Method, system, device and medium for cardiac deformation field estimation based on a set of registration points

Abstract

The application discloses a kind of based on registration point set cardiac deformation field estimation method, system, device and medium, solve the precision of image registration in prior art needs to be improved, method includes: receiving diastolic end and systolic end cardiac image and its segmentation map;The segmentation map of diastolic end and systolic end cardiac image is used to extract the contour of heart, and the registration point set corresponding to the contour of heart image center is obtained using farthest point sampling algorithm;Diastolic end and systolic end cardiac image and the registration point set are input into the variational neural network trained in advance, and the distribution of latent variable is obtained;The distribution of latent variable is input into the radial basis function constructed in advance, and the deformation field of diastolic end image is decoded to obtain, and the deformed diastolic end image is obtained according to the deformation field.This application improves the image registration precision of cardiac deformation field.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing, and particularly relates to a cardiac deformation field estimation method, system and device based on a registration point set and a medium. BACKGROUND

[0002] Image registration is crucial for estimating cardiac motion, and the purpose is to estimate the displacement between the anatomical structures of the heart at different time points. Existing generative models focus on data distribution and tend to model the underlying pattern or data distribution. However, the accuracy of image registration needs to be improved. SUMMARY

[0003] The main purpose of the present application is to provide a cardiac deformation field estimation method, system, device and medium based on a registration point set, aiming to improve the accuracy of cardiac image registration.

[0004] To achieve the above purpose, the first aspect of the present application provides a cardiac deformation field estimation method based on a registration point set, comprising: receiving end-diastolic and end-systolic cardiac images and their segmentation maps; extracting the contour of the heart using the segmentation maps of the end-diastolic and end-systolic cardiac images, and obtaining the registration point set corresponding to the contour of the heart using the farthest point sampling algorithm; inputting the end-diastolic and end-systolic cardiac images and the registration point set into a pre-trained variational neural network to obtain the distribution of latent variables; inputting the distribution of latent variables into a pre-constructed radial basis function to decode and obtain the deformation field of the end-diastolic image, and deforming to obtain the deformed end-diastolic image according to the deformation field.

[0005] Further, the extraction of the contour of the heart using the segmentation maps of the end-diastolic and end-systolic cardiac images comprises: calculating the gradients in the X and Y directions of the segmentation maps of the cardiac images, and the positions with non-zero gradients constitute the contour of the heart.

[0006] Further, the registration point set corresponding to the contour of the heart image is obtained using the farthest point sampling algorithm, which comprises: taking the point set on the contour of the heart as a candidate point set, randomly selecting an initial point on the contour as an initial point, and obtaining the final registration point set using the farthest point sampling algorithm.

[0007] Further, the variational neural network comprises at least one Transformer layer for extracting image features of the end-diastolic and end-systolic heart images, the Transformer layer comprising an encoder and a decoder; two PRNet layers, one PRNet layer for extracting image correlation features between the image features output by the Transformer layer, and one PRNet layer for learning spatial correlation features between the registration point sets of the end-diastolic and end-systolic heart images, the PRNet layer comprising a fusion module, a multilayer perceptron, a max-pooling layer and a correlation module; and one multilayer perceptron for fusing the image correlation features and the spatial correlation features to infer distribution parameters of the deformation parameters of the control points.

[0008] Further, the Transformer layer has an encoder and a decoder, the end-diastolic heart image takes a plurality of image blocks centered on the control points, the number of image blocks being the same as the number of control points, the end-systolic heart image takes a plurality of image blocks centered on the control points, the number of image blocks being the same as the number of control points, and the image blocks are input into the encoder layer and the decoder layer of the Transformer, respectively.

[0009] Further, the PRNet layer comprises a fusion module, a multilayer perceptron, a max-pooling layer and a correlation module; the fusion module completes feature splicing, the multilayer perceptron and the max-pooling layer learn features, and the correlation module measures the correlation between features.

[0010] The second aspect of the present application provides a heart deformation field estimation system based on registration point sets, comprising: a receiving module for receiving end-diastolic and end-systolic heart images and their segmentation maps; a point sampling module for extracting heart contours of the end-diastolic and end-systolic heart images and obtaining corresponding registration point sets of the heart images by using a farthest point sampling algorithm; a processing module for inputting the end-diastolic and end-systolic heart images and the registration point sets into a pre-trained variational neural network to obtain a distribution of latent variables; and a determination module for decoding the latent variables by using a radial basis function to output a deformation field of the end-diastolic image and determining a registered end-diastolic heart image according to the deformation field.

[0011] The third aspect of the present application provides an electronic device, comprising a memory and a processor, the memory storing a computer program executable on the processor, and the processor implements the heart deformation field estimation method based on registration point sets according to any one of the above aspects when executing the computer program.

[0012] The fourth aspect of the present application provides a computer readable storage medium storing a computer program, wherein the computer program is executed by a processor to implement the heart deformation field estimation method based on registration point sets according to any one of the above aspects.

[0013] The application provides a cardiac deformation field estimation method, system and device based on a registration point set and a medium, which has the beneficial effect of integrating a point set-based method and an intensity-based method, using spatial information of the point set and intensity information of an image, and improving the accuracy of image registration. BRIEF DESCRIPTION OF DRAWINGS

[0014] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or prior art description will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor on the basis of these drawings.

[0015] Figure 1 A flowchart of the cardiac deformation field estimation method based on the registration point set of the embodiment of the present application;

[0016] Figure 2 A network structure diagram of the estimation model of the cardiac deformation field estimation method based on the registration point set of the embodiment of the present application;

[0017] Figure 3 An operation diagram of the fusion module of the PRNet module of the estimation model of the cardiac deformation field estimation method based on the registration point set of the embodiment of the present application;

[0018] Figure 4 An operation diagram of the correlation module of the PRNet module of the estimation model of the cardiac deformation field estimation method based on the registration point set of the embodiment of the present application;

[0019] Figure 5 A module composition diagram of the cardiac deformation field estimation system based on the registration point set of the embodiment of the present application;

[0020] Figure 6 A structural schematic diagram of the electronic device of the embodiment of the present application. DETAILED DESCRIPTION

[0021] In order to make the inventive purposes, features and advantages of the present application more obvious and easy to understand, the technical solutions in the embodiments of the present application will be described clearly and completely below in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, but not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0022] Embodiment 1

[0023] The method for estimating a cardiac deformation field based on a registration point set provided by the embodiment comprises the following steps as shown in the figure: Figure 1

[0024] Step S1: receiving an end-diastolic cardiac image and an end-systolic cardiac image and segmentation maps thereof.

[0025] In the embodiment, a training sample set is constructed by using the end-diastolic cardiac image and the end-systolic cardiac image and the segmentation maps thereof. After obtaining the training sample, