Information processing methods, computer programs, and information processing devices.

Abstract

The objective is to provide an information processing method, a computer program, and an information processing device that can output radiation therapy planning information that takes into account the patient's condition. [Solution] The present invention relates to an information processing method in which an information processing device 1 stores information on the location and cross-sectional contour of the affected area, acquires the patient's electronic medical record information TD, and inputs the acquired electronic medical record information TD and the information on the location and cross-sectional contour of the affected area to a learning model that is trained to output a radiation dose distribution suitable for the patient in response to the input of information on the location and cross-sectional contour of the affected area and the electronic medical record information TD, acquires the dose distribution output by the learning model, and outputs an ideal dose distribution based on said dose distribution.

Description

Technical Field

[0001] The present invention relates to an information processing method, a computer program, and an information processing apparatus for outputting radiation treatment plan information in radiation treatment.

Background Art

[0002] In recent radiation treatments, intensity-modulated radiation therapy (IMRT), volumetric-modulated radiation therapy (VMRT), etc. that not only efficiently concentrate radiation on a target tumor but also reduce irradiation to surrounding normal tissues have become widespread. To perform such advanced radiation treatments, it is necessary for medical staff to create in advance precise radiation treatment plan information including the desired dose distribution, etc., and at present, a lot of time and labor are required until the medical staff determines the ideal radiation treatment plan information (for example, dose distribution, contours of targets and organs, irradiation parameters, etc.) that they consider optimal.

[0003] As a solution to this, a system has been proposed (Patent Document 1) that uses a learning model obtained by pre-training deep learning with anatomical image information such as MRI images, CT images, and contour data of tumors and organs, and creates radiation treatment plan information (dose distribution) that medical staff desire based on the anatomical image information of the target patient. With such a system, the time required to create radiation treatment plan information has been significantly reduced.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] However, these machine learning-based systems have the drawback of outputting radiation therapy planning information based solely on anatomical image information, and therefore failing to output radiation therapy planning information that takes into account the patient's circumstances, such as the patient's wishes or medical history.

[0006] Therefore, in view of the above circumstances, the present invention aims to provide an information processing method, a computer program, and an information processing device that enable medical professionals to output radiation therapy planning information that is more appropriate to the patient by taking the patient's condition into greater consideration when creating radiation therapy planning information. [Means for solving the problem]

[0007] In light of the current situation, the inventors of this invention have conducted thorough research and conceived the idea that by using electronic medical record information, it is possible to output radiation therapy planning information that better reflects the patient's condition.

[0008] In other words, by incorporating not only direct anatomical image information but also indirect patient information into the radiation therapy planning information, the burden on medical professionals can be reduced. Furthermore, by considering the patient's circumstances, such as the patient's wishes and medical history, it becomes possible to create radiation therapy planning information that is appropriate for the patient. This discovery led to the completion of the present invention.

[0009] In other words, the present invention encompasses the following inventions. (1) An information processing method comprising an information processing device that acquires information on an affected area based on an anatomical image, acquires electronic medical record information which is information based on the patient's electronic medical record, and outputs radiation therapy planning information based on the information on the affected area and the electronic medical record information. Information processing methods.

[0010] (2) An information processing method comprising: an information processing device storing information on the location and cross-sectional contour of an affected area; acquiring electronic medical record information which is information based on the patient's electronic medical record; inputting the acquired electronic medical record information and the information on the location and cross-sectional contour of the affected area into a learning model that is trained to output a radiation dose distribution suitable for the patient in response to inputs of information on the location and cross-sectional contour of an affected area and electronic medical record information which is information based on the patient's electronic medical record; acquiring the dose distribution output by the learning model; and outputting an ideal dose distribution based on said dose distribution.

[0011] (3) An information processing method comprising: an information processing device storing image information of the location and cross-section of an affected area; acquiring electronic medical record information of a patient, which is information based on the patient's electronic medical record; inputting the acquired electronic medical record information and the image information of the location and cross-section of the affected area into a learning model that is trained to output target and organ contours within a range suitable for the patient in response to the input of image information of the location and cross-section of an affected area and electronic medical record information, which is information based on the patient's electronic medical record; acquiring target and organ contours within a range output by the learning model; and outputting said contours.

[0012] (4) The information processing method according to (2) or (3), wherein the input of the electronic medical record information to the learning model is done via a large-scale language model (LLM).

[0013] (5) The information processing method described in (4), comprising: generating text prompts from the information of the patient's electronic medical record that should be reflected in the radiotherapy plan information; using the large-scale language model, dividing the text prompts into tokens; and extracting the tokens that should be reflected in the radiotherapy plan information.

