Deformation registration method and system, adaptive treatment planning method and treatment planning system

Abstract

The invention relates to the technical field of medical treatment, in particular to a deformation registration method and system, a self-adaptive treatment planning method and a treatment planning system which are used for improving the accuracy of image deformation registration, and the method comprises the steps: obtaining a plan image, a fractional image and dose distribution data corresponding to an initial treatment plan of a patient; inputting the plan image, the graded image and the dose distribution data corresponding to the initial treatment plan of the patient into a trained deformation registration model, performing deformation registration on the graded image and the plan image, and outputting a deformation field; wherein the trained deformation registration model is obtained based on dose distribution consistency loss training between dose distribution data corresponding to the initial treatment plan and deformation dose distribution data corresponding to the deformation field.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical treatment, and in particular, to a deformation registration method and system, an adaptive treatment planning method and a treatment planning system. BACKGROUND

[0002] In the clinical scenario of adaptive radiation therapy (ART), the deformation registration of the cone beam CT (CBCT) and the planning CT is a key technical link to achieve precise dose delivery. Due to the factors such as the change of patient's position between fractions, the physiological movement of organs, and the change of anatomical structure during treatment, the spatial positional relationship of the target region and the organs at risk may change significantly.

[0003] However, the existing registration algorithm relies on image grayscale information and ignores the influence of physical constraint factors on deformation field, which makes it difficult to ensure the registration accuracy. SUMMARY

[0004] The purpose of the present application is to provide a deformation registration method and system, an adaptive treatment planning method and a treatment planning system for improving the accuracy of image deformation registration.

[0005] In a first aspect, a deformation registration method is provided. The method comprises: obtaining a planning image, a fraction image and dose distribution data corresponding to an initial treatment plan of a patient; inputting the planning image, the fraction image and the dose distribution data corresponding to the initial treatment plan of the patient into a trained deformation registration model, performing deformation registration on the fraction image and the planning image, and outputting a deformation field; wherein the trained deformation registration model is trained based on dose distribution consistency loss between the dose distribution data corresponding to the initial treatment plan and deformation dose distribution data corresponding to the deformation field.

[0006] In a possible implementation, the method further comprises: generating a deformed image based on the deformation field and the planning image; and driving or terminating the iterative optimization of the deformation field by the trained deformation registration model based on the similarity between the deformed image and the fraction image.

[0007] In a possible implementation, the method further comprises: generating a deformed image based on the deformation field and the planning image; generating an optimized treatment plan based on the deformation field and the initial treatment plan; and driving or terminating the iterative optimization of the deformation field by the trained deformation registration model based on the similarity between the deformed image and the fraction image and the quality of the optimized treatment plan.

[0008] In a possible implementation, the deformation registration model comprises a dose-aware network and a deformation registration network; and the method comprises: inputting, into the trained deformation registration model, the plan image, the fraction image, and the dose distribution data corresponding to the initial treatment plan, performing deformation registration on the fraction image and the plan image, and outputting a deformation field; inputting the dose distribution data corresponding to the initial treatment plan and the plan image into the dose-aware network to obtain an anatomic dose feature output by the dose-aware network; the anatomic dose feature is used to reflect a correlation between an anatomic feature in the target image and a dose distribution; and inputting the anatomic dose feature, a feature map of the plan image, and a feature map of the fraction image into the deformation registration network, performing deformation registration on the fraction image and the plan image by using the deformation registration network, and outputting the deformation field.

[0009] In a possible implementation, the deformation registration model is obtained based on dose distribution consistency loss between dose distribution data corresponding to the initial treatment plan and deformation dose distribution data corresponding to the deformation field, and based on simultaneous training of the dose-aware network and the deformation registration network.

[0010] In a possible implementation, the deformation registration model further comprises a feature extraction network; and before the anatomic dose feature, the feature map of the plan image, and the feature map of the fraction image are input into the deformation registration network, the method further comprises: inputting the plan image and the fraction image into the feature extraction network, performing feature extraction on the plan image and the fraction image by using the feature extraction network, and outputting the feature map of the plan image and the feature map of the fraction image.

[0011] In a possible implementation, the deformation registration model further comprises a denoising network and a feature extraction network; and before the anatomic dose feature, the feature map of the plan image, and the feature map of the fraction image are input into the deformation registration network, the method further comprises: inputting the fraction image into the denoising network, performing denoising processing on the fraction image by using the denoising network, performing feature extraction on the fraction image after the denoising processing, and outputting the feature map of the fraction image; and inputting the plan image into the feature extraction network, performing feature extraction on the plan image by using the feature extraction network, and outputting the feature map of the plan image.

[0012] In a possible implementation, the deformation registration model is obtained based on dose distribution consistency loss between dose distribution data corresponding to the initial treatment plan and deformation dose distribution data corresponding to the deformation field, based on simultaneous training of the dose-aware network and the deformation registration network, and based on similarity loss between the fraction image and a deformation image corresponding to the deformation field.

[0013] In a possible implementation, the deformation image is generated based on the deformation field and the plan image, comprising: inputting the deformation field and the plan image into a spatial conversion network, performing spatial conversion on the plan image based on the deformation field by using the spatial conversion network, and obtaining the deformation image.

[0014] In a second aspect, an adaptive treatment planning method is provided. The method comprises: obtaining a planning image, a fraction image and an initial treatment plan of a patient; performing deformation registration on the fraction image and the planning image by using the deformation registration method provided in the first aspect to obtain a deformation field; and optimizing the initial treatment plan based on the deformation field to obtain an adaptive treatment plan.

[0015] In a third aspect, a deformation registration apparatus is provided. The deformation registration apparatus comprises an obtaining module and a processing module. The obtaining module is configured to obtain a planning image, a fraction image and dose distribution data corresponding to an initial treatment plan of a patient. The processing module is configured to input the planning image, the fraction image and the dose distribution data corresponding to the initial treatment plan into a trained deformation registration model, perform deformation registration on the fraction image and the planning image by using the trained deformation registration model, and output a deformation field. The trained deformation registration model is trained based on dose distribution consistency loss between the dose distribution data corresponding to the initial treatment plan and deformation dose distribution data corresponding to the deformation field.

[0016] In a possible implementation, the processing module is configured to generate a deformed image based on the deformation field and the planning image, and drive or terminate iterative optimization of the deformation field by the trained deformation registration model based on similarity between the deformed image and the fraction image.

[0017] In a possible implementation, the processing module is configured to generate a deformed image based on the deformation field and the planning image, generate an optimized treatment plan based on the deformation field and the initial treatment plan, and drive or terminate iterative optimization of the deformation field by the trained deformation registration model based on similarity between the deformed image and the fraction image and quality of the optimized treatment plan.

