CN117456315A

CN117456315A - Training method and device for dose prediction model and computer readable storage medium

Info

Publication number: CN117456315A
Application number: CN202311803165.8A
Authority: CN
Inventors: 周琦超; 李梓荣
Original assignee: Manteia Data Technology Co ltd In Xiamen Area Of Fujian Pilot Free Trade Zone
Current assignee: Manteia Data Technology Co ltd In Xiamen Area Of Fujian Pilot Free Trade Zone
Priority date: 2023-12-26
Filing date: 2023-12-26
Publication date: 2024-01-26
Anticipated expiration: 2043-12-26
Also published as: CN117456315B

Abstract

The application discloses a training method and device of a dose prediction model and a computer readable storage medium. Wherein the method comprises the following steps: randomly generating K target dose areas; combining the K target dose areas with Z medical images in the image library to obtainThe images to be processed; according toF target radiotherapy plans are generated by the images to be processed, first dose information corresponding to each target radiotherapy plan is calculated through a first dose algorithm, second dose information corresponding to each target radiotherapy plan is calculated through a second dose algorithm, and the first dose is calculatedThe calculation accuracy of the algorithm is higher than that of the second dose algorithm; and training according to the first dose information and the second dose information corresponding to each target radiotherapy plan to obtain a dose prediction model. The method and the device solve the technical problems that in the prior art, due to the fact that a large amount of training data are difficult to obtain, the training difficulty of a dose prediction model is high, and the training effect is poor.

Description

Training method and device for dose prediction model and computer readable storage medium

Technical Field

The present application relates to the field of medical science and technology, and in particular, to a method and apparatus for training a dose prediction model, and a computer readable storage medium.

Background

In the prior art, prediction of radiotherapy dosage and optimization of radiotherapy dosage can be performed by training a neural network model, however, because the training of a dosage prediction model requires treatment data of patients, and the patient data are generally difficult to obtain due to the patient privacy, in the training process of the existing dosage prediction model, the technical problems of high training difficulty and poor training effect of the dosage prediction model caused by the difficulty in acquiring a large amount of training data are generally faced.

In view of the above problems, no effective solution has been proposed at present.

Disclosure of Invention

The application provides a training method, a device and a computer readable storage medium of a dose prediction model, which at least solve the technical problems of high training difficulty and poor training effect of the dose prediction model caused by difficulty in acquiring a large amount of training data in the prior art.

According to one aspect of the present application, there is provided a training method of a dose prediction model, comprising: randomly generating K target dose areas, wherein K is an integer greater than 1, and the target dose areas are areas which are irradiated by rays; combining the K target dose areas with Z medical images in the image library to obtain The medical images in the image library at least comprise medical images generated according to N preset materials and medical images generated according to various organ tissues, wherein Z and N are integers larger than 1; according to->F target radiotherapy plans are generated by the images to be processed, first dose information corresponding to each target radiotherapy plan is calculated through a first dose algorithm, and second dose information corresponding to each target radiotherapy plan is calculated through a second dose algorithmWherein F is an integer greater than the product of K and Z, the calculation accuracy of the first dosage algorithm is higher than that of the second dosage algorithm, and the calculation speed of the first dosage algorithm is slower than that of the second dosage algorithm; and training according to the first dose information and the second dose information corresponding to each target radiotherapy plan to obtain a dose prediction model.

Optionally, the training method of the dose prediction model further comprises: combining the K target dose areas with Z medical images in the image library to obtainBefore the images to be processed, generating one medical image corresponding to each preset material in N preset materials to obtain N medical images; m times of target operation are executed based on N preset materials to obtain M medical images, wherein M is an integer larger than 1, each target operation in the M times of target operation is used for randomly selecting at least two preset materials from the N preset materials as a material combination, and medical images corresponding to the material combination are generated; acquiring Y medical images based on each of X human body regions of a human body to obtain +. >The method comprises the steps of performing medical imaging, wherein X and Y are integers larger than 1, and organ tissues corresponding to different human body areas in X human body areas are different; n medical images, M medical images and +.>Storing the medical images into an image library, wherein +_>。

Optionally, the training method of the dose prediction model further comprises: generating a virtual sphere in a three-dimensional form with random size; constructing a unit deformation field around the virtual sphere, and randomly setting force application angles in a plurality of preset directions respectively; and respectively carrying out K times of force application operations on the virtual sphere in the unit deformation field according to the force application angle to obtain K target dose areas, wherein the virtual sphere after each time of force application operation in the K times of force application operations serves as one target dose area in the K target dose areas, the force application size and the force application direction of each time of force application operation are randomly arranged, and the force application direction represents outward force application or inward force application on the virtual sphere according to the force application angle.

Optionally, the training method of the dose prediction model further comprises: in accordance withBefore F target radiotherapy plans are generated by the images to be processed, randomly setting L isocenter points, wherein L is an integer larger than 1, and each isocenter point in the L isocenter points represents the isocenter point corresponding to the rotation of the accelerator in the process of implementing the radiotherapy plans; randomly setting H pieces of field information, wherein H is an integer greater than 1, the number of fields included in each piece of field information is a random value, and the angle of each field included in each piece of field information is a random value; g prescription dose information is randomly set, wherein G is an integer greater than 1.

Optionally, the training method of the dose prediction model further comprises: associating each of the L isocenter points withCombining each of the images to be processed to obtain +.>The personal parameter data; associating each of the H pieces of portal information with +.>Each of the parameter data is combined to obtain +.>Target parameter data; according to each of the G prescribed dose information and +.>Each of the target parameter dataGenerating an initial radiotherapy plan by the target parameter data to obtain +.>An initial radiotherapy plan, wherein->The method comprises the steps of carrying out a first treatment on the surface of the And carrying out plan optimization operation on each initial radiotherapy plan in the F initial radiotherapy plans, taking each initial radiotherapy plan completed with the plan optimization operation as a target radiotherapy plan, and obtaining F target radiotherapy plans, wherein the plan optimization operation is used for realizing that the volume of a preset proportion in a target dosage area meets prescription dosage information by adjusting the radiation field information.

Optionally, the training method of the dose prediction model further comprises: splitting the target radiotherapy plan into a plurality of sub-plans according to the radiation field information corresponding to each target radiotherapy plan, wherein each sub-plan in the plurality of sub-plans corresponds to one radiation field or sub-field corresponding to the target radiotherapy plan; calculating first sub-dose information corresponding to each sub-plan through a first dose algorithm; taking all first sub-dose information corresponding to the multiple sub-plans as first dose information corresponding to the target radiotherapy plan; calculating second sub-dose information corresponding to each sub-plan through a second dose algorithm; and taking all second sub-dose information corresponding to the plurality of sub-plans as second dose information corresponding to the target radiotherapy plan.

Optionally, the training method of the dose prediction model further comprises: representing the first dose information by a first dose profile and the second dose information by a second dose profile; resampling the first dose distribution map and the second dose distribution map respectively according to the preset training lattice point number, wherein the resampling is used for extracting J subgraphs corresponding to the first dose distribution map and J subgraphs corresponding to the second dose distribution map, the sizes of the subgraphs are all preset sizes, and J is the training lattice point number; taking J sub-graphs corresponding to the first dose distribution map as training samples and J sub-graphs corresponding to the second dose distribution map as training labels; and carrying out iterative training on the neural network according to the training sample and the training label, and taking the neural network after the iterative training is finished as a dose prediction model.

Optionally, the training method of the dose prediction model further comprises: after training according to the first dose information and the second dose information corresponding to each target radiotherapy plan to obtain a dose prediction model, acquiring a medical image corresponding to a target object; dividing a medical image corresponding to a target object into J first sub-images with the same size, and calculating a target ratio between the size of the first sub-images and a preset size; according to the target ratio, adjusting the medical image corresponding to the target object into a target medical image, wherein the target medical image can be divided into J second sub-images, and the sizes of the second sub-images are preset sizes; and generating a radiotherapy plan of the target object according to the target medical image.

Optionally, the training method of the dose prediction model further comprises: after a radiotherapy plan of a target object is generated according to a target medical image, splitting the radiotherapy plan of the target object into a plurality of sub-plans to be processed, wherein each sub-plan to be processed in the plurality of sub-plans to be processed corresponds to one radiation field or sub-field corresponding to the radiotherapy plan of the target object; calculating the corresponding dose information of each sub-plan to be processed through a second dose algorithm, and converting the corresponding dose information of each sub-plan to be processed into an initial dose distribution map corresponding to the sub-plan to be processed; inputting an initial dose distribution map corresponding to each sub-plan to be processed into a dose prediction model, predicting and generating an intermediate dose distribution map corresponding to the sub-plan to be processed through the dose prediction model according to priori knowledge learned by first dose information and second dose information, wherein the intermediate dose distribution map corresponding to the sub-plan to be processed represents a dose distribution map correspondingly obtained when the sub-plan to be processed is calculated by using a first dose algorithm.

Optionally, the training method of the dose prediction model further comprises: after an initial dose distribution map corresponding to each sub-plan to be processed is input into a dose prediction model, a prior knowledge obtained by learning the first dose information and the second dose information through the dose prediction model is used for predicting and generating an intermediate dose distribution map corresponding to the sub-plan to be processed, and then the intermediate dose distribution map corresponding to each sub-plan to be processed is adjusted according to a target ratio, so that a target dose distribution map corresponding to the sub-plan to be processed is obtained, wherein the image size ratio between the target dose distribution map and the intermediate dose distribution map is the target ratio; and superposing the target dose distribution map corresponding to each sub-plan to be processed to obtain the integral dose distribution map corresponding to the radiotherapy plan of the target object.

According to another aspect of the present application, there is also provided a training device of a dose prediction model, wherein the training device of the dose prediction model includes: the generation unit is used for randomly generating K target dose areas, wherein K is an integer greater than 1, and the target dose areas are areas which are irradiated by rays; a combination unit for combining the K target dose regions with the Z medical images in the image library to obtainThe medical images in the image library at least comprise medical images generated according to N preset materials and medical images generated according to various organ tissues, wherein Z and N are integers larger than 1; a computing unit for according to->F target radiotherapy plans are generated by the images to be processed, first dose information corresponding to each target radiotherapy plan is calculated through a first dose algorithm, second dose information corresponding to each target radiotherapy plan is calculated through a second dose algorithm, wherein F is an integer larger than the product of K and Z, the calculation accuracy of the first dose algorithm is higher than that of the second dose algorithm, and the calculation speed of the first dose algorithm is slower than that of the second dose algorithm; and the training unit is used for training to obtain a dose prediction model according to the first dose information and the second dose information corresponding to each target radiotherapy plan.

According to another aspect of the present application, there is also provided a computer readable storage medium, wherein the computer readable storage medium has a computer program stored therein, and wherein the computer program when run controls a device in which the computer readable storage medium is located to perform the training method of the dose prediction model of any one of the above.

In the application, a mode of randomly generating F target radiotherapy plans for training a dose prediction model is adopted, K target dose areas are randomly generated first, wherein K is an integer greater than 1, and the target dose areas are areas which receive radiation irradiation. Then, combining the K target dose areas with the Z medical images in the image library to obtainThe medical images in the image library at least comprise medical images generated according to N preset materials and medical images generated according to various organ tissues, and Z and N are integers larger than 1. Subsequently, according to->F target radiotherapy plans are generated by the images to be processed, first dose information corresponding to each target radiotherapy plan is calculated through a first dose algorithm, second dose information corresponding to each target radiotherapy plan is calculated through a second dose algorithm, F is an integer larger than the product of K and Z, the calculation accuracy of the first dose algorithm is higher than that of the second dose algorithm, and the calculation speed of the first dose algorithm is slower than that of the second dose algorithm. And finally, training according to the first dose information and the second dose information corresponding to each target radiotherapy plan to obtain a dose prediction model.

As can be seen from the above, the present application combines the medical images generated according to N preset materials and the medical images generated according to multiple organ tissues in the image library to obtain a large number of medical image samples (i.e.The images to be processed), and then generates a larger number of radiotherapy plan samples (F target radiotherapy plans) based on a large number of medical image samples, thereby solving the problem that a large amount of training data is difficult to obtain in the prior art, and further reducing the difficulty in training of a dose prediction modelAnd the training effect of the dose prediction model is improved.

In addition, the method calculates first dose information corresponding to each target radiotherapy plan through a first dose algorithm, calculates second dose information corresponding to each target radiotherapy plan through a second dose algorithm, and trains according to the first dose information and the second dose information corresponding to each target radiotherapy plan to obtain a dose prediction model. Because the calculation accuracy of the first dose algorithm is higher than that of the second dose algorithm, the calculation speed of the first dose algorithm is slower than that of the second dose algorithm, so that a model which can convert dose information with poor accuracy into dose information with high accuracy is obtained through training, and further, in the use process of the dose prediction model, only the second dose algorithm is needed to be used for quickly calculating to obtain dose information with poor accuracy, the accuracy of the dose information can be improved by combining the dose prediction model, and the overall determination efficiency of the dose information is improved under the condition that the accuracy of the dose information is ensured.

Therefore, through the technical scheme of the application, the purpose of automatically generating a large number of training samples of the dose prediction model is achieved, the technical effect of reducing the acquisition difficulty of the model training samples is achieved, and the technical problems that the training difficulty of the dose prediction model is high and the training effect is poor due to the fact that a large number of training data are difficult to acquire in the prior art are solved.

Drawings

The accompanying drawings, which are included to provide a further understanding of the application and are incorporated in and constitute a part of this application, illustrate embodiments of the application and together with the description serve to explain the application and do not constitute an undue limitation to the application. In the drawings:

FIG. 1 is a flow chart of an alternative method of training a dose prediction model according to an embodiment of the present application;

FIG. 2 is a flowchart of an alternative construction of an image library according to an embodiment of the present application;

FIG. 3 is a flowchart of an alternative randomly generated K target dose regions according to an embodiment of the present application;

FIG. 4 is a flow chart of an alternative dose information calculation process according to an embodiment of the present application;

FIG. 5 is a training flow diagram of an alternative dose prediction model according to an embodiment of the present application;

FIG. 6 is a schematic diagram of an alternative dose prediction model training device according to an embodiment of the present application.

Detailed Description

In order to make the present application solution better understood by those skilled in the art, the following description will be made in detail and with reference to the accompanying drawings in the embodiments of the present application, it is apparent that the described embodiments are only some embodiments of the present application, not all embodiments. All other embodiments, which can be made by one of ordinary skill in the art based on the embodiments herein without making any inventive effort, shall fall within the scope of the present application.

It should be noted that the terms "first," "second," and the like in the description and claims of the present application and the above figures are used for distinguishing between similar objects and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used may be interchanged where appropriate such that embodiments of the present application described herein may be implemented in sequences other than those illustrated or otherwise described herein. Furthermore, the terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.

It should be further noted that, related information (including but not limited to user equipment information, user personal information, etc.) and data (including but not limited to data for presentation, analyzed data, etc.) related to the present application are information and data authorized by the user or sufficiently authorized by each party. For example, an interface is provided between the system and the relevant user or institution, before acquiring the relevant information, the system needs to send an acquisition request to the user or institution through the interface, and acquire the relevant information after receiving the consent information fed back by the user or institution.

Example 1

In accordance with embodiments of the present application, there is provided an embodiment of a method of training a dose prediction model, it being noted that the steps illustrated in the flowchart of the figures may be performed in a computer system, such as a set of computer executable instructions, and, although a logical order is illustrated in the flowchart, in some cases, the steps illustrated or described may be performed in an order other than that illustrated herein.

FIG. 1 is a flowchart of an alternative method of training a dose prediction model according to an embodiment of the present application, as shown in FIG. 1, comprising the steps of:

In step S101, K target dose regions are randomly generated.

In step S101, K is an integer greater than 1, and the target dose region is a region to be irradiated with radiation.

In an alternative embodiment, a dose information determining system may be used as an execution body of the training method of the dose prediction model in the embodiments of the present application, where the dose information determining system may be a software system or an embedded system combining software and hardware. In addition, the dose information determining system not only can train to obtain at least one dose prediction model through the training method of the dose prediction model in the embodiment of the application, but also can call the trained dose prediction model to carry out operations such as formulation, optimization and the like of dose information on a target object, wherein the target object is an object which needs to receive radiotherapy.

Alternatively, the dose information determining system may randomly generate K target dose regions in a virtual three-dimensional space, where each target dose region is a region that is supposed to be irradiated with radiation, and is analogous to a target region of radiotherapy that is supposed to be irradiated with radiation during actual radiotherapy and a jeopardized organ that is to be irradiated with radiation.

Step S102, combining the K target dose areas with Z medical images in the image library to obtainAnd (5) processing the images to be processed.

In step S102, the medical images in the image library at least include medical images generated according to N preset materials and medical images generated according to a plurality of organ tissues, where Z and N are integers greater than 1.

Alternatively, the Z medical images in the image library may be various types of medical images, such as CT images, CBCT images, MR images, and the like. In addition, the medical images in the image library may also be randomly generated images, including but not limited to medical images generated based on each of the N preset materials, medical images generated based on a combination of at least two of the N preset materials.

Optionally, the N preset materials include, but are not limited to, water, air, bone, liver, white matter, gray matter, blood, muscle, kidney, cerebrospinal fluid, fat, and various types of metal materials, and after acquiring the N preset materials, the dose information determining system establishes a material density table for recording a density value of each preset material, and establishes a HU table for recording a CT value corresponding to each preset material, wherein the HU value is a unit of the CT value.

Alternatively, the dose information determination system may combine the K target dose regions with the Z medical images in the image library, resulting in at mostAnd (5) processing the images to be processed.

Step S103, according toF target radiotherapy plans are generated by the images to be processed, first dose information corresponding to each target radiotherapy plan is calculated through a first dose algorithm, and second dose information corresponding to each target radiotherapy plan is calculated through a second dose algorithm.

In step S103, F is an integer greater than the product of K and Z, and the calculation accuracy of the first dose algorithm is higher than that of the second dose algorithm, and the calculation speed of the first dose algorithm is slower than that of the second dose algorithm.

Alternatively, the process may be carried out in a single-stage,and the first and second dosing algorithms are two different algorithms, for example, the first dosing algorithm may be a MC (Monte Carlo Meteod monte carlo) dosing algorithm and the second dosing algorithm may be a PB (Pencil-Beam) dosing algorithm.

Step S104, training according to the first dose information and the second dose information corresponding to each target radiotherapy plan to obtain a dose prediction model.

Optionally, in step S104, the dose information determining system may use the first dose information corresponding to each target radiotherapy plan as a model training tag, use the second dose information corresponding to each target radiotherapy plan as model training data, and then input the model training tag and the model training data into the deep neural network for iterative training, and finally train to obtain the dose prediction model.

Among the existing radiotherapy, intensity-modulated radiotherapy (IMRT) and volume-modulated radiotherapy (VMAT) are the most commonly used techniques. Wherein, the radiotherapy plans of IMRT and VMAT are generally obtained by inverse optimization in a radiotherapy planning system. While the speed and quality of the optimization is generally dependent on the dose calculation algorithm.

Among them, the MC algorithm is a very accurate dose calculation algorithm in the industry. However, the MC algorithm requires a simulation of the real transfer process of a large number of particles, and thus requires a lot of work to be completed quickly, resulting in a problem of slow calculation speed. In addition, dose calculations based on analytical algorithms (e.g., PB dose algorithms) are typically faster to calculate in addition to MC algorithms, but these dose algorithms are less accurate than MC algorithms.

At present, although some schemes in the prior art propose to use a deep learning model to accelerate the calculation speed of the MC algorithm, since the training effect of the deep learning model is closely related to the number of training data sets, and the training data is difficult to obtain because of the patient privacy, this makes the training difficulty of the dose prediction model in the prior art higher, even if the model is obtained by training according to a small sample, the generalization ability of the model obtained by training is also poor, and it is difficult to make accurate predictions.

In the present application, a plurality of medical image samples are obtained by randomly generating K target dose regions and combining medical images generated according to N preset materials and medical images generated according to various organ tissues in an image library (i.e.The images to be processed), and further generates a plurality of radiotherapy plan samples (F target radiotherapy plans) based on a plurality of medical image samples, thereby solving the problem that a large amount of training data are difficult to acquire in the prior art, further reducing the training difficulty of the dose prediction model and improving the training effect of the dose prediction model.

In an alternative embodiment, the K target dose regions are combined with the Z medical images in the image library to obtainBefore each image to be processed, the dose information determining system builds an image library by the steps shown in fig. 2.

Step S201, generating a medical image corresponding to each preset material in the N preset materials to obtain N medical images.

Optionally, the dose information determining system generates a medical image corresponding to each preset material according to the preset material, for example, generates a bone CT image for bones in the N preset materials, and generates a liver CT image according to livers in the N preset materials.

Step S202, M target operations are executed based on N preset materials, and M medical images are obtained.

In step S202, M is an integer greater than 1, each of the M target operations is used to randomly select at least two preset materials from the N preset materials as a material combination, and generate a medical image corresponding to the material combination.

For example, in one target operation, the dose information determining system may select a material combination of water and blood in N preset materials, and generate a CT image corresponding to the material combination; in another target operation, the dose information determining system may select a material combination of muscle, blood and fat in N preset materials, and generate a CT image corresponding to the material combination.

Step S203, based on each of the X human body regions of the human bodyCollecting Y medical images to obtainAnd (3) a medical image.

In step S203, X and Y are integers greater than 1, and the organ tissues corresponding to different human body regions in the X human body regions are different.

Optionally, the X human body regions include, but are not limited to, regions of the head, neck, chest, upper abdomen, lower abdomen, etc. of the human body, based on each human body region, the dose information determination system may acquire at least one medical image from multiple angles, respectively, and finally acquire Y medical images based on each human body region, thereby obtaining a human body-based dose information determination systemAnd (3) a medical image.

Step S204, N medical images, M medical images are obtainedStoring the medical images into an image library, wherein +_ >。

Optionally, the dose information determining system stores all the medical images acquired in step S201, step S202 and step S203 in the image library, thereby obtaining Z medical images in the image library.

From the above, the method and the device achieve the aim of automatically generating a large number of medical image samples, and avoid the technical problems of difficult training and poor training effect of the dose prediction model caused by the difficulty in obtaining the medical image samples.

In an alternative embodiment, fig. 3 is a flowchart of an alternative randomly generated K target dose regions according to an embodiment of the present application, as shown in fig. 3, comprising the steps of:

in step S301, a virtual sphere with a random three-dimensional form is generated.

Alternatively, the dose information determining system may generate a virtual sphere of random size in a virtual three-dimensional space in the form of a software program.

Alternatively, in practical applications, in addition to generating virtual spheres, three-dimensional objects of other shapes may be generated, e.g., virtual cylinders, virtual cones, etc.

In step S302, a unit deformation field is constructed around the virtual sphere, and force application angles are randomly set in a plurality of preset directions, respectively.

In step S302, the plurality of preset directions include, but are not limited to, an x-axis direction, a y-axis direction, and a z-axis direction of the three-dimensional space, and the force application angle randomly set in each preset direction is any one angle of 0 degrees to 360 degrees.

Step S303, performing force application operation on the virtual sphere for K times in the unit deformation field according to the force application angle, so as to obtain K target dose areas.

In step S303, the virtual sphere after each of the K application operations is set as one of the K target dose regions, and the application size and the application direction of each application operation are randomly set, and the application direction characterizes the application of force to the virtual sphere according to the application angle.

Alternatively, the magnitude of the applied force corresponding to each applied force angle is set randomly in each applied force operation, so after each applied force operation, the virtual sphere can be changed into an irregular three-dimensional object due to the fact that the virtual sphere can receive forces with uneven magnitudes from different angles, and the purpose of the operation is to fully simulate the irregularity of a radiotherapy target region or a jeopardized organ.

The larger the deformation force applied to the virtual sphere, the higher the degree of deformation of the virtual sphere.

Optionally, the dose information determining system further randomly sets a force application direction in each force application operation, and applies force to the virtual sphere outwards, so that the force application is a thrust force, and the volume of the virtual sphere is enlarged; the virtual sphere is forced inwards, which means that the virtual sphere is forced to be pressure, so that the volume of the virtual sphere is reduced.

Optionally, each force application operation is performed on the virtual sphere that is not changed, and the virtual sphere after each force application operation is used as one target dose region of the K target dose regions, and K target dose regions can be obtained after the virtual sphere is repeatedly subjected to the force application operation for K times.

According to the method, the purpose of automatically generating a plurality of dosage areas is achieved through the steps, and therefore the problems that a dosage prediction model is high in training difficulty and poor in training effect due to the fact that dosage area data of a patient are difficult to acquire are solved.

In an alternative embodiment, the method is based onBefore the F target radiotherapy plans are generated by the images to be processed, the dose information determining system needs to set the following data randomly, including:

l isocenter points are randomly set, wherein L is an integer greater than 1, and each isocenter point in the L isocenter points represents the corresponding isocenter point when the accelerator rotates in the process of implementing the radiotherapy plan.

H pieces of field information are randomly set, wherein H is an integer greater than 1, the number of fields included in each piece of field information is a random value, and the angle of each field included in each piece of field information is a random value.

G prescription dose information is randomly set, wherein G is an integer greater than 1.

Alternatively, to generate F target radiotherapy plans, the dose information determination system first associates each of the L isocenters withCombining each of the images to be processed to obtain +.>And data of the parameters. Then, the dose information determination system associates each of the H pieces of portal information with +.>Each of the parameter data is combined to obtain +.>Target parameter data. Subsequently, the dose information determining system determines the dose information and +.>Each of the target parameter data generates an initial radiotherapy plan to obtain +.>An initial radiotherapy plan, wherein->. And finally, performing plan optimization operation on each initial radiotherapy plan in the F initial radiotherapy plans by the dose information determining system, and taking each initial radiotherapy plan which completes the plan optimization operation as a target radiotherapy plan to obtain F target radiotherapy plans, wherein the plan optimization operation is used for realizing that the preset proportion of the volume in the target dose area meets the prescription dose information by adjusting the radiation field information.

Optionally, after the L isocenters are randomly set, the dose information determination system associates each of the L isocenters withEach of the images to be processed is combined to obtain +.>Each parameter data corresponds to an image to be processed and an isocenter.

Optionally, after obtainingAfter the data of the parameters, the dose information determining system sets up H radiation randomlyEach of the field information is associated with +.>Each of the parameter data is combined to obtain +.>Each target parameter data corresponds to an image to be processed, an isocenter and a portal message.

Optionally, the dose information determining system further determines a dose information of each of the G prescribed dose information based on the prescribed dose information and the G prescribed dose informationEach of the target parameter data generates an initial radiotherapy plan to obtain +.>An initial radiotherapy plan, wherein->。

The prescription dose information is dose information which should be satisfied as much as possible in the radiotherapy process, the prescription dose information is aimed as much as possible, and the dose information determining system determines the prescription dose information according to each of the G prescription dose information and the G prescription dose information Each target parameter data of the target parameter data generates an initial radiotherapy plan, and finally obtainsAn initial radiotherapy plan due to ∈ ->Thus, F initial radiotherapy plans are finally obtained.

Optionally, the dose information determining system may further perform a plan optimization operation on each of the F initial radiotherapy plans, and take each initial radiotherapy plan for which the plan optimization operation is completed as a target radiotherapy plan, to obtain the F target radiotherapy plans, where the plan optimization operation is used to implement that a preset proportion of the volume in the target dose region satisfies the prescription dose information by adjusting the portal information.

Alternatively, the planned optimization operation may adjust the number of sub-fields and/or the sub-field angle included under the field information when adjusting the field information, and the optimization target of the planned optimization operation may be set as follows:

95% of the volume of the target dose region (corresponding to the preset proportion of the volume in the target dose region) satisfies the prescription dose information, and the optimization weights are randomly set to 200-500;

dose information outside the target dose region is limited to 0, and the optimization weights are set to 10-100.

In an alternative embodiment, fig. 4 is a flowchart of an alternative dose information calculation process according to an embodiment of the present application, as shown in fig. 4, comprising the steps of:

Step S401, splitting each target radiotherapy plan into a plurality of sub-plans according to the corresponding radiation field information of the target radiotherapy plan.

In step S401, each of the plurality of sub-plans corresponds to a field or sub-field corresponding to the target radiotherapy plan.

Alternatively, for each target radiotherapy plan, the dose information determination system may split each target radiotherapy plan into a single-field plan (i.e. into sub-plans corresponding to each field), or further into sub-field plans (i.e. into sub-plans corresponding to each sub-field).

In step S402, first sub-dose information corresponding to each sub-plan is calculated by a first dose algorithm.

In step S403, all the first sub-dose information corresponding to the plurality of sub-plans is used as the first dose information corresponding to the target radiotherapy plan.

Alternatively, for each sub-plan obtained by splitting, the dose information determining system may calculate first sub-dose information corresponding to each sub-plan by using a first dose algorithm (for example, MC algorithm), and take all the first sub-dose information of the plurality of sub-plans corresponding to each target radiotherapy plan as the first dose information corresponding to the target radiotherapy plan.

In step S404, second sub-dose information corresponding to each sub-plan is calculated by a second dose algorithm.

In step S405, all the second sub-dose information corresponding to the plurality of sub-plans is used as the second dose information corresponding to the target radiotherapy plan.

Alternatively, for each sub-plan obtained by splitting, the dose information determining system may calculate second sub-dose information corresponding to each sub-plan through a second dose algorithm (such as an analytical algorithm such as PB algorithm), and take all the second sub-dose information of the plurality of sub-plans corresponding to each target radiotherapy plan as the second dose information corresponding to the target radiotherapy plan.

In an alternative embodiment, fig. 5 is a training flowchart of an alternative dose prediction model according to an embodiment of the present application, as shown in fig. 5, comprising the steps of:

in step S501, the first dose information is represented by a first dose distribution map and the second dose information is represented by a second dose distribution map.

Alternatively, both the first and second Dose profiles may be in the form of DVH (Dose and Volume).

Step S502, resampling the first dose distribution map and the second dose distribution map respectively according to the preset training lattice point number.

In step S502, resampling is used to extract J subgraphs corresponding to the first dose distribution map and J subgraphs corresponding to the second dose distribution map, where the subgraphs are all of a preset size, and J is the number of training lattice points.

Alternatively, the number of training lattice points may be custom set, e.g., set to. In addition, the dosage information determining system can determine the number of training lattice points according to the preset number of training lattice pointsThe first dose distribution map and the second dose distribution map are resampled respectively, and spacing (corresponding to the preset size) used in resampling can be set to 0.1-4, and the step size can be set to 0.1-1.

It should be noted that, in the training process of the neural network, a training image with a fixed size is usually required, but the training image with a fixed size cannot meet the requirement of training to obtain a general model with strong generalization capability (because in an actual model usage scenario, a user cannot always fix the grid size of the dose image). In order to achieve training to obtain a general model with strong generalization capability, the training data is further subjected to difference processing through the content of the step S502, namely, the preset size of the subgraph is set, so that the size of the training image can be adjusted in a self-defined mode, the purpose of obtaining training images with more various sizes is achieved, the model with more general and stronger generalization capability is obtained through training, and the requirement of fixing the size of the input image is not limited in the use process of the model.

In step S503, the J sub-graphs corresponding to the first dose distribution chart are used as training samples, and the J sub-graphs corresponding to the second dose distribution chart are used as training labels.

And step S504, performing iterative training on the neural network according to the training sample and the training label, and taking the neural network after the iterative training as a dose prediction model.

Alternatively, the neural network includes, but is not limited to, a CNN network, a transducer attention network, a UNET network, and an LSTM network.

In an alternative embodiment, the dose information determining system may further represent the first sub-dose information through a first sub-dose distribution map, and represent the second sub-dose information through a second sub-dose distribution map, and then resample the first sub-dose distribution map and the second sub-dose distribution map according to a preset training lattice number, where the resampling is used to extract J sub-graphs corresponding to the first sub-dose distribution map and J sub-graphs corresponding to the second sub-dose distribution map, where the sub-graphs are all of a preset size, and J is the training lattice number. Then, the dose information determining system may take the J sub-graphs corresponding to the first sub-dose distribution chart as training samples and the J sub-graphs corresponding to the second sub-dose distribution chart as training labels, and finally perform iterative training on the neural network according to the training samples and the training labels, and take the neural network after the iterative training as a dose prediction model.

In an alternative embodiment, after training to obtain a dose prediction model according to the first dose information and the second dose information corresponding to each target radiotherapy plan, the dose information determines that a medical image corresponding to a target object can be obtained, divides the medical image corresponding to the target object into J first sub-images with the same size, and calculates a target ratio between the size of the first sub-images and a preset size. Then, the dose information determining system adjusts the medical image corresponding to the target object into a target medical image according to the target ratio, wherein the target medical image can be divided into J second sub-images, and the sizes of the second sub-images are preset sizes. Finally, the dose information determination system generates a radiation therapy plan for the target object from the target medical image.

Optionally, the target object is an object to be subjected to radiotherapy, in order to determine dose information corresponding to the target object according to a dose prediction model obtained by training in the present application, a medical image corresponding to the target object may be divided into J first sub-images with the same size, where the size of each first sub-image may be the same as or different from a preset size (i.e., spacing used in resampling during model training), and based on this, the dose information determining system calculates a target ratio between the size of the first sub-image and the preset size.

Optionally, if the target ratio is 1, it is indicated that the size of the first sub-graph is the same as the preset size, and the medical image corresponding to the target object may be directly used as the target medical image; if the target ratio is smaller than 1, it is indicated that the size of the first sub-image is smaller than the preset size (which can be understood as that the grid point size of the medical image corresponding to the target object is smaller, and the number of sub-images divided according to the preset size is smaller than the number of preset training grid points), so that the medical image corresponding to the target object needs to be amplified in equal proportion according to the target ratio to obtain the target medical image. In addition, in addition to the equal-scale magnification, the purpose of adjusting the medical image to the target medical image may be achieved by filling an area other than the medical image.

Optionally, if the target ratio is greater than 1, it is indicated that the size of the first sub-image is greater than a preset size (which may also be understood as that the grid point size of the medical image corresponding to the target object is too large, and the number of sub-images divided according to the preset size is greater than the number of preset training grid points), so that the medical image corresponding to the target object needs to be scaled down according to the target ratio to obtain the target medical image.

Optionally, after the target medical image is obtained, a radiotherapy plan for the target object may be generated based on the target medical image.

In an alternative embodiment, after generating the radiotherapy plan of the target object from the target medical image, the dose information determination system may split the radiotherapy plan of the target object into a plurality of sub-plans to be processed, wherein each of the plurality of sub-plans to be processed corresponds to a portal or sub-field to which the radiotherapy plan of the target object corresponds. Then, the dose information determining system may calculate dose information corresponding to each sub-plan to be processed through a second dose algorithm, and convert the dose information corresponding to each sub-plan to be processed into an initial dose distribution map corresponding to the sub-plan to be processed. And finally, inputting an initial dose distribution map corresponding to each sub-plan to be processed into a dose prediction model by the dose information determination system, and predicting and generating an intermediate dose distribution map corresponding to the sub-plan to be processed according to priori knowledge learned by the first dose information and the second dose information by the dose prediction model, wherein the intermediate dose distribution map corresponding to the sub-plan to be processed represents the dose distribution map correspondingly obtained when the first dose algorithm is used for calculating the sub-plan to be processed.

Alternatively, the dose information determining system may split the radiotherapy plan of the target object into a plurality of sub-plans to be processed according to the radiation field or the sub-field, then calculate dose information corresponding to each sub-plan to be processed using a second dose algorithm (e.g., an analytical algorithm such as PB algorithm), and convert the dose information corresponding to each sub-plan to be processed into an initial dose distribution map corresponding to the sub-plan to be processed. Finally, the dose information determining system inputs the initial dose distribution map corresponding to each sub-plan to be processed into a dose prediction model, and the dose prediction model generates an intermediate dose distribution map corresponding to the sub-plan to be processed according to the initial dose distribution map prediction corresponding to each sub-plan to be processed according to the training process of the dose prediction model, wherein the intermediate dose distribution map corresponding to the sub-plan to be processed represents the dose distribution map correspondingly obtained when the sub-plan to be processed is calculated by using a first dose algorithm (such as an MC algorithm).

In an alternative embodiment, the dose information determining system further adjusts the intermediate dose distribution map corresponding to each sub-plan to be processed according to a target ratio to obtain a target dose distribution map corresponding to the sub-plan to be processed, where an image size ratio between the target dose distribution map and the intermediate dose distribution map is the target ratio. And then, the dose information determining system superimposes the target dose distribution map corresponding to each sub-plan to be processed to obtain the whole dose distribution map corresponding to the radiotherapy plan of the target object.

Optionally, because there is a ratio difference of the target ratio between the target medical image and the medical image corresponding to the original target object, the intermediate dose distribution map is not a dose distribution map matched with the medical image corresponding to the original target object, in order to obtain an accurate dose distribution map corresponding to the target object, the intermediate dose distribution map corresponding to each sub-plan to be processed needs to be adjusted according to the target ratio to obtain a target dose distribution map corresponding to the sub-plan to be processed, for example, if the ratio between the target medical image and the medical image corresponding to the original target object is 3/4, the intermediate dose distribution map corresponding to each sub-plan to be processed needs to be amplified by 4/3 times, so as to obtain a target dose distribution map corresponding to each sub-plan to be processed; if the ratio between the target medical image and the medical image corresponding to the original target object is 4/3, the intermediate dose distribution map corresponding to each sub-plan to be processed needs to be reduced by 3/4 times, so that the target dose distribution map corresponding to each sub-plan to be processed is obtained.

And finally, superposing the target dose distribution map corresponding to each sub-plan to be processed (namely superposing the doses of all the radiation fields or the sub-fields), thereby obtaining the integral dose distribution map corresponding to the radiotherapy plan of the target object.

As can be seen from the above, the present application combines the medical images generated according to N preset materials and the medical images generated according to multiple organ tissues in the image library to obtain a large number of medical image samples (i.e.The images to be processed), and further generates a plurality of radiotherapy plan samples (F target radiotherapy plans) based on a plurality of medical image samples, thereby solving the problem that a large amount of training data are difficult to acquire in the prior art, further reducing the training difficulty of the dose prediction model and improving the training effect of the dose prediction model.

Example 2

According to an embodiment of the present application, an embodiment of a training device of a dose prediction model is provided. Fig. 6 is a schematic diagram of an alternative training apparatus for a dose prediction model according to an embodiment of the present application, as shown in fig. 6, the training apparatus for a dose prediction model includes: a generating unit 601, a combining unit 602, a calculating unit 603 and a training unit 604.

The generating unit 601 is configured to randomly generate K target dose areas, where K is an integer greater than 1, and the target dose areas are areas that receive radiation irradiation; a combining unit 602 for combining the K target dose regions with the Z medical images in the image library to obtainThe medical images in the image library at least comprise medical images generated according to N preset materials and medical images generated according to various organ tissues, wherein Z and N are integers larger than 1; a calculation unit 603 for calculating according to- >F target radiotherapy plans are generated by the images to be processed, first dose information corresponding to each target radiotherapy plan is calculated through a first dose algorithm, second dose information corresponding to each target radiotherapy plan is calculated through a second dose algorithm, wherein F is an integer larger than the product of K and Z, the calculation accuracy of the first dose algorithm is higher than that of the second dose algorithm, and the calculation speed of the first dose algorithm is slower than that of the second dose algorithm; the training unit 604 is configured to train to obtain a dose prediction model according to the first dose information and the second dose information corresponding to each target radiotherapy plan. />

Optionally, the training device of the dose prediction model further comprises: the device comprises a first generation unit, a first execution unit, a first acquisition unit and a storage unit. The first generation unit is used for generating one medical image corresponding to each preset material in the N preset materials to obtain N medical images; the first execution unit is used for executing M times of target operations based on N preset materials to obtain M medical images, wherein M is an integer larger than 1, each target operation in the M times of target operations is used for randomly selecting at least two preset materials from the N preset materials as a material combination, and generating a medical image corresponding to the material combination; a first acquisition unit for acquiring Y medical images based on each of X human body regions of a human body to obtain The method comprises the steps of performing medical imaging, wherein X and Y are integers larger than 1, and organ tissues corresponding to different human body areas in X human body areas are different; a storage unit for storing N medical images, M medical images>Storing the medical images into an image library, wherein +_>。

Optionally, the generating unit 601 includes: the device comprises a first generation subunit, a setting subunit and an operation subunit, wherein the first generation subunit is used for generating a three-dimensional virtual sphere with random size; the setting subunit is used for constructing a unit deformation field around the virtual sphere and randomly setting force application angles in a plurality of preset directions respectively; and the operation subunit is used for respectively carrying out force application operation on the virtual sphere for K times in the unit deformation field according to the force application angle to obtain K target dose areas, wherein the virtual sphere after each force application operation in the K force application operations serves as one target dose area in the K target dose areas, the force application size and the force application direction of each force application operation are randomly arranged, and the force application direction represents outward force application or inward force application on the virtual sphere according to the force application angle.

Optionally, the training device of the dose prediction model further comprises: the first setting subunit, the second setting subunit and the third setting subunit. The first setting subunit is used for randomly setting L isocenter points, wherein L is an integer larger than 1, and each isocenter point in the L isocenter points represents the isocenter point corresponding to the rotation of the accelerator in the process of implementing the radiotherapy plan; a second setting subunit, configured to set H pieces of portal information at random, where H is an integer greater than 1, the number of the portals included in each piece of portal information is a random value, and the angle of each portal included in each piece of portal information is a random value; and a third setting subunit for randomly setting G prescription dose information, where G is an integer greater than 1.

Optionally, the computing unit 603 includes: a first processing subunit, a second processing subunit, a third processing subunit, and a fourth processing subunit, where the first processing subunit is configured to combine each of the L isocenter withCombining each of the images to be processed to obtain +.>The personal parameter data; a second processing subunit for associating each of the H pieces of portal information with +.>Each parameter data in the individual parameter data is combined to obtainTarget parameter data; a third processing subunit for processing the G prescription dose information and +.>Each of the target parameter data generates an initial radiotherapy plan to obtain +.>An initial radiotherapy plan, wherein->The method comprises the steps of carrying out a first treatment on the surface of the And the fourth processing subunit is used for carrying out plan optimization operation on each initial radiotherapy plan in the F initial radiotherapy plans, taking each initial radiotherapy plan which completes the plan optimization operation as a target radiotherapy plan, and obtaining F target radiotherapy plans, wherein the plan optimization operation is used for realizing that the preset proportion of the volume in the target dosage area meets the prescription dosage information by adjusting the radiation field information.

Optionally, the computing unit 603 includes: the device comprises a plan splitting sub-unit, a first calculating sub-unit, a fifth processing sub-unit, a second calculating sub-unit and a sixth processing sub-unit, wherein the plan splitting sub-unit is used for splitting each target radiotherapy plan into a plurality of sub-plans according to the radiation field information corresponding to the target radiotherapy plan, and each sub-plan in the plurality of sub-plans corresponds to one radiation field or sub-field corresponding to the target radiotherapy plan; the first calculating subunit is used for calculating first sub-dose information corresponding to each sub-plan through a first dose algorithm; a fifth processing subunit, configured to use all the first sub-dose information corresponding to the multiple sub-plans as first dose information corresponding to the target radiotherapy plan; a second calculation subunit, configured to calculate second sub-dose information corresponding to each sub-plan through a second dose algorithm; and the sixth processing subunit is used for taking all second sub-dose information corresponding to the plurality of sub-plans as second dose information corresponding to the target radiotherapy plan.

Optionally, training unit 604 includes: a seventh processing subunit, a resampling subunit, an eighth processing subunit, and a training subunit. Wherein the seventh processing subunit is configured to represent the first dose information by a first dose distribution map and represent the second dose information by a second dose distribution map; the resampling subunit is used for resampling the first dose distribution diagram and the second dose distribution diagram respectively according to the preset training lattice point number, wherein the resampling is used for extracting J subgraphs corresponding to the first dose distribution diagram and J subgraphs corresponding to the second dose distribution diagram, the sizes of the subgraphs are all preset sizes, and J is the training lattice point number; an eighth processing subunit, configured to use J sub-graphs corresponding to the first dose distribution chart as training samples, and J sub-graphs corresponding to the second dose distribution chart as training labels; and the training subunit is used for carrying out iterative training on the neural network according to the training sample and the training label, and taking the neural network after the iterative training is finished as a dose prediction model.

Optionally, the training device of the dose prediction model further comprises: the device comprises an acquisition unit, a first calculation unit, a first adjustment unit and a radiotherapy plan generation unit. The acquisition unit is used for acquiring medical images corresponding to the target objects; the first calculating unit is used for dividing the medical image corresponding to the target object into J first subgraphs with the same size and calculating a target ratio between the size of the first subgraphs and a preset size; the first adjusting unit is used for adjusting the medical image corresponding to the target object into a target medical image according to the target ratio, wherein the target medical image can be divided into J second sub-images, and the sizes of the second sub-images are preset sizes; and the radiotherapy plan generation unit is used for generating a radiotherapy plan of the target object according to the target medical image.

Optionally, the training device of the dose prediction model further comprises: the device comprises a splitting unit, a second calculating unit and an input unit. The splitting unit is used for splitting the radiotherapy plan of the target object into a plurality of sub-plans to be processed, wherein each sub-plan to be processed in the plurality of sub-plans to be processed corresponds to one radiation field or sub-field corresponding to the radiotherapy plan of the target object; the second calculation unit is used for calculating the dose information corresponding to each sub-plan to be processed through a second dose algorithm and converting the dose information corresponding to each sub-plan to be processed into an initial dose distribution map corresponding to the sub-plan to be processed; the input unit is used for inputting an initial dose distribution map corresponding to each sub-plan to be processed into the dose prediction model, and generating an intermediate dose distribution map corresponding to the sub-plan to be processed through the dose prediction model according to priori knowledge obtained through learning of the first dose information and the second dose information, wherein the intermediate dose distribution map corresponding to the sub-plan to be processed represents the dose distribution map corresponding to the sub-plan to be processed when the first dose algorithm is used for calculating the sub-plan to be processed.

Optionally, the training device of the dose prediction model further comprises: the second adjusting unit and the superposition unit are used for adjusting the intermediate dose distribution map corresponding to each sub-plan to be processed according to the target ratio to obtain a target dose distribution map corresponding to the sub-plan to be processed, wherein the image size ratio between the target dose distribution map and the intermediate dose distribution map is the target ratio; and the superposition unit is used for superposing the target dose distribution map corresponding to each sub-plan to be processed to obtain the integral dose distribution map corresponding to the radiotherapy plan of the target object.

Example 3

According to another aspect of the embodiments of the present application, there is also provided a computer readable storage medium, including a stored computer program, where the computer program when executed controls a device in which the computer readable storage medium is located to perform the training method of the dose prediction model according to any one of the above embodiments 1.

The foregoing embodiment numbers of the present application are merely for describing, and do not represent advantages or disadvantages of the embodiments.

In the foregoing embodiments of the present application, the descriptions of the embodiments are emphasized, and for a portion of this disclosure that is not described in detail in this embodiment, reference is made to the related descriptions of other embodiments.

In the several embodiments provided in the present application, it should be understood that the disclosed technology content may be implemented in other manners. The above-described embodiments of the apparatus are merely exemplary, and the division of units may be a logic function division, and there may be another division manner in actual implementation, for example, multiple units or components may be combined or integrated into another system, or some features may be omitted, or not performed. Alternatively, the coupling or direct coupling or communication connection shown or discussed with each other may be through some interfaces, units or modules, or may be in electrical or other forms.

The units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed over a plurality of units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

In addition, each functional unit in each embodiment of the present application may be integrated in one processing unit, or each unit may exist alone physically, or two or more units may be integrated in one unit. The integrated units may be implemented in hardware or in software functional units.

The integrated units, if implemented in the form of software functional units and sold or used as stand-alone products, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present application may be embodied in essence or a part contributing to the prior art or all or part of the technical solution, in the form of a software product stored in a storage medium, including several instructions to cause a computer device (which may be a personal computer, a server or a network device, etc.) to perform all or part of the steps of the methods of the embodiments of the present application. And the aforementioned storage medium includes: a U-disk, a Read-Only Memory (ROM), a random access Memory (RAM, random Access Memory), a removable hard disk, a magnetic disk, or an optical disk, or other various media capable of storing program codes.

The foregoing is merely a preferred embodiment of the present application and it should be noted that modifications and adaptations to those skilled in the art may be made without departing from the principles of the present application and are intended to be comprehended within the scope of the present application.

Claims

1. A method of training a dose prediction model, comprising:

randomly generating K target dose areas, wherein K is an integer greater than 1, and the target dose areas are areas which are irradiated by rays;

combining the K target dose areas with Z medical images in an image library to obtainThe medical images in the image library at least comprise medical images generated according to N preset materials and medical images generated according to various organ tissues, wherein Z and N are integers larger than 1;

according to the describedF target radiotherapy plans are generated by the images to be processed, first dose information corresponding to each target radiotherapy plan is calculated through a first dose algorithm, second dose information corresponding to each target radiotherapy plan is calculated through a second dose algorithm, F is an integer larger than the product of K and Z, the calculation accuracy of the first dose algorithm is higher than that of the second dose algorithm, and the calculation speed of the first dose algorithm is slower than that of the second dose algorithm;

and training according to the first dose information and the second dose information corresponding to each target radiotherapy plan to obtain a dose prediction model.