[0014] (6) The information processing method according to (2) or (3), wherein the learning model is a model that has learned at least one of the patient's basic information and medical information from the text information contained in the electronic medical record information as learning data.

[0015] (7) The information processing method according to (6), wherein the learning model is a model that has been learned based on electronic medical record information which includes at least one of the following as basic information of the patient: a medical history record about the patient's medical history, a medication record about the medications the patient is taking, a lifestyle record about the patient's lifestyle, and a preference record about the patient's preferences regarding invasiveness.

[0016] (8) The information processing method according to (6), wherein the learning model is a model that has been learned based on the input of electronic medical record information which includes at least one of the following as medical information of the patient: a record of examination findings, a record of diagnosis, a record of consideration, and a record of treatment plan by a healthcare professional.

[0017] (9) The information processing method described in (1), wherein the electronic medical record information includes the initial consultation record and the progress notes.

[0018] (10) The information processing method according to (2), wherein the dose distribution output by the learning model is input into a large-scale language model (LLM) to be verbalized, and the verbalized dose distribution information is fed back to the learning model.

[0019] (11) A computer program that causes a computer to perform information processing, which involves acquiring information about an affected area based on anatomical images, acquiring electronic medical record information which is information based on the patient's electronic medical record, and outputting radiation therapy planning information based on the information about the affected area and the electronic medical record information.

[0020] (12) A computer program that causes a computer to store information on the location and cross-sectional contour of an affected area, acquire electronic medical record information which is information based on the patient's electronic medical record, input the acquired electronic medical record information and the information on the location and cross-sectional contour of the affected area into a learning model which is machine-learned to output a radiation dose distribution suitable for the patient in response to inputs of information on the location and cross-sectional contour of the affected area and the electronic medical record information which is information based on the patient's electronic medical record, acquire the dose distribution output by the learning model, and output an ideal dose distribution based on said dose distribution.

[0021] (13) A computer program that causes a computer to store image information of the location and cross-section of an affected area, acquire electronic medical record information which is information based on the patient's electronic medical record, input the acquired electronic medical record information and the image information of the location and cross-section of the affected area into a contour drawing model which is machine-learned to output contours of targets and organs within a range suitable for the patient, acquires contours of targets and organs within a range output by the learning model, and outputs the contours.

[0022] (14) An information processing device comprising: a diseased area information acquisition unit that acquires information about a diseased area based on an anatomical image; a medical record information acquisition unit that acquires electronic medical record information which is information based on the patient's electronic medical record; and an output unit that outputs radiation therapy planning information based on the diseased area information and the electronic medical record information.

[0023] (15) An information processing apparatus, comprising: a storage unit that stores information on the position of the affected part and the contour of the cross-section; a medical record information acquisition unit that acquires medical record information which is information based on the electronic medical record of the patient; a dose distribution acquisition unit that inputs the acquired medical record information, the information on the position of the affected part and the contour of the cross-section into a learning model that is machine-learned to output a dose distribution of radiation suitable for the patient, and acquires the dose distribution output by the learning model; and an output unit that outputs an ideal dose distribution based on the dose distribution.

[0024] (16) An information processing apparatus, comprising: a storage unit that stores information on the position of the affected part and the image information of the cross-section; a medical record information acquisition unit that acquires medical record information which is information based on the electronic medical record of the patient; a contour acquisition unit that inputs the acquired medical record information, the information on the position of the affected part and the image information of the cross-section into a learning model that is machine-learned to output the contours of the target and organs within a range suitable for the patient, and acquires the contours of the target and organs within the range output by the learning model; and an output unit that outputs the contours. [Effect of the Invention]

[0025] According to the information processing method, computer program, and information processing apparatus according to the present invention as described above, not only direct anatomical image information but also indirect information regarding the patient is reflected in the radiation therapy plan information, thereby reducing the burden on medical staff and creating radiation therapy plan information suitable for the patient by considering the patient's situation such as the patient's wishes and medical history. [Brief Description of the Drawings]

[0026] [Figure 1] A block diagram showing an information processing apparatus according to an embodiment of the present invention. [Figure 2] An explanatory diagram showing the operation of a processing unit in the information processing apparatus. [Figure 3] Similarly, this is an explanatory diagram showing the various processing operations within the processing unit. [Figure 4] A flowchart showing the processing steps up to outputting the contour. [Figure 5] A flowchart showing the processing steps up to outputting the dose distribution. [Figure 6] A diagram illustrating an example of context included in electronic medical record information. [Figure 7] A diagram illustrating an example of context included in electronic medical record information. [Figure 8] A diagram illustrating an example of context included in electronic medical record information. [Modes for carrying out the invention]

[0027] Hereinafter, typical embodiments of the present invention will be described in detail with reference to the attached drawings.