[0018] In a possible implementation, the deformation registration model comprises a dose-aware network and a deformation registration network. The processing module is configured to input the dose distribution data corresponding to the initial treatment plan and the planning image into the dose-aware network to obtain anatomical dose features output by the dose-aware network. The anatomical dose features are used to reflect a correlation between anatomical features in the target image and dose distribution. The processing module is further configured to input the anatomical dose features, a feature map of the planning image and a feature map of the fraction image into the deformation registration network, perform deformation registration on the fraction image and the planning image by using the deformation registration network, and output the deformation field.

[0019] In a possible implementation, the deformation registration model is trained based on dose distribution consistency loss between the dose distribution data corresponding to the initial treatment plan and deformation dose distribution data corresponding to the deformation field, and simultaneously trains the dose-aware network and the deformation registration network.

[0020] In a possible implementation, the deformation registration model further comprises a feature extraction network, and before the anatomical dose feature, the feature map of the planning image, and the feature map of the fraction image are input into the deformation registration network, the processing module is configured to input the planning image and the fraction image into the feature extraction network, perform feature extraction on the planning image and the fraction image through the feature extraction network, and output the feature map of the planning image and the feature map of the fraction image.

[0021] In a possible implementation, the deformation registration model further comprises a denoising network and a feature extraction network; before the anatomical dose feature, the feature map of the planning image, and the feature map of the fraction image are input into the deformation registration network, the processing module is configured to input the fraction image into the denoising network, perform denoising processing on the fraction image through the denoising network, and perform feature extraction on the fraction image after the denoising processing, and output the feature map of the fraction image; and input the planning image into the feature extraction network, perform feature extraction on the planning image through the feature extraction network, and output the feature map of the planning image.

[0022] In a possible implementation, the deformation registration model is obtained based on dose distribution consistency loss between dose distribution data corresponding to an initial treatment plan and deformation dose distribution data corresponding to a deformation field, simultaneous training of a dose perception network and a deformation registration network, and similarity loss between a fraction image and a deformation image corresponding to the deformation field, and simultaneous training of a denoising network and the deformation registration network.

[0023] In a possible implementation, the processing module is configured to input the deformation field and the planning image into a spatial conversion network, perform spatial conversion on the planning image based on the deformation field through the spatial conversion network, and obtain a deformation image.

[0024] In a fourth aspect, a deformation registration system is provided, comprising: a processor and a memory configured to store processor-executable instructions; wherein the processor is configured to execute the instructions to implement any of the deformation registration methods in the first aspect.

[0025] In a fifth aspect, a treatment planning system is provided, comprising: a processor and a memory configured to store processor-executable instructions; wherein the processor is configured to execute the instructions to implement any of the deformation registration methods in the first aspect.

[0026] In a sixth aspect, a computer-readable storage medium is provided, and the computer-readable storage medium stores instructions, when the instructions in the computer-readable storage medium are executed by a device, the device is enabled to execute any of the deformation registration methods in the first aspect.

[0027] In a seventh aspect, there is provided a computer program product comprising computer instructions which, when executed on a processor of a device, enable the device to perform the deformation registration method of any one of the first aspect described above.

[0028] The deformation registration method provided by the embodiments of the present application enables the trained deformation registration model to perform deformation registration between the planning image and the fraction image by taking the dose distribution data corresponding to the initial treatment plan as a basis to constrain the deformation field output by the deformation registration model, so as to ensure the rationality of the deformed dose field, focus the dose on the target region, and at the same time, realize accurate protection of the region of the organs at risk around the target region, improve the accuracy of deformation registration, and further ensure the dosimetric accuracy of the optimized radiotherapy plan. BRIEF DESCRIPTION OF DRAWINGS

[0029] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0030] Figure 1 A structural schematic diagram of a deformation registration system provided by the embodiments of the present application;

[0031] Figure 2 A structural schematic diagram of a treatment planning system provided by the embodiments of the present application;

[0032] Figure 3 A schematic diagram of a radiotherapy system provided by the embodiments of the present application;

[0033] Figure 4 A structural schematic diagram of another radiotherapy system provided by the embodiments of the present application;

[0034] Figure 5 A flowchart of a deformation registration method provided by the embodiments of the present application;

[0035] Figure 6 A flowchart of another deformation registration method provided by the embodiments of the present application;

[0036] Figure 7 A flowchart of another deformation registration method provided by the embodiments of the present application;

[0037] Figure 8 A flowchart of another deformation registration method provided by the embodiments of the present application;

[0038] Figure 9 A flowchart of another deformation registration method provided in an embodiment of the present application is shown in FIG. 8;

[0039] Figure 10 A flowchart of another deformation registration method provided in an embodiment of the present application is shown in FIG. 8;

[0040] Figure 11 A flowchart of another deformation registration method provided in an embodiment of the present application is shown in FIG. 8;

[0041] Figure 12 A flowchart of another deformation registration method provided in an embodiment of the present application is shown in FIG. 8;

[0042] Figure 13 A flowchart of another deformation registration method provided in an embodiment of the present application is shown in FIG. 8;

[0043] Figure 14 A flowchart of another adaptive treatment planning method provided in an embodiment of the present application is shown in FIG. 8;

[0044] Figure 15 A structural diagram of a deformation registration apparatus provided in an embodiment of the present application is shown in FIG. 8;

[0045] Figure 16 A structural diagram of an electronic device provided in an embodiment of the present application is shown in FIG. 8. DETAILED DESCRIPTION

[0046] In the embodiments of the present application, the terms "first", "second", "third", etc. are used only for the purpose of description, and should not be understood as indicating or implying relative importance or implying the number of the technical features indicated. Therefore, the features defined with "first", "second", etc. can explicitly or implicitly include one or more of the features.

[0047] In the embodiments of the present application, the terms "including", "containing" or any other variants thereof are intended to cover non-exclusive inclusion, so that the processes, methods, articles or devices including a series of elements not only include those elements, but also include other elements not explicitly listed or inherent to such processes, methods, articles or devices. Without more limitations, the element defined by the statement "including a" does not exclude the presence of another identical element in the process, method, article or device including the element.

[0048] "A and / or B" includes the following three combinations: only A, only B, and a combination of A and B.

[0049] In the clinical scenario of adaptive radiotherapy, the deformation registration of cone beam CT (CBCT) and planning CT is a key technology for precise dose delivery. Due to the factors such as the change of patient's position between fractions, organ physiological movement and anatomical structure change during treatment, the spatial position relationship of target and organs at risk may change significantly. However, the existing registration algorithm relies on image gray information and ignores the influence of physical constraint factors on deformation field, which makes it difficult to ensure the registration accuracy.

[0050] Based on the above technical problems, the embodiment of the present application provides a deformation registration method which can improve the accuracy of image deformation registration. The method comprises: obtaining the planning image, the fraction image and the dose distribution data corresponding to the initial treatment plan of the patient; inputting the planning image, the fraction image and the dose distribution data corresponding to the initial treatment plan of the patient into a trained deformation registration model, performing deformation registration on the fraction image and the planning image, and outputting a deformation field; wherein the trained deformation registration model is obtained based on the dose distribution consistency loss between the dose distribution data corresponding to the initial treatment plan and the deformation dose distribution data corresponding to the deformation field.

