Deep learning image registration method, device, electronic device and storage medium

Abstract

The embodiment of the present invention discloses a deep learning image registration method, device, electronic device and storage medium, the method comprising: obtaining a deformation field of a first fixed image and a moving image through a first algorithm; obtaining a second fixed image according to the deformation field and the moving image; training a deep neural network according to the moving image, the second fixed image and the deformation field to obtain a trained deep neural network. The deep learning image registration method provided by the embodiment of the present invention generates a simulated deformation field by using a generation algorithm and is not manually labeled, so as to better train the registration network, solves the problem of relying on manually set registration metrics to define the registration standard and drive the registration process in the prior art, and realizes a greatly increased amount of training data in the neural network, further improving the registration accuracy and the effect of fast calculation speed.

Description

Technical Field

[0001] The embodiments of the present invention relate to image processing technology, and in particular to a deep learning image registration method, device, electronic device and storage medium. Background Art

[0002] Medical image registration in recent years can be divided into two categories: (1) using deep learning networks to estimate the similarity measure of two images and drive iterative optimization; (2) directly using deep regression networks to predict transformation parameters. The former only uses deep learning for similarity measurement and still requires traditional registration methods for iterative optimization. It does not fully utilize the advantages of deep learning, takes a long time, and is difficult to achieve real-time registration. According to the type of deep learning used, it can be divided into two categories: supervised learning-based registration and unsupervised learning-based registration. The input images are F and M (M has been affine aligned at this time), and CNN (convolutional neural network) is used to establish a function gθ(F,M), output the displacement deformation field DVF, and combine the spatial transformation to transform M into Then The similarity is measured with F and the loss is calculated. The existing unsupervised DNN (deep neural network) registration has the following main disadvantages: the network is directly trained by minimizing the loss (registration metric), and the loss is preset manually. The potential of deep neural networks is not fully utilized. Only the framework of deep neural networks is used, and neural networks are used as optimization tools. The existing registration adopts an iterative optimization method, and the registration speed is slow. Summary of the invention

[0003] The present invention provides a deep learning image registration method, device, electronic device and storage medium to greatly increase the amount of training data in a neural network, further improve the registration accuracy and increase the calculation speed.

[0004] In a first aspect, an embodiment of the present invention provides a deep learning image registration method, comprising:

[0005] Acquire the deformation field of the first fixed image and the moving image by using a first algorithm;

[0006] Acquire a second fixed image according to the deformation field and the moving image;

[0007] A deep neural network is trained according to the moving image, the second fixed image and the deformation field to obtain a trained deep neural network.

[0008] Optionally, the first algorithm includes:

[0009]

[0010] in Refers to any image indicator, I i For a 3D fixed image, I j is a 3D moving image, τ() is the regularity of the metric deformation field, λ is the weight factor, and D i j is the deformation field.

[0011] Optionally, training a deep neural network according to the moving image, the second fixed image, and the deformation field to obtain a trained deep neural network includes:

[0012] Generate a large number of simulated deformation fields according to the generation algorithm and the deformation field;

[0013] A large amount of training data is obtained according to the large amount of simulated deformation fields to train the deep neural network to obtain a trained deep neural network.

[0014] Optionally, generating a large number of simulated deformation fields according to the generation algorithm and the deformation field includes:

[0015]

[0016] D k is the simulation deformation field, I i For moving images, as a fixed image.

[0017] Optionally, training a deep neural network according to the moving image, the second fixed image, and the deformation field to obtain a trained deep neural network includes:

[0018] Taking the second fixed image and the moving image as input;

[0019] The deformation field is used as an output to train an untrained registration network to obtain a trained registration network.

[0020] Optionally, after using the deformation field as an output to train an untrained registration network to obtain a trained registration network, the method further includes:

[0021] Perform a registration test on the trained registration network.

[0022] Optionally, the performing a registration test on the trained registration network includes:

[0023] Randomly selecting image blocks of fixed sizes from the second fixed image and the moving image for multiple times and inputting them into the trained registration network to output deformation field image blocks of the same size;

[0024] The deformation fields are stacked in the voting and averaging manner to obtain a complete deformation field.