[0028] As shown in Figure 1, the information processing device 1 according to this embodiment is a server device comprising a processing unit 10, a storage unit 20, and a communication unit 30. The information processing device 1 may be composed of a single computer, or it may be composed of multiple computers that are communicated with each other. In this example, the contour drawing process and the dose distribution generation process using the created contour are shown as being performed by a single server device, but it is also preferable that the contour drawing process and the dose distribution generation process be performed by separate server devices. In this example, the storage and execution of each learning model, which will be described later, are shown as being performed by a single server device, but the server device that stores and executes each learning model may be different for each learning model.

[0029] The processing unit 10 can be composed of an arithmetic processing unit including a circuit consisting of a CPU (Central Processing Unit), GPU (Graphics Processing Unit), ASIC (Application Specific Integrated Circuit), PLD (Programmable Logic Device), or FPGA (Field Programmable Gate Array). In other words, the functions of the processing unit 10 are realized by the arithmetic processing unit. The storage unit 20 can be composed of a storage device consisting of an HDD or SSD, and RAM, etc. The communication unit 30 can communicate with various devices via a network N such as a mobile phone communication network, wireless LAN (Local Area Network), and the Internet.

[0030] The information processing device 1 communicates via the network N with an image database containing anatomical images such as CT images and MRI images, and an electronic medical record database containing electronic medical records, through the communication unit 30, and can acquire necessary anatomical images and electronic medical record data. In the following, anatomical images will be referred to as image information, and an example will be given where the anatomical image is a CT image. Furthermore, electronic medical record information may be all of the information recorded in the electronic medical record, or only a part of it. For example, electronic medical record information may be a portion of the information recorded in the electronic medical record of the target patient, taking into consideration the patient's personal information, or information resulting from processing that removes personal information based on the information recorded in the electronic medical record using a predetermined processing model (for example, statistical information resulting from statistical processing or classification information resulting from classification processing). Furthermore, electronic medical record information may be information linked to the electronic medical record of the target patient, for example, information such as conference information for the target patient.

[0031] The memory unit 20 includes at least a program memory unit 21 in which a computer program for the information processing device 1 to perform an information processing method is stored, a learning model memory unit 22 in which a learning model for creating contours and dose distributions is stored, and a data memory unit 23 in which a tuning dataset for refining and optimizing the learning model is stored.

[0032] The learning models stored in the learning model memory unit 22 include at least a contour drawing model M1, a dose distribution generation model M2, a contour prompt generation model M3, a contour language processing model M4, a dose distribution prompt generation model M5, and a dose distribution language processing model M6.

[0033] As shown in Figure 2, the contour drawing model M1 is a machine learning model that takes medical images, particularly CT images and anatomical image information GD1 of tumors and organs, and text information based on electronic medical records as input, and outputs contour image information GD2 which includes target and organ contour segments within the range deemed appropriate for the patient. The dose distribution generation model M2 is a machine learning model that takes contour image information GD2 output by the contour drawing model M1 and text information based on electronic medical records as input, and outputs dose distribution image information GD3 which includes radiation dose distributions deemed appropriate for the patient.

[0034] As shown in Figure 2, the contour prompt generation model M3 and the dose distribution prompt generation model M5 are machine learning models that extract context for contour drawing / dose distribution generation from the contents of the electronic medical record information TD and generate text prompts TP to be input to the contour language processing model M4 / dose distribution language processing model M6. Note that " / " is used to mean "or".

[0035] As shown in Figure 2, the contour language processing model M4 and the dose distribution language processing model M6 are machine learning models that receive text prompts TP generated by the contour prompt generation model M3 and the dose distribution prompt generation model M5, and output tokens to be reflected in the contour drawing model M1 and the dose distribution generation model M2. The machine learning models for contour language processing model M4 and dose distribution language processing model M6 can be adapted by applying a pre-trained large-scale language model (LLM) that has learned text information related to radiotherapy, such as Llama2-7B. Alternatively, any known language processing model that can extract tokens from text prompts TP based on electronic medical record information that are preferable to be reflected in the outputs of the contour drawing model M1 and the dose distribution generation model M2 can be widely applied. Note that in this invention, learning in the learning model may refer to the process of tuning the parameters of the learning model based on the learning data, or it may refer to the act of preconditioning the learning model with prompts or the like.