[0051] In some embodiments, the present application provides a deformation registration system, as shown in Figure 1 The deformation registration system 100 comprises a communication device 101 and a processing device 102, wherein the communication device 101 is in communication connection with the processing device 102.

[0052] The communication device 101 is used to obtain the planning image, the fraction image and the dose distribution data corresponding to the initial treatment plan of the patient.

[0053] In some embodiments, the planning image is used for the formulation of the initial treatment plan. The planning image is usually an image taken by a special imaging device when the patient first visits a doctor and the doctor needs to diagnose the disease for the first time.

[0054] For example, the planning image can be a CT image or an MRI image.

[0055] In some embodiments, the fraction image is an image taken by the imaging function of the treatment device before fraction treatment.

[0056] For example, the fraction image can be a CBCT image, a CT image or an MRI image.

[0057] In some embodiments, the dose distribution data corresponding to the initial treatment plan is obtained based on the planning image and the initial treatment plan parameters. For example, the electron density is calculated based on the tissue density reflected by the planning image, and the dose distribution result is calculated by using a Monte Carlo simulation or a convolution / superposition algorithm in combination with the initial treatment plan parameters (e.g., the ray source parameters (energy, shape, etc.)), the patient target region geometric information and the tissue density, to obtain the dose value of each point in the three-dimensional space of the patient target region, thereby obtaining the dose distribution result.

[0058] In some embodiments, the dose distribution data corresponding to the initial treatment plan is in the form of a fluence map, that is, the calculated dose distribution result is presented in the form of pseudo-color or isodose line to generate a two-dimensional or three-dimensional dose distribution map. For example, the high-dose area can be displayed as red and the low-dose area can be displayed as blue to form an image that visually displays the dose gradient.

[0059] In some embodiments, the trained deformation registration model is configured in the processing device 102, and the processing device 102 is used to input the planning image of the patient, the fraction image and the dose distribution data corresponding to the initial treatment plan into the trained deformation registration model, perform deformation registration on the fraction image and the planning image, and output the deformation field.

[0060] In some embodiments, the trained deformation registration model is trained based on the dose distribution consistency loss between the dose distribution data corresponding to the initial treatment plan and the deformation dose distribution data corresponding to the deformation field.

[0061] In some embodiments, the deformation field can be output in the form of a vector field or a displacement field. For example, the deformation field can be represented as a two-dimensional or three-dimensional vector field, and each vector represents the displacement vector of the corresponding pixel or voxel in the image in the registration process.

[0062] It should be noted that the deformation field reflects the displacement information of the human tissues of the patient, especially the displacement information of the organs at risk and the target region from the planning image to the fraction image, such as the expansion, contraction, rotation or distortion of the organs, which can be used for subsequent image deformation and analysis.

[0063] In some embodiments, the regularization loss is used to constrain the output deformation field to ensure the smoothness of the deformation field and prevent overfitting of the deformation registration model.

[0064] In some embodiments, the planning image and the fraction image are preprocessed before being input into the trained deformation registration model, such as image normalization processing, grayscale adjustment, image cropping processing, etc., so as to ensure the quality and consistency of the input images and improve the quality of the deformation field output by the deformation registration model.

[0065] In some embodiments, the present application provides a treatment planning system, and the treatment planning system 200 has a deformation registration function and a treatment planning function.

[0066] In some embodiments, the deformation registration function comprises performing deformation registration on the fraction image and the planning image of the patient, and outputting a deformation field.

[0067] In some embodiments, the treatment planning function comprises generating an initial treatment plan according to the planning image of the patient.

[0068] In some embodiments, the treatment planning function further comprises optimizing the initial treatment plan according to the deformation field to generate an adaptive treatment plan.

[0069] As shown in Figure 2 , the treatment planning system 200 comprises a communication device 201 and a processing device 202, which are communicatively connected.

[0070] The processing device 202 is configured to implement the deformation registration function and the treatment planning function of the treatment planning system 200.

[0071] The communication device 201 is configured to enable the treatment planning system 200 to establish communication with other devices or systems and perform data transmission, such as transmitting the fraction image and the planning image of the patient, the initial treatment plan and the adaptive treatment plan, etc.

[0072] In some embodiments, the present application provides a radiotherapy system for performing radiotherapy on a patient.

[0073] Figure 3 A structural schematic diagram of a radiotherapy system provided by an embodiment of the present application is shown in Figure 3 , which comprises a treatment planning system 200 (including a communication device 201 and a processing device 202), an image acquisition device 301 and a radiotherapy device 302. The communication device 201 of the treatment planning system 200 is communicatively connected to the image acquisition device 301 and the radiotherapy device 302, respectively.

[0074] In some embodiments, Figure 3 , the treatment planning system 200 comprises a treatment planning system 200 as shown in Figure 2 , which is configured to implement the treatment planning function and the deformation registration function.

[0075] Figure 4 A structural schematic diagram of another radiotherapy system provided by an embodiment of the present application is shown in Figure 4As shown, the radiotherapy system 300 includes the deformation registration system 100 (including the communication device 101 and the processing device 102), the image acquisition device 301, the radiotherapy device 302 and the treatment planning device 303. Among them, the image acquisition device 301 is in communication connection with the communication device 101 and the treatment planning device 303 respectively, and the radiotherapy device 302 is in communication connection with the communication device 101 and the treatment planning device 303 respectively.

[0076] In some embodiments, Figure 4 The deformation registration system 100 included in the above Figure 1 The deformation registration system 100 shown in FIG. 1 is used to achieve the role of the deformation registration system 100 shown in FIG. 1. Figure 1

[0077] In some embodiments, Figure 4 The treatment planning device 303 included in the above

[0078] In some embodiments, the treatment planning device 303 can include a client and a server.

[0079] Among them, the client can be at least one of a smart phone, a desktop computer, a laptop computer, a virtual reality terminal, an augmented reality terminal, a wireless terminal and a laptop computer. For example, the user runs the treatment planning system on the server through the client, which can trigger the server to execute the treatment planning optimization process and display the optimized adaptive treatment plan. In this way, the user's time can be effectively saved, and the optimized treatment plan can be more intuitively presented to the user for evaluation of the radiotherapy plan.

[0080] Among them, the server can be at least one of an independent physical server, or a server cluster or distributed file system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud service, cloud database, cloud computing, cloud function, cloud storage, network service, cloud communication, middleware service, domain name service, security service, content distribution network, and big data or artificial intelligence platform. The number of servers can be more or less in some embodiments, which is not limited in the embodiments of the present application.

[0081] The image acquisition device 301 in the above Figure 3 and Figure 4 is used to obtain the planning image of the patient.

[0082] For example, the planning image can be a CT image or an MRI image.

[0083] ​For example, the image acquisition device 301 can be a CT device or an MRI device.

[0084] The above Figure 3 and Figure 4 The radiotherapy device 302 can be used to irradiate a tumor or other diseased tissue with high-energy rays to achieve the purpose of treating diseases.

[0085] In some embodiments, the radiotherapy device 302 itself has an image acquisition function, and the radiotherapy device 302 is further configured to capture a fraction image of the patient before or during fraction treatment.

[0086] For example, the fraction image can be a CBCT image, a CT image, or an MRI image.

[0087] For example, the radiotherapy device 302 can be a medical linear accelerator, a gamma knife, an integrated treatment device integrating a medical linear accelerator and a gamma knife, or a CyberKnife device.

[0088] In some embodiments, the radiotherapy device 302 is further configured to perform radiotherapy on the patient based on an adaptive treatment plan.

[0089] The system architecture and business scenarios described in the embodiments of the present disclosure are used to more clearly illustrate the technical solutions of the embodiments of the present disclosure, and do not constitute a limitation on the technical solutions provided by the embodiments of the present disclosure. Those skilled in the art can know that, as the network architecture evolves and new business scenarios appear, the technical solutions provided by the embodiments of the present disclosure are also applicable to similar technical problems.

[0090] In some embodiments, the present application provides a deformation registration method applied to the processing device 102 in the deformation registration system 100, as shown in Figure 5 The method includes the following steps S101-S102:

[0091] S101, obtaining a planning image, a fraction image, and dose distribution data corresponding to an initial treatment plan of a patient.

[0092] In some embodiments, the planning image is used for the formulation of the initial treatment plan. The planning image is usually an image captured by a special imaging device when the patient first visits a doctor and the doctor first needs to diagnose the patient's condition.

[0093] For example, the planning image can be a CT image or an MRI image.

[0094] In some embodiments, the fraction image is an image captured by the imaging function of the treatment device before fraction treatment.

[0095] Exemplarily, the fractionated image can be a CBCT image, a CT image, or an MRI image.

[0096] In some embodiments, the dose distribution data corresponding to the initial treatment plan is obtained based on the planning image and the initial treatment plan parameters. Exemplarily, the electron density is calculated based on the tissue density reflected by the planning image, and a Monte Carlo simulation or a convolution / superposition algorithm is used to calculate the dose value of each point in the three-dimensional space of the patient target region in combination with the initial treatment plan parameters (such as the ray source parameters (energy, shape, etc.)), the patient target region geometric information, and the tissue density, to obtain the dose distribution result.

[0097] In some embodiments, the dose distribution data corresponding to the initial treatment plan is in the form of a fluence map, that is, the calculated dose distribution result is presented in the form of pseudo-color or isodose line to generate a two-dimensional or three-dimensional dose distribution map. For example, a high-dose area can be displayed as red and a low-dose area as blue to form an image that visually displays the dose gradient.

[0098] S102, input the planning image, the fractionated image, and the dose distribution data corresponding to the initial treatment plan of the patient into the trained deformation registration model, perform deformation registration on the fractionated image and the planning image, and output a deformation field.

[0099] In some embodiments, the trained deformation registration model is trained based on the dose distribution consistency loss between the dose distribution data corresponding to the initial treatment plan and the deformation dose distribution data corresponding to the deformation field.

[0100] In some embodiments, the deformation field can be output in the form of a vector field or a displacement field, and exemplarily, the deformation field can be represented as a two-dimensional or three-dimensional vector field, each vector representing the displacement vector of the corresponding pixel or voxel in the image in the registration process.

[0101] It should be noted that the deformation field reflects the displacement information of the human tissues of the patient, especially the organs at risk and the target region, from the planning image to the fractionated image, such as the expansion, contraction, rotation, or distortion of the organs, which can be used for subsequent image deformation and analysis.

[0102] In some embodiments, the deformation field output is constrained by the regularization loss to ensure the smoothness of the deformation field and prevent overfitting of the deformation registration model.

[0103] In some embodiments, the deformation dose distribution data corresponding to the deformation field refers to the deformed dose distribution data obtained by applying the deformation field to the dose distribution data corresponding to the initial treatment plan.

[0104] In some embodiments, the fraction image and the planning image are preprocessed before being input into the trained deformation registration model, such as image normalization, grayscale adjustment, image cropping, etc., so as to ensure the quality and consistency of the input images and improve the quality of the deformation field output by the deformation registration model.

[0105] It can be understood that the deformation registration method provided by the embodiments of the present application inputs the planning image, the fraction image and the dose distribution data corresponding to the initial treatment plan into the trained deformation registration model, so that the deformation registration model, when performing deformation registration between the planning image and the fraction image, takes the dose distribution data corresponding to the initial plan as a basis to constrain the deformation field output by the deformation registration model, ensures the rationality of the deformed dose field, can focus the dose on the target region, at the same time, realizes accurate protection of the region around the target region, improves the accuracy of deformation registration, and further ensures the dosimetric accuracy of the optimized radiotherapy plan.

[0106] In some embodiments, the accuracy of the deformation field obtained by single registration may be low, and the deformation registration model optimizes the final output deformation field through iterative processing.

[0107] A possible implementation manner is shown in Figure 6 The method further includes the following processes:

[0108] S201, generating a deformed image based on the deformation field and the planning image.

[0109] In some embodiments, the deformed image is generated based on the deformation field and the planning image by using a spatial conversion network. The deformation field and the planning image are input into the spatial conversion network, the spatial conversion network uses the displacement information provided by the deformation field and the initial position information of the target region of the patient provided by the planning image to perform spatial conversion on the planning image, applies the deformation field on the planning image, and obtains the deformed image.

[0110] For example, the spatial conversion network uses a bilinear interpolation or a cubic spline interpolation method to calculate the position and grayscale value of each pixel point in the planning image after deformation according to the displacement vector of the deformation field, and generates the deformed image.

[0111] S202, driving or terminating the iterative optimization of the deformation field of the trained deformation registration model based on the similarity between the deformed image and the fraction image.

[0112] In some embodiments, the similarity between the deformed image and the fraction image is determined based on multiple indexes, including image matching degree, coverage conformity (key organ Dice coefficient), target region conformity index, etc.

[0113] In some embodiments, the similarity threshold is determined based on actual clinical requirements, which is not limited in the present application.

[0114] For example, if the similarity between the morphed image and the fraction image is greater than or equal to the similarity threshold, the morphing registration model is terminated to optimize the morphing field; if the similarity between the morphed image and the fraction image is less than the similarity threshold, the morphing registration model is driven to optimize the morphing field.

[0115] In some embodiments, the similarity between the morphed image and the fraction image is calculated by a similarity loss function, for example, the similarity loss function can be mean square error (MSE), mutual information (MI), correlation coefficient (CC) or structural similarity (SSIM), etc.

[0116] For example, as shown in Figure 7 the planning image, the fraction image and the dose distribution data are input into the morphing registration model, the morphing registration model outputs the morphing field, and the intelligent evaluation includes calculating the similarity (such as image matching degree) between the morphed image and the fraction image based on the morphing field, if the similarity is not up to standard, the planning image is updated to the morphed image, the dose distribution data is updated to the morphed dose distribution data based on the morphing field and the dose distribution data, and the morphing registration model is re-input with the fraction image to optimize the morphing field.

[0117] Another possible implementation, as shown in Figure 8 the method further includes the following processes:

[0118] S301, generating a morphed image based on the morphing field and the planning image.

[0119] In some embodiments, the morphed image is generated by a spatial conversion network based on the morphing field and the planning image. The morphing field and the planning image are input into the spatial conversion network, the spatial conversion network uses the displacement information provided by the morphing field and the initial position information of the patient target area provided by the planning image to perform spatial conversion on the planning image, applies the morphing field on the planning image, and obtains the morphed image.

[0120] For example, the spatial conversion network uses bilinear interpolation or cubic spline interpolation method to calculate the position and gray value of each pixel point in the planning image after morphing according to the displacement vector of the morphing field, and generates the morphed image.

[0121] S302, generating an optimized treatment plan based on the morphing field and the initial treatment plan.

[0122] In some embodiments, the planning parameters of the initial treatment plan are mapped onto a fractional image using a deformation field to obtain an optimized treatment plan.

[0123] S303. Based on the similarity between the deformed image and the fractional image, and the quality of the optimized treatment plan, drive or terminate the trained deformed registration model to iteratively optimize the deformed field.

[0124] In some embodiments, the similarity between deformed images and fractionated images is determined based on multiple indicators, including image matching degree, coverage conformity (Dice coefficient for critical organs), target conformity index, etc. The quality of optimized treatment plans is determined based on multiple quality indicators, such as target dose coverage (e.g., prescription dose-to-volume ratio), dose received by organs at risk (e.g., maximum dose, average dose, etc.), and uniformity of dose distribution. Treatment plan evaluation software or custom quality evaluation functions can be used to calculate various quality indicators of optimized treatment plans.

[0125] In this embodiment, a comprehensive judgment criterion is set for the similarity between the deformed image and the segmented image and the quality of the optimized treatment plan. Based on the comprehensive judgment criterion, the trained deformation registration model is driven or terminated to iteratively optimize the deformation field.

[0126] In some embodiments, by setting a similarity threshold and a quality threshold, the iterative optimization of the deformation field by the trained deformation registration model is determined. If the similarity between the deformed image and the fractional image is greater than or equal to the similarity threshold, and the quality of the optimized treatment plan is greater than or equal to the quality threshold, the iterative optimization of the deformation field by the deformation registration model is terminated. If the similarity between the deformed image and the fractional image is less than the similarity threshold, and / or the quality of the optimized treatment plan is less than the quality threshold, the iterative optimization of the deformation field by the deformation registration model is driven. The similarity threshold and quality threshold are determined based on actual clinical needs, and this application does not impose specific limitations on them.

[0127] For example, such as Figure 9 As shown, the planning image, fractionation image, and dose distribution data are input into the deformation registration model. The deformation registration model outputs a deformation field and obtains a new optimized treatment plan based on the deformation field. Through intelligent evaluation, including calculating the similarity (such as image matching degree and coverage conformity) between the deformation image and the fractionation image obtained based on the deformation field, as well as the quality of the optimized treatment plan, if the similarity and / or quality do not meet the standards, the planning image is updated to a deformation image, and the dose distribution data is updated to deformation dose distribution data obtained based on the deformation field and dose distribution data. These are then re-input into the deformation registration model along with the fractionation image to optimize the deformation field.

[0128] In some embodiments, the deformation registration model comprises a dose-aware network and a deformation registration network, the dose-aware network is configured to obtain the anatomical dose feature, and the deformation registration network is configured to perform deformation registration on the fraction image and the planning image to output the deformation field.

[0129] Correspondingly, as shown in Figure 10 The step S102 can be specifically implemented as steps S401-S402.

[0130] S401, input the dose distribution data corresponding to the initial treatment plan and the planning image into the dose-aware network to obtain the anatomical dose feature output by the dose-aware network.

[0131] The anatomical dose feature is used to reflect the correlation between the anatomical feature in the target image and the dose distribution, i.e., to reflect the mapping relationship between different positions of the anatomical structure in the planning image and the dose information.

[0132] In some embodiments, before inputting the dose distribution data corresponding to the initial treatment plan into the dose-aware network, the dose distribution data is preprocessed to obtain a dose weight matrix, and then the dose weight matrix and the planning image are input into the dose-aware network to obtain the anatomical dose feature output by the dose-aware network.

[0133] S402, input the anatomical dose feature, the feature map of the planning image, and the feature map of the fraction image into the deformation registration network, perform deformation registration on the fraction image and the planning image through the deformation registration network, and output the deformation field.

[0134] In some embodiments, the deformation registration model is obtained by training the dose-aware network and the deformation registration network based on the dose distribution consistency loss between the dose distribution data corresponding to the initial treatment plan and the deformation dose distribution data corresponding to the deformation field.

[0135] For example, the dose-aware network and the deformation registration network are trained based on the dose distribution consistency loss between the dose distribution data corresponding to the initial treatment plan and the deformation dose distribution data corresponding to the deformation field. The dose distribution consistency loss can be measured by calculating the mean square error (MSE) or the structural similarity index (SSIM) between the initial dose distribution and the deformation dose distribution, and the network parameters of the dose-aware network and the deformation registration network are updated by an optimization algorithm (such as gradient descent method) to minimize the dose distribution consistency loss.

[0136] This application embodiment is based on dose distribution consistency loss and simultaneously trains a dose-sensing network and a deformation registration network. On the one hand, the dose-sensing network is trained using dose distribution data to establish an accurate correlation between dose distribution and anatomical features. On the other hand, the deformation registration process is constrained using dose distribution data, enabling the deformation registration network to focus on high-dose target area features and protect the boundaries of sensitive organs. This provides feature guidance for deformation registration that conforms to both anatomical rules and clinical dosimetry requirements, forcing the dose-sensing network and the deformation registration network to maintain dosimetric fidelity, thereby improving the accuracy of deformation registration.

[0137] In some embodiments, the deformation registration model further includes a feature extraction network, which is used to extract features from the planned image and the segmented images, thereby obtaining feature maps of the planned image and the segmented images.

[0138] like Figure 11 As shown, before step S102, the method further includes:

[0139] S501. Input the planned image and the segmented images into the feature extraction network, extract features from the planned image and the segmented images through the feature extraction network, and output the feature map of the planned image and the feature map of the segmented image.

[0140] In some embodiments, the deformation registration model further includes a denoising network and a feature extraction network. The denoising network is used to denoise the segmented images and extract features to obtain feature maps of the segmented images. The feature extraction network is used to extract features of the planned images to obtain feature maps of the planned images.

[0141] In some embodiments, such as Figure 12 As shown, before step S102, the method further includes the following steps S601-S602:

[0142] S601. Input the segmented image into the denoising network, perform denoising processing on the segmented image through the denoising network, extract features from the denoised segmented image, and output the feature map of the segmented image.

