Medical image morphing registration method and system thereof

Abstract

The application discloses a medical image morphing registration method and system, and belongs to the technical field of medical image processing. The application aims to solve the problems of insufficient precision, topological structure damage and limited cross-scale modeling capability of existing registration methods when processing large-scale non-rigid deformation. The proposed method combines the semantic perception ability of the diffusion model and the local evolution characteristics of the neural cellular automaton. By constructing a multi-scale image pyramid, a conditional diffusion model is introduced at each scale to extract potential semantic scores, guiding the NCA module to iteratively model local regions and generate a deformation field that maintains topological consistency. Finally, through the layer-by-layer progressive method between scales, a high-precision non-rigid registration result is obtained.

Description

Technical Field

[0001] This invention belongs to the field of medical image processing technology, specifically relating to a medical image deformation registration method and system. Background Technology

[0002] The main purpose of medical image registration is to establish pixel-level or voxel-level non-rigid correspondences between images of the same anatomical structure at different times, modalities, or viewpoints. It has important applications in clinical scenarios such as disease diagnosis, preoperative planning, intraoperative navigation, and efficacy evaluation. Especially in soft tissue structures, due to their non-rigid, non-linear, and highly deformable characteristics, accurately and stably restoring the correspondences between images is crucial for improving the effectiveness of clinical interventions.

[0003] Traditional non-rigid registration methods primarily rely on optimization-driven strategies, combining predefined similarity metrics with spatial transformation models to iteratively solve for the optimal deformation field. However, these traditional methods have certain drawbacks. Traditional optimization methods are highly dependent on the initial parameter settings, and when deformation is large or anatomical differences are significant, they are prone to converging to a non-optimal solution. Secondly, for high-resolution 3D medical images, traditional methods require repeated image interpolation and gradient calculations, resulting in long registration times that are difficult to meet the real-time and batch processing requirements of practical applications. Furthermore, most methods do not explicitly constrain the generated deformation field, easily leading to non-physical deformations such as folding and tearing, failing to achieve topological consistency, and reducing the clinical interpretability of the results.

[0004] To overcome the aforementioned limitations, deep learning methods have been widely adopted in medical image registration in recent years, gradually becoming a mainstream research direction. Typical methods such as VoxelMorph and TransMorph employ an end-to-end training model, learning a direct mapping from image pairs to deformation fields without requiring manually designed similarity metrics, thus possessing stronger nonlinear modeling capabilities and higher prediction efficiency. These methods are typically based on U-Net or its variants, utilizing an encoder-decoder structure to extract multi-scale semantic features and compensating for the loss of detail information through skip connections, achieving leading performance in multiple registration benchmarks.

[0005] Although deep learning registration models have improved efficiency and accuracy to some extent, the main technical problems are: current models mostly focus on image similarity optimization, ignoring the geometric consistency constraints of the deformation field, which can easily lead to irreversible deformation and damage to anatomical structures; and they lack the ability to model cross-level structural information in large-scale deformation scenes. Summary of the Invention

[0006] This invention provides a medical image deformation registration method and system, aiming to solve the problems of insufficient accuracy, topological structure destruction, and limited cross-scale modeling ability of existing registration methods when dealing with large-scale non-rigid deformations.

[0007] This invention is achieved through the following technical solution:

[0008] A medical image deformation registration method, the method comprising the following steps:

[0009] Step 1: Construct a multi-scale pyramid based on a set of three-dimensional medical images to be registered, wherein the set of three-dimensional medical images to be registered includes moving images and fixed images;

[0010] Step 2: Extract the semantic differences at the current scale from the multi-scale pyramid in Step 1, and output a latent feature score tensor through a multi-step backdiffusion process;

[0011] Step 3: Input the latent feature score tensor generated from the diffusion model in Step 2, the input image pair, and the deformation field output from the previous scale into the NCA module to perform local region deformation modeling, generate the deformation field, and output it.

[0012] Step 4: At each scale, apply the deformation field generated in Step 3 to the moving image to obtain the registered image at that scale, and calculate the image similarity loss between the registered image and the fixed image at that scale.

[0013] Step 5: Repeat steps 2 to 4 until the preset scale level is reached to complete the generation of the final deformation field, and apply the final deformation field to the moving image to obtain the final registered 3D medical image.

[0014] Furthermore, step 1 specifically involves obtaining moving images at different scales. and fixed image for:

[0015] (1)

[0016] Among them, scale level The coarsest scale, scale level For the finest scale, For the first Scale-level moving images, For the first Fixed image at scale level.

[0017] Furthermore, from the original resolution Begin layer-by-layer downsampling:

[0018] (2)

[0019] (3)

[0020] in, For activation function, This is a 3D convolution operation. In scale The convolution kernel for the moving image, The stride for the convolution operation is 2. For bias terms, For the first Moving images at a scale of -1 For the first -1 scale level fixed image;

[0021] By repeating formulas (2) and (3), a multi-scale image pyramid is constructed layer by layer, namely:

[0022] (4)

[0023] in, For the moving image at scale level 1, For fixed images at the first scale level, and so on, For the moving image at the L-th scale level, This is a fixed image at the L-th scale level.

[0024] Furthermore, step 2 specifically involves the latent feature score tensor. The formula is:

[0025] (5)

[0026] in, This serves as the semantic guidance basis for deformation modeling at this layer. As an independent conditional diffusion network, t For time steps.

[0027] Furthermore, in step 3, the NCA module performs local region deformation modeling on the latent feature score tensor generated from the diffusion model to generate a deformation field output.

[0028] Furthermore, in step 4, the spatial transformation process at each scale is abstracted into a formula:

[0029] (6)

[0030] in, Indicated in scale The image after registration, Indicated in scale Upward moving image, Indicated in scale The deformation field generated on it;

[0031] Next, the image similarity loss between the registered image and the fixed image at each scale is calculated, expressed by the formula:

[0032] (7)

[0033] Furthermore, the NCA module includes a learnable state iteration function. Perform on a discrete spatial grid Step-by-step evolution, the initial state tensor of each layer is defined as:

[0034] (8)

[0035] in, This is the latent feature score tensor for this layer. For the first Scale-level moving images, For the first Fixed images at different scale levels The deformation field output from the previous scale is upsampled to the current scale, and `concat` is used to concatenate them along the channel dimension to form the state tensor. .

[0036] Furthermore, state evolution is perceptually performed within a local neighborhood, recursively according to the following rules:

[0037] (9)

[0038] in, This represents a state iteration function whose parameters are learnable. Indicates the first Iterate up to the 1st level The state representation at each step, Indicates the first Iterate up to the 1st level State representation during a step;

[0039] Final deformation field From a stable state Decoded from Chinese, implemented using an MLP decoder:

[0040] (10)

[0041] Each scale layer independently generates a deformation field, which serves as the optimal estimate of the alignment for that layer and as the input to the conditional diffusion model for the next scale.

[0042] A medical image deformation registration system, the system using a medical image deformation registration method as described above, comprising:

[0043] Multi-scale pyramid building blocks: Based on a set of three-dimensional medical images to be registered, i.e., moving and stationary images, a multi-scale pyramid is constructed.

[0044] Latent Feature Score Tensor Generation Unit: Extracts the semantic differences at the current scale from the multi-scale pyramid obtained by the multi-scale pyramid construction unit, and outputs a latent feature score tensor through a multi-step back diffusion process;

[0045] Deformation field generation unit: It inputs the latent feature score tensor generated from the diffusion model, the input image pair, and the deformation field output from the previous scale into the NCA module to perform local region deformation modeling, generate the deformation field, and output it.

[0046] Similarity loss calculation unit: At each scale, the deformation field generated by the deformation field generation unit is applied to the moving image to obtain the registered image at that scale, and the image similarity loss between the registered image and the fixed image at that scale is calculated.

[0047] The medical image deformation registration unit repeats the working process of the latent feature score tensor generation unit, the deformation field generation unit, and the similarity loss calculation unit until the preset scale level is reached, to complete the generation of the final deformation field, and applies the final deformation field to the moving image to obtain the final registered 3D medical image.