[0036] The algorithms used in these learning models are not limited to machine learning algorithms based on neural networks. For example, machine learning algorithms based on linear regression analysis or logistic regression may also be used. Other methods such as statistical analysis and mathematical optimization may also be used as appropriate. Expert systems may also be used, as well as rule-based algorithms that extract text containing predetermined keywords.

[0037] As shown in Figure 1, the processing unit 10 includes an image information acquisition unit 11A, a medical record information acquisition unit 12A, an image processing unit 13A, a prompt processing unit 14A, a text processing unit 15A, a modality processing unit 16A, and an output unit 17A for contour drawing. The processing unit 10 also includes a contour image information acquisition unit 11B, a medical record information acquisition unit 12B, an image processing unit 13B, a prompt processing unit 14B, a text processing unit 15B, a modality processing unit 16B, and an output unit 17B for dose distribution generation. The processing unit 10 functions using a computer program stored in the storage unit 20, utilizing a contour drawing model M1, a dose distribution generation model M2, a contour prompt generation model M3, a contour language processing model M4, a dose distribution prompt generation model M5, and a dose distribution language processing model M6.

[0038] (Each processing unit for outline drawing) As shown in Figure 3, the image information acquisition unit 11A for contour drawing performs the process of acquiring image information GD1 to be used by the image processing unit 13A. More specifically, the image information acquisition unit 11A acquires one or more image information GD1 (CT images) from an external image database (not shown) via the communication unit 30 and stores them in the data storage unit 23.

[0039] As shown in Figure 3, the medical record information acquisition unit 12A performs the process of acquiring electronic medical record information TD to be used by the text processing unit 15A. More specifically, the medical record information acquisition unit 12A acquires electronic medical record information TD from an external electronic medical record database (not shown) via the communication unit 30.

[0040] As shown in Figure 3, the image processing unit 13A inputs image information GD1 and electronic medical record information (vectors) converted by the modality processing unit 16A (described later) into the contour drawing model M1, and acquires and outputs the output contour image information GD2. More specifically, the image processing unit 13A uses the contour drawing model M1 to extract features of the image information GD1 acquired by the image information acquisition unit 11A via encoder processing, and then, through decoder processing based on these features, outputs contour image information GD2 consisting of contours of targets and organs within a range suitable for the patient, taking into account the patient's circumstances, such as the patient's wishes and medical history.

[0041] As shown in Figure 3, the prompt processing unit 14a analyzes the content that needs to be reflected in the contour image information GD2 from the electronic medical record information TD using the contour prompt generation model M3, extracts the context from the content, and generates a text prompt TP. The content in the electronic medical record information is usually unstructured natural language text. More specifically, the prompt processing unit 14a generates the context of the content to be reflected in the contour image information GD2 into a text prompt TP using the contour prompt generation model M3. Examples of the context of the content to be reflected include basic patient information such as medical history records, medication records, lifestyle records, and preference records regarding the patient's wishes regarding invasiveness; patient medical information such as examination findings, diagnostic records, discussion records, and treatment plan records by healthcare professionals; and vital information from clinical examinations and examination information collected by medical devices.

[0042] As shown in Figure 3, the text processing unit 15a uses the contour language processing model M4 to divide the text prompt TP into tokens TK, and then identifies and extracts the tokens TK necessary to reflect them in the contour image information GD2 output by the image processing unit 13a. More specifically, the contour language processing model M4 analyzes the text prompt TP and divides it into multiple tokens TK, and the text processing unit 15a extracts tokens TK from the multiple tokens TK divided by the contour language processing model M4 that are preferable to reflect in the target and contour segmentation by the contour drawing model M1.

[0043] As shown in Figure 3, the modality processing unit 16a converts the token TK extracted by the text processing unit 15A into a data format that can be reflected in the contour drawing model M1, such as a vector, and the image processing unit 13A inputs (reflects) this into the contour drawing model M1. As a result, the image processing unit 13a extracts the features of the image information GD1 via encoder processing by the contour drawing model M1, and outputs contour image information GD2 via decoder processing by the contour drawing model M1 based on these features and the input vector (information based on the electronic medical record). The contour image information GD2 reflects the contents of the electronic medical record information TD.

[0044] As shown in Figure 3, the output unit 17A outputs contour image information GD2 output by the image processing unit 13A. In this embodiment, the image information acquisition unit 11B for dose distribution generation, which will be described later, acquires the contour image information GD2 created by the image processing unit 13A. However, the output unit 17A may also output it to the data storage unit 23 or to an external source as data when creating radiation therapy planning information, and is not limited to this.

[0045] Furthermore, the image processing unit 13A extracts features of contour image information GD2 through encoder processing by the dose distribution generation model M2, and outputs dose distribution image information GD3 through decoder processing by the dose distribution generation model M2 based on these features and vectors.

[0046] (Each processing unit for generating dose distribution) Furthermore, the image information acquisition unit 11B for dose distribution generation performs the process of acquiring contour image information GD2 (contour image) to be used by the image processing unit 13B, as shown in Figure 3. More specifically, the image information acquisition unit 11B acquires the contour image information GD2 created by the image processing unit 13A and stores it in the data storage unit 23. However, the present invention is not limited thereto, and contour images created separately outside the information processing device 1, rather than by the image processing unit 13A, may be acquired by the communication unit 30.

[0047] As shown in Figure 3, the medical record information acquisition unit 12B performs the process of acquiring electronic medical record information TD to be used by the text processing unit 15B. More specifically, the medical record information acquisition unit 12B acquires electronic medical record information TD from an external medical record database (not shown) via the communication unit 30.

[0048] As shown in Figure 3, the image processing unit 13B inputs contour image information GD2 and electronic medical record information (vectors) converted by the modality processing unit 16B (described later) into the dose distribution generation model M2, and performs processing to acquire and output the output dose distribution image information GD3. More specifically, the image processing unit 13B extracts features of the contour image information GD2 via encoder processing using the dose distribution generation model M2, and then outputs dose distribution image information GD3 consisting of a radiation dose distribution suitable for the patient by considering the patient's situation, such as the patient's wishes and medical history, through decoder processing based on these features.

[0049] As shown in Figure 3, the prompt processing unit 14B uses the dose distribution prompt generation model M5 to analyze the content of the electronic medical record information TD that needs to be reflected in the dose distribution image information GD3, extracts the context from the content, and generates a text prompt TP. The content in the electronic medical record information is usually unstructured natural language text. More specifically, the prompt processing unit 14B generates the context of the content to be reflected in the dose distribution image information GD3 into a text prompt TP using the dose distribution prompt generation model M5. Examples of the context of the content to be reflected include basic patient information such as medical history records, medication records, lifestyle records, and preference records regarding the patient's wishes regarding invasiveness; patient medical information such as examination findings, diagnostic records, discussion records, and treatment plan records by healthcare professionals; and vital information from clinical examinations and examination information collected by medical devices.

[0050] As shown in Figure 3, the text processing unit 15B uses the dose distribution language processing model M6 to divide the text prompt TP into tokens TK, and then identifies and extracts the tokens TK necessary to reflect them in the dose distribution image information GD3 output by the image processing unit 13B. More specifically, the text prompt TP is analyzed by the dose distribution language processing model M6 and divided into multiple tokens TK, and the text processing unit 15B extracts tokens TK from the multiple tokens TK divided by the dose distribution language processing model M6 that are preferable to reflect in the plotting of the dose distribution by the dose distribution generation model M2. Note that the preferred tokens TK in the dose distribution generation model M2 may be different from the tokens TK in the contour drawing model M1, or tokens TK with the same content may be used.

[0051] As shown in Figure 3, the modality processing unit 16B converts the token TK extracted by the text processing unit 15B into a data format that can be reflected in the dose distribution generation model M2, such as a vector, and the image processing unit 13B inputs (reflects) this into the dose distribution generation model M2. As a result, the image processing unit 13B extracts the features of the image information GD2 via encoder processing by the dose distribution generation model M2, and outputs the dose distribution image information GD3 via decoder processing by the dose distribution generation model M2 based on these features and the vector (information based on the electronic medical record). The dose distribution image information GD3 reflects the contents of the electronic medical record information TD.

[0052] As shown in Figure 3, the output unit 17b processes the dose distribution image information GD3 output by the image processing unit 13b to output the dose distribution ideal for medical professionals as radiation therapy planning information to an external device via the communication unit 30. Here, it is also preferable for the image processing unit 13B to input the dose distribution output by the output unit 17B into the dose distribution language processing model M6 to verbalize it, and then feed back (reflect) the verbalized dose distribution information to the dose distribution generation model M2. This makes it possible to bring the output dose distribution closer to ideal radiation therapy planning information.

[0053] The completed dose distribution can be sent via the communication unit 30, etc., to a conventionally known radiotherapy planning system (TPS) that plans and determines the dose distribution to be irradiated in radiotherapy such as intensity-modulated radiation therapy (IMRT) and volume-modulated radiation therapy (VMRT). Furthermore, it is preferable that the output units 17A and 17B output image information GD1, contour image information GD2, dose distribution image information GD3, text prompt TP, and token TK, etc., input to and output from each learning model to the data storage unit 23, and that the data storage unit 23 stores this information as a dataset for fine-tuning each learning model. The process of performing fine-tuning is a well-known technique, so a description will be omitted.

[0054] The following describes, using radiation therapy for head and neck cancer as an example, the processing steps for drawing the contour using the information processing device 1, and the processing steps for creating the dose distribution, based on Figures 4 and 5.

[0055] (Processing steps for outline drawing) First, the image information acquisition unit 11A acquires anatomical image information GD1 showing the location and cross-sectional contour of the affected area (S401), and stores the image information GD1 in the data storage unit 23 (S402).

[0056] Next, the medical record information acquisition unit 12A acquires the patient's electronic medical record information TD, such as the initial consultation record shown in Figure 6 and the progress notes shown in Figures 7 and 8 (S403), and stores the electronic medical record information TD in the data storage unit 23 (S404). At this point, the prompt processing unit 14A uses the contour prompt generation model M3 to extract the context necessary for the subsequent processing that outputs the contours of targets and organs within a range suitable for the patient, and the radiation dose distribution suitable for the patient (S405), and stores each context in the data storage unit 23 (S406).

[0057] Specifically, the prompt processing unit 14 extracts contexts CT11, CT12, CT13, CT15, CT17, CT18, and CT19 from the electronic medical record information as shown in Figures 6 to 8. These contexts are extracted because they contain text content that should be reflected in the contours of targets and organs within a range appropriate for the patient.

[0058] Furthermore, the prompt processing unit 14A generates text prompts TP based on contexts CT11, CT12, CT13, CT15, CT17, CT18, and CT19 using the contour prompt generation model M3. For example, based on context CT11, the prompt processing unit 14A generates a text prompt TP to be input to the contour drawing model M1, such as "Since it is sphenoid sinus cancer (cT4bN0M0), lymph nodes will not be included in the treatment area, and the treatment area will be limited to the extent of tumor progression." Also, based on context CT12, for example, the prompt processing unit 14A determines that "the tumor will shrink because radiotherapy will be performed after chemotherapy (TPF)," and generates a text prompt TP to be input to the contour drawing model M1, such as "reduce the range of gross tumor volume (GTV) and clinical target volume (CTV)."

[0059] Next, the image processing unit 13A inputs image information GD1 to the contour drawing model M1 (S407). At the same time, the text processing unit 15A inputs a text prompt TP based on the electronic medical record information TD to the contour language processing model M4 (S408), and extracts multiple tokens TK identified by the contour language processing model M4 (S409).

[0060] Next, the modality processing unit 16A converts the multiple tokens TK into a data format that can be input to the contour drawing model M1 (S410) and inputs it to the contour drawing model M1 (S411). Next, the image processing unit 13A acquires contour image information GD2 that shows the contours of the target and organs in the range output by the image information GD1 and the multiple tokens input to the contour drawing model M1 (S412). Once the image processing unit 13A acquires the contour image information GD2, it stores it in the data storage unit 23 together with the image information GD1 and the multiple tokens input to the contour drawing model M1 (S413).

[0061] (Processing procedure for generating dose distribution) First, the image information acquisition unit 11B acquires the contour image information GD2 from the data storage unit 23 (S501). The medical record information acquisition unit 12B acquires the patient's electronic medical record information TD, such as the initial consultation record shown in Figure 6 and the progress notes shown in Figures 7 and 8 (S502), and stores the electronic medical record information TD in the data storage unit 23 (S503). At this point, the prompt processing unit 14B uses the dose distribution prompt generation model M5 to extract the context necessary for the subsequent processing that outputs a radiation dose distribution suitable for the patient (S504), and stores each context in the data storage unit 23 (S505).

[0062] Specifically, the prompt processing unit 14B extracts contexts CT12, CT13, CT14, CT16, CT17, CT18, and CT19 from the electronic medical record information as shown in Figures 6 to 8. These contexts are extracted because they contain text content that should be reflected in the radiation dose distribution appropriate for the patient.

