate dose distribution information relating to an organ based on first information that can identify areas in the organ that are at risk of radiation exposure. According to a second aspect of the present invention, the information processing device includes a processing unit that calculates dose distribution information relating to an organ based on first information that can identify locations in the organ that are dangerous to be exposed to radiation. According to a third aspect of the present invention, a method performed by a computer includes calculating dose distribution information relating to an organ based on first information that can identify areas in the organ that are at risk of exposure. According to a fourth aspect of the present invention, a program to be executed by a computer causes the computer to display on a display device information to assist the user in creating dose distribution information about organs, including information about locations in organs where exposure to radiation is dangerous. According to a fifth aspect of the present invention, the information processing device includes a processing unit that performs processing to display on a display device information for assisting a user in creating dose distribution information about organs, including information about locations in organs where exposure to radiation is dangerous. According to a sixth aspect of the present invention, the method performed by the computer includes displaying on a display device information to assist the user in creating dose distribution information about organs, including information about locations in organs where exposure to radiation is dangerous. [Effects of the Invention]

[0007] According to the present invention, dose distribution information for organs can be obtained, taking into account areas within the organ that are at risk of radiation exposure. [Brief explanation of the drawing]

[0008] [Figure 1] A diagram showing an example of the functional configuration related to calculations in an information processing device. [Figure 2] A diagram showing an example of the system configuration of an information processing system. [Figure 3] Diagram illustrating the radiation therapy plan. [Figure 4] Figure illustrating an example of a method for calculating dose distribution information. [Figure 5] Figure illustrating an example of a method for calculating dose distribution information. [Figure 6] Figure illustrating an example of a method for calculating dose distribution information. [Figure 7] Figure illustrating an example of a method for calculating dose distribution information. [Figure 8] Figure illustrating an example of a method for calculating dose distribution information. [Figure 9] Flowchart illustrating an example of a processing flow executed by an information processing apparatus. [Figure 10] Explanatory diagram of a lung function percentile region and a hazard map. [Figure 11] Figure illustrating an example of a two-dimensional pneumonia onset hazard map. [Figure 12] Figure illustrating an example of a three-dimensional pneumonia onset hazard map. [Figure 13] Figure illustrating an example of experimental results. [Figure 14] Figure illustrating an example of experimental results. [Figure 15] Figure illustrating an example of experimental results. [Figure 16] Figure illustrating an example of experimental results. [Figure 17] Figure illustrating an example of experimental results. [Figure 18] Figure illustrating an example of an information processing system in an example. [Figure 19] Flowchart illustrating an example of a processing flow executed by each apparatus in an example. [Figure 20] Figure illustrating an example of a screen displayed on a display unit of a radiation therapy planning apparatus in an example. [Figure 21] Figure illustrating an example of a screen displayed on a display unit of a radiation therapy planning apparatus in an example. [Figure 22] Figure illustrating an example of information stored in a storage unit of a radiation therapy planning apparatus in a modified example. [Figure 23] Figure illustrating an example of an auxiliary screen for creating dose distribution information in a modified example. [Figure 24]A flowchart showing an example of the process for assisting in the creation of dose distribution information in a modified example. [Figure 25] A figure showing an example of a deep learning model in a modified example. [Modes for carrying out the invention]

[0009] Hereinafter, an example of an embodiment for carrying out the present invention will be described with reference to the drawings. In addition, in the descriptions of the drawings, the same reference numeral is used for identical elements, and redundant explanations may be omitted. Furthermore, the components described in this embodiment are merely illustrative and are not intended to limit the scope of the present invention to them.

[0010] [Embodiment] The following describes an example of an embodiment for realizing the information processing technology of the present invention. However, the embodiments to which the present invention can be applied are not limited to the embodiments described below.

[0011] In this specification, "information" may include various types of information, such as numerical (scalar) and vector information (numerical information, vector information), image information (image information) derived from numerical or vector information, and textual information.

[0012] One of the objectives of the present invention is to calculate the dose distribution (dose distribution information) for an organ based on information that can identify dangerous locations in the organ where exposure is likely to cause adverse events (for example, information predicting dangerous locations in the organ where exposure is likely to cause adverse events).

[0013] In the following, we will use "hazard maps" as an example of information that can identify areas in organs where exposure to radiation is dangerous, or more specifically, information that can identify areas in organs where exposure to radiation is likely to cause adverse events (first type of information). Furthermore, among hazard maps, those that target the lungs (adverse events: for example, radiation pneumonitis) will be referred to as "pneumonia hazard maps." The first piece of information only needs to be information that can identify areas in organs that are at risk of radiation exposure; for example, it could be a safe map, which is the opposite concept to a hazard map.

[0014] Figure 1 shows an example of a functional unit corresponding to the calculation unit (calculation unit) among the functional units of the information processing device 1 in this embodiment. The information processing device 1 may also be called a computing device.

[0015] The information processing device 1 may have a dose distribution information calculation unit 11, as shown in Figure 1(a) as the information processing device 1A. This unit may be a functional unit of the processing unit (or control unit) as the calculation unit (calculation unit) of the information processing device 1, and may be configured with processing circuits such as a CPU (Central Processing Unit), GPU (Graphics Processing Unit), DSP (Digital Signal Processor), ASIC (Application Specific Integrated Circuit), or FPGA (Field Programmable Gate Array).

[0016] Furthermore, the information processing device 1 may be configured as, for example, the information processing device 1B shown in Figure 1(b), and the information processing device 1B may have a hazard map calculation unit 13 in addition to the dose distribution information calculation unit 11. This may also be, for example, a functional unit of the processing unit (or control unit) as the calculation unit (calculation unit) of the information processing device 1.

[0017] Furthermore, the information processing device 1 (information processing device 1A or information processing device 1B) may include, in addition to the processing unit (or control unit), functional units such as an operation unit, an output unit (display unit, sound output unit, etc.), a communication unit, a clock unit, a memory unit, etc., and may be configured as a standalone computer device.

[0018] Furthermore, the information processing device 1 may be provided in a radiotherapy planning device 300, which is a computer device for planning radiotherapy, as shown in Figure 1(c). Alternatively, the radiotherapy planning device 300 may also be the information processing device 1. The following example illustrates a case where the information processing device 1 is installed in the radiation therapy planning device 300.

[0019] Figure 2 shows an example of the system configuration of the information processing system 1000A in this embodiment. The information processing system 1000A includes, for example, a medical image diagnostic device 100, a medical image database 200, a radiation therapy planning device 300, and a radiation therapy device 400. These devices may be configured to communicate with each other, for example, via a bus or a communication unit. The communication method of the communication unit may be wired communication or wireless communication.

[0020] The medical imaging diagnostic device 100 may include, for example, devices for acquiring and evaluating morphological information (information on anatomical structures), such as a simple X-ray device, an X-ray CT (Computed Tomography) device, and an MRI (Magnetic Resonance Imaging) device, as well as devices for acquiring and evaluating functional information (physiological functional information), such as a PET (Positron Emission Tomography) device and a SPECT (Single Photon Emission Computed Tomography) device (devices for nuclear medicine examinations). Alternatively, it may be a medical imaging diagnostic device composed of any of these various devices, or a combination of these various devices. It may also be a PET-CT device or a SPECT-CT device, etc. Morphological and functional information can be considered, for example, a type of mathematical information related to the subject's organs (hereinafter referred to as "mathematical information").

[0021] Medical image data acquired by the medical imaging diagnostic device 100, such as volume data and image data obtained by imaging it, may be stored in the medical image database 200 in association with information such as the subject's identification information (ID, name, etc.) and date / time.

[0022] Patients may be included among the subjects. Furthermore, as will be described later in the modified examples, the method of this embodiment is also applicable to animals (excluding humans), and these can also be referred to as subjects.

[0023] The radiation therapy planning device 300 is a computer device that creates (generates) a radiation therapy plan, which is a plan for radiation therapy performed by the radiation therapy device 400. Typically, before administering radiation therapy, a radiation therapy plan is created to determine the treatment method. This treatment plan is sometimes called a radiation simulation. In this embodiment, as shown in Figure 1(c), the information processing device 1 is provided in the radiation therapy planning device 300 (or the radiation therapy planning device 300 is the information processing device 1), and the radiation therapy planning device 300 (information processing device 1) performs radiation simulation. The radiation therapy planning device 300 may include, in addition to the processing unit (or control unit), functional units such as an operation unit, an output unit (display unit, sound output unit, etc.), a communication unit, a clock unit, a memory unit, and so on.