[0048] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the above-described DiffuseNCA-based medical image deformation registration method.

[0049] The beneficial effects of this invention are:

[0050] This invention enables cross-scale registration modeling from coarse to fine and from global to local, effectively improving the accuracy and robustness of registration.

[0051] This invention employs a learnable state iteration update mechanism, enabling the registration process to adaptively adjust without relying on fixed optimization rules. This mechanism can dynamically learn and update strategies based on different anatomical structures, thereby improving the method's adaptability to complex scenarios and individual differences. Attached Figure Description

[0052] Figure 1 This is a flowchart of the method of the present invention.

[0053] Figure 2 This is a schematic diagram of the conditional diffusion network in this invention.

[0054] Figure 3 This is a flowchart of the NCA module in this invention.

[0055] Figure 4 This is a schematic diagram of the result of the first type of image of the present invention, wherein (a) is a moving image of the first type of image, (b) is a fixed image of the first type of image, (c) is a registered image of the first type of image, and (d) is a deformed long image of the first type of image.

[0056] Figure 5 This is a schematic diagram of the result of the second type of image of the present invention, wherein (a) is the moving image of the second type of image, (b) is the fixed image of the second type of image, (c) is the registered image of the second type of image, and (d) is the deformed long image of the second type of image. Detailed Implementation

[0057] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods are omitted so as not to obscure the description of this application with unnecessary detail.

[0058] It should be understood that, when used in this specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0059] It should also be understood that the terminology used in this application specification is for the purpose of describing particular embodiments only and is not intended to limit the application. As used in this application specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0060] The following is in conjunction with the appendix to this application specification. Figure 1-5 The technical solutions in the embodiments of this application are clearly and completely described. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0061] Many specific details are set forth in the following description in order to provide a full understanding of this application. However, this application may also be implemented in other ways different from those described herein. Those skilled in the art can make similar extensions without departing from the spirit of this application. Therefore, this application is not limited to the specific embodiments disclosed below.

[0062] Implementation Method 1

[0063] This embodiment provides a medical image deformation registration method, specifically a medical image deformation registration method based on DiffuseNCA. The method of this embodiment combines a diffusion model and a neural cell automata (NCA) to improve registration accuracy while effectively maintaining the topological consistency of the deformation field.

[0064] This invention aims to achieve high-precision registration of soft tissue structures in medical images. Combining the high-quality generation capabilities of diffusion models with the local adaptive evolution characteristics of neural automata, a soft tissue registration method integrating diffusion models and multi-scale structures of neural automata is designed. This method enables cross-scale registration modeling from coarse to fine and from global to local, effectively improving the accuracy and robustness of registration.

[0065] Fixed and Moving Images: In medical image registration, the fixed image serves as a reference image, its spatial position remaining unchanged, used to define the target coordinate system. The moving image, on the other hand, is the image that needs to be mapped to the coordinate system of the fixed image through spatial transformation. The core task of registration is to estimate the optimal spatial mapping, ensuring that the moving image achieves maximum consistency with the fixed image at the anatomical or structural level, thereby providing a unified spatial framework for subsequent analysis and processing.

[0066] Deformation field: A mapping function used to describe the spatial correspondence between images. It defines the corresponding position of each voxel point in the fixed image coordinate system in the moving image. In other words, the deformation field assigns a displacement vector to each point in the fixed image, indicating how that point should translate or deform in the moving image to achieve registration. Through the deformation field, non-rigid registration can be achieved, finely aligning complex anatomical differences, thus ensuring consistency between images across time or modalities in the same spatial coordinate system.

[0067] Topological consistency refers to maintaining the topological properties of anatomical structures in an image without disruption during registration. The deformable mapping after registration should ensure that the relative connectivity between structures, the closure of boundaries, and the adjacency relationships of regions remain consistent with the original anatomical space.