2. The method of claim 1, wherein the combination of the K target dose regions with the Z medical images in the image library yieldsBefore the images to be processed, the training method of the dose prediction model further comprises the following steps:

generating a medical image corresponding to each preset material in the N preset materials to obtain N medical images;

m times of target operations are executed based on the N preset materials to obtain M medical images, wherein M is an integer larger than 1, each target operation in the M times of target operations is used for randomly selecting at least two preset materials from the N preset materials as a material combination, and medical images corresponding to the material combination are generated;

acquiring Y medical images based on each of X human body regions of a human body to obtainThe method comprises the steps of performing medical imaging, wherein X and Y are integers larger than 1, and organ tissues corresponding to different human body areas in X human body areas are different;

the N medical images, the M medical images and theStoring the individual medical images into the image library, wherein +_>。

3. The method of training a dose prediction model according to claim 1, wherein randomly generating K target dose regions comprises:

Generating a virtual sphere in a three-dimensional form with random size;

constructing a unit deformation field around the virtual sphere, and randomly setting force application angles in a plurality of preset directions respectively;

and respectively carrying out K times of force application operations on the virtual sphere in the unit deformation field according to the force application angles to obtain K target dose areas, wherein the virtual sphere after each time of force application operation in the K times of force application operations serves as one target dose area in the K target dose areas, the force application size and the force application direction of each time of force application operation are randomly arranged, and the force application direction represents outward force application or inward force application on the virtual sphere according to the force application angles.

4. A method of training a dose predictive model as claimed in claim 1, in accordance with the followingBefore the F target radiotherapy plans are generated by the images to be processed, the training method of the dose prediction model further comprises the following steps:

randomly setting L isocenter points, wherein L is an integer greater than 1, and each isocenter point in the L isocenter points represents the isocenter point corresponding to the rotation of the accelerator in the process of implementing the radiotherapy plan;

Randomly setting H pieces of field information, wherein H is an integer greater than 1, the number of fields included in each piece of field information is a random value, and the angle of each field included in each piece of field information is a random value;

5. The method of training a dose predictive model as set forth in claim 4, wherein, in accordance with saidGenerating F target radiotherapy plans by the images to be processed comprises the following steps:

associating each of the L isocenter points with theCombining each of the images to be processed to obtain +.>The personal parameter data;

associating each of the H pieces of portal information with the portal informationEach of the parameter data is combined to obtain +.>Target parameter data;

according to each of the G prescription dose information and the prescription dose informationEach of the target parameter data generates an initial radiotherapy plan to obtain +.>An initial radiation therapy plan, wherein,；

and carrying out plan optimization operation on each initial radiotherapy plan in the F initial radiotherapy plans, taking each initial radiotherapy plan completed with the plan optimization operation as a target radiotherapy plan, and obtaining the F target radiotherapy plans, wherein the plan optimization operation is used for realizing that the preset proportion of the volume in the target dose area meets the prescription dose information by adjusting the radiation field information.

6. The method of claim 1, wherein calculating the first dose information corresponding to each target radiotherapy plan by a first dose algorithm and calculating the second dose information corresponding to each target radiotherapy plan by a second dose algorithm comprises:

splitting the target radiotherapy plan into a plurality of sub-plans according to the radiation field information corresponding to each target radiotherapy plan, wherein each sub-plan in the plurality of sub-plans corresponds to one radiation field or sub-field corresponding to the target radiotherapy plan;

calculating first sub-dose information corresponding to each sub-plan through the first dose algorithm;

taking all the first sub-dose information corresponding to the plurality of sub-plans as first dose information corresponding to the target radiotherapy plan;

calculating second sub-dose information corresponding to each sub-plan through the second dose algorithm;

and taking all second sub-dose information corresponding to the plurality of sub-plans as second dose information corresponding to the target radiotherapy plan.

7. The method for training a dose prediction model according to claim 1, wherein training to obtain the dose prediction model according to the first dose information and the second dose information corresponding to each target radiotherapy plan comprises:

Representing the first dose information by a first dose profile and the second dose information by a second dose profile;

resampling the first dose distribution map and the second dose distribution map respectively according to the preset training lattice point number, wherein the resampling is used for extracting J subgraphs corresponding to the first dose distribution map and J subgraphs corresponding to the second dose distribution map, the sizes of the subgraphs are preset, and J is the training lattice point number;

taking J sub-graphs corresponding to the first dose distribution map as training samples and J sub-graphs corresponding to the second dose distribution map as training labels;

and carrying out iterative training on the neural network according to the training sample and the training label, and taking the neural network after the iterative training is finished as the dose prediction model.

8. The method of claim 7, wherein after training to obtain a dose prediction model according to the first dose information and the second dose information corresponding to each target radiotherapy plan, the method further comprises:

acquiring a medical image corresponding to a target object;

Dividing the medical image corresponding to the target object into J first sub-images with the same size, and calculating a target ratio between the size of the first sub-images and the preset size;

according to the target ratio, the medical image corresponding to the target object is adjusted to be a target medical image, wherein the target medical image can be divided into J second sub-images, and the size of the second sub-images is the preset size;

and generating a radiotherapy plan of the target object according to the target medical image.

9. The method of claim 8, further comprising, after generating a radiation therapy plan for the target subject from the target medical image:

splitting the radiotherapy plan of the target object into a plurality of sub-plans to be processed, wherein each sub-plan to be processed in the plurality of sub-plans to be processed corresponds to one field or sub-field corresponding to the radiotherapy plan of the target object;

calculating the dose information corresponding to each sub-plan to be processed through the second dose algorithm, and converting the dose information corresponding to each sub-plan to be processed into an initial dose distribution map corresponding to the sub-plan to be processed;

And inputting the initial dose distribution map corresponding to each sub-plan to be processed into the dose prediction model, and predicting and generating an intermediate dose distribution map corresponding to the sub-plan to be processed according to priori knowledge learned by the first dose information and the second dose information through the dose prediction model, wherein the intermediate dose distribution map corresponding to the sub-plan to be processed represents the dose distribution map corresponding to the sub-plan to be processed when the first dose algorithm is used for calculating the sub-plan to be processed.

10. The method according to claim 9, wherein after inputting the initial dose distribution map corresponding to each sub-plan to be processed into the dose prediction model, the method further comprises:

adjusting the intermediate dose distribution map corresponding to each sub-plan to be processed according to the target ratio to obtain a target dose distribution map corresponding to the sub-plan to be processed, wherein the image size ratio between the target dose distribution map and the intermediate dose distribution map is the target ratio;

And superposing the target dose distribution map corresponding to each sub-plan to be processed to obtain an overall dose distribution map corresponding to the radiotherapy plan of the target object.

11. A training device for a dose prediction model, comprising:

the generation unit is used for randomly generating K target dose areas, wherein K is an integer greater than 1, and the target dose areas are areas which are irradiated by rays;

a combining unit for combining the K target dose regions with Z medical images in the image library to obtainThe medical images in the image library at least comprise medical images generated according to N preset materials and medical images generated according to various organ tissues, wherein Z and N are integers larger than 1;

a computing unit for according to theF target radiotherapy plans are generated by the images to be processed, first dose information corresponding to each target radiotherapy plan is calculated through a first dose algorithm, second dose information corresponding to each target radiotherapy plan is calculated through a second dose algorithm, F is an integer larger than the product of K and Z, the calculation accuracy of the first dose algorithm is higher than that of the second dose algorithm, and the calculation speed of the first dose algorithm is slower than that of the second dose algorithm A calculation speed of the second dose algorithm;

and the training unit is used for training to obtain a dose prediction model according to the first dose information and the second dose information corresponding to each target radiotherapy plan.

12. A computer readable storage medium, wherein a computer program is stored in the computer readable storage medium, and wherein the computer program, when run, controls a device in which the computer readable storage medium is located to perform the method of training the dose prediction model according to any one of claims 1 to 10.