[0024] The dose distribution information calculated by the radiation therapy planning device 300 may be stored as data in the memory unit of the radiation therapy planning device 300, for example.

[0025] Unlike this embodiment, the information processing device 1 may be configured as a separate device from the radiation therapy planning device 300.

[0026] The radiotherapy device 400 is a device for performing radiotherapy, and may be an IMRT (Intensity Modulated Radiation Therapy) device including, for example, TomoTherapy, a Multi-Leaf Collimator (MLC), or CyberKnife. IMRT may also include Volumetric Modulated Arc Therapy (VMAT). However, the system is not limited to this, and for example, devices for performing three-dimensional conformal radiotherapy (3D-CRT), stereotactic radiotherapy (SRT), etc. may also be used. Alternatively, a combination of these may be used.

[0027] Furthermore, as radiation therapy, treatment using X-rays or gamma rays, which are classified as electromagnetic waves, may be applied, or treatment using alpha rays, beta rays, electron beams, proton beams, heavy ion beams, or neutron beams, which are classified as particle beams, may be applied. For example, proton beam therapy, heavy ion beam therapy, neutron beam therapy, etc., may be applied, and these devices may be used as the radiation therapy apparatus 400. In addition, any combination of the above may be applied.

[0028] Figure 3 is a diagram illustrating the overview of IMRT, a type of radiation therapy. IMRT is a type of radiation therapy that allows for the concentration of radiation on a tumor by varying the intensity of the radiation within the irradiation field. In this diagram, the organ is the lung, and for example, if the area indicated by the hatched circle is the tumor portion of lung cancer, the radiation range and dose can be set in various patterns as shown in the diagram.

[0029] Returning to Figure 1, the dose distribution information calculation unit 11 of the information processing device 1 performs radiation simulation processing and calculates dose distribution information. This makes it possible to calculate dose distribution information for, for example, a patient scheduled to receive radiation therapy. The dose distribution information may be, for example, information that associates positional information within organs (which may be a concept including position and region) with the dose of radiation.

[0030] In this embodiment, the dose distribution information calculation unit 11 calculates dose distribution information based on an artificial intelligence model for predicting dose distribution information (hereinafter simply referred to as the "prediction model"), which is a trained prediction model (hereinafter simply referred to as the "trained model").

[0031] Furthermore, in this embodiment, the hazard map calculation unit 13 calculates the hazard map based on a method described in detail later.

[0032] <Method for calculating dose distribution information> The following are examples of methods for calculating dose distribution information. In this embodiment, dose distribution information is calculated using a deep learning model, such as a fully convolutional network (FCN) model like U-Net (a model with an encoder-decoder structure), as the artificial intelligence model. The information (e.g., images) used in the method described below can basically be three-dimensional information (e.g., three-dimensional images), and for example, a 3DU-Net, which is a three-dimensional extension of U-Net, may be used. Specifically, a model may be applied in which a three-dimensional input image corresponding to the input information described below is subjected to feature extraction by convolution operation using an encoder, and then an inverse convolution operation is performed using a decoder to output an image of a map (in this embodiment, a dose distribution map) of the same size as the input image. Note that two-dimensional information may also be used, and for example, a 2DU-Net may be used. Alternatively, you may use CNN (convolutional neural network) models such as ResNet, or neural network models such as ViT (Vision Transformer). However, the artificial intelligence models shown here are merely examples and are not the only ones that are relevant.

[0033] Furthermore, for convenience, the following explanation assumes that the information processing device 1 performs model learning and inference, but the information processing device 1 may infer dose distribution information based on a pre-acquired, trained model. The same applies to hazard maps. Further details will be provided later.

[0034] One of the ideas behind this invention is to perform radiation simulations to ensure that tumors (affected areas) are appropriately irradiated with radiation, while avoiding (or reducing) exposure to normal or dangerous tissues. Dangerous tissue often refers to tissue within an organ that is prone to causing adverse events when exposed to radiation, and can be identified, for example, from a hazard map. When considering the lungs, dangerous lung tissue often refers to lung tissue that is prone to causing radiation pneumonitis when exposed to radiation, and can be identified, for example, from a pneumonia hazard map. The following calculation method is based on these principles.

[0035] Below, we will explain two main calculation methods, which are distinguished by differences in the information used for learning.

[0036] (1) First calculation method Figures 4 to 7 illustrate the learning and inference processes of the prediction model in the first calculation method.

[0037] In the learning phase of the first calculation method, for example, for one subject, ·Form information • Contour information of the area of ​​interest Hazard map • Dose distribution information based on hazard maps (training data) This is used as a single dataset, and a predictive model is generated using pre-acquired datasets for multiple subjects as training data.

[0038] Morphological information, contour information, and hazard maps can be considered explanatory variables. Furthermore, dose distribution information based on hazard maps corresponds to the dependent variable and can be considered training data. These combined constitute the training data.

[0039] Morphological information may be, for example, information acquired by the aforementioned medical imaging diagnostic device 100 capable of acquiring morphological information. Furthermore, the morphological information does not necessarily have to be the morphological information of the entire organ; it may be the morphological information of only a part of the organ.

[0040] The region of interest may be, for example, the area including the tumor site (target site) and surrounding normal tissue, and its contour may be detected by known methods based on morphological information (for example, the contour can be drawn from morphological images such as CT images). While it is not necessary to include normal tissue, it may be included to avoid exposure to normal tissue. The tumor site may be determined based on medical definitions, for example, by gross tumor volume (GTV) or clinical target volume (CTV). It may also be determined based on planning target volume (PTV), etc.

[0041] In this embodiment, dose distribution information (training data) based on a hazard map may be, for example, dose distribution information artificially created by a specialist such as a physician by referring to a hazard map (hereinafter referred to as "handmade" or "handmade dose distribution information" as appropriate). This may include information created by a specialist such as a physician using a radiation therapy planning device 300, etc.

[0042] As will be explained later in the section on modifications, it is also possible to calculate the dose distribution information used as training data using a computer without human intervention, and the information obtained in this way can also be considered dose distribution information based on hazard maps.

[0043] Furthermore, when creating dose distribution information manually, medical professionals and other experts may refer not only to hazard maps but also to contour information of areas of interest. Specifically, for example, • Refer to contour information to ensure that the appropriate dose of radiation is delivered to the tumor. • Refer to contour information to avoid irradiating normal tissue with radiation. Refer to hazard maps to avoid exposure of lung tissue to radiation. The dose distribution may be adjusted accordingly. Furthermore, morphological information may be used to calculate the attenuation of radiation.

[0044] Furthermore, as will be described later in the section on modifications, the handmade dose distribution information may include, for example, dose distribution information calculated by the radiation therapy planning device 300 based on information input by the user, such as dose distribution optimization.

[0045] As shown in Figure 4, in the inference phase, the information processing device 1 acquires morphological information, contour information, and a hazard map corresponding to the target subject as input information, and uses this acquired information to infer dose distribution information (dose distribution information based on the hazard map) corresponding to the target subject using a trained model. In this case, for example, the radiotherapy planning device 300 can acquire morphological information, contour information, and a hazard map corresponding to the target subject based on user operation.

[0046] Here, acquiring information may include, for example, receiving information transmitted from an external device or reading information stored in an external or internal storage device. The same may apply hereafter.

[0047] For example, as shown in Figure 5, in the inference phase, the information processing device 1 acquires morphological information and contour information corresponding to the target subject as input information, and the hazard map calculation unit 13 calculates a hazard map based on the acquired morphological information. The dose distribution information calculation unit 11 may then use the acquired morphological information and contour information and the calculated hazard map to infer dose distribution information corresponding to the target subject (dose distribution information based on the hazard map) using a trained model. In this method, the information processing device 1 internally calculates a hazard map based on the acquired morphological information, eliminating the need to prepare a hazard map in advance for the target subject. Furthermore, it is not necessary to output the hazard map externally.

[0048] Furthermore, as shown in Figure 6, for example, in the inference phase, the information processing device 1 acquires morphological information and a hazard map corresponding to the target subject as input information, and calculates contour information based on the acquired morphological information. The dose distribution information calculation unit 11 may then use the acquired morphological information and hazard map, along with the calculated contour information, to infer dose distribution information (dose distribution information based on the hazard map) corresponding to the target subject using a trained model. In this method, the information processing device 1 internally calculates contour information based on the acquired morphological information, eliminating the need to pre-prepare contour information corresponding to the target subject. Furthermore, it is not necessary to output the contour information externally.

