Image registration model training method, image registration method, device and medium

By training the image registration model and combining the deformation field, the accuracy problem of traditional medical image registration algorithms in soft tissues and organs with larger deformations is solved, and higher image registration accuracy is achieved.

CN114359356BActive Publication Date: 2024-11-26SHANGHAI UNITED IMAGING INTELLIGENCE CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202111625645.0
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2021-12-28
Publication Date
2024-11-26
Estimated Expiration
2041-12-28

AI Technical Summary

Technical Problem

Traditional medical imaging registration algorithms have poor registration results when dealing with soft tissues and organs with large deformations, making it difficult to meet high-precision requirements.

Method used

By obtaining the images to be registered and referenced, the image registration model is trained, the two deformation fields are output, and the deformation fields are combined to reduce deformation and improve registration accuracy. The U-shaped neural network structure is used to accelerate the training process.

Benefits of technology

By reducing the deformation size, the registration difficulty between large images is reduced and the accuracy of medical image registration is improved.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114359356B_ABST
    Figure CN114359356B_ABST
Patent Text Reader

Abstract

The present invention discloses a training method, an image registration method, a device and a medium for an image registration model, wherein the training method comprises: obtaining a training image to be registered and a reference training image corresponding to the training image to be registered; inputting the training image to be registered and the corresponding reference training image into the image registration model to be trained, outputting a first training deformation field corresponding to the training image to be registered and a second training deformation field corresponding to the corresponding reference training image; calculating a first loss between an image obtained after the training image to be registered is registered with the first training deformation field and an image obtained after the corresponding reference training image is registered with the second training deformation field; and training the image registration model according to the first loss. The present application simultaneously registers each pair of input images to a certain intermediate state, outputs two deformation fields and merges them into a total deformation field, thereby reducing the deformation size between the images to be registered, reducing the difficulty of registration between images with larger deformations, and improving the registration accuracy.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of medical image processing, and in particular to a training method for an image registration model, an image registration method, a device and a medium. Background Art

[0002] Medical image registration refers to seeking one or a series of spatial transformations for a medical image so that its corresponding points can be spatially consistent with those of another medical image. This consistency means that the same anatomical point on the human body has the same spatial position in two matching images. Medical image registration technology is the basis of medical image processing and plays a very important role in image information fusion, auxiliary diagnosis, surgical planning, surgical navigation, and basic medical theory research.

[0003] The most representative non-rigid registration method for elastic transformation of images among traditional methods is the Demons method, which is a registration method based on image grayscale. According to the theory of optical flow field, the images to be registered are regarded as two frames of a continuously moving image sequence, and then the size of the deformation field is calculated to achieve image registration. However, for most medical images, especially soft tissue organs, the deformation is large, and in most medical image registration applications, the registration requirements for soft tissue organs are high. In this case, the registration results of the Demons algorithm and most other traditional registration algorithms are poor. Summary of the invention

[0004] The technical problem to be solved by the present invention is to overcome the above-mentioned defects in the prior art and to provide a training method for an image registration model, an image registration method, a device and a medium.

[0005] The present invention solves the above technical problems through the following technical solutions:

[0006] A training method for an image registration model, the training method comprising:

[0007] Acquire a training image to be registered and a reference training image corresponding to the training image to be registered;

[0008] Inputting the training image to be registered and the corresponding reference training image into the image registration model to be trained, and outputting a first training deformation field corresponding to the training image to be registered and a second training deformation field corresponding to the corresponding reference training image;

[0009] Calculating a first loss between an image obtained after the training image to be registered is registered with the first training deformation field and an image obtained after the corresponding reference training image is registered with the second training deformation field;

[0010] The image registration model is trained according to the first loss.

[0011] Preferably, the training method further comprises:

[0012] Combining the first training deformation field and the second training deformation field to obtain a total training deformation field; the total training deformation field includes a first total training deformation field in which the training image to be registered is registered to the corresponding reference training image, and / or a second total training deformation field in which the corresponding reference training image is registered to the training image to be registered;

[0013] Calculating a second loss between an image obtained after the training image to be registered is registered with the first total training deformation field and the corresponding reference training image; and / or calculating a third loss between an image obtained after the corresponding reference training image is registered with the second total training deformation field and the training image to be registered;

[0014] The step of training the image registration model according to the first loss specifically includes:

[0015] The image registration model is trained according to the second loss and / or the third loss, and the first loss.

[0016] Preferably, the training method further comprises:

[0017] Calculating a first smoothing loss of the first training deformation field and a second smoothing loss of the second training deformation field;

[0018] The step of training the image registration model according to the first loss specifically includes:

[0019] The image registration model is trained according to the first loss, the first smoothness loss, and the second smoothness loss.

[0020] Preferably, the calculating of the first smooth loss of the first training deformation field and the second smooth loss of the second training deformation field specifically includes:

[0021] The first-order gradients are calculated for the first training deformation field and the second training deformation field respectively to obtain the first smooth loss and the second smooth loss.

[0022] Preferably, in the training method, the first loss, the second loss and / or the third loss are calculated based on mean square error loss and / or normalized cross correlation loss.

[0023] Preferably, the training method obtains the image registration model based on U-shaped neural network structure training.

[0024] An image registration method, comprising:

[0025] Acquire an image to be registered and a reference image corresponding to the image to be registered;

[0026] Inputting the image to be registered and the corresponding reference image into the image registration model trained by the above-mentioned image registration model training method, and outputting a first deformation field corresponding to the image to be registered and a second deformation field corresponding to the corresponding reference image;

[0027] Combining the first deformation field and the second deformation field to obtain a total deformation field;

[0028] The image to be registered is registered to a reference image corresponding to the image to be registered based on the total deformation field.

[0029] Preferably, the combining the first deformation field and the second deformation field to obtain a total deformation field specifically includes:

[0030] Inverting the second deformation field to obtain an inverse deformation field;

[0031] The first deformation field is converted into the coordinate system of the inverse deformation field and then superimposed with the inverse deformation field to obtain the total deformation field.

[0032] An electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned training method for the image registration model when executing the computer program;

[0033] and / or,

[0034] The processor implements the above-mentioned image registration method when executing the computer program.

[0035] A computer-readable medium having computer instructions stored thereon, wherein the computer instructions, when executed by a processor, implement the above-mentioned training method of the image registration model;

[0036] and / or,

[0037] The computer instructions implement the above-mentioned image registration method when executed by a processor.

[0038] The positive progressive effect of the present invention is that: the present application inputs each pair of images to be registered into the image registration model to be trained, simultaneously registers each pair of input images to a certain intermediate state, outputs two deformation fields, and then combines the two deformation fields as the total deformation field for registering the image to be registered to the corresponding reference image. The above-mentioned intermediate state refers to the middle position of the corresponding positions of each pair of input images. By reducing the size of the deformation between the images to be registered in disguise, the difficulty of registration between images with larger deformations is reduced, so as to improve the subsequent registration accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 The present invention is a flowchart of a method for training an image registration model according to an embodiment of the present invention.

[0040] Figure 2 The present invention is a flowchart of a preferred embodiment of the method for training the image registration model of the present invention.

[0041] Figure 3 The present invention is a flowchart of another preferred embodiment of the method for training the image registration model of the present invention.

[0042] Figure 4 The present invention is a flowchart of an image registration method according to another embodiment of the present invention.

[0043] Figure 5 It is a registration schematic diagram of a specific example of the image registration method of the present invention.

[0044] Figure 6 A structural schematic diagram of an electronic device for the image registration model training method or the image registration method of the present invention. DETAILED DESCRIPTION

[0045] The present invention is further described below by way of examples, but the present invention is not limited to the scope of the examples.

[0046] In order to overcome the above-mentioned defects that currently exist, the present embodiment provides a training method for an image registration model, the training method comprising: obtaining a training image to be registered and a reference training image corresponding to the training image to be registered; inputting the training image to be registered and the corresponding reference training image into the image registration model to be trained, and outputting a first training deformation field corresponding to the training image to be registered and a second training deformation field corresponding to the corresponding reference training image; calculating a first loss between an image obtained after the training image to be registered is registered with the first training deformation field and an image obtained after the corresponding reference training image is registered with the second training deformation field; and training the image registration model according to the first loss.

[0047] Based on the above-mentioned training method, this embodiment also provides an image registration method, including: acquiring an image to be registered and a reference image corresponding to the image to be registered; inputting the image to be registered and the corresponding reference image into an image registration model trained by the above-mentioned image registration model training method, and outputting a first deformation field corresponding to the image to be registered and a second deformation field corresponding to the corresponding reference image; merging the first deformation field and the second deformation field to obtain a total deformation field; and registering the image to be registered to the reference image corresponding to the image to be registered based on the total deformation field.