[0068] See appendix Figure 1 A medical image deformation registration method, the method comprising the following steps:

[0069] Step 1: Construct a multi-scale pyramid based on a set of 3D medical images to be registered, wherein the set of 3D medical images to be registered includes moving images. and fixed image ;

[0070] Step 2: As Figure 2 As shown, for the multi-scale pyramid in step 1, an independent conditional diffusion network is introduced at each scale. It is used to extract semantic differences at the current scale. This conditional diffusion network accepts image pairs. and random noise disturbance time step As input, a latent feature score tensor is output through a multi-step backdiffusion process. ;

[0071] Step 3: As Figure 3 As shown, the latent feature score tensor generated from the diffusion model in step 2, the input image pair, and the deformation field output from the previous scale are input into the NCA module to perform local region deformation modeling, generate the deformation field, and output it.

[0072] Step 4: At each scale, apply the deformation field generated in Step 3 to the moving image to obtain the registered image at that scale, and calculate the image similarity loss between the registered image and the fixed image at that scale.

[0073] Step 5: Repeat steps 2 to 4 until the preset scale level is reached to complete the generation of the final deformation field, and apply the final deformation field to the moving image to obtain the final registered 3D medical image.

[0074] Furthermore, step 1 specifically involves obtaining moving images at different scales. and fixed image for:

[0075] (1)

[0076] Among them, scale level The coarsest scale, scale level For the finest scale, For the first Scale-level moving images, For the first Fixed image at scale level.

[0077] Furthermore, from the original resolution Begin layer-by-layer downsampling:

[0078] (2)

[0079] (3)

[0080] Formula (2) is for moving images. For activation function, This is a 3D convolution operation. In scale The convolution kernel for the moving image, The stride for the convolution operation is 2. For bias terms, For the first Moving images at a scale of -1 For the first -1 scale level fixed image;

[0081] Formula (3) applies to fixed images and works on the same principle;

[0082] By repeating formulas (2) and (3), a multi-scale image pyramid is constructed layer by layer, namely:

[0083] (4)

[0084] in, For the moving image at scale level 1, For fixed images at the first scale level, and so on, For the moving image at the L-th scale level, This is a fixed image at the L-th scale level.

[0085] Furthermore, step 2 specifically involves the latent feature score tensor. The formula is:

[0086] (5)

[0087] in, , The latent feature score tensor between image pairs is the deformation direction or trend required to recover a fixed image in the current state; The residual and gradient information of the transformation from a moving image to a fixed image are encoded. This feature serves as the semantic guidance basis for deformation modeling in this layer. As an independent conditional diffusion network, t For time steps.

[0088] Furthermore, in step 3, the NCA module performs local region deformation modeling on the latent feature score tensor generated from the diffusion model to generate a deformation field output.

[0089] Furthermore, in step 4, the spatial transformation process at each scale is abstracted into a formula:

[0090] (6)

[0091] in, Indicated in scale The image after registration, Indicated in scale Upward moving image, Indicated in scale The deformation field generated on it.

[0092] Next, the image similarity loss between the registered image and the fixed image at each scale is calculated, expressed by the formula:

[0093] (7)

[0094] Furthermore, the NCA module includes a learnable state iteration function. Perform on a discrete spatial grid Step-by-step evolution, the initial state tensor of each layer is defined as:

[0095] (8)

[0096] in, This is the latent feature score tensor for this layer. For the first Scale-level moving images, For the first Fixed images at different scale levels The deformation field output from the previous scale is upsampled to the current scale, and `concat` is used to concatenate them along the channel dimension to form the state tensor. .

[0097] Furthermore, state evolution is perceptually performed within a local neighborhood, recursively according to the following rules:

[0098] (9)

[0099] in, This represents a state iteration function whose parameters are learnable. Indicates the first Iterate up to the 1st level The state representation at each step, Indicates the first Iterate up to the 1st level State representation during a step.

[0100] Each pixel interacts only with the state of its local neighborhood during evolution. After several steps, the state tends to stabilize, resulting in the final deformed field. From a stable state Decoded from Chinese, implemented using an MLP decoder:

[0101] (10)

[0102] Each scale layer independently generates a deformation field, which serves as the optimal estimate of the alignment for that layer and as the input to the conditional diffusion model for the next scale.

