WO2023178527A1

WO2023178527A1 - Generation method and generation apparatus for tumor radiotherapy region

Info

Publication number: WO2023178527A1
Application number: PCT/CN2022/082331
Authority: WO
Inventors: 梁晓坤; 产银萍; 谢耀钦; 张楚龙; 何文丰
Original assignee: 中国科学院深圳先进技术研究院
Priority date: 2022-03-22
Filing date: 2022-03-22
Publication date: 2023-09-28

Abstract

The present application discloses a generation method and generation apparatus for a tumor radiotherapy region. The generation method for a tumor radiotherapy region comprises: acquiring a first CT image actually obtained during a treatment process of a target object and a second CT image initially obtained before the treatment; in response to an external input instruction, marking the position of a first control region on the second CT image to generate a third CT image; obtaining a first translation parameter and a first rotation parameter by means of inputting the first CT image and the third CT image into a dense network model; transforming the first CT image according to the first translation parameter and the first rotation parameter to generate a fourth CT image, wherein the fourth CT image is an image provided with a target control region for the radiotherapy of the target object.

Description

Method and device for generating tumor radiotherapy area

Technical field

The present application relates to the field of medical imaging, and in particular to a method and device for generating a tumor radiotherapy area.

Background technique

For radiotherapy of tumors with complex anatomy (e.g., head and neck cancer), precise positioning is crucial, and image guidance is a key step to achieve high-precision positioning. At present, computed tomography (CT) is a commonly used tool in image guidance, which is generally performed by obtaining the optimal transformation parameters that maximize the similarity between the initially obtained CT image and the actual CT image obtained during the treatment process. Registration for precise positioning.

In the related art, although automatic registration algorithms have been implemented into image-guided radiotherapy (IGRT) systems, due to the presence of artifacts in images and variations in anatomical structures during treatment , may be affected by factors such as deviations in the global image similarity measure between the initially obtained CT image and the actually obtained CT image, resulting in the positioning of the radiotherapy area for tumor patients still requiring manual intervention for correction.

Contents of the invention

This application discloses a method and device for generating a tumor radiotherapy area to solve the problem that the positioning of a tumor patient's radiotherapy area requires manual intervention for correction.

In order to solve the above problems, this application adopts the following technical solutions:

In a first aspect, embodiments of the present application provide a method for generating a tumor radiotherapy area, including: acquiring a first CT image actually obtained by a target object during treatment and a second CT image initially obtained before treatment; in response to External input instructions mark the position of the first control area on the second CT image to generate a third CT image; by inputting the first CT image and the third CT image into a dense network model, obtain the first translation parameters and a first rotation parameter; according to the first translation parameter and the first rotation parameter, the first CT image is transformed to generate a fourth CT image, wherein the fourth CT image is a Image of the target control area for radiotherapy treatment of the target subject.

In a second aspect, embodiments of the present application provide a device for generating a tumor radiotherapy area, including: a first acquisition module for acquiring the first CT image actually obtained by the target object during the treatment and the initial CT image obtained before the treatment. a second CT image; a marking module configured to mark the position of the first control area on the second CT image in response to an external input instruction to generate a third CT image; a second acquisition module configured to generate a third CT image by A CT image and the third CT image are input into a dense network model to obtain a first translation parameter and a first rotation parameter; a generation module is used to generate the first translation parameter and the first rotation parameter according to the first translation parameter and the first rotation parameter. A CT image is transformed to generate a fourth CT image, wherein the fourth CT image is an image with a target control area for radiotherapy of the target object.

In a third aspect, embodiments of the present application provide an electronic device. The electronic device includes a processor and a memory. The memory stores programs or instructions that can be run on the processor. The programs or instructions are processed by the processor. When the processor is executed, the steps of the method described in the first aspect are implemented.

In a fourth aspect, embodiments of the present application provide a readable storage medium. Programs or instructions are stored on the readable storage medium. When the programs or instructions are executed by a processor, the steps of the method described in the first aspect are implemented. .