[0063] Furthermore, the prompt processing unit 14B generates text prompts TP based on contexts CT12, CT13, CT14, CT16, CT17, CT18, and CT19 using the dose distribution prompt generation model M5. For example, based on context CT12, the prompt processing unit 14B generates a text prompt TP to be input to the dose distribution generation model M2, such as "reduce the irradiation range to the tumor and reduce the prescribed radiation dose." Also, based on context CT14, for example, the prompt processing unit 14B determines "invasion to the left ocular apex and pterygoid base" and "the patient intends to preserve their vision," and generates a text prompt TP to be input to the dose distribution generation model M2, such as "reduce the radiation dose to the optic nerve and eyeball."

[0064] Next, the image processing unit 13B inputs contour image information GD2 into the dose distribution generation model M2 (S506). At the same time, the text processing unit 15B inputs a text prompt TP based on the electronic medical record information TD into the dose distribution language processing model M6 (S507) and extracts multiple tokens TK identified by the language processing model (S508). Next, the modality processing unit 16B converts the multiple tokens TK into a data format that can be input into the dose distribution generation model M2 (S509), and the image processing unit 13B inputs it into the dose distribution generation model M2 (S510).

[0065] Next, the image processing unit 13 acquires (S511) dose distribution image information GD3, which shows the dose distribution of radiation output based on the contour image information GD2 and multiple tokens input to the dose distribution generation model M2, and outputs it. Once the image processing unit 13 acquires the dose distribution image information GD3, it stores it in the data storage unit 23 along with the multiple tokens input to the dose distribution generation model M2 (S512).

[0066] The information processing device 1 that implements the information processing method according to this embodiment performs two information processing operations—drawing the outlines of tumors and organs and generating radiation dose distributions—within a single computer using a learning model, which facilitates information consistency and makes the process more efficient.

[0067] Although embodiments of the present invention have been described above, the present invention is not limited in any way to these embodiments, and can be implemented in various forms without departing from the spirit of the invention. For example, in this embodiment, the information processing device 1 was described as performing two processes according to the present invention: contour drawing processing and dose distribution generation processing, but it is also possible to perform only one of these processes. Furthermore, although the prompt generation model and language processing model were described as separate for contour drawing and dose distribution generation, the present invention is not limited to this, and it is preferable to use a common model, especially for the language processing model. In addition, although the prompt generation model and language processing model were used to reflect electronic medical record information in the contour drawing model and dose distribution generation model, the present invention is not limited to this, and methods for reflecting electronic medical record information in the contour drawing model and dose distribution generation model by other known methods are also included. [Explanation of Symbols]

[0068] 1. Information Processing Device 10 Processing Unit 11A, 11B Image information acquisition section 12A, 12B Medical Record Information Acquisition Department 13A, 13B Image Processing Unit 14A, 14B Prompt Processing Unit 15A, 15B Text Processing Unit 16A, 16B Modality Processing Unit 17A, 17B output section 20 Memory section 21 Program Storage Unit 22 Learning Model Memory Unit 23 Data Storage Unit 30 Communications Department CT11~CT19 Context GD1 Image Information GD2 contour image information GD3 Dose Distribution Image Information M1 Contour Drawing Model M2 dose distribution generation model M3 contour prompt generation model M4 Contour Language Processing Model M5 Dose Distribution Prompt Generation Model M6 Language processing model for dose distribution N Network TD Electronic Medical Record Information TK Token TP Text Prompt

Claims

1. Information processing method, Information processing device, We obtain information about the affected area based on anatomical images. We obtain electronic medical record information, which is information based on the patient's electronic medical record. Based on the information about the affected area and the electronic medical record information, radiation therapy plan information is output. Information processing methods.

2. Information processing method, Information processing device, It stores information about the location of the affected area and the contour of the cross-section. We obtain electronic medical record information, which is information based on the patient's electronic medical record. A machine learning model is used to output a radiation dose distribution suitable for a patient, based on inputs of information on the location and cross-sectional contour of the affected area, and electronic medical record information based on the patient's electronic medical record. The acquired electronic medical record information and the information on the location and cross-sectional contour of the affected area are input to the machine learning model. The dose distribution output by the aforementioned learning model is obtained, Based on the given dose distribution, an ideal dose distribution is output. Information processing methods.

3. Information processing method, Information processing device, The location and cross-sectional image information of the affected area are stored, We obtain patient electronic medical record information, which is information based on the patient's electronic medical record. The acquired electronic medical information and the image information of the affected area and its cross-section are input to a machine learning model that outputs target and organ contours within a range appropriate for the patient, based on the input of image information of the location and cross-section of the affected area and electronic medical record information based on the patient's electronic medical record. The target and organ contours within the range output by the aforementioned learning model are obtained. Output the contour, Information processing methods.