[0048] Specifically, as an implementation method, Figure 1 As shown, the training method of the image registration model provided in this embodiment specifically includes:

[0049] Step 11: Acquire a training image to be registered and a reference training image corresponding to the training image to be registered;

[0050] Step 12: input the training image to be registered and the corresponding reference training image into the image registration model to be trained, and output a first training deformation field corresponding to the training image to be registered and a second training deformation field corresponding to the corresponding reference training image;

[0051] Exemplarily, taking a set of images A and B as an example, in the process of registering image A to image B, B is the reference image and A is the floating image. A and B are input into the initial image registration model, and the spatial transformation relationship of image A registered to an intermediate form image and the spatial transformation relationship of image B registered to the same intermediate form image are output. The spatial transformation relationship is further reflected by the deformation field, that is, the first training deformation field f that reflects the coordinate correspondence between image A and the intermediate form image. A , and the second training deformation field f that reflects the coordinate correspondence between image B and the intermediate morphological image B .

[0052] Step 13, calculating a first loss between an image obtained after the training image to be registered is registered with the first training deformation field and an image obtained after the corresponding reference training image is registered with the second training deformation field;

[0053] Among them, the first training deformation field f A Apply it to image A to obtain the intermediate form Am after image A registration. Similarly, the second training deformation field f B Act on image B to obtain the intermediate state Bm after image B is registered. In this embodiment, the training goal is to make the intermediate state Am and the intermediate state Bm the same state, that is, Am is the same as Bm. During the training process, the registration accuracy loss of Am and Bm, that is, the first loss, is calculated. The first loss can be calculated based on the mean square error loss algorithm or the normalized cross correlation loss algorithm.

[0054] Step 14: Train the image registration model according to the first loss.

[0055] Among them, after obtaining the first loss, back propagation is performed to update the parameters of the image registration model, and then the training image is output again to the updated image registration model, and the final image registration model is obtained by iterative training. In addition, in the present application, the image registration model is preferably trained based on the U-type neural network structure. For example, the UNet network structure is adopted, wherein the entire network structure of UNet is divided into two parts: a compression path and a non-compression path, that is, the feature mapping of the medical image is reduced and expanded, and the low-level features of the reduced end are sent to the corresponding position of the enlarged end, so as to provide more effective information for the training of the model and accelerate the convergence of the model during the training process.

[0056] In the above implementation, during the training process, the two input images are simultaneously registered to a certain intermediate state, and the deformation field of image A registered to the intermediate state and the deformation field of image B registered to the intermediate state are obtained. The intermediate state refers to the intermediate position between the corresponding positions of the two input images. By reducing the size of the deformation between the images to be registered in disguise, the difficulty of registration between images with larger deformations is reduced, so as to improve the subsequent registration accuracy.

[0057] As a preferred implementation of the training method of the image registration model, Figure 2 As shown, before step 14, the training method further includes:

[0058] Step 131: merging the first training deformation field and the second training deformation field to obtain a total training deformation field; the total training deformation field includes a first total training deformation field in which the training image to be registered is registered to the corresponding reference training image, and / or a second total training deformation field in which the corresponding reference training image is registered to the training image to be registered;

[0059] Specifically, take the first training deformation field f A and the first training deformation field f B Merge to the first total training deformation field f for registering image A to image B AB For example:

[0060] First, the second training deformation field f B Inverse and get the second training f B The inverse deformation field f B -1 , the first training deformation field f A Transform to the inverse deformation field f B -1 In the spatial coordinate system, the deformation field obtained after the transformation is then superimposed on the inverse deformation field f B -1 The first total training deformation field f is obtained by registering image A to image B. AB .

[0061] Among them, the inverse deformation field fB -1 For the first training deformation field f A The specific calculation process of spatial transformation is as follows: First, the inverse deformation field f is calculated. B -1 The specific coordinate position L(x+dx,y+dy) at the position (x,y) where (dx,dy)=f B -1 (x, y). Since position L is likely to appear in the first training deformation field f A The first training deformation field f is calculated using the interpolation algorithm at the non-grid point A The value at position L is used as the first training deformation field f after geometric transformation A The value at the position (x, y) finally obtains the deformation field after the spatial geometric transformation.