[0025] In a second aspect, an embodiment of the present invention further provides a deep learning image registration device, the device comprising:

[0026] An acquisition module, used for acquiring a deformation field of a first fixed image and a moving image by using a first algorithm;

[0027] A generating module, used for acquiring a second fixed image according to the deformation field and the moving image;

[0028] An output module is used to train a deep neural network according to the moving image, the second fixed image and the deformation field to obtain a trained deep neural network.

[0029] In a third aspect, an embodiment of the present invention further provides an electronic device, the electronic device comprising:

[0030] one or more processors;

[0031] a storage device for storing one or more programs,

[0032] When the one or more programs are executed by the one or more processors, the one or more processors implement any of the deep learning image registration methods described above.

[0033] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium on which a computer program is stored, wherein the computer program includes program instructions, which, when executed by a processor, implement a deep learning image registration method as described above.

[0034] The embodiment of the present invention discloses a deep learning image registration method, device, electronic device and storage medium, the method comprising: obtaining a deformation field of a first fixed image and a moving image through a first algorithm; obtaining a second fixed image according to the deformation field and the moving image; training a deep neural network according to the moving image, the second fixed image and the deformation field to obtain a trained deep neural network. The deep learning image registration method provided by the embodiment of the present invention generates a simulated deformation field by using a generation algorithm and is not manually labeled, so as to better train the registration network, solves the problem of relying on manually set registration metrics to define the registration standard and drive the registration process in the prior art, and realizes a greatly increased amount of training data in the neural network, further improving the registration accuracy and the effect of fast calculation speed. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 A schematic diagram of a process flow of a deep learning image registration method provided in Embodiment 1 of the present invention;

[0036] Figure 2 A schematic diagram of a flow chart of a deep learning image registration method provided in Embodiment 2 of the present invention;

[0037] Figure 3 A schematic diagram of the structure of a deep learning image registration device provided in Embodiment 3 of the present invention;

[0038] Figure 4 A schematic diagram of the structure of an electronic device provided in Embodiment 4 of the present invention. DETAILED DESCRIPTION

[0039] The present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. It is to be understood that the specific embodiments described herein are only used to explain the present invention, rather than to limit the present invention. It should also be noted that, for ease of description, only parts related to the present invention, rather than all structures, are shown in the accompanying drawings.

[0040] It should be mentioned before discussing the exemplary embodiments in more detail that some exemplary embodiments are described as processes or methods depicted as flow charts. Although the flow charts describe the steps as sequential processes, many of the steps therein can be implemented in parallel, concurrently or simultaneously. In addition, the order of the steps can be rearranged. The process can be terminated when its operation is completed, but can also have additional steps not included in the accompanying drawings. The process can correspond to a method, function, procedure, subroutine, subprogram, etc.

[0041] In addition, the terms "first", "second", etc. can be used in this article to describe various directions, actions, steps or elements, but these directions, actions, steps or elements are not limited by these terms. These terms are only used to distinguish a first direction, action, step or element from another direction, action, step or element. For example, without departing from the scope of the present application, the first module can be the second module, and similarly, the second module can be referred to as the first module. Both the first module and the second module are modules, but they are not the same module. The terms "first", "second", etc. cannot be understood as indicating or implying relative importance or implicitly indicating the number of indicated technical features. Thus, the features defined as "first" and "second" can expressly or implicitly include one or more of the features. In the description of the present invention, the meaning of "multiple" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0042] Embodiment 1

[0043] Figure 1 A flowchart of a deep learning image registration method provided by an embodiment of the present invention is provided. The deep learning image registration method provided by this embodiment is applicable to a deep neural network. The method can be executed by an electronic device. Specifically, the deep learning image registration method provided by this embodiment includes the following steps:

[0044] Step 100: Obtain deformation fields of a first fixed image and a moving image using a first algorithm.