[0143] In some embodiments, the denoising network can be constructed using a deep convolutional neural network (CNN) as the backbone, with feature extraction branches added at different depth locations.

[0144] For example, a U-Net architecture is used as the backbone of the denoising network, including an encoder and a decoder part. In the encoder part, the features of the sub-iteration images are extracted step by step through convolutional layers and pooling layers. The convolutional layers use small-sized convolutional kernels (such as 3x3 or 5x5) to perform convolution operations on the images to extract local features of the images. The pooling layers reduce the spatial resolution of the feature maps, reduce the amount of calculation and the number of parameters, and at the same time preserve important anatomical feature information. The decoder part gradually restores the spatial resolution of the sub-iteration images through deconvolutional layers or up-sampling layers, and outputs images after denoising processing with the same size as the original images. Then, feature extraction branches are added at different depth positions in the encoder part. These branches are used to output feature maps from different levels of convolutional layers. For example, branches are added after the 2nd, 3rd, and 4th layers of the encoder. Each branch can include one or more convolutional layers for further processing of the features extracted from the backbone network to generate feature maps at different levels. The feature maps output by each feature extraction branch can be fused by simply splicing or attention mechanism weighted fusion to splice the feature maps at different levels in the channel dimension to obtain a comprehensive feature map.

[0145] In some embodiments, the feature extraction network is based on a deep learning network, and the network architecture can include multiple convolutional layers, pooling layers, and fully connected layers, etc.

[0146] In some embodiments, the feature extraction network is based on a deep learning network, and the network architecture can include multiple convolutional layers, pooling layers, and fully connected layers, etc.

[0147] In some embodiments, after the planning image is input into the feature extraction network, the convolutional kernels in the convolutional layers of the feature extraction network perform convolution operations on the planning image to extract local features such as edges, lines, and local textures in the planning image. The pooling layers down-sample the feature maps output by the convolutional layers to reduce the spatial resolution of the feature maps while preserving the most important feature information. Then, the fully connected layers integrate the features extracted before to form higher-level semantic feature representations. Finally, the feature extraction network outputs the feature map of the planning image.

[0148] In some embodiments, the feature extraction network is trained and optimized based on a large amount of data sets, including original planning images and planning images with feature annotations, and has feature extraction capabilities for specific types or specific tasks, such as feature extraction capabilities for object outlines, textures, color distributions, etc. in images.

[0149] In some embodiments, the deformation registration model is obtained based on a dose distribution consistency loss between the dose distribution data corresponding to the initial treatment plan and the deformed dose distribution data corresponding to the deformation field, while training the dose perception network and the deformation registration network, and based on a similarity loss between the fraction image and the deformed image corresponding to the deformation field, while training the denoising network and the deformation registration network.

[0150] For example, the dose perception network and the deformation registration network are trained based on a dose distribution consistency loss between the dose distribution data corresponding to the initial treatment plan and the deformed dose distribution data corresponding to the deformation field. The dose distribution consistency loss can be measured by calculating the mean square error (MSE) or the structural similarity index (SSIM) between the initial dose distribution and the deformed dose distribution, and the network parameters of the dose perception network and the deformation registration network are updated by an optimization algorithm (such as gradient descent method) to minimize the dose distribution consistency loss.

[0151] For example, the denoising network and the deformation registration network are trained based on a similarity loss between the fraction image and the deformed image corresponding to the deformation field. The similarity loss can be measured by calculating the mutual information (MI), normalized mutual information (NMI) or Pearson correlation coefficient between the fraction image and the deformed image, and the network parameters of the denoising network and the deformation registration network are updated by an optimization algorithm to minimize the similarity loss.

[0152] In the embodiments of the present application, the dose perception network and the deformation registration network are trained based on the dose distribution consistency loss, which forces the dose perception network and the deformation registration network to maintain dose fidelity. At the same time, the denoising network and the deformation registration network are trained based on the similarity loss between the fraction image and the deformed image, which makes the denoising process retain the key geometric information required for registration, and at the same time, the registration requirement guides the feature extraction strategy of the denoising network, forming a virtuous cycle of denoising quality and registration accuracy, and fundamentally solving the problem of error accumulation caused by inconsistent features in traditional denoising and registration step-by-step processing.

[0153] The training process of the deformation registration model is introduced below through a complete embodiment, such as Figure 13As shown, the deformation registration model includes a dose-aware network, a feature extraction network, a denoising network, and a deformation registration network. Before the dose distribution data, the planning image, and the fraction image are input into the deformation registration model, the dose distribution data, the planning image, and the fraction image are respectively preprocessed to obtain a dose weight matrix, a preprocessed planning image, and a preprocessed fraction image. Then, the preprocessed data are input into the deformation registration model. The deformation registration model inputs the dose weight matrix and the planning image into the dose-aware network to obtain an anatomical dose feature map, inputs the planning image into the feature extraction network to obtain a feature map of the planning image, and inputs the fraction image into the denoising network to obtain a feature map of the fraction image. Then, the three kinds of feature maps are input into the deformation registration network to output a deformation field (a displacement field). The deformation field is subjected to dose constraint loss and regularization loss operations to make the deformation field transition smooth. Then, the processed deformation field is input into a spatial conversion network to obtain a deformed dose weight matrix and a deformed image. The dose distribution consistency loss of the deformed dose weight matrix and the dose weight matrix is calculated. According to the calculation result, the parameters of the dose-aware network and the deformation registration network are adjusted. Meanwhile, the similarity loss of the deformed image and the fraction image is calculated. According to the calculation result, the parameters of the denoising network and the deformation registration network are adjusted. The denoising loss of the fraction image feature map and the fraction image is calculated. According to the calculation result, the parameters of the denoising network are adjusted. Iterative optimization is performed to minimize the dose distribution consistency loss, the similarity loss, and the denoising loss, so as to complete the training of the deformation registration model.

[0154] In some embodiments, the present application also provides a self-adaptive treatment planning method, as shown in Figure 14 As shown, the self-adaptive treatment planning method includes the following steps:

[0155] S701, obtaining a planning image, a fraction image, and an initial treatment plan of a patient.

[0156] In some embodiments, the planning image and the initial treatment plan of the patient can be called from a tumor information management system of a hospital, or directly obtained from a medical imaging device or a treatment planning system.

[0157] In some embodiments, the fraction image is obtained in real time from a radiotherapy device.

[0158] S702, performing deformation registration on the fraction image and the planning image by using a deformation registration method to obtain a deformation field.

[0159] In some embodiments, the specific implementation of the deformation registration method is as described in the foregoing steps S101-S102, which will not be repeated here.

[0160] S703, optimizing the initial treatment plan based on the deformation field to obtain a self-adaptive treatment plan.

[0161] In some embodiments, the initial treatment plan parameters are mapped onto the fraction images using the deformation field by an interpolation algorithm, such as a trilinear interpolation method, to obtain an adaptive treatment plan.

[0162] The above mainly introduces the scheme provided by the embodiments of the application from the perspective of method. In order to implement the above functions, the deformation registration device includes hardware structures and / or software modules corresponding to the execution of each function. Those skilled in the art should easily realize that, in combination with the units and algorithm steps of the examples described in the embodiments disclosed herein, the application can be implemented in the form of hardware or a combination of hardware and computer software. Whether a certain function is implemented in hardware or computer software driven hardware depends on the specific application and design constraints of the technical solution. The skilled person can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the application.

[0163] The embodiments of the application can divide the deformation registration device into functional modules according to the above method. For example, the deformation registration device can include functional modules corresponding to each function, or two or more functions can be integrated into one processing module. The integrated module can be implemented in the form of hardware or software functional module. It should be noted that the division of modules in the embodiments of the application is illustrative, and is only a logical functional division. Actual implementation can have another division method.

[0164] Figure 15 A structural schematic diagram of a deformation registration device provided by the embodiments of the application is shown in FIG. 6. Figure 15 The deformation registration device 600 includes an acquisition module 601 and a processing module 602.

[0165] The acquisition module 601 is configured to acquire the plan image, the fraction image and the dose distribution data corresponding to the initial treatment plan of the patient. The processing module 602 is configured to input the plan image, the fraction image and the dose distribution data corresponding to the initial treatment plan of the patient into a trained deformation registration model, perform deformation registration on the fraction image and the plan image, and output a deformation field. The trained deformation registration model is obtained based on the dose distribution consistency loss between the dose distribution data corresponding to the initial treatment plan and the deformation dose distribution data corresponding to the deformation field.

[0166] In one possible implementation, the processing module 602 is configured to generate a deformation image based on the deformation field and the plan image, and drive or terminate the iterative optimization of the deformation field by the trained deformation registration model based on the similarity between the deformation image and the fraction image.

[0167] In a possible implementation, the processing module 602 is configured to generate a deformation image based on the deformation field and the planning image; generate an optimized treatment plan based on the deformation field and the initial treatment plan; and drive or terminate the trained deformation registration model to iteratively optimize the deformation field based on similarity between the deformation image and the fraction image and quality of the optimized treatment plan.

[0168] In a possible implementation, the deformation registration model comprises a dose-aware network and a deformation registration network; and the processing module 602 is configured to input, into the dose-aware network, dose distribution data corresponding to the initial treatment plan and the planning image, to obtain an anatomic dose feature output by the dose-aware network; the anatomic dose feature is used to reflect a correlation between an anatomic feature in the target image and a dose distribution; and the processing module 602 is configured to input, into the deformation registration network, the anatomic dose feature, a feature map of the planning image, and a feature map of the fraction image, to perform deformation registration on the fraction image and the planning image by using the deformation registration network, and to output the deformation field.

[0169] In a possible implementation, the deformation registration model is obtained based on dose distribution consistency loss between dose distribution data corresponding to the initial treatment plan and deformation dose distribution data corresponding to the deformation field, and based on simultaneous training of the dose-aware network and the deformation registration network.

[0170] In a possible implementation, the deformation registration model further comprises a feature extraction network; and before the anatomic dose feature, the feature map of the planning image, and the feature map of the fraction image are input into the deformation registration network, the processing module 602 is configured to input the planning image and the fraction image into the feature extraction network, to perform feature extraction on the planning image and the fraction image by using the feature extraction network, and to output the feature map of the planning image and the feature map of the fraction image.

[0171] In a possible implementation, the deformation registration model further comprises a denoising network and a feature extraction network; and before the anatomic dose feature, the feature map of the planning image, and the feature map of the fraction image are input into the deformation registration network, the processing module 602 is configured to input the fraction image into the denoising network, to perform denoising processing on the fraction image by using the denoising network, to perform feature extraction on the fraction image after the denoising processing, and to output the feature map of the fraction image; and the processing module 602 is configured to input the planning image into the feature extraction network, to perform feature extraction on the planning image by using the feature extraction network, and to output the feature map of the planning image.

[0172] In a possible implementation, the deformation registration model is obtained based on dose distribution consistency loss between dose distribution data corresponding to the initial treatment plan and deformation dose distribution data corresponding to the deformation field, and based on simultaneous training of the dose-aware network and the deformation registration network, and based on similarity loss between the fraction image and a deformation image corresponding to the deformation field, and based on simultaneous training of the denoising network and the deformation registration network.

[0173] In a possible implementation, the processing module 602 is configured to input the deformation field and the planning image into a space conversion network, and perform space conversion on the planning image based on the deformation field through the space conversion network to obtain a deformation image.

[0174] Figure 16 A structural schematic diagram of an electronic device is provided in an embodiment of the present application. As shown in the Figure 16 The electronic device 700 includes, but is not limited to, a processor 701 and a memory 702.

[0175] The memory 702 is configured to store executable instructions of the processor 701. It can be understood that the processor 701 is configured to execute the instructions to implement the method in the above embodiment.

[0176] It should be noted that those skilled in the art can understand, Figure 15 The electronic device structure shown in the above embodiment does not constitute a limitation on the electronic device. The electronic device can include more or fewer components than those shown in the Figure 16 or combine certain components, or different component arrangements.

[0177] The processor 701 is the control center of the electronic device, connects all parts of the electronic device through various interfaces and lines, executes software programs and / or modules stored in the memory 702 and data stored in the memory 702, processes data, and thus monitors the whole electronic device. The processor 701 can include one or more processing units. Optionally, the processor 701 can integrate an application processor and a modem processor, wherein the application processor mainly processes the operating system, user interface, and application programs, and the modem processor mainly processes wireless communication. It can be understood that the modem processor can also not be integrated into the processor 701.

[0178] The memory 702 can be used to store software programs and various data. The memory 702 can mainly include a program storage area and a data storage area, wherein the program storage area can store the operating system, application programs (such as determination units, processing units, etc.) required by at least one functional module, etc. In addition, the memory 702 can include a high-speed random access memory, and can also include a non-volatile memory, for example, at least one magnetic disk storage device, a flash memory device, or other volatile solid-state memory device.

[0179] In an exemplary embodiment, a computer-readable storage medium including instructions is also provided, for example, the memory 702 including instructions, which can be executed by the processor 701 of the electronic device 700 to implement the method in the above embodiment.

[0180] In actual implementation,Figure 15 The acquisition module 601 and the processing module 602 in the method can be implemented by the processor 701 in the device. Figure 16 The processor 701 in the device can invoke the computer program stored in the memory 702 to implement. The specific implementation process can refer to the description of the method part in the above embodiment, and will not be described here.

[0181] Optionally, the computer readable storage medium can be a non-transitory computer readable storage medium, for example, the non-transitory computer readable storage medium can be a read-only memory (Read-Only Memory, ROM), a random access memory (Random Access Memory, RAM), a CD-ROM, a magnetic tape, a floppy disk and an optical data storage device, etc.

[0182] In the example embodiment, the embodiment of the application also provides a computer program product including one or more instructions, which can be executed by the processor 701 of the electronic device to complete the method in the above embodiment.