[0062] Step 132: Calculate a second loss between an image obtained after the training image to be registered is registered with the first total training deformation field and the corresponding reference training image; and / or calculate a third loss between an image obtained after the corresponding reference training image is registered with the second total training deformation field and the training image to be registered;

[0063] Specifically, the second loss and the third loss can be calculated based on the mean square error loss algorithm or the normalized cross correlation loss algorithm to obtain the first total training deformation field f AB After that, the first total training deformation field f AB Apply it to image A to obtain the result image A' of A registered to image B, and calculate the mean square error L between B and A' BA ', and similarly the first training deformation field f A and the second training deformation field f B Merge into the second total training deformation field (f BA ), the second total training deformation field f BA Apply it to image B to obtain the result image B' of image B registered to image A, and calculate the mean square error L between A and B' AB ', during the training process, only one of the second loss and the third loss can be used for back propagation to update the parameters of the image registration model, or the two can be combined for reverse adjustment and numerical processing of the two, such as L BA 'With L AB The weighted sum of ' is taken as the total registration accuracy loss L between image A and image B AB .

[0064] Furthermore, step 14 specifically includes:

[0065] Step 141: train an image registration model according to the second loss and / or the third loss, and the first loss.

[0066] Among them, after obtaining the first, second and third losses, back propagation is performed to update the parameters of the image registration model, and then the training image is output again to the updated image registration model, and the final image registration model is obtained by iterative training.

[0067] In the above implementation, the first loss between the image obtained after the training image to be registered is registered with the first training deformation field and the image obtained after the corresponding reference training image is registered with the second training deformation field, the second loss between the training image to be registered and the corresponding reference training image, and the third loss between the reference training image and the training image to be registered are used as back-propagation data to further update the optimized parameters and improve the training accuracy of the image registration model.

[0068] As another preferred implementation of the training method of the image registration model, Figure 3 As shown, before step 14, the training method further includes:

[0069] Step 133, calculating a first smoothing loss of the first training deformation field and a second smoothing loss of the second training deformation field;

[0070] Specifically, in order to ensure the local smoothness of the deformation field, smoothness means that the topological structure of the image before and after registration remains unchanged, the method of the present invention constrains the deformation field and uses it as part of the loss function for gradient feedback to update the parameters of the image registration model. Preferably, considering the model training accuracy and efficiency, in this embodiment, the first-order gradient is calculated for the first training deformation field and the second training deformation field respectively to obtain the first smooth loss and the second smooth loss.

[0071] Furthermore, step 14 specifically includes:

[0072] Step 142: train the image registration model according to the first loss, the first smooth loss, and the second smooth loss.

[0073] Among them, after obtaining the first loss, the first smooth loss and the second smooth loss, the weighted sum of the two types of loss functions can be used as the final loss value, and then back propagation can be performed to update the parameters of the image registration model, and then the training image is output again to the updated image registration model, and the final image registration model is obtained by iterative training.

[0074] In addition, it should be noted that in the process of back-propagation to update the parameters of the image registration model, the present application can also perform the weighted sum of the above-mentioned first, second, and third losses and the first and second smoothing losses as the final loss value, and then perform back-propagation to update the parameters of the image registration model, and then output the training image again to the updated image registration model, and iterate the training to obtain the final image registration model.

[0075] In the above implementation, the first loss between the image obtained after the training image to be registered is registered with the first training deformation field and the image obtained after the corresponding reference training image is registered with the second training deformation field, the first smoothing loss of the first training deformation field, and the second smoothing loss of the second training deformation field are used as back-propagation data to further update the optimized parameters and improve the training accuracy of the image registration model.

[0076] This embodiment also provides another implementation method. Figure 4 As shown, the image registration method provided in this embodiment specifically includes:

[0077] Step 21, obtaining the image to be registered and a reference image corresponding to the image to be registered;

[0078] Step 22: input the image to be registered and the corresponding reference image into the image registration model trained by the above image registration model training method, and output a first deformation field corresponding to the image to be registered and a second deformation field corresponding to the corresponding reference image;

[0079] Step 23, combining the first deformation field and the second deformation field to obtain a total deformation field;

[0080] Specifically, the second deformation field is inverted to obtain the inverse deformation field; the first deformation field is converted to the coordinate system of the inverse deformation field and then superimposed with the inverse deformation field to obtain the total deformation field. The specific merging method can refer to the above-mentioned process of merging the first training deformation field and the second training deformation field to obtain the total training deformation field.