[0045] In this embodiment, this embodiment is mainly applicable to the field of nuclear magnetic resonance imaging. Specifically, nuclear magnetic resonance imaging (NMRI), also known as spin imaging, also known as magnetic resonance imaging (MRI), is based on the principle of nuclear magnetic resonance. According to the different attenuation of the released energy in different structural environments inside the material, the electromagnetic waves emitted can be detected by applying an external gradient magnetic field to know the position and type of the atomic nuclei constituting the object, and the internal structure image of the object can be drawn accordingly. Applying this technology to the imaging of the internal structure of the human body will produce a revolutionary medical diagnostic tool. Nuclear magnetic resonance has been applied to imaging diagnosis of various systems throughout the body. The best effect is the brain, spinal cord, heart and blood vessels, joint bones, soft tissues and pelvic cavity. The application of rapidly changing gradient magnetic fields has greatly accelerated the speed of nuclear magnetic resonance imaging, making the application of this technology in clinical diagnosis and scientific research a reality, and greatly promoting the rapid development of medicine, neurophysiology and cognitive neuroscience. Image registration has many practical applications in medical image processing and analysis. With the advancement of medical imaging equipment, images containing accurate anatomical information such as CT and MRI can be collected for the same patient; at the same time, images containing functional information such as SPECT can also be collected. However, diagnosis by observing different images requires spatial imagination and the doctor's subjective experience. The use of the correct image registration method can accurately integrate a variety of information into the same image, allowing doctors to more conveniently and accurately observe lesions and structures from all angles. At the same time, by registering dynamic images collected at different times, the changes in lesions and organs can be quantitatively analyzed, making medical diagnosis, surgical planning, and radiotherapy planning more accurate and reliable.

[0046] In this embodiment, a 3D fixed image I is first obtained by obtaining a first algorithm. i and 3D Moving Image I j , the displacement deformation field (DVF) D is obtained by the traditional method i j Specifically, the first algorithm in this embodiment is

[0047]

[0048] in, Refers to any image indicator, I i For a 3D fixed image, I jis a 3D moving image, τ() is the regularity of the metric deformation field, γ is the weight factor, and D i j is the deformation field.

[0049] Step 110: Acquire a second fixed image according to the deformation field and the moving image.

[0050] In this embodiment, the deformation field obtained in step 100 is applied to the moving image to obtain a second fixed image. The second fixed image and the moving image are perfectly matched with the deformation field. By inputting this group of data into an untrained neural network, accurate results can be obtained, which meets the expected effect.

[0051] Step 120: Train a deep neural network according to the moving image, the second fixed image, and the deformation field to obtain a trained deep neural network.

[0052] In this embodiment, a complete deformation field of the image is obtained according to the moving image, the second fixed image and the deformation field and a pre-trained deep neural network in step 110. Specifically, training the deep neural network according to the moving image, the second fixed image and the deformation field to obtain the trained deep neural network includes: generating a large number of simulated deformation fields according to a generation algorithm and the deformation field; obtaining a large number of training data according to the large number of simulated deformation fields to train the deep neural network to obtain the trained deep neural network.

[0053] In this embodiment, specifically, using the D obtained in step 100 i j Train the generative network to generate D k .

[0054] Definable

[0055]

[0056] D k is the simulation deformation field, I i For moving images, As a fixed image, D k It's I i As a moving image The real displacement deformation field (DVF) as a fixed image transformation.

[0057] In this embodiment, a large amount of training data is obtained based on a large number of simulated deformation fields. For example, a fixed image is obtained by applying a simulated deformation field to a moving image. For example, if there are 40 moving images and simulated deformation fields, 1,600 sets of training data can be obtained to train the deep neural network to obtain a trained deep neural network.

[0058] This embodiment discloses a deep learning image registration method, the method comprising: obtaining a deformation field of a first fixed image and a moving image through a first algorithm; obtaining a second fixed image according to the deformation field and the moving image; training a deep neural network according to the moving image, the second fixed image and the deformation field to obtain a trained deep neural network. A deep learning image registration method provided by an embodiment of the present invention generates a simulated deformation field by using a generation algorithm and is not manually labeled, so as to better train the registration network, solve the problem of relying on manually set registration metrics in the prior art to define the registration standard and drive the registration process, and realizes a greatly increased amount of training data in the neural network, further improving the registration accuracy and the effect of fast calculation speed.

[0059] Embodiment 2

[0060] Figure 2 A method flow chart of a deep learning image registration method provided in Embodiment 2 of the present invention. The deep learning image registration method provided in this embodiment is applicable to a deep neural network. This embodiment is expanded on the basis of Embodiment 1, and specifically includes the following steps:

[0061] Step 200: Obtain deformation fields of a first fixed image and a moving image using a first algorithm.

[0062] Step 210: Acquire a second fixed image according to the deformation field and the moving image.

[0063] Step 220: Take the second fixed image and the moving image as input.

[0064] Step 230: Use the deformation field as output to train an untrained registration network to obtain a trained registration network.