[0183] It should be noted that the instructions in the above computer readable storage medium or one or more instructions in the computer program product are executed by the processor of the electronic device to realize each process of the above method embodiment, and the same technical effect as the above method can be achieved. To avoid repetition, it will not be described here.

[0184] Through the description of the above embodiments, those skilled in the art can clearly understand that, for the convenience and brevity of description, only the above division of each functional module is taken as an example for illustration, and in actual application, the above functions can be completed by different functional modules according to needs, that is, the internal structure of the device is divided into different functional modules to complete the above described full classification part or part of the function.

[0185] In several embodiments provided in the application, it should be understood that the disclosed device and method can be implemented by other ways. For example, the above described device embodiment is only schematic, for example, the division of the module or unit is only a logical function division, and actual implementation can have another division manner, for example, a plurality of units or components can be combined or integrated into another device, or some features can be ignored or not executed. Another point, the coupling or direct coupling or communication connection between the displayed or discussed each other can be through some interface, indirect coupling or communication connection between the devices or units, which can be electrical, mechanical or other forms.

[0186] The units described as separate components may or may not be physically separate, and the components displayed as units may be a physical unit or multiple physical units, that is, may be located in one place, or also can be distributed to multiple different places. Part or all of the classified units can be selected according to actual needs to achieve the purpose of the embodiment of the present application.

[0187] In addition, each functional unit in each embodiment of the present application can be integrated in one processing unit, or each unit can be physically present alone, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware or in the form of a software functional unit.

[0188] The integrated unit, if realized in the form of a software functional unit and sold or used as an independent product, can be stored in a readable storage medium. Based on such understanding, the technical solutions of the embodiments of the present application essentially or the part of the prior art that contributes to the technical solutions or the whole classification or part of the technical solutions can be embodied in the form of a software product. The software product is stored in a storage medium, including a plurality of instructions for causing an apparatus (which can be a single-chip microcomputer, a chip, etc.) or a processor to execute all or part of the steps of the method of each embodiment of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a ROM, a RAM, a magnetic disk or an optical disk, and various program code storage media.

[0189] In the description of the embodiments of the present application, specific features, structures, materials or characteristics can be combined in any one or more embodiments or examples in a suitable manner.

[0190] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of changes or replacements within the technical scope disclosed in the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A deformation registration method, characterized in that, The method includes: Acquire the patient's treatment plan images, fractionation images, and dose distribution data corresponding to the initial treatment plan; The patient's planned image, fractionated image, and dose distribution data corresponding to the initial treatment plan are input into a trained deformation registration model. Deformation registration is performed on the fractionated image and the planned image, and the deformation field is output. The trained deformation registration model is trained based on the dose distribution consistency loss between the dose distribution data corresponding to the initial treatment plan and the deformation dose distribution data corresponding to the deformation field.

2. The method according to claim 1, characterized in that, The method further includes: Based on the deformation field and the planned image, a deformation image is generated; Based on the similarity between the deformed image and the segmented image, the trained deformation registration model is driven or terminated to iteratively optimize the deformation field.

3. The method according to claim 1, characterized in that, The method further includes: Based on the deformation field and the planned image, a deformation image is generated; Based on the deformation field and the initial treatment plan, an optimized treatment plan is generated; Based on the similarity between the deformed image and the fractional image and the quality of the optimized treatment plan, the trained deformation registration model is driven or terminated to iteratively optimize the deformation field.

4. The method according to claim 1, characterized in that, The deformation registration model includes a dose-sensing network and a deformation registration network; the step of inputting the patient's planned image, fractionated image, and dose distribution data corresponding to the initial treatment plan into the trained deformation registration model, performing deformation registration on the fractionated image and the planned image, and outputting a deformation field includes: The dose distribution data corresponding to the initial treatment plan and the plan image are input into the dose-sensing network to obtain the anatomical dose features output by the dose-sensing network; the anatomical dose features are used to reflect the correlation between anatomical features and dose distribution in the target image; The anatomical dose characteristics, the feature map of the planning image, and the feature map of the fractionated image are input into the deformation registration network. The deformation registration network performs deformation registration on the fractionated image and the planning image and outputs the deformation field.

5. The method according to claim 4, characterized in that, The deformation registration model is obtained by training the dose-sensing network and the deformation registration network simultaneously, based on the dose distribution consistency loss between the dose distribution data corresponding to the initial treatment plan and the deformation dose distribution data corresponding to the deformation field.

6. The method according to claim 4, characterized in that, The deformation registration model further includes a feature extraction network. Before inputting the anatomical dose features, the feature map of the planning image, and the feature map of the fractionation image into the deformation registration network, the method further includes: The planned image and the segmented images are input into the feature extraction network, and the feature extraction network extracts features from the planned image and the segmented images, outputting feature maps of the planned image and the segmented images.

7. The method according to claim 4, characterized in that, The deformation registration model further includes a denoising network and a feature extraction network; before inputting the anatomical dose features, the feature map of the planning image, and the feature map of the fractionation image into the deformation registration network, the method further includes: The segmented image is input into the denoising network, the segmented image is denoised by the denoising network, and features are extracted from the denoised segmented image to output the feature map of the segmented image. The planned image is input into the feature extraction network, which extracts features from the planned image and outputs a feature map of the planned image.

8. The method according to claim 7, characterized in that, The deformation registration model is obtained by training the dose-aware network and the deformation registration network simultaneously based on the dose distribution consistency loss between the dose distribution data corresponding to the initial treatment plan and the deformation dose distribution data corresponding to the deformation field, and by training the denoising network and the deformation registration network simultaneously based on the similarity loss between the fractional image and the deformation image corresponding to the deformation field.

9. The method according to claim 2 or 3, characterized in that, The step of generating a deformation image based on the deformation field and the planned image includes: The deformation field and the planned image are input into a spatial transformation network. The spatial transformation network performs spatial transformation on the planned image based on the deformation field to obtain the deformation image.

10. An adaptive treatment planning method, characterized in that, The method includes: Acquire the patient's planning images, fractionation images, and initial treatment plan; Using the deformation registration method as described in any one of claims 1 to 9, deformation registration is performed on the segmented image and the planned image to obtain a deformation field; Based on the deformation field, the initial treatment plan is optimized to obtain an adaptive treatment plan.

11. A deformation registration system, characterized in that, include: processor; A memory configured to store processor-executable instructions; The processor is configured to execute the instructions to implement the deformation registration method as described in any one of claims 1 to 9.

12. A treatment planning system, characterized in that, include: processor; A memory configured to store processor-executable instructions; The processor is configured to execute the instructions to implement the deformation registration method as described in any one of claims 1 to 9.

13. A non-volatile storage medium, characterized in that, The non-volatile storage medium stores computer instructions that, when executed on a computer, cause the computer to perform the deformation registration method as described in any one of claims 1 to 9, or the adaptive treatment planning method as described in claim 10.

Images