[0081] Step 24: register the image to be registered to the reference image corresponding to the image to be registered based on the total deformation field.

[0082] See also Figure 5, providing a specific example, the breathing state difference between the reference image a and the floating image b is relatively large, wherein the reference image a is in the exhalation state and the floating image b is in the inhalation state, and it is difficult to directly perform deformation registration. The intermediate state c is the breathing state between the reference image a and the floating image b (you can refer to the position of the crosshairs), and the breathing state of the registration result image d is consistent with the reference image. It should be noted that the intermediate state c here is for the purpose of illustrating the registration method of this embodiment. In actual applications, it is not necessary to output the intermediate state. After inputting a pair of images, the corresponding registration result image will be directly output.

[0083] In the above implementation, an image registration model is obtained by training based on the training method of the image registration model. In practical applications, a pair of images to be registered are input into the above image registration model, and two deformation fields are output. Then, the two deformation fields are combined as the total deformation field for registering the image to be registered to the corresponding reference image. By reducing the size of the deformation between the images to be registered in disguised form, the difficulty of registration between images with larger deformations is reduced, thereby improving the subsequent registration accuracy.

[0084] Figure 6 This is a schematic diagram of the structure of an electronic device provided in this embodiment. The electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned training method of the image registration model when executing the computer program;

[0085] And / or, the processor implements the above-mentioned image registration method when executing the computer program. Figure 6 A block diagram of an exemplary electronic device 90 suitable for use in implementing embodiments of the present invention is shown. Figure 6 The electronic device 90 shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention.

[0086] like Figure 6 As shown, the electronic device 90 may be in the form of a general-purpose computing device, for example, it may be a server device. The components of the electronic device 90 may include, but are not limited to: at least one processor 91, at least one memory 92, and a bus 93 connecting different system components (including the memory 92 and the processor 91).

[0087] The bus 93 includes a data bus, an address bus, and a control bus.

[0088] The memory 92 may include a volatile memory, such as a random access memory (RAM) 921 and / or a cache memory 922 , and may further include a read-only memory (ROM) 923 .

[0089] The memory 92 may also include a program tool 925 having a set (at least one) of program modules 924, such program modules 924 including but not limited to: an operating system, one or more application programs, other program modules and program data, each of which or some combination may include the implementation of a network environment.

[0090] The processor 91 executes various functional applications and data processing by running computer programs stored in the memory 92 .

[0091] The electronic device 90 may also communicate with one or more external devices 94 (e.g., keyboards, pointing devices, etc.). Such communication may be performed via an input / output (I / O) interface 95. Furthermore, the electronic device 90 may also communicate with one or more networks (e.g., a local area network (LAN), a wide area network (WAN), and / or a public network, such as the Internet) via a network adapter 96. The network adapter 96 communicates with other modules of the electronic device 90 via a bus 93. It should be understood that, although not shown in the figure, other hardware and / or software modules may be used in conjunction with the electronic device 90, including but not limited to: microcode, device drivers, redundant processors, external disk drive arrays, RAID (disk array) systems, tape drives, and data backup storage systems, etc.

[0092] It should be noted that although several units / modules or sub-units / modules of the electronic device are mentioned in the above detailed description, this division is merely exemplary and not mandatory. In fact, according to the embodiments of the present application, the features and functions of two or more units / modules described above can be embodied in one unit / module. Conversely, the features and functions of one unit / module described above can be further divided into multiple units / modules to be embodied.

[0093] This embodiment also provides a computer-readable medium having computer instructions stored thereon, wherein the computer instructions implement the above-mentioned image registration model training method when executed by a processor;

[0094] And / or, the computer instructions implement the above-mentioned image registration method when executed by the processor. The readable storage medium may include, but is not limited to, a portable disk, a hard disk, a random access memory, a read-only memory, an erasable programmable read-only memory, an optical storage device, a magnetic storage device, or any suitable combination thereof.

[0095] In a possible implementation, the present invention can also be implemented in the form of a program product, which includes a program code. When the program product is run on a terminal device, the program code is used to enable the terminal device to execute the training method for implementing the above-mentioned image registration model, and / or the image registration method.