[0065] In this embodiment, the deep neural network is a registration network, and an untrained registration network is pre-trained through various parameter sets to obtain a trained registration network.

[0066] In this embodiment, As input, D k As output, randomly select image patches to train DNN (deep neural network). The solution of this embodiment is also applicable to multimodal registration. Specifically, in some public data sets, multimodal images of the same subject are included. i , there will be an image J i can be almost aligned with it, because they are acquired in almost the same conditions but in different orders. i is the MR image at T1, Image J i It is the MR image at T2.

[0067] Same as (2), we can define

[0068]

[0069] at this time Can be used as a set of training data.

[0070] By using a deep neural network as a registration network and also for generating real data, the advantages of deep neural networks are fully utilized. One of the biggest advantages of deep neural networks is that they discover image features through learning. The method of this patent fully utilizes this feature of deep neural networks and learns the features and patterns of image matching based on perfectly matched image pairs and the DVF between them.

[0071] Step 240: Perform a registration test on the trained registration network.

[0072] In this embodiment, step 240 further includes:

[0073] Step 241 : randomly selecting image blocks of fixed sizes from the second fixed image and the moving image for multiple times and inputting them into the trained registration network to output deformation field image blocks of the same size.

[0074] Step 242: stack the deformation fields by adopting the voting and averaging method to obtain a complete deformation field.

[0075] In this embodiment, the DVFs predicted by the network are all patches. When restoring the original image size, the average is taken when stacking. The definition of patch domain is k=1,...,R. After images I and J are input into the trained network model, patch U is obtained respectively. k and V k , thus obtaining the DVF patch W k To get a complete DVF, use a voting method when stacking patches:

[0076]

[0077]

[0078] Where K represents the number of patch overlaps. Finally, the DVF of image I and image J can be obtained from the following formula:

[0079]

[0080] This embodiment discloses a deep learning image registration method, the method comprising: obtaining a deformation field of a first fixed image and a moving image through a first algorithm; obtaining a second fixed image according to the deformation field and the moving image; training a deep neural network according to the moving image, the second fixed image and the deformation field to obtain a trained deep neural network. A deep learning image registration method provided by an embodiment of the present invention generates a simulated deformation field by using a generation algorithm and is not manually labeled, so as to better train the registration network, solve the problem of relying on manually set registration metrics in the prior art to define the registration standard and drive the registration process, and realizes a greatly increased amount of training data in the neural network, further improving the registration accuracy and the effect of fast calculation speed.

[0081] Embodiment 3

[0082] The deep learning image registration device of the embodiment of the present invention can implement the deep learning image registration method provided by any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method. Figure 3 3 is a schematic diagram of a deep learning image registration device 300 in an embodiment of the present invention. Figure 3 , the deep learning image registration device 300 provided in the embodiment of the present invention may specifically include:

[0083] An acquisition module 300, configured to acquire a deformation field of a first fixed image and a moving image by using a first algorithm;

[0084] A generating module 310, configured to acquire a second fixed image according to the deformation field and the moving image;

[0085] The output module 320 is used to train a deep neural network according to the moving image, the second fixed image and the deformation field to obtain a trained deep neural network.

[0086] Optionally, the first algorithm includes:

[0087]

[0088] in Refers to any image indicator, I i For a 3D fixed image, I j is a 3D moving image, τ() is the regularity of the metric deformation field, λ is the weight factor, and D i j is the deformation field.

[0089] Optionally, training a deep neural network according to the moving image, the second fixed image, and the deformation field to obtain a trained deep neural network includes:

[0090] Generate a large number of simulated deformation fields according to the generation algorithm and the deformation field;

[0091] A large amount of training data is obtained according to the large amount of simulated deformation fields to train the deep neural network to obtain a trained deep neural network.

[0092] Optionally, generating a large number of simulated deformation fields according to the generation algorithm and the deformation field includes:

[0093]

[0094] D k is the simulation deformation field, I i For moving images, as a fixed image.

[0095] Optionally, training a deep neural network according to the moving image, the second fixed image, and the deformation field to obtain a trained deep neural network includes:

[0096] Taking the second fixed image and the moving image as input;

[0097] The deformation field is used as an output to train an untrained registration network to obtain a trained registration network.

[0098] Optionally, after using the deformation field as an output to train an untrained registration network to obtain a trained registration network, the method further includes:

[0099] Perform a registration test on the trained registration network.