[0049] Furthermore, as described above, hazard maps and contour information can be calculated based on morphological information. For example, as shown in Figure 7, in the inference phase, the information processing device 1 acquires morphological information corresponding to the target subject as input information, calculates contour information based on the acquired morphological information, and the hazard map calculation unit 13 calculates a hazard map based on the acquired morphological information. The dose distribution information calculation unit 11 may then use the acquired morphological information and the calculated hazard map and contour information to infer dose distribution information corresponding to the target subject (dose distribution information based on the hazard map) using a trained model.

[0050] In the first calculation method, dose distribution information based on hazard maps (for example, handmade dose distribution information) is used as training data for learning. Separately, the hazard maps themselves are also used for learning. This is intended to improve the accuracy of the model.

[0051] Furthermore, since contour information can be calculated by a computer based on morphological information, the information processing device 1 does not need to acquire contour information as input information, as shown in Figures 6 and 7, for example. However, the information processing device 1 may acquire contour information as input information in addition to morphological information, as shown in Figures 4 and 5, for example. Contour information calculated by a computer is not always accurate, and it may be more reliable for a doctor or other professional to visually confirm it. Therefore, for example, a user may use some computer device to draw a contour based on morphological information, or modify the contour based on contour information calculated by a computer based on morphological information, and the information processing device 1 may acquire the contour information obtained in this way as input information. Furthermore, this may apply not only to contour information for a single subject used in the inference phase, but also to contour information for multiple subjects used in the learning phase.

[0052] (2) Second calculation method Figure 8 is an explanatory diagram of the training and inference of the prediction model in the second calculation method.

[0053] In the learning phase of the second calculation method, for example, for one subject, ·Form information • Contour information • Dose distribution information based on hazard maps (training data) This is used as a single dataset, and a predictive model is generated using pre-obtained datasets for multiple subjects as training data.

[0054] The dose distribution information based on the hazard map may be calculated using the same method as in the first calculation method. However, unlike the first calculation method, the second calculation method does not use the hazard map itself for learning.

[0055] In the inference phase, the information processing device 1 acquires, for example, morphological information and contour information corresponding to the target subject as input information, and uses this acquired information to infer dose distribution information (dose distribution information based on a hazard map) corresponding to the target subject using a trained model. In this case, the contour information used as input information may be contour information obtained through human intervention, similar to the first calculation method.

[0056] In the inference phase, the information processing device 1 may acquire morphological information corresponding to the target subject as input information and calculate contour information based on the acquired morphological information. The dose distribution information calculation unit 11 may then use the acquired morphological information and the calculated contour information to infer dose distribution information (dose distribution information based on a hazard map) corresponding to the target subject using a trained model.

[0057] A key feature of both the first and second calculation methods is that dose distribution information based on hazard maps is used as training data for learning. This allows the calculation of dose distribution information to avoid exposure to dangerous lung tissue. Furthermore, by using contour information in the learning process, dose distribution information is calculated to ensure that radiation is appropriately irradiated to tumors while avoiding exposure to normal tissue.

[0058] <Other> In the first calculation method, it is not mandatory to use morphological information in the learning and inference phases; dose distribution information may be calculated without using morphological information. Similarly, in the second calculation method, it is not mandatory to use morphological information in the learning and inference phases; dose distribution information may be calculated without using morphological information. In other words, morphological information is not essential for any of the calculation methods.

[0059] Furthermore, morphological information and contour information may be treated as separate information, or contour information may be linked to morphological information. For example, an image in which contours have been input into a morphological image such as a CT image may be used.

[0060] Furthermore, information regarding the location of the region of interest within an organ may be any information that allows for the identification of the location of the region of interest within the organ. For example, it may be information obtained by applying a mask process to extract only the region of interest (e.g., a masked image from which the region of interest has been extracted). This can also be considered a type of contour information.

[0061] <Processing> Figure 9 is a flowchart showing an example of the processing flow performed by the information processing device 1 in this embodiment. This process is achieved, for example, by the processing unit of the information processing device 1 reading the program code stored in a memory unit (not shown) into a RAM (Random Access Memory) (not shown) and executing it.

[0062] The flowchart described below is merely one example of an information processing procedure; other steps may be added, or some steps may be deleted. Furthermore, some of the steps in the flowchart may be rearranged before execution.

[0063] First, the information processing device 1 acquires the aforementioned input information based on, for example, user input (A1).

[0064] Then, the dose distribution information calculation unit 11 uses the acquired input information to perform radiation simulation processing using one of the above calculation methods and calculates dose distribution information (A3).

[0065] Then, the information processing device 1 outputs the calculated dose distribution information (A5) and terminates the process.

[0066] <How hazard maps are calculated> Next, we will explain how hazard maps are calculated. Here, we will use the lungs as the organ and illustrate a method for calculating a pneumonia hazard map.

[0067] The hazard maps used in the training phase of the aforementioned prediction model and the hazard maps used as input information in the inference phase may be calculated by some device using the method described below. Furthermore, as mentioned above, when the information processing device 1 calculates a hazard map during the inference phase, it may calculate the hazard map using the method described below.

[0068] The following example illustrates the use of 4DCT information (4DCT data) as morphological information. 4DCT information can be, for example, time-series three-dimensional volume data, and can represent information about the dynamics of a subject's respiration. Specifically, it can be information composed of 3DCT information (3DCT data) at two points in time, such as the subject's expiratory position (peak expiratory position, etc.) and inspiratory position (peak inspiratory position). Furthermore, when applying CT, it is also possible to perform the processing described below based not only on 4DCT information, but also on 3DCT information and 2DCT information.

[0069] (A) Quantification of functional values First, by performing DIR (Deformable Image Registration) and quantitative evaluation (quantitative analysis) on 4DCT information (e.g., 4DCT images), lung function information (e.g., lung function images) that quantifies lung ventilation is generated.

[0070] In this case, a quantitative evaluation method such as the Hounsfield Unit (HU) method may be applied, and for example, the following equation (1) can be used.

number

[0071] Furthermore, as a quantitative evaluation method, for example, a method that analyzes the extent of lung volume change from exhalation to inhalation using the Jacobian matrix (Jacobian-based metric) may be used.

[0072] Furthermore, the following processes only require the use of lung function information (e.g., lung function values), and the method of obtaining this information is not specified. For example, lung function values ​​obtained through the aforementioned nuclear medicine examination may be used.

[0073] (B) Percentile value conversion process Next, a percentile conversion process is performed to convert the quantified lung function values ​​from process (A) into percentile values ​​(e.g., 0 to 100). The percentile values ​​obtained in this way are conveniently referred to as "lung function percentile values".

[0074] Furthermore, noise reduction processing (noise removal processing) may be performed on the pulmonary function values ​​(pulmonary function images: for example, 4DCT ventilation images composed of pulmonary function values ​​quantified by HU) obtained through the processing in (A). Specifically, filtering processing using a median filter (for example, a median filter with a width of 3 x 3 x 3 voxels) may be performed. Alternatively, filtering may be performed using smoothing filters such as moving average filters or Gaussian filters.

[0075] (C) Fractionation Next, based on the pre-set upper and lower thresholds for the lung function percentile values ​​obtained in process (or filtering) (B), the data, for example, that associates location with lung function percentile values, is fractionated (divided) into multiple data sets. The thresholds may be set automatically by the device (automatic setting) or by the device based on user input (manual setting). The same applies to various settings below.

[0076] Specifically, for example, lower cutoff limits are set at intervals of 20 from the 0th to the 80th position, and upper cutoff limits are set at intervals of 20 from the 20th to the 100th position. Then, the lung function percentile values ​​(lung function percentile values ​​associated with location) included in the intervals defined by the lower and upper cutoff limits are extracted. For convenience, each of the fractionated intervals is called a "fractionation interval," and the number of fractionation intervals is called the "number of fractions." This results in data corresponding to five fractional intervals: "0-20th", "20-40th", "40-60th", "60-80th", and "80-100th". These data represent lung function classified into five stages based on lung function percentile values, and the data extracted represents the lung regions belonging to each stage of lung function (hereinafter referred to as "lung function percentile regions").