[0096] The program code for executing the present invention may be written in any combination of one or more programming languages, and may be executed entirely on a user device, partially on a user device, as an independent software package, partially on a user device and partially on a remote device, or entirely on a remote device.

[0097] Although the specific embodiments of the present invention are described above, it should be understood by those skilled in the art that this is only for illustration and the protection scope of the present invention is defined by the appended claims. Those skilled in the art may make various changes or modifications to these embodiments without departing from the principles and essence of the present invention, but these changes and modifications all fall within the protection scope of the present invention.

Claims

1. A training method for an image registration model, It is characterized in that The training method comprises: Acquire a training image to be registered and a reference training image corresponding to the training image to be registered; Inputting the training image to be registered and the corresponding reference training image into the image registration model to be trained, and outputting a first training deformation field corresponding to the training image to be registered to an intermediate form image and a second training deformation field corresponding to the corresponding reference training image to the same intermediate form image; The deformation of the training image to be registered to the intermediate form image is smaller than the deformation of the training image to be registered directly to the reference training image; the deformation of the reference training image to be registered to the intermediate form image is smaller than the deformation of the training image to be registered directly to the reference training image; Calculating a first loss between an image obtained after the training image to be registered is registered with the first training deformation field and an image obtained after the corresponding reference training image is registered with the second training deformation field; The image registration model is trained according to the first loss.

2. The method for training the image registration model according to claim 1, It is characterized in that The training method further comprises: Combining the first training deformation field and the second training deformation field to obtain a total training deformation field; the total training deformation field includes a first total training deformation field in which the training image to be registered is registered to the corresponding reference training image, and / or a second total training deformation field in which the corresponding reference training image is registered to the training image to be registered; Calculating a second loss between an image obtained after the training image to be registered is registered with the first total training deformation field and the corresponding reference training image; and / or calculating a third loss between an image obtained after the corresponding reference training image is registered with the second total training deformation field and the training image to be registered; The step of training the image registration model according to the first loss specifically includes: The image registration model is trained according to the second loss and / or the third loss, and the first loss.

3. The method for training the image registration model according to claim 1, It is characterized in that The training method further comprises: Calculating a first smoothing loss of the first training deformation field and a second smoothing loss of the second training deformation field; The step of training the image registration model according to the first loss specifically includes: The image registration model is trained according to the first loss, the first smoothness loss, and the second smoothness loss.

4. The method for training the image registration model according to claim 3, It is characterized in that The calculating a first smooth loss of the first training deformation field and a second smooth loss of the second training deformation field specifically includes: The first-order gradients are calculated for the first training deformation field and the second training deformation field respectively to obtain the first smooth loss and the second smooth loss.

5. The method for training the image registration model according to claim 2, It is characterized in that In the training method, the first loss, the second loss and / or the third loss are calculated based on mean square error loss and / or normalized cross correlation loss.

6. The method for training an image registration model according to claim 1, It is characterized in that The training method obtains the image registration model based on U-shaped neural network structure training.

7. An image registration method, It is characterized in that include: Acquire an image to be registered and a reference image corresponding to the image to be registered; Inputting the image to be registered and the corresponding reference image into an image registration model trained by the training method of the image registration model according to any one of claims 1 to 6, and outputting a first deformation field corresponding to the image to be registered and a second deformation field corresponding to the corresponding reference image; Combining the first deformation field and the second deformation field to obtain a total deformation field; The image to be registered is registered to a reference image corresponding to the image to be registered based on the total deformation field.

8. The image registration method according to claim 7, It is characterized in that The combining the first deformation field and the second deformation field to obtain a total deformation field specifically includes: Inverting the second deformation field to obtain an inverse deformation field; The first deformation field is converted into the coordinate system of the inverse deformation field and then superimposed with the inverse deformation field to obtain the total deformation field.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, It is characterized in that When the processor executes the computer program, the training method of the image registration model according to any one of claims 1 to 6 is implemented; or, When the processor executes the computer program, the image registration method according to claim 7 or 8 is implemented.

10. A computer readable medium having computer instructions stored thereon, It is characterized in that The computer instructions, when executed by a processor, implement the training method of the image registration model as described in any one of claims 1 to 6; or, When the computer instructions are executed by a processor, the image registration method according to claim 7 or 8 is implemented.