[0100] Optionally, the performing a registration test on the trained registration network includes:

[0101] Randomly selecting image blocks of fixed sizes from the second fixed image and the moving image for multiple times and inputting them into the trained registration network to output deformation field image blocks of the same size;

[0102] The deformation fields are stacked in the voting and averaging manner to obtain a complete deformation field.

[0103] The present embodiment discloses a deep learning image registration device, which includes: an acquisition module for acquiring a deformation field of a first fixed image and a moving image through a first algorithm; a generation module for acquiring a second fixed image according to the deformation field and the moving image; and an output module for training a deep neural network according to the moving image, the second fixed image and the deformation field to obtain a trained deep neural network. A deep learning image registration method provided by an embodiment of the present invention generates a simulated deformation field by using a generation algorithm and is not manually labeled, so as to better train the registration network, solve the problem in the prior art of relying on manually set registration metrics to define the registration standard and drive the registration process, and achieves a greatly increased amount of training data in the neural network, further improving the registration accuracy and the effect of fast calculation speed.

[0104] Embodiment 4

[0105] Figure 4 A schematic diagram of the structure of an electronic device provided by an embodiment of the present invention is shown in FIG. Figure 4 As shown, the electronic device includes a memory 410 and a processor 420. The number of processors 420 in the device may be one or more. Figure 4 A processor 420 is taken as an example; the memory 410 and the processor 420 in the server can be connected via a bus or other means. Figure 4 The example of connecting through bus is taken in the following.

[0106] The memory 410, as a computer-readable storage medium, can be used to store software programs, computer executable programs and modules, such as program instructions / modules corresponding to the deep learning image registration method in the embodiment of the present invention (for example, the acquisition module 310, the generation module 320, and the output module 330 in the deep learning image registration device 300). The processor 420 executes various functional applications and data processing of the server / terminal / server by running the software programs, instructions and modules stored in the memory 410, thereby realizing the above-mentioned deep learning image registration method.

[0107] The processor 420 is used to run the computer program stored in the memory 410 to implement the following steps:

[0108] Acquire the deformation field of the first fixed image and the moving image by using a first algorithm;

[0109] Acquire a second fixed image according to the deformation field and the moving image;

[0110] A deep neural network is trained according to the moving image, the second fixed image and the deformation field to obtain a trained deep neural network.

[0111] In one embodiment, an electronic device provided by an embodiment of the present invention, whose computer program is not limited to the above method operations, can also perform related operations in the deep learning image registration method provided by any embodiment of the present invention.

[0112] The memory 410 may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system and at least one application required for a function; the data storage area may store data created according to the use of the terminal, etc. In addition, the memory 410 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some instances, the memory 410 may further include a memory remotely arranged relative to the processor 420, and these remote memories may be connected to the server / terminal / server via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0113] The present embodiment discloses a deep learning image registration electronic device for executing the following method: obtaining a deformation field of a first fixed image and a moving image through a first algorithm; obtaining a second fixed image according to the deformation field and the moving image; training a deep neural network according to the moving image, the second fixed image and the deformation field to obtain a trained deep neural network. A deep learning image registration method provided by an embodiment of the present invention generates a simulated deformation field by using a generation algorithm and is not manually labeled, so as to better train the registration network, solve the problem of relying on manually set registration metrics in the prior art to define the registration standard and drive the registration process, and realizes a greatly increased amount of training data in the neural network, further improving the registration accuracy and the effect of fast calculation speed.

[0114] Embodiment 5

[0115] Embodiment 5 of the present invention further provides a storage medium containing computer executable instructions, wherein the computer executable instructions are used to perform a deep learning image registration method when executed by a computer processor, the method comprising:

[0116] Acquire the deformation field of the first fixed image and the moving image by using a first algorithm;

[0117] Acquire a second fixed image according to the deformation field and the moving image;

[0118] A deep neural network is trained according to the moving image, the second fixed image and the deformation field to obtain a trained deep neural network.

[0119] Of course, the storage medium containing computer executable instructions provided by an embodiment of the present invention, whose computer executable instructions are not limited to the method operations described above, can also execute related operations in a deep learning image registration method provided by any embodiment of the present invention.