[0077] Figure 10 shows an example of a two-dimensional image obtained by binarizing these five data points. In the binarized image corresponding to each divided section N (N=1 to 5) on the right side of the drawing, the region shown in white indicates the pulmonary function percentile region.

[0078] To the information of the pulmonary function percentile region corresponding to each divided section N (N=1 to 5) obtained in this manner (hereinafter referred to as "pulmonary function percentile region information"), the corresponding coefficient "α N " obtained by the method described below is multiplied, and these are three-dimensionally summed up, thereby obtaining a pneumonia occurrence hazard map.

[0079] Here, in order to obtain the pneumonia occurrence hazard map, it is necessary to obtain the coefficient "α N ". An example of a method for calculating the coefficient "α N " is given below.

[0080] For example, the pneumonia occurrence prediction model represented by the following formula (2) is formulated.

Mathematical Expression

[0081] "D N " (=D1 to D5) on the right-hand side is, for example, the exposure dose (average exposure dose) to the pulmonary function percentile region in divided sections "80-100th", "60-80th", "40-60th", "20-40th", and "0-20th", respectively. The coefficient "α N " is a coefficient corresponding to each divided section. "f(D)" on the left-hand side is, for example, a value representing the possibility of pneumonia occurrence (an example of a value representing the possibility of an adverse event occurrence).

[0082] In the learning phase of the pneumonia occurrence prediction model, for a single patient, for example, by superimposing the dose distribution image used in the patient's radiation therapy (for example, an image created using the radiation therapy planning apparatus 300) on each of the five images of the pulmonary function percentile regions, the exposure dose "D NThe values ​​(=D1~D5) are calculated. Then, for that patient, a value indicating whether or not pneumonia actually occurred (for example, pneumonia occurred: 1, pneumonia not occurred: 0) is used as the training data for the left side "f(D)", and this training data is used with the radiation dose in the lung function percentile region "D N The combination of "[ ]" will be used as the training dataset. Then, using the training dataset obtained from multiple patients, the pneumonia occurrence prediction model of equation (2) is trained, and the coefficient "α" is used. N We find (=α1~α5).

[0083] Furthermore, when inferring the occurrence of pneumonia using a pneumonia prediction model, the radiation dose in the lung function percentile range "D" obtained for the target subjects is used. N Substitute this into equation (2) to infer "f(D)". Then, for example, if "f(D)" is greater than or equal to the cutoff value, it is determined that "pneumonia occurred", and if "f(D)" is less than the cutoff value, it is determined that "pneumonia did not occur".

[0084] Alternatively, the above cutoff value may also be included as a parameter in the pneumonia outbreak prediction model and determined through learning. Furthermore, thresholds (upper and lower thresholds) for the lung function percentile values ​​that define the fractional intervals may also be included as parameters in the pneumonia prediction model and determined through learning.

[0085] Furthermore, in experiments conducted by the inventor using sample data, the coefficient "α5" corresponding to the fractionation interval "0-20th" and the coefficient "α4" corresponding to the fractionation interval "20-40th" were found to be "0". For this reason, the following equation (3) may be used instead of equation (2) as a pneumonia occurrence prediction model.

number

[0086] Furthermore, as described above, if "pneumonia has occurred" is determined when "f(D)" is equal to or greater than the cutoff value, then "pneumonia has not occurred" will be determined if "f(D)" is less than the cutoff value. Therefore, the radiation dose in the lung function percentile range "D" should be set so that "f(D)" is less than the cutoff value. N By adjusting the settings, it is possible to calculate dose distribution information that results in "no pneumonia cases."

[0087] The training of a pneumonia outbreak prediction model can be implemented, for example, using the Lasso method. The Lasso method is a technique that simultaneously determines the coefficient "α" and selects features based on an objective function that incorporates the L1 norm. Using the training dataset, the hyperparameter representing the magnitude of the penalty (λ of the L1 regularization term) and the coefficient "α" are determined. N This allows us to calculate the following. Note that the Lasso method is publicly known, so a detailed explanation will be omitted. One reason for using the Lasso method is to eliminate unnecessary features through feature selection. This is more effective when the number of fractions is large.

[0088] Alternatively, instead of the above, multiple regression analysis methods such as Ridge regression, Elastic Net regression, Bayesian linear regression, robust regression, maximum likelihood estimation, and gradient descent may be used. Furthermore, advanced machine learning techniques such as deep learning (an evolution of neural networks) and MLP (Multi-Layer Perceptron) may be applied (although deep learning and MLP are considered types of machine learning in this context). Genetic algorithms may also be used.

[0089] The coefficient "α" obtained in this way N By multiplying this value by the corresponding number "N" in the lung function percentile region information and summing them up, a pneumonia hazard map can be obtained.

[0090] The lower part of Figure 10 shows an example of a two-dimensional pneumonia hazard map obtained as described above, and Figure 11 shows an enlarged version of this map. In this pneumonia hazard map, areas with dark hatched lines indicate "very dangerous" areas, which are the most dangerous areas for lung exposure. Areas with light hatched lines indicate "dangerous" areas, which are the next most dangerous areas for lung exposure. For clarity, a section of the lung area on the right side of the diagram that is classified as "dangerous" is indicated with a leader line. You can think of the "very dangerous" area as representing the so-called red zone, and the "dangerous" area as representing the so-called yellow zone. When using color to represent the situation, for example, you could depict the "very dangerous" areas in red and the "dangerous" areas in yellow or orange.

[0091] In addition to the two categories mentioned above, classifications based on the values ​​(output values) of the pneumonia hazard map are also possible. For example, classifications such as "moderate risk," "low risk," and "very low risk" are possible. In this case, a color corresponding to the classified risk can be set, and the pneumonia hazard map can be color-coded for rendering. For example, if classifying into five levels, a pneumonia hazard map can be generated and displayed with the following colors: "very high risk = red," "high risk = orange," "moderate risk = yellow," "low risk = green," and "very low risk = blue."

[0092] Furthermore, the risk of radiation exposure is measured by the coefficient "α N It can also be thought of as being represented by ", and the region corresponding to the fractionation interval "α=0" can be considered a "safe" region. Experiments conducted by the inventor of this application showed that using the Lasso method resulted in the coefficients "α5" for the fractionation interval "0-20th" and "α4" for the fractionation interval "20-40th" becoming "0". Therefore, the regions "0-20th" and "20-40th" can be considered "safe" regions. Provided, however, that if there is no significant difference in the value of the coefficient "α" corresponding to each segmented interval, it cannot necessarily be concluded that the region in the segmented interval is "safe" just because the calculated coefficient "α" is small. However, at least the region of the segmented interval where "α=0" may be considered as a "safe" region.

[0093] It should be noted that in the hazard map image, each region may be displayed by being distinguished by hatching instead of color. In addition, in the hazard map image, for example, characters or icons such as "Danger" (危) may be displayed in association with a region having particularly high risk, or the outline of a region having particularly high risk may be displayed in an emphasized manner. Furthermore, the hazard map may be displayed in any format of monochrome, black and white, grayscale, or color.

[0094] When three-dimensional volume data is used, for example, as shown in FIG. 12, a pneumonia occurrence hazard map is generated as three-dimensional data. It should be noted that the same applies even when 4DCT information is used, since three-dimensional lung function information can be obtained.

[0095] <Another Example of Calculation Method for Dose Distribution Information (Teacher Data) Based on Hazard Map> In view of the above, the dose distribution information (teacher data) based on the hazard map in the aforementioned method for calculating dose distribution information may be calculated by a computer using, for example, the following method. Specifically, (a) Compliance with guideline values regarding radiation irradiation dose in radiation therapy for lung cancer (b) A region of a certain size or larger is determined as "no pneumonia occurrence" the dose distribution information (teacher data) based on the hazard map may be calculated such that the two conditions of (a) and (b) are satisfied.

[0096] The condition in (a) may be, for example, that the percentage of lung volume exposed to a specific exposure dose (a set threshold dose: for example, 20 Gy) exceeding a certain dose (e.g., V20 Gy) for the entire lung is less than a predetermined percentage (e.g., 35%). In other words, for example, "V20 Gy < 35%" may be set as the condition in (a). Condition (b) may be defined as, for example, that the proportion (or number) of regions where f(D) is less than the cutoff value is greater than or equal to a certain value.

[0097] In this way, the dose distribution information used as training data in the dose distribution prediction model can be calculated by computer.