[0103] The loss function used in the training process of the method in this application is explained below:

[0104] Image similarity loss:

[0105] (11)

[0106] Deformation field smoothness regularization term:

[0107] (12)

[0108] Topological consistency constraints:

[0109] (13)

[0110] in, for In position Jacobian determinant at the location.

[0111] Appendix Figure 4 and Figure 5 The results are illustrated in the diagrams. Each diagram includes a fixed image, a moving image, a registered image, and a deformation field.

[0112] Implementation Method 2

[0113] This embodiment provides a medical image deformation registration system, wherein the system uses a medical image deformation registration method as described in Embodiment 1, including:

[0114] Multi-scale pyramid building blocks: Based on a set of three-dimensional medical images to be registered, i.e., moving and stationary images, a multi-scale pyramid is constructed.

[0115] Latent Feature Score Tensor Generation Unit: Extracts the semantic differences at the current scale from the multi-scale pyramid obtained by the multi-scale pyramid construction unit, and outputs a latent feature score tensor through a multi-step back diffusion process;

[0116] Deformation field generation unit: It inputs the latent feature score tensor generated from the diffusion model, the input image pair, and the deformation field output from the previous scale into the NCA module to perform local region deformation modeling, generate the deformation field, and output it.

[0117] Similarity loss calculation unit: At each scale, the deformation field generated by the deformation field generation unit is applied to the moving image to obtain the registered image at that scale, and the image similarity loss between the registered image and the fixed image at that scale is calculated.

[0118] The medical image deformation registration unit repeats the working process of the latent feature score tensor generation unit, the deformation field generation unit, and the similarity loss calculation unit until the preset scale level is reached, to complete the generation of the final deformation field, and applies the final deformation field to the moving image to obtain the final registered 3D medical image.

[0119] Implementation Method 3

[0120] This invention provides an electronic device including a memory, a processor, and a computer program stored in the memory and executable on the processor. The memory stores software programs and modules, and the processor executes various functional applications and data processing by running the software programs and modules stored in the memory. The memory and processor are connected via a bus. Specifically, the processor implements any step in Embodiment 1 by running the computer program stored in the memory.

[0121] It should be understood that, in the embodiments of the present invention, the processor may be a Central Processing Unit (CPU), but it may also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor.

[0122] Memory may include read-only memory, flash memory, and random access memory, and provides instructions and data to the processor. Some or all of the memory may also include non-volatile random access memory.

[0123] It should be understood that if the integrated modules / units described above are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods described above can also be implemented by a computer program instructing related hardware. This computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content contained in the computer-readable storage medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction.

[0124] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0125] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the above device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this invention. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0126] It should be noted that the methods and detailed examples provided in the above embodiments can be incorporated into the apparatus and devices provided in the embodiments for mutual reference, and will not be repeated here.

[0127] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0128] In the embodiments provided by this invention, it should be understood that the disclosed apparatus / terminal devices and methods can be implemented in other ways. For example, the apparatus / device embodiments described above are merely illustrative. For instance, the division of modules or units described above is merely a logical functional division, and in actual implementation, it can be divided in other ways. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed.

[0129] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A medical image deformation registration method, characterized in that, The method includes the following steps: Step 1: Construct a multi-scale pyramid based on a set of three-dimensional medical images to be registered, wherein the set of three-dimensional medical images to be registered includes moving images and fixed images; Step 2: Extract the semantic differences at the current scale from the multi-scale pyramid in Step 1, and output a latent feature score tensor through a multi-step backdiffusion process; Step 3: Input the latent feature score tensor generated from the diffusion model in Step 2, the input image pair, and the deformation field output from the previous scale into the NCA module to perform local region deformation modeling, generate the deformation field, and output it. Step 4: At each scale, apply the deformation field generated in Step 3 to the moving image to obtain the registered image at that scale, and calculate the image similarity loss between the registered image and the fixed image at that scale. Step 5: Repeat steps 2 to 4 until the preset scale level is reached to complete the generation of the final deformation field, and apply the final deformation field to the moving image to obtain the final registered 3D medical image.