[0120] The computer-readable storage medium of the embodiment of the present invention can adopt any combination of one or more computer-readable media. The computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples (non-exhaustive list) of computer-readable storage media include: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this document, a computer-readable storage medium can be any tangible medium containing or storing a program, which can be used by an instruction execution system, device or device or used in combination with it.

[0121] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, which carry computer-readable program code. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer-readable signal media may also be any computer-readable medium other than a computer-readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0122] The program code contained on the storage medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

[0123] Computer program code for performing the operations of the present invention may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or terminal. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., through the Internet using an Internet service provider).

[0124] This embodiment discloses an electronic storage medium for deep learning image registration, which is used to perform the following method: obtaining a deformation field of a first fixed image and a moving image through a first algorithm; obtaining a second fixed image according to the deformation field and the moving image; training a deep neural network according to the moving image, the second fixed image and the deformation field to obtain a trained deep neural network. A deep learning image registration method provided by an embodiment of the present invention generates a simulated deformation field by using a generation algorithm and is not manually labeled, so as to better train the registration network, solve the problem of relying on manually set registration metrics in the prior art to define the registration standard and drive the registration process, and achieves a greatly increased amount of training data in the neural network, further improving the registration accuracy and the effect of fast calculation speed.

[0125] Note that the above are only preferred embodiments of the present invention and the technical principles used. Those skilled in the art will understand that the present invention is not limited to the specific embodiments described herein, and that various obvious changes, readjustments and substitutions can be made by those skilled in the art without departing from the scope of protection of the present invention. Therefore, although the present invention has been described in more detail through the above embodiments, the present invention is not limited to the above embodiments, and may include more other equivalent embodiments without departing from the concept of the present invention, and the scope of the present invention is determined by the scope of the appended claims.

Claims

1. A deep learning image registration method, characterized in that: include: Acquire the deformation field of the first fixed image and the moving image by using a first algorithm; Acquire a second fixed image according to the deformation field and the moving image; Generate a simulated deformation field according to the generation algorithm and the deformation field; Acquire training data according to the simulated deformation field to train a deep neural network to obtain a trained deep neural network; Wherein, the first algorithm includes: in, Refers to any image indicator, I i For a 3D fixed image, I j is a 3D moving image, τ() is the regularity of the metric deformation field, λ is the weight factor, and D i j is the deformation field; The generating a simulated deformation field according to the generating algorithm and the deformation field comprises: D k is the simulation deformation field, I i For moving images, as a fixed image.

2. A deep learning image registration method according to claim 1, characterized in that: Training a deep neural network according to the moving image, the second fixed image, and the deformation field to obtain a trained deep neural network includes: Taking the second fixed image and the moving image as input; The deformation field is used as an output to train an untrained registration network to obtain a trained registration network.

3. A deep learning image registration method according to claim 2, characterized in that: After using the deformation field as output to train an untrained registration network to obtain a trained registration network, the method further includes: Perform a registration test on the trained registration network.

4. A deep learning image registration method according to claim 3, characterized in that: The registration test of the trained registration network comprises: Randomly selecting image blocks of fixed sizes in the second fixed image and the moving image for multiple times and inputting them into the trained registration network to output deformation field image blocks of the same size; The deformation fields are stacked by voting and averaging to obtain a complete deformation field.

5. A deep learning image registration device, characterized in that: include: An acquisition module, used for acquiring a deformation field of a first fixed image and a moving image by using a first algorithm; A generating module, used for acquiring a second fixed image according to the deformation field and the moving image; An output module, used for generating a simulated deformation field according to a generation algorithm and the deformation field; Acquire training data according to the simulated deformation field to train a deep neural network to obtain a trained deep neural network; Wherein, the first algorithm includes: in, Refers to any image indicator, I i For a 3D fixed image, I j is a 3D moving image, τ() is the regularity of the metric deformation field, λ is the weight factor, and D i j is the deformation field; The output module is specifically used to generate a simulated deformation field according to the following algorithm: D k is the simulation deformation field, I i For moving images, as a fixed image.

6. An electronic device, characterized in that: The electronic device comprises: one or more processors; a storage device for storing one or more programs, When the one or more programs are executed by the one or more processors, the one or more processors implement the deep learning image registration method as described in any one of claims 1-4.

7. A computer-readable storage medium having a computer program stored thereon, the computer program comprising program instructions, characterized in that: When the program instructions are executed by a processor, the deep learning image registration method as described in any one of claims 1 to 4 is implemented.

Images