[0098] <Other> The method for calculating the hazard map described above is merely one example and is not the only way to do so. Other methods, such as using deep learning, may also be used, and the input information is not limited to the information described above.

[0099] Furthermore, in the method for calculating the hazard map described above, the source information used for processing can be any mathematical information (mathematical information) related to the subject's organs, and it is not relevant what aspect of the organ it represents. This mathematical information may include morphological information or functional information. Furthermore, mathematical information may be, for example, information represented in real space relating to organs (3-dimensional information, or 4-dimensional information if the time axis is considered). Alternatively, it may be information represented on a real plane (2-dimensional information).

[0100] Furthermore, mathematical information may be acquired by any method. While specialized equipment may be used, mathematical information may also be acquired without such equipment. For example, 2D information can be converted into pseudo-3D to obtain morphological information such as 3DCT information. In addition to the method described in this embodiment, functional information may also be obtained using a deep learning model such as a CNN (Convolutional Neural Network). For example, “A deep learning method for translating 3DCT to SPECT ventilation imaging: First comparison with 81mKr-gas SPECT ventilation imaging” (https: / / doi.org / 10.1002 / mp.15697) discloses a method for generating functional information from 3DCT information using a deep learning model, and this method may be applied, for example. Alternatively, methods using GANs (Generative Adversarial Networks) or diffusion models may be applied.

[0101] Furthermore, mathematical information may be obtained using different types of information. For example, mathematical information may be obtained using functional information acquired by some method and morphological information acquired by some method. In this case, for example, a deep learning model (hereinafter referred to as the "mathematical information model") may be generated that takes functional information and morphological information as input and outputs mathematical information.

[0102] In this case, if we train a mathematical information model to prioritize morphological information over functional information, we can make it output information that, while not outwardly representing morphology, appears as if it does. In other words, we can obtain information that, while not outwardly morphological, can be considered morphological information. For example, the information obtained in this way may be used as morphological information, and the information obtained in this way may be treated as morphological information.

[0103] Furthermore, as mentioned above, hazard maps can be calculated using functional information. Functional information can be calculated based on morphological information, but it is also possible to calculate hazard maps directly from functional information. For example, in the mathematical information model described above, if the model is trained to prioritize functional information over morphological information, it can output information that, while not outwardly representing functionality, appears to represent functionality. In other words, it is possible to obtain information that, while not outwardly functional, can be considered functional information. For example, the information obtained in this way may be used as functional information to calculate a hazard map, or the information obtained in this way may be treated as functional information.

[0104] Furthermore, as mentioned above, morphological information is not mandatory for calculating dose distribution information, but it may be used. In this case, the morphological information may also be acquired in the same manner as described above.

[0105] <Experimental Results> Next, we will show the results of an experiment in which dose distribution information was calculated based on the method of this embodiment. Here, we will show the results using 80 cases of lung cancer patients at the inventor's hospital, with 64 cases (80% of the cases) used for model creation and 16 cases (the remaining 20%) used for model testing. Here, we will illustrate the test results (Result C) of the model when the first method described above was applied and the model was trained using manually created dose distribution information as training data.

[0106] For comparison, we also show the results of generating dose distribution information using only CT images, a method widely used in current clinical practice (Result A), and the results of generating dose distribution information manually by referring to CT images linked with contour information and a pneumonia hazard map (Result B).

[0107] Figure 13 shows a dose distribution image, which represents dose distribution information corresponding to result A. The areas indicated by thick white lines are areas in the lungs where exposure is dangerous, based on the hazard map (hereinafter referred to as "dangerous areas" for convenience), and are displayed overlaid on the CT image for clarity. The distribution of radiation is shown with regular white lines (thinner than the thick white lines above), and the levels of radiation dose are represented by three different types of hatching. The hatching is in order of increasing dose: diagonal lines, vertical lines, and horizontal lines. Additionally, a magnified view of the R1 dangerous area is shown below. These results show that moderate levels of radiation are distributed throughout the entire danger zone R1. A similar trend is observed in other danger zones. Therefore, there is a high probability of increased exposure to dangerous lung tissue and the development of pneumonia.

[0108] Figure 14 shows a dose distribution image, which represents dose distribution information corresponding to result B. The interpretation of this figure is the same as in Figure 13. These results show that although moderate to low intensity radiation is distributed throughout the entire danger zone R1, the results are better than those in result A. This demonstrates the usefulness of the handmade method.

[0109] Figure 15 shows a dose distribution image, which represents dose distribution information corresponding to result C. The interpretation of this figure is the same as in Figure 13. This result is a dose distribution image obtained by applying 3DU-Net (described above) as a prediction model, performing training, and then performing inference using the trained model. These results show that the radiation exposure range and dose in the hazardous region R1 are significantly reduced compared to results A and B, indicating a favorable outcome. This demonstrates the usefulness of the method using the artificial intelligence model.

[0110] Figure 16 is a table showing the results of calculating the mean ± standard deviation of exposure dose for each of the "very high," "high," and "medium" risk areas based on the hazard map, and shows the results corresponding to Result B in comparison to Result A above. The unit of exposure dose is "Gy". The results of verifying whether there is a significant difference between Result A and Result B using p-values ​​are also shown. This table shows that in Result B, the radiation dose is lower in all areas compared to Result A, and since the "p value < 0.01", it is clear that there is a statistically significant difference.

[0111] Figure 17 shows a table similar to Figure 16, corresponding to result C. This table shows that in Result C, the radiation dose is lower in all areas compared to Result A, and since the "p value < 0.01", it is clear that there is a statistically significant difference.

[0112] <Examples> Next, as an example of an embodiment applying the above method, we will describe an embodiment that applies a client-server system and uses an application for performing radiation simulations.

[0113] Figure 18 shows an example of the system configuration of the information processing system 1000B in this embodiment. The information processing system 1000B is a communication system comprising, for example, a server 2 and multiple radiation therapy planning devices 300 (400a, 400b, 400c, ...), which are connected via a network N. Although not shown in the diagram, the radiation therapy planning device 300 may be configured as an element of the information processing system 1000A shown in Figure 2, for example, in the hospital or other facility where the radiation therapy planning device 300 is introduced, and may be connected to the medical image diagnostic device 100, the medical image database 200, the radiation therapy device 400, etc.

[0114] Server 2 is an information processing device that has the function of providing predetermined services to the user's terminal. In this embodiment, server 2 provides a radiation simulation service to the user's terminal, for example, through a radiation simulation application. The user's terminal may be any device the user can use, but in this embodiment, the radiation therapy planning device 300 is used as the user's terminal. Note that Server 2 may consist of multiple units rather than just one.

[0115] Server 2 includes, for example, a control unit 21, an input / output unit 22, a communication unit 25, a clock unit 27, and a storage unit 29. In this embodiment, unlike the example described above, Server 2 includes the information processing device 1 (or Server 2 corresponds to the information processing device 1).

[0116] The radiation therapy planning device 300 includes, for example, a control unit 310, an input / output unit 320 including an operation unit 330 and a display unit 340, a communication unit 350, a clock unit 370, and a storage unit 390. The operation unit 330 may have a touch panel, and the display unit 340 may be configured as a touchscreen.

[0117] <Processing> Figure 19 is a flowchart showing an example of the processing flow performed by each device in this embodiment. In this figure, the left side shows an example of the processing performed by the control unit 310 of the radiation therapy planning device 300, and the right side shows an example of the processing performed by the control unit 21 of the server 2. This process is executed by the control unit 310 of the radiation therapy planning device 300 and the control unit 21 of the server 2, respectively, based on programs stored in their respective memory units.

[0118] In this embodiment, for simplicity, it is assumed that for each patient, the medical image data stored in the medical image database 200 and various data used as input information have already been uploaded to the server 2. However, the embodiment is not limited to this.

[0119] The control unit 310 of the radiation therapy planning device 300 selects, for example, a subject to be sent to the server 2 for radiation simulation based on user input, and transmits subject selection information, including the identification information of the subject, to the server 2 via the communication unit 350 (R1). Furthermore, the control unit 310 of the radiation therapy planning device 300 selects, for example, the type of model and the type of information to be used as input information based on user input, and transmits the selected information to the server 2 via the communication unit 350 (R3).

[0120] When the communication unit 25 receives this information, the control unit 21 of the server 2 transmits candidate information to the radiotherapy planning device 300 via the communication unit 25, which includes, for example, images that can be selected as input information by the user of the radiotherapy planning device 300 from among various images of the subject that have been previously stored in the storage unit 29 (S1).