Embodiments of the present application provide a method for generating a tumor radiotherapy area by acquiring the first CT image actually obtained by the target object during the treatment and the second CT image initially obtained before the treatment, and then in response to an external input instruction, The position of the first control area is marked on the second CT image to generate a third CT image, and then the first CT image and the third CT image are input into the dense network model to obtain the first translation parameter and the first rotation parameter, and then according to the A translation parameter and a first rotation parameter are used to transform the first CT image to generate a fourth CT image with a target control area for radiotherapy of the target object, thereby achieving positioning of the tumor radiotherapy area. In addition, registration is achieved through the artifact-free first control area manually selected by the doctor, which can avoid the impact of uneven image information in global registration, thereby reducing potential uncertainty errors.

Description of the drawings

Figure 1 is a schematic flow chart of a method for generating a tumor radiotherapy area disclosed in an embodiment of the present application;

Figure 2 is a schematic structural diagram of a device for generating a tumor radiotherapy area disclosed in an embodiment of the present application;

FIG. 3 is a schematic structural diagram of an electronic device disclosed in an embodiment of the present application.

Detailed ways

The technical solutions in the embodiments of the present application will be clearly described below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, but not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art fall within the scope of protection of this application.

The terms "first", "second", etc. in the description and claims of this application are used to distinguish similar objects and are not used to describe a specific order or sequence. It is to be understood that the figures so used are interchangeable under appropriate circumstances so that the embodiments of the present application can be practiced in orders other than those illustrated or described herein, and that "first," "second," etc. are distinguished Objects are usually of one type, and the number of objects is not limited. For example, the first object can be one or multiple. In addition, "and/or" in the description and claims indicates at least one of the connected objects, and the character "/" generally indicates that the related objects are in an "or" relationship.

The method for generating a tumor radiation treatment area provided by embodiments of the present application will be described in detail below with reference to the accompanying drawings through specific embodiments and application scenarios.

Figure 1 is a schematic flowchart of a method for generating a tumor radiotherapy area disclosed in an embodiment of the present application. The method can be executed by an electronic device. In other words, the method can be executed by software or hardware installed on the electronic device, as shown in Figure 1 As shown, the method includes the following steps.

S120. Obtain the first CT image actually obtained by the target object during the treatment and the second CT image initially obtained before the treatment.

S140. In response to the external input instruction, mark the position of the first control area on the second CT image to generate a third CT image.

That is to say, the doctor manually selects and marks the position of the first control area on the second CT image to generate a third CT image, where the position of the first control area can be an artifact-free area that is easily identifiable by the doctor.

S160. Obtain the first translation parameter and the first rotation parameter by inputting the first CT image and the third CT image into the dense network model.

That is, by inputting the first CT image and the second CT image marked with the position of the first control area into the dense network model, the first translation parameter and the first rotation parameter of the treatment bed are obtained.

S180. According to the first translation parameter and the first rotation parameter, transform the first CT image to generate a fourth CT image, where the fourth CT image is an image with a target control area for radiotherapy of the target object.

By transforming the first CT image to generate a fourth CT image according to the first translation parameter and the first rotation parameter, registration is achieved, and the fourth CT image and the second CT image can be consistent in a spatial and anatomical sense, That is, the fourth CT image and the second CT image are most similar.

Taking head and neck cancer (HNC) as an example, in the HNC area, the skeletal anatomy is not a simple rigid object. Therefore, for robustness, the positions of multiple first control areas can usually be marked on the second CT image. , and when there are multiple positions of the first control areas marked on the second CT image, by inputting the first CT image and the third CT image into the dense network model, the first translation parameter and the first rotation parameter are obtained, It may include: inputting the first CT image and the third CT image into a dense network model to obtain a plurality of first translation parameters and a plurality of first rotation parameters; and calculating respectively according to the plurality of first translation parameters and the plurality of first rotation parameters. The average value of the first translation parameter and the average value of the first rotation parameter. That is to say, when there are multiple positions of the first control areas marked on the second CT image, multiple first translation parameters can be obtained by inputting the first CT image and the third CT image into the dense network model. and a plurality of first rotation parameters, by calculating the average of the first translation parameter and the first rotation parameter respectively, to determine the conversion parameter, thereby improving the robustness of the model.

In the embodiment of the present application, before inputting the first CT image and the third CT image into the dense network model, it may also include: acquiring multiple sets of data sets of the first object, wherein the data set includes the data sets actually obtained during the treatment process. The fifth CT image, the sixth CT image initially obtained before treatment, and the position of the second control region in the sixth CT image. The position of the second control region is a randomly generated position in the sixth CT image; according to multiple sets of A training sample set is generated from a data set of an object; the training sample set is input into the dense network model to be trained for iterative training processing to obtain a trained dense network model. By using the position of the second control area randomly generated in the sixth CT image, this application eliminates the need for manual annotation of data and can complete the training of dense network models without relying on annotation data sets, greatly saving manpower.

In one implementation, inputting the training sample set into the dense network model to be trained for iterative training processing may include: marking the position of the second control area on the sixth CT image to generate a seventh CT image; The CT image and the seventh CT image are input into the dense network model to obtain the second translation parameter and the second rotation parameter; according to the second translation parameter and the second rotation parameter, the fifth CT image is transformed to generate the eighth CT image; according to The position of the second control area, respectively extract the first target control area in the sixth CT image and the second target control area in the eighth CT image; determine the first target control area and the second target control area through normalized cross-correlation Local similarity between regions; determine the global similarity between the sixth CT image and the eighth CT image through normalized cross-correlation; determine the sixth CT image and the eighth CT image based on the local similarity and global similarity. loss function between.

That is to say, in the process of training the dense network model, after the fifth CT image is transformed to generate the eighth CT image according to the second translation parameter and the second rotation parameter, according to the second control marked on the sixth CT image The position of the area, respectively extract the first target control area in the sixth CT image and the second target control area in the eighth CT image, and determine the relationship between the first target control area and the second target control area through normalized cross-correlation local similarity, determine the global similarity between the sixth CT image and the eighth CT image through normalized cross-correlation, and then determine the relationship between the sixth CT image and the eighth CT image based on the local similarity and global similarity. loss function. And in order to avoid the global impact of local artifacts or anatomical differences and improve the registration accuracy of the doctor's area of interest, the local similarity between the first target control area and the second target control area is used as the main part of the loss function. When the loss function between the sixth CT image and the eighth CT image is minimum, the image similarity between the sixth CT image and the eighth CT image is maximum.

Among them, the loss function between the sixth CT image and the eighth CT image can be expressed by the following formula: L (θ; I _d , I _p , v) = L _G (θ; I _p , I _d ) + λL _CV (θ; I _{p, CV} ; I′ _{d, CV} ), where _LG represents the global similarity between the sixth CT image and the eighth CT image, and L _CV represents the first target control area and the second target control The local similarity between regions, λ represents the constant parameter of the relative weight of global similarity and local similarity, θ represents the learning parameter of the dense network model, v represents the position of the second control region in the sixth CT image, I _d represents the third In the fifth CT image, I _p represents the sixth CT image, I _p,CV represents the first target control area in the sixth CT image, and I′ _d,CV represents the second target control area in the eighth CT image.

The local similarity between the first target control area and the second target control area, and the global similarity between the sixth CT image and the eighth CT image are calculated using negative normalized cross-correlation. The specific calculation is The formula is as follows:

Among them, I ₁ and I ₂ respectively represent two images or two target control areas, p is the index of the voxel,

and

It represents the average gray level of the voxels in I ₁ and I ₂ .

In the embodiment of the present application, after determining the loss function between the sixth CT image and the eighth CT image, it may also include: performing iterative training processing through the Adam (Adaptive Moment Estimation) algorithm, and saving the dense corresponding to the minimum loss function. Learning parameters of the network model. In other words, the Adam algorithm can be used to continuously iteratively train the dense network model to minimize the loss function, and save the weight parameters corresponding to the minimum loss function and the learning parameters of the dense network model, so that when the dense network model is used later, By directly substituting the data, the corresponding results can be obtained without iterating again, which can speed up the process. For example, in order to reduce the loss function, the learning rate in the Adam algorithm can be initially set to 10 ^-2 and then gradually reduced to 10 ^-6 , using a total of 10 ⁵ iterations to train the dense network model. In addition, when performing iterative training processing, a large amount of initial data can be generated by randomly translating and rotating the sixth CT image initially obtained before treatment, and by transforming the sixth CT image, and the dense network model to be trained can be processed through a large amount of initial data. Training can make the dense network model obtained by training more robust and avoid over-fitting.

In the embodiment of this application, the dense network model includes a convolution layer and three dense layers. Each dense layer is followed by a transition layer. The last dense layer is followed by a pooling layer and a linear layer. The dense layer The layer consists of a convolutional layer, a batch normalization layer and a sequence of rectified linear units, and the transition layer consists of a convolutional layer and a pooling layer. Among them, the convolutional layer of the dense network model is used to extract different features of the input image. The first convolutional layer extracts some low-level features such as edges, lines, corners, etc., and the multi-layer convolutional layer obtains deeper features and outputs Becomes a feature map; the pooling layer is used to compress the input feature map. On the one hand, it makes the feature map smaller and simplifies the network calculation complexity. On the other hand, it compresses the features and extracts the main features. It can also be used to prevent overfitting; The batch normalization layer is used to make the distribution of input data relatively stable and accelerate the model learning speed; the corrected linear unit sequence is used to provide the nonlinear expression modeling capability of the network; the transition layer is used to reduce the feature mapping between adjacent dense blocks The dimensionality of the layer; the linear layer is used to implement a linear combination or linear transformation of the previous layer, that is, convert their input features into output features.

Taking HNC as an example, during the experimental verification, for each patient, the CT images actually obtained during the treatment process, the CT images initially obtained before treatment, and the manually selected control area in the initially obtained CT images were collected. For robustness, four control regions are usually selected at the spinous process of the second cervical vertebra, the spinous process of the sixth cervical vertebra, the mandible and the skull. The final transformation parameters are obtained by applying the 4 translation parameters and 4 rotation parameters output by the dense network model respectively. A calculated average was determined using eight known and easily identifiable anatomical points marked by an experienced radiation oncologist as a reference.

In this application, the translation parameters and rotation parameters of the treatment bed using bone landmark alignment can be determined by the following formula:

Among them, R represents the rotation matrix of the treatment bed in the yaw, pitch and roll directions, T represents the translation matrix of the treatment bed in the front and rear, left and right and up and down directions,

Represents the control area in the CT image actually obtained during the treatment process,

Represents the control area in the CT image initially obtained before treatment. The value of R corresponding to when the formula obtains the minimum value is determined as the rotation parameter, and the value of T corresponding to when the formula obtains the minimum value is determined as the translation parameter.

Since this application gives more weight to the artifact-free control area in the loss function (that is, taking local similarity as the main part of the loss function), compared with the entire cone beam CT image when using the traditional registration method This application eliminates this misalignment by moving it rearward relative to its intended position. In addition, after registration through traditional methods, although the error of anatomical mapping in the control area is reduced, it still cannot be matched well due to metal image artifacts. However, through the method provided by this application, the anatomical structure can be matched very well. .

Table 1 is a summary of the translation and rotation errors of three different methods: the traditional registration method, the deep learning (DL) model without control volumes (CVs), and the DL model with CVs.

Table 1 Summary of translation and rotation errors of different methods

Through Table 1, it can be seen that in translation and rotation, the systematic/random positioning error between the measurement and reference of this method (DL model with CVs) is less than 0.47/1.13mm and 0.17/0.29° respectively. And, within the predefined clinically acceptable tolerance (2.0mm/1.0°), the test scores of the proposed method were 87.88% and 100.00% in translation and rotation, respectively. The proportion of HNC cases within the clinically acceptable range increased from 66.67% with the traditional method to 90.91% with this method. And, through quantitative comparison of deep learning-based registration results, we found that 90.91% of cases were clinically acceptable with CVs, which significantly improved the acceptable ratio of 63.64% without CVs.

The above data can show that the method proposed in this application can effectively achieve high-precision patient positioning.

For the method for generating a tumor radiation treatment area provided by an embodiment of the present application, the execution subject may be a device for generating a tumor radiation treatment area. In the embodiments of the present application, a method for generating a tumor radiotherapy area performed by a device for generating a radiation treatment area for tumors is used as an example to illustrate the device for generating a radiation treatment area for tumors provided by embodiments of the application.

Figure 2 is a schematic structural diagram of a device for generating a tumor radiation treatment area disclosed in an embodiment of the present application. As shown in FIG. 2 , the device 200 for generating a tumor radiation treatment area includes: a first acquisition module 210 , a marking module 220 , a second acquisition module 230 and a generation module 240 .

In this application, the first acquisition module 210 is used to acquire the first CT image actually obtained by the target object during the treatment and the second CT image initially obtained before the treatment; the marking module 220 is used to respond to external input instructions. , mark the position of the first control area on the second CT image, and generate a third CT image; the second acquisition module 230 is used to input the first CT image and the third CT image into a dense network model , obtain the first translation parameter and the first rotation parameter; the generation module 240 is configured to transform the first CT image according to the first translation parameter and the first rotation parameter to generate a fourth CT image, wherein, The fourth CT image is an image with a target control area for radiation therapy of the target subject.

In one implementation, when there are multiple positions of the first control areas marked on the second CT image, the second acquisition module 230 combines the first CT image and the The third CT image is input into the dense network model, and a first translation parameter and a first rotation parameter are obtained, including: inputting the first CT image and the third CT image into the dense network model, and obtaining a plurality of first translation parameters. and a plurality of first rotation parameters; according to a plurality of the first translation parameters and a plurality of the first rotation parameters, the average value of the first translation parameters and the average value of the first rotation parameters are respectively calculated.

In one implementation, the second acquisition module 230 is further configured to: acquire multiple sets of data sets of the first object before inputting the first CT image and the third CT image into the dense network model. , wherein the data set includes the fifth CT image actually obtained during the treatment, the sixth CT image initially obtained before the treatment, and the position of the second control area in the sixth CT image, and the second control area The position of the region is a randomly generated position in the sixth CT image; a training sample set is generated according to the data sets of the multiple groups of first objects; and the training sample set is input into the dense network model to be trained for iteration Training process to obtain the dense network model that has been trained.

In one implementation, the second acquisition module 230 inputs the training sample set into the dense network model to be trained for iterative training processing, including: marking the second control on the sixth CT image The position of the area, generate a seventh CT image; by inputting the fifth CT image and the seventh CT image into the dense network model, obtain the second translation parameter and the second rotation parameter; according to the second translation parameter and the second rotation parameter, transform the fifth CT image to generate an eighth CT image; according to the position of the second control area, extract the first target control area and the first target control area in the sixth CT image respectively. the second target control area in the eighth CT image; determine the local similarity between the first target control area and the second target control area through normalized cross-correlation; determine the local similarity between the first target control area and the second target control area through normalized cross-correlation The global similarity between the sixth CT image and the eighth CT image; determine the loss between the sixth CT image and the eighth CT image according to the local similarity and the global similarity. function.

In one implementation, the second acquisition module 230 is further configured to: after determining the loss function between the sixth CT image and the eighth CT image, perform iterative training processing through the Adam algorithm. , save the learning parameters of the dense network model corresponding to the minimum loss function.

The device for generating a tumor radiation treatment area provided by the embodiments of the present application can implement various processes implemented by the method embodiments for generating a tumor radiation treatment area. To avoid duplication, they will not be described again here.

Optionally, as shown in Figure 3, this embodiment of the present application also provides an electronic device 300, including a processor 301 and a memory 302. The memory 302 stores programs or instructions that can be run on the processor 301. When the program or instruction is executed by the processor 301, each step of the above embodiment of the method for generating a tumor radiation treatment area is implemented, and the same technical effect can be achieved. To avoid duplication, the details will not be described here.

Embodiments of the present application also provide a readable storage medium, with a program or instructions stored on the readable storage medium. When the program or instructions are executed by a processor, each process of the above embodiment of the method for generating a tumor radiation treatment area is implemented. And can achieve the same technical effect. To avoid repetition, they will not be described again here.

Wherein, the processor is the processor in the electronic device described in the above embodiment. The readable storage medium includes computer readable storage media, such as computer read-only memory ROM, random access memory RAM, magnetic disk or optical disk, etc.

It should be noted that, in this document, the terms "comprising", "comprising" or any other variations thereof are intended to cover a non-exclusive inclusion, such that a process, method, article or device that includes a series of elements not only includes those elements, It also includes other elements not expressly listed or inherent in the process, method, article or apparatus. Without further limitation, an element defined by the statement "comprises a..." does not exclude the presence of other identical elements in the process, method, article or device that includes the element. In addition, it should be pointed out that the scope of the methods and devices in the embodiments of the present application is not limited to performing functions in the order shown or discussed, but may also include performing functions in a substantially simultaneous manner or in reverse order according to the functions involved. Functions may be performed, for example, the methods described may be performed in an order different from that described, and various steps may be added, omitted, or combined. Additionally, features described with reference to certain examples may be combined in other examples.

Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus the necessary general hardware platform. Of course, it can also be implemented by hardware, but in many cases the former is better. implementation. Based on this understanding, the technical solution of the present application can be embodied in the form of a computer software product that is essentially or contributes to the existing technology. The computer software product is stored in a storage medium (such as ROM/RAM, disk , optical disk), including several instructions to cause a terminal (which can be a mobile phone, computer, server, or network device, etc.) to execute the methods described in various embodiments of this application.

The embodiments of the present application have been described above in conjunction with the accompanying drawings. However, the present application is not limited to the above-mentioned specific implementations. The above-mentioned specific implementations are only illustrative and not restrictive. Those of ordinary skill in the art will Inspired by this application, many forms can be made without departing from the purpose of this application and the scope protected by the claims, all of which fall within the protection of this application.

Claims

A method for generating a tumor radiotherapy area, including:

Obtaining the first CT image actually obtained by the target object during the treatment and the second CT image initially obtained before the treatment;

In response to an external input instruction, mark the position of the first control area on the second CT image to generate a third CT image;

By inputting the first CT image and the third CT image into a dense network model, a first translation parameter and a first rotation parameter are obtained;

According to the first translation parameter and the first rotation parameter, the first CT image is transformed to generate a fourth CT image, wherein the fourth CT image is with a target for radiotherapy of the target object. Image of the control area.
The generation method according to claim 1, wherein when there are multiple positions of the first control areas marked on the second CT image, the first CT image and the The third CT image is input into the dense network model to obtain the first translation parameter and the first rotation parameter, including:

Input the first CT image and the third CT image into the dense network model to obtain a plurality of first translation parameters and a plurality of first rotation parameters;

According to a plurality of the first translation parameters and a plurality of the first rotation parameters, an average value of the first translation parameters and an average value of the first rotation parameters are respectively calculated.
The generation method according to claim 1, wherein before inputting the first CT image and the third CT image into the dense network model, it further includes:

Acquire multiple sets of data sets of the first object, wherein the data set includes a fifth CT image actually obtained during the treatment, a sixth CT image initially obtained before treatment, and a second control in the sixth CT image. The position of the region, the position of the second control region is a randomly generated position in the sixth CT image;