4. The input of the electronic medical record information to the learning model is done via a large-scale language model (LLM). The information processing method according to claim 2 or 3.

5. A text prompt is generated from the information in the patient's electronic medical record that should be reflected in the radiation therapy plan. Using the aforementioned large-scale language model, the text prompt is divided into tokens, and each token is extracted to be reflected in the radiotherapy planning information. The information processing method according to claim 4.

6. The aforementioned learning model, The model was trained using electronic medical record information, which includes at least one of the patient's basic information and medical information, as training data. The information processing method according to claim 2 or 3.

7. The aforementioned learning model uses the patient's basic information as follows: This model is trained on electronic medical record information that includes at least one of the following records: a medical history record about the patient's medical history, a medication record about the patient's current medications, a lifestyle record about the patient's lifestyle habits, and a preference record about the patient's wishes regarding invasive procedures. The information processing method according to claim 6.

8. The aforementioned learning model uses the patient's medical information as follows: This model is trained based on input from electronic medical record information, which includes at least one of the following records from healthcare professionals: examination findings, diagnostic records, discussion records, and treatment plan records. The information processing method according to claim 6.

9. The aforementioned electronic medical record information, Including initial consultation records and progress notes, The information processing method according to claim 1.

10. The dose distribution output by the learning model is input into a large-scale language model (LLM) to be verbalized, and this verbalized dose distribution information is fed back to the learning model. The information processing method according to claim 2.

11. It is a computer program, On the computer, We obtain information about the affected area based on anatomical images. We obtain electronic medical record information, which is information based on the patient's electronic medical record. Based on the information about the affected area and the electronic medical record information, radiation therapy plan information is output. A computer program that performs information processing.

12. It is a computer program, On the computer, It stores information about the location of the affected area and the contour of the cross-section. We obtain electronic medical record information, which is information based on the patient's electronic medical record. A machine learning model is used to output a radiation dose distribution suitable for a patient, based on inputs of information on the location and cross-sectional contour of the affected area, and electronic medical record information based on the patient's electronic medical record. The acquired electronic medical record information and the information on the location and cross-sectional contour of the affected area are input to the machine learning model. The dose distribution output by the aforementioned learning model is obtained, Based on the given dose distribution, an ideal dose distribution is output. A computer program that executes a process.

13. It is a computer program, On the computer, The location and cross-sectional image information of the affected area are stored, We obtain electronic medical record information, which is information based on the patient's electronic medical record. The acquired electronic medical information and the image information of the affected area and its cross-section are input to a machine learning model that outputs target and organ contours within a range appropriate for the patient, based on the input of image information of the location and cross-section of the affected area and electronic medical record information based on the patient's electronic medical record. The target and organ contours within the range output by the aforementioned learning model are obtained. Output the contour, A computer program that executes a process.

14. An information processing device, A unit for acquiring information about the affected area based on anatomical images, A medical record information acquisition unit acquires electronic medical record information, which is information based on the patient's electronic medical record, An information processing device comprising an output unit that outputs radiation therapy plan information based on the information of the affected area and the electronic medical record information.

15. An information processing device, A memory unit that stores information about the location and cross-sectional contour of the affected area, A medical record information acquisition unit acquires electronic medical record information, which is information based on the patient's electronic medical record, A dose distribution acquisition unit acquires a dose distribution output by a learning model that takes information on the location and cross-sectional contour of the affected area, as well as electronic medical record information based on the patient's electronic medical record, as input to the learning model, and takes the acquired electronic medical record information and the information on the location and cross-sectional contour of the affected area as input to the learning model. An output unit that outputs an ideal dose distribution based on the dose distribution, An information processing device equipped with the following features.

16. An information processing device, A memory unit that stores the location and cross-sectional image information of the affected area, A medical record information acquisition unit acquires electronic medical record information, which is information based on the patient's electronic medical record, A learning model that is trained to output target and organ contours within a range appropriate for the patient, in response to input of image information of the location and cross-section of the affected area and electronic medical record information based on the patient's electronic medical record, comprises a contour acquisition unit that inputs the acquired electronic medical record information and the image information of the location and cross-section of the affected area into the learning model and acquires target and organ contours within the range output by the learning model, An output unit that outputs the contour, An information processing device equipped with the following features.

Images