[0121] When candidate information is received by the communication unit 350, the control unit 310 of the radiation therapy planning device 300 displays the received candidate information on the display unit 340 (R5). When input information is selected from candidate information based on user input, the control unit 310 of the radiotherapy planning device 300 transmits selection information, including identification information of the selected input information, to the server via the communication unit 350 (R7).

[0122] When the communication unit 25 receives identification information of the input information, the control unit 21 of the server 2 performs radiation simulation processing using the input information corresponding to the received identification information and calculates dose distribution information (S3). Then, the control unit 21 of server 2 transmits the calculated dose distribution information to the radiation therapy planning device 300 via the communication unit 25 (S5).

[0123] When the communication unit 350 receives dose distribution information, the control unit 310 of the radiation therapy planning device 300 displays the received dose distribution information on the display unit 340 (R9).

[0124] Furthermore, after R9, it may be possible to modify the dose distribution information based on user input. In this case, for example, user operation information may be sent to server 2, server 2 may perform a process to modify the dose distribution information, update the dose distribution information in its memory, and send the modified dose distribution information to the radiation therapy planning device 300. The radiation therapy planning device 300 may then display the modified dose distribution information.

[0125] Furthermore, as mentioned above, it is also possible to configure the device to calculate contour information and hazard maps using morphological information as input. In this case, the control unit 21 of server 2 should calculate this information in the processing of S3. Alternatively, in this case, the contour information and hazard maps calculated by server 2 may be transmitted to the radiation therapy planning device 300, and the radiation therapy planning device 300 may display this information on the display unit 340.

[0126] <Display screen example> Figures 20 and 21 show examples of screens displayed on the display unit 340 of the radiation therapy planning device 300.

[0127] The upper part of Figure 20 shows the model selection screen for radiation simulation in the radiation simulation application. In this example, the top of the screen is configured to display the application name, which in this example is "Radiation Simulation Application". Below that, information about the patient selected as the target patient is configured, and in this example, the patient's name is displayed. Note that this explanation assumes the patient has already been selected, but it may be possible to allow the patient to be selected later.

[0128] Below that, along with the text "Please select a radiation simulation model," buttons corresponding to two models (hazard map input model and hazard map non-input model) are displayed. For convenience, the "hazard map input model" corresponds to the first calculation method shown in Figures 4 to 7, and the "hazard map non-input model" corresponds to the second calculation method shown in Figure 8.

[0129] For example, when the "Hazard Map Input Model" is selected via the control unit 330, the screen shown below in Figure 20 is displayed. This screen is for the user to select input information. In the center of the screen, along with the text "Please select the type of image to input," buttons corresponding to four different patterns (CT only, CT & hazard map, CT & contour, CT & contour & hazard map) are displayed. For convenience, "CT only" corresponds to the method shown in Figure 7, "CT & hazard map" corresponds to the method shown in Figure 6, "CT & contour" corresponds to the method shown in Figure 5, and "CT & contour & hazard map)" corresponds to the method shown in Figure 4. As mentioned earlier, morphological information is not essential for calculating dose distribution information, so "CT" can be omitted.

[0130] For example, when "CT & Contour" is selected via the control unit 330, an image selection screen (not shown) is displayed. On this image selection screen, the user selects an image to be used as input information, and the server 2 performs radiation simulation processing. Although not shown in the diagram, dose distribution information is then transmitted from the server 2 to the radiation therapy planning device 300 and displayed on the display unit 340. Here, the dose distribution information may be calculated as three-dimensional information, as described above, and a corresponding dose distribution image may be displayed on the display unit 340.

[0131] The upper part of Figure 21 shows an example of a screen that appears when, for example, a user performs an operation to display a predetermined tomographic image in two dimensions. On this screen, the left side of the drawing contains an area where UI elements such as buttons are displayed (hereinafter referred to as the "operation area"). In this example, a "Hazard Map" button for displaying the hazard map, a "Contour" button for displaying the contour (CT & contour), a "Modify" button for correcting the dose based on this dose distribution image, and a "Back" button for returning to the previous screen are displayed. On the right side of the screen, there is an area where the image is displayed (hereinafter referred to as the "image area"). On this screen, a two-dimensional dose distribution image is displayed.

[0132] For example, when the "Hazard Map" button is pressed via the control unit 330, the display changes to the screen shown in the center of Figure 21, for instance. On this screen, the "Hazard Map" button in the operation area disappears, and the image area displays two images side-by-side horizontally: the dose distribution image and the hazard map generated by Server 2 to produce this dose distribution image. Note that, unlike this example, they may also be displayed side-by-side vertically. This allows users to view not only dose distribution images but also hazard maps. While displaying hazard maps is not always necessary, it is convenient to be able to display them alongside the images, especially when conducting conferences about cases.

[0133] Furthermore, the above images may be displayed in a way that allows them to be enlarged or reduced within the image area based on user zoom operations, rotated within the image area based on user rotation operations, or moved within the image area based on user movement operations. Two or more images may also be displayed superimposed on each other. In addition, displays based on various commonly performed operations may be made possible.

[0134] For example, when the "Contour" button is operated via the control unit 330, the display changes to the lower screen shown in Figure 21. This screen displays the dose distribution image and hazard map from the central screen of Figure 21, as well as the contour image (CT & contour image).

[0135] Although not shown in the diagram, when the "Modify" button is selected via the operation unit 330, a UI for modifying the dose may be displayed in the operation area. Based on the user's modification operation, the operation information may be sent to the server 2, and the dose distribution information may be modified by the server 2. The modified dose distribution image may then be sent to the radiation therapy planning device 300 and displayed on the display unit 340. Alternatively, the radiation therapy planning device 300 may perform correction processing and send the correction results to the server 2.

[0136] Note that the display screen can be configured similarly if, for example, "Hazard Map Non-Input Mode" is selected on the upper screen of Figure 20, or if, for example, "CT Only," "CT & Hazard Map," or "CT & Contour & Hazard Map" is selected on the lower screen of Figure 20, so illustrations and explanations are omitted.

[0137] The above describes an embodiment of the client-server system, but for example, the radiotherapy planning device 300 may perform the above processing using Stanalon. Furthermore, the radiotherapy planning device 300 may perform some of the processing that the server 2 performs, and the server 2 may perform some of the processing that the radiotherapy planning device 300 performs.

[0138] <Effects and Effects of the Embodiment> The information processing device (for example, information processing device 1, server 2, and radiation therapy planning device 300) calculates dose distribution information for organs based on first information (for example, a hazard map) that can identify areas in organs where exposure is dangerous. This allows for the appropriate calculation of dose distribution information for organs, taking into account areas in organs where exposure would be dangerous.

[0139] In this case, the information processing device may calculate dose distribution information for an organ based on an artificial intelligence model based on first information and second information (e.g., contour information) relating to at least the location of the region of interest in the organ. This allows for the simple and accurate calculation of dose distribution information related to organs.

[0140] Furthermore, the information processing device may calculate dose distribution information for organs based on an artificial intelligence model based on the first information and at least morphological information (e.g., CT information) related to the organs. This allows for the simple and accurate calculation of dose distribution information related to organs.

[0141] <Modified examples of embodiments> (1) In the above embodiment, information to assist (support) the user in manually creating dose distribution information based on a hazard map (hereinafter referred to as "auxiliary information") may be displayed on the display device. Here, we will illustrate how this can be achieved using the above-mentioned radiation therapy planning device 300.

[0142] Figure 22 shows an example of the information stored in the memory unit 390 of the radiotherapy planning device 300 in this modified example. The memory unit 390 stores, for example, a dose distribution information creation assistance program 491, which is read by the control unit 310 and executed as a dose distribution information creation assistance process, hazard map data 493, which is reference hazard map data, and dose distribution data 495, which is the data of the created dose distribution information.

[0143] Figure 23 shows an example of a dose distribution information creation assistance screen displayed on the display unit 340 in this modified example. Here, it is shown as an example of the screen for the radiation simulation application described above. Although a 2D display is shown here, a 3D display may also be possible. In this example, the hazard map image (in this example, a pneumonia hazard map image) and the created dose distribution image (in this example, a lung dose distribution image) are displayed side-by-side vertically in the right-hand area below the patient's name. The dose distribution image also displays information indicating the intensity of the dose using hatching (or color coding), and the hazard map image displays information indicating the risk of pneumonia development using hatching (or color coding). Additionally, a DVH (Dose Volume Histogram) is displayed. Furthermore, morphological information such as CT images and contour information of the region of interest may be displayed.

[0144] In this example, input information is displayed on the left side of the screen as a UI (user interface) for the user to input information regarding the radiation dose via the operation unit 330. In other words, in this example, the system is configured so that the user can input information regarding the radiation dose while viewing the hazard map. In this example, the input information includes items for radiation dose based on definitions of volumes related to radiotherapy, such as macroscopic tumor volume (GTV), as well as items that allow for the collectively specifying radiation doses for high-risk areas on the hazard map. In this example, the item SRH, indicated as "Risk High" in the "Type" column, corresponds to this. For example, the user can enter a value in this item that they consider to be appropriate for high-risk areas on the hazard map. As shown in this figure, it may also be possible to input lower and upper limits for the radiation dose. Alternatively, a similar approach could be taken by creating a category (for example, "Risk Middle" or "Risk Low") that allows for the collectively specifying radiation doses for areas with moderate or low risk on the hazard map.

[0145] Hazard maps and input information are related to locations where exposure to radiation in organs is dangerous, and therefore can be considered a type of information regarding locations where exposure to radiation in organs is dangerous.

[0146] When the user inputs (specifies) the radiation dose using various items including the SRH item, the control unit 310 performs a dose distribution optimization process, which optimizes the dose distribution based on, for example, a hazard map and its radiation dose (or numerical range of the radiation dose), and displays the resulting dose distribution information on the display unit 340. In the dose distribution optimization process of this method, for example, the dose distribution optimization calculation may be performed to optimize the overall dose based on the areas identified as dangerous in the hazard map and the user-inputted irradiation dose for those areas, while also considering the irradiation dose for other areas.

[0147] Alternatively, the system may display dose distribution information that directly reflects the radiation dose entered by the user.

[0148] Furthermore, for example, it may be possible to request (command) the system to collectively reduce the radiation dose in dangerous areas on the hazard map and optimize the dose distribution, without directly inputting numerical values ​​for the radiation dose. Furthermore, for example, the system may be designed to allow users to directly input radiation doses for dangerous areas in a hazard map image, thereby enabling users to individually specify radiation doses for dangerous areas or to request (command) a reduction in radiation doses for those areas.

[0149] Furthermore, by displaying the input information described above along with the DVH (Dose Volume Histogram), users can create a radiation therapy plan without having to view images such as morphological information, contour information, hazard maps, and dose distributions. Therefore, these images do not necessarily need to be displayed.

[0150] In this way, it is possible to create handmade dose distribution information (dose distribution information based on hazard maps). Furthermore, the dose distribution information of a subject created in this way (dose distribution information based on hazard maps) can be used for the subject's actual radiation therapy. Furthermore, the dose distribution information of subjects created in this manner (dose distribution information based on hazard maps) may be used as training data for the artificial intelligence model described in the above embodiment to train the model.

[0151] Figure 24 is a flowchart showing an example of the flow of the process for assisting in the creation of dose distribution information. The control unit 310, for example, displays the various auxiliary information mentioned above on the display unit 340 (R21). Next, the control unit 310 receives user input, for example, via the operation unit 330 (R23). Specifically, it receives input for the irradiation dose for the specified information displayed as auxiliary information. Then, the control unit 310 performs, for example, dose distribution optimization processing and saves the obtained dose distribution information in the storage unit 390 as dose distribution data 495 (R25). Then, the control unit 21 displays the dose distribution information on the display unit 340 (R27) and terminates the process.

[0152] For example, the control unit 310 may determine whether the dose in some or all of the areas on the hazard map where the risk of pneumonia occurrence is above a certain level is appropriate, and display information (such as "OK / NG") based on that determination. Such information may also be considered a type of auxiliary information. Specifically, for example, the control unit may set it to determine "pneumonia occurrence" when f(D) is equal to or greater than the cutoff value, using the aforementioned equation (2) or equation (3). Then, the control unit 310 may calculate f(D) using the dose in the target area (e.g., average exposure dose) as the exposure dose D based on the created dose distribution information, and if it determines "pneumonia occurrence," it may display information indicating that the dose in the target area is not appropriate. Also, for example, if there is at least one area determined to have "pneumonia occurrence," or if there is a certain number (or a certain percentage or more) of such areas, it may display information indicating that the dose distribution is not appropriate. Furthermore, the above determinations and displays may be made during the creation of the dose distribution information, or they may be made after the creation of the dose distribution information is complete.

[0153] Furthermore, morphological and functional information (which may also be mathematical information), contour information, and other information related to the aforementioned organs may also be displayed. Furthermore, it may be possible to display various types of information that can generally be displayed on a radiation therapy planning system. Alternatively, the above can be implemented in a standalone configuration or in a client-server system.

[0154] (2) In the above embodiment, a neural network model or deep learning model, which is a type of artificial intelligence model, may be used to calculate the hazard map.

[0155] Figure 25 shows an example of the model configuration of the deep learning model DM1, which is an example of a deep learning model in this case. The deep learning model DM1 may consist of, for example, a CNN model such as ResNet or a neural network model such as ViT (Vision Transformer). The deep learning model DM1, for example, takes N features as input and outputs a value representing the probability of an adverse event (radiation pneumonitis in the case of lung disease) occurring, according to the trained neural network model. The output layer of the deep learning model DM1 may have multiple output nodes to enable multi-class classification, or it may have one output node to enable binary classification (for example, presence or absence of adverse event: occurrence "1", absence "0"). The output may also be a probability (0 to 1).

[0156] Functionally speaking, the deep learning model DM1 includes, for example, a hazard map generation model unit and a harmful event occurrence probability information calculation model unit. The hazard map generation model takes, for example, N features as input to the input layer, multiplies each feature by a pre-determined weight corresponding to that feature, and outputs the sum of these results (more precisely, information transformed by an activation function) from the output layer. This output information corresponds to a hazard map. The adverse event probability information calculation model unit calculates exposure doses according to the risk classification in the hazard map, for example, based on the hazard map output from the output layer of the hazard map generation model unit and separately input dose distribution information. Then, it multiplies the calculated exposure doses by weights that have been determined in advance as weights corresponding to each risk classification, and outputs the sum of these (more precisely, information transformed by an activation function) from the output layer. This output information corresponds to adverse event probability information.

[0157] In the lung example mentioned earlier, a pneumonia hazard map can be obtained as a hazard map, and pneumonia risk information can be obtained as adverse event risk information.

[0158] Additionally, a cutoff value for the probability of adverse events occurring may be included as a model parameter and explored during the learning phase. Alternatively, thresholds (upper and lower thresholds) for percentile values ​​that define fractionation intervals (for example, the lung function percentile values ​​mentioned above) may also be included as model parameters and explored during the learning phase.

[0159] (3) In the above embodiment, organ function information was used as a feature to calculate a hazard map and a value representing the likelihood of adverse events occurring, but the embodiment is not limited to this. The information used as features may include not only functional information (including information that can be considered functional information) but also morphological information (including information that can be considered morphological information), and a combination of functional information and morphological information may also be used.

[0160] In this case, for example, the predictive model represented by equation (4) below may be used as the predictive model corresponding to equation (2).

number

[0161] This prediction model calculates the exposure dose corresponding to each of the M features (M exposure doses D M This is a model in which the explanatory variables are ) and the predictive model is trained in the same manner as in the embodiment described above, and the weights α (M weights α) corresponding to each of the M explanatory variables are used. M You may calculate ). Furthermore, when using morphological information as a feature, for example, morphological information (e.g., CT information) can be used to determine the percentile region in the same way as the functional percentile region described above, and this can then be used as a feature.

[0162] The same procedure can be applied when using deep learning, as explained in Modification (2), so a detailed explanation is omitted.

[0163] (4) In the above embodiment, the fractionation interval was divided into five parts, but the invention is not limited to this. The number of fractionation intervals can be set as appropriate; for example, numbers such as "10," "15," or "20" may be set. This setting can be done automatically by the computer or based on user input. However, if the number of fractionation intervals is too small, information will be lost, so it is best to set the value to something like "3" to "10." Furthermore, thresholds (upper and lower thresholds) for percentile values ​​defining fractionation intervals can be determined through learning by including them in the model parameters as described above, or they can be set arbitrarily.

[0164] (5) In radiation therapy, at least a portion of an organ other than the target organ (the lung in the example of the above embodiment) (hereinafter referred to as "other organs") may be included in the radiation field, which may cause specific adverse events in that organ. Such organs are called organs at risk (OAR).

[0165] Therefore, dose distribution information to be used as training data may be calculated while also considering the effects on other organs. In this case, for example, a hazard map similar to the lung hazard map described above may be calculated for other organs adjacent to the lungs, based on a method similar to that of the embodiment described above, and dose distribution information may be calculated by taking into account the hazard maps calculated for the other organs as well. In this case, for example, a hazard map for the organ can be calculated by quantifying functional information based on morphological information obtained for that organ and performing the same processing as described above. If functional information can be directly obtained through nuclear medicine examinations, etc., that information may also be used.

[0166] In this case, for example, the irradiation range can be set such that the dose in the area where the hazard map value exceeds the threshold is relatively lower than the dose in the other areas.

[0167] Generally, the approach is to exclude other organs (especially vital organs) from the irradiation area. However, if the area is expected to have a low risk of adverse events, it may be acceptable to include other organs in the irradiation area. In this case, for example, the hazard map values ​​can be modified based on the lung hazard map values, the hazard map values ​​for other organs, and the corresponding location information. Specifically, the following processes can be performed, for example: • Determine specific body parts from volume data, etc. • Change the hazard map values ​​for body parts other than specific body parts to values ​​that do not exceed the threshold.

[0168] The specific area may be, for example, a site including a lung tumor, and a site set as a predetermined range of the vicinity of that site (hereinafter referred to as the "predetermined range of the vicinity"), and a site of another organ, in this case, • Specific areas (areas within a predetermined vicinity & areas of other organs): Areas where the hazard map value does not exceed the threshold will have a relatively higher dose, while areas where the value exceeds the threshold will have a relatively lower dose. • Areas excluding specific areas: The dose is relatively high because it does not exceed the threshold. It can be done that way. In this case, other organs may also be included in the irradiation range, but the dose can be made relatively lower for areas where the hazard map value exceeds the threshold, thus reducing exposure to areas with a high risk of adverse events. For example, if other organs are designated as vital organs, exposure to areas of those vital organs with a high risk of adverse events can be reduced.

[0169] Furthermore, if the approach is to exclude other organs from the irradiation range, then, for example, the hazard map values ​​for other organs may be changed to values ​​exceeding the threshold. In this case, for example, if other organs are designated as vital organs, it would be possible to reduce or minimize exposure to these vital organs to zero.

[0170] Furthermore, for example, among other organs, those for which adverse events are unlikely to occur, or for which the risk is considered low even if adverse events do occur, may be included in the irradiation range, at least a portion of them.

[0171] Furthermore, within the organ targeted by radiation therapy (the lungs in the above embodiment), it is also possible to consider avoiding radiation exposure to specific areas. In this case, for example, the hazard map values ​​for areas within the organ, excluding those in a predetermined vicinity, may be changed to values ​​that do not exceed a threshold. In this case, within the organ being treated, • Areas within a predetermined vicinity: Locations where the hazard map value does not exceed the threshold will have relatively higher doses, while locations where the value exceeds the threshold will have relatively lower doses. • Other body parts: The dose is relatively high because it does not exceed the threshold. It can be done that way.

[0172] For example, the dose distribution information calculated in this way may be used as training data in a dose distribution prediction model.

[0173] (6) The methods of the above embodiments may be applied similarly to medical procedures such as examinations and diagnoses using radiation, in addition to radiation therapy. In other words, they may be applied to all aspects of radiation medicine, including treatment, examinations, and diagnoses, and are applicable to all cases in which adverse events may occur due to radiation exposure.

[0174] Furthermore, it is certainly not limited to the lungs (lung cancer), and other organs may also be targeted.

[0175] Furthermore, the method of the above embodiment may be applied to only a portion of the target organ, rather than the entire organ. It is sufficient to perform the processing on at least the region including the affected area, and information on the entire target organ is not essential.

[0176] (7) The methods of the above embodiments may be applied similarly to animals (excluding humans) as well as humans, and may be applied to subjects including humans and animals. In this case, the subjects may be, for example, subjects of the same type. Specifically, when applied to animals, for example, if it is dogs, the model may be trained using training data from multiple dogs. [Explanation of Symbols]

[0177] 1 (1A, 1B) Information Processing Device 2 servers 100 Medical imaging diagnostic equipment 200 Medical Image Databases 300 Radiation therapy planning system 400 Radiation therapy equipment

Claims

1. A program to be executed by a computer, A program for causing a computer to calculate dose distribution information for an organ based on first information that can identify areas in the organ that are at risk of radiation exposure.

2. The first information is information calculated based on mathematical information relating to the organ. The program according to claim 1.

3. The first piece of information described above can be calculated using an artificial intelligence model. The program according to claim 1.

4. The dose distribution information is calculated by the computer based on an artificial intelligence model based on the first information and second information relating to at least the location of the region of interest in the organ. The program according to claim 1.

5. The artificial intelligence model is a model trained using the dose distribution information based on the first information as training data. The program according to claim 4.

6. The artificial intelligence model is a model trained using training data that includes the training data and at least the second information. The program according to claim 5.

7. The training data includes morphological information relating to the organs, The program according to claim 6, which causes the computer to calculate dose distribution information relating to the organs of the subject based on the artificial intelligence model, the second information of the subject, and the morphological information of the subject.

8. The program according to claim 7, which causes the computer to calculate the second information of the subject based on the morphological information of the subject.

9. The training data includes the first information, The program according to claim 6, which causes the computer to calculate the dose distribution information relating to the organs of the subject based on the artificial intelligence model, the second information of the subject, and the first information of the subject.

10. The training data includes morphological information relating to the organs, The program according to claim 9, which causes the computer to calculate dose distribution information relating to the organs of the subject based on the artificial intelligence model, the second information of the subject, the first information of the subject, and the morphological information of the subject.

11. The program according to claim 10, which causes the computer to calculate the second information of the subject based on the morphological information of the subject.

12. The program according to claim 10, which causes the computer to calculate the first information of the subject based on the morphological information of the subject.

13. The dose distribution information is calculated by the computer based on an artificial intelligence model based on the first information and morphological information relating to at least the organs. The program according to claim 1.

14. The artificial intelligence model is a model trained using the dose distribution information based on the first information as training data. The program according to claim 13.

15. The artificial intelligence model is a model trained using training data that includes the training data and at least the morphological information. The program according to claim 14.

16. The dose distribution information is calculated by the computer based on the first information and user input information regarding the radiation dose to the hazardous location. The program according to claim 1.

17. The first piece of information is information that identifies the first location as the dangerous place, The input information is information for reducing the radiation dose at the first location. L as described in claim 16.

18. The first information is information that identifies the first location and the second location as the dangerous locations, The input information is information for reducing the radiation dose for the first location and the second location. L as described in claim 16.

19. A program to be executed by a computer, A program to cause the computer to display information on a display device to assist the user in creating dose distribution information for said organs, including information about areas in said organs that are at risk of radiation exposure.

20. The information relating to the hazardous location includes user input information relating to the radiation dose to the hazardous location. The program according to claim 19.

21. The aforementioned input information includes character information representing the dangerous location. The program according to claim 20.

22. The information relating to the hazardous location includes first information that can identify the hazardous location. The program according to claim 19.

23. An information processing device, An information processing device comprising a processing unit that calculates dose distribution information for an organ based on first information that can identify areas in the organ that are at risk of radiation exposure.

24. The information processing apparatus according to claim 23, A display unit that displays the dose distribution information, A radiation therapy planning system equipped with the following features.

25. The information processing apparatus according to claim 23, A communication unit that transmits the dose distribution information, A server equipped with the following features.

26. A method used by computers, A method comprising calculating dose distribution information for an organ based on first information that allows for the identification of areas in the organ that are at risk of radiation exposure.

27. An information processing device, An information processing device comprising a processing unit that performs processing to display on a display device information to assist a user in creating dose distribution information for an organ, including information about areas in the organ that are dangerous to be exposed to radiation.

28. A method used by computers, A method comprising displaying information on a display device to assist a user in creating dose distribution information for an organ, including information about locations in the organ where exposure to radiation is dangerous.

Images