2. The medical image deformation registration method according to claim 1, characterized in that, Step 1 specifically involves obtaining moving images at different scales. and fixed image for: （1） Among them, scale level The coarsest scale, scale level For the finest scale, For the first Scale-level moving images, For the first Fixed image at scale level.

3. The medical image deformation registration method according to claim 2, characterized in that, From the original resolution Start downsampling layer by layer: （2） （3） in, For activation function, This is a 3D convolution operation. In scale The convolution kernel for the moving image, The stride for the convolution operation is 2. For bias terms, For the first Moving images at a scale of -1 For the first -1 scale level fixed image; By repeating formulas (2) and (3), a multi-scale image pyramid is constructed layer by layer, namely: （4） in, For the moving image at scale level 1, For fixed images at the first scale level, and so on, For the moving image at the L-th scale level, This is a fixed image at the L-th scale level.

4. The medical image deformation registration method according to claim 1, characterized in that, Step 2 specifically involves the latent feature score tensor. The formula is: （5） in, As the semantic guidance basis for deformation modeling at this layer, As an independent conditional diffusion network, t For time steps.

5. The medical image deformation registration method according to claim 1, characterized in that, The NCA module in step 3 will perform local region deformation modeling on the latent feature score tensor generated from the diffusion model to generate a deformation field output.

6. The medical image deformation registration method according to claim 1, characterized in that, In step 4, the spatial transformation process at each scale is abstracted into a formula: （6） in, Indicated in scale The image after registration, Indicated in scale Upward moving image, Indicated in scale The deformation field generated on it; Next, the image similarity loss between the registered image and the fixed image at each scale is calculated, expressed by the formula: （7）。 7. The medical image deformation registration method according to claim 5, characterized in that, The NCA module contains a learnable state iteration function. Perform on a discrete spatial grid Step evolution, initial state tensor of each layer Defined as: （8） in, This is a diffusion characteristic of this layer. For the first Scale-level moving images, For the first Fixed images at different scale levels The deformation field output from the previous scale is upsampled to the current scale, and `concat` is used to concatenate them along the channel dimension to form the state tensor. .

8. The medical image deformation registration method according to claim 6, characterized in that, State evolution is based on the local neighborhood as the perception range and is recursively derived according to the following rules: （9） in, This represents a state iteration function whose parameters are learnable. Indicates the first Iterate up to the 1st level The state representation at each step, Indicates the first Iterate up to the 1st level State representation during a step; Final deformation field From a stable state Decoded from Chinese, implemented using an MLP decoder: （10） Each scale layer independently generates a deformation field, which serves as the optimal estimate of the alignment for that layer and as the input to the conditional diffusion model for the next scale.

9. A medical image deformation registration system, characterized in that, The system uses a medical image deformation registration method as described in any one of claims 1-8, comprising: Multi-scale pyramid building blocks: Based on a set of three-dimensional medical images to be registered, i.e., moving and stationary images, a multi-scale pyramid is constructed. Latent Feature Score Tensor Generation Unit: Extracts the semantic differences at the current scale from the multi-scale pyramid obtained by the multi-scale pyramid construction unit, and outputs a latent feature score tensor through a multi-step back diffusion process; Deformation field generation unit: It inputs the latent feature score tensor generated from the diffusion model, the input image pair, and the deformation field output from the previous scale into the NCA module to perform local region deformation modeling, generate the deformation field, and output it. Similarity loss calculation unit: At each scale, the deformation field generated by the deformation field generation unit is applied to the moving image to obtain the registered image at that scale, and the image similarity loss between the registered image and the fixed image at that scale is calculated. The medical image deformation registration unit repeats the working process of the latent feature score tensor generation unit, the deformation field generation unit, and the similarity loss calculation unit until the preset scale level is reached, to complete the generation of the final deformation field, and applies the final deformation field to the moving image to obtain the final registered 3D medical image.

10. A computer device, characterized in that, It includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements a medical image deformation registration method as described in any one of claims 1-8.

Images