Generate a training sample set according to the data sets of the plurality of first objects;

The training sample set is input into the dense network model to be trained for iterative training processing to obtain the trained dense network model.
The generation method according to claim 3, wherein said inputting the training sample set into the dense network model to be trained for iterative training processing includes:

Mark the position of the second control area on the sixth CT image to generate a seventh CT image;

By inputting the fifth CT image and the seventh CT image into the dense network model, a second translation parameter and a second rotation parameter are obtained;

Transform the fifth CT image to generate an eighth CT image according to the second translation parameter and the second rotation parameter;

According to the position of the second control area, respectively extract the first target control area in the sixth CT image and the second target control area in the eighth CT image;

Determining the local similarity between the first target control area and the second target control area by normalized cross-correlation;

Determine the global similarity between the sixth CT image and the eighth CT image by normalized cross-correlation;

According to the local similarity and the global similarity, a loss function between the sixth CT image and the eighth CT image is determined.
The generation method according to claim 4, wherein, after determining the loss function between the sixth CT image and the eighth CT image, further comprising:

Iterative training is performed through the Adam algorithm, and the learning parameters of the dense network model corresponding to the minimum loss function are saved.
A device for generating a tumor radiotherapy area, including:

The first acquisition module is used to acquire the first CT image actually obtained by the target object during the treatment and the second CT image initially obtained before treatment;

A marking module, configured to mark the position of the first control area on the second CT image in response to an external input instruction, and generate a third CT image;

a second acquisition module, configured to acquire the first translation parameter and the first rotation parameter by inputting the first CT image and the third CT image into a dense network model;

Generating module, configured to transform the first CT image to generate a fourth CT image according to the first translation parameter and the first rotation parameter, wherein the fourth CT image is a Image of the target control area of a subject receiving radiation therapy.
The generating device according to claim 6, wherein when there are multiple positions of the first control area marked on the second CT image, the second acquisition module The image and the third CT image are input into the dense network model to obtain the first translation parameter and the first rotation parameter, including:

Input the first CT image and the third CT image into the dense network model to obtain a plurality of first translation parameters and a plurality of first rotation parameters;

According to a plurality of the first translation parameters and a plurality of the first rotation parameters, an average value of the first translation parameters and an average value of the first rotation parameters are respectively calculated.
The generating device according to claim 6, wherein the second acquisition module is also used for:

Before inputting the first CT image and the third CT image into the dense network model, a plurality of data sets of the first object are obtained, wherein the data set includes a fifth CT actually obtained during the treatment process. image, the sixth CT image initially obtained before treatment and the position of the second control region in the sixth CT image, where the position of the second control region is a randomly generated position in the sixth CT image;

Generate a training sample set according to the data sets of the plurality of first objects;

The training sample set is input into the dense network model to be trained for iterative training processing to obtain the trained dense network model.
The generating device according to claim 8, wherein the second acquisition module inputs the training sample set into the dense network model to be trained for iterative training processing, including:

Mark the position of the second control area on the sixth CT image to generate a seventh CT image;

By inputting the fifth CT image and the seventh CT image into the dense network model, a second translation parameter and a second rotation parameter are obtained;

Transform the fifth CT image to generate an eighth CT image according to the second translation parameter and the second rotation parameter;

According to the position of the second control area, respectively extract the first target control area in the sixth CT image and the second target control area in the eighth CT image;

Determining the local similarity between the first target control area and the second target control area by normalized cross-correlation;

Determine the global similarity between the sixth CT image and the eighth CT image by normalized cross-correlation;

According to the local similarity and the global similarity, a loss function between the sixth CT image and the eighth CT image is determined.
The generating device according to claim 9, wherein the second acquisition module is also used for:

After the loss function between the sixth CT image and the eighth CT image is determined, an iterative training process is performed through the Adam algorithm, and the learning parameters of the dense network model corresponding to the minimum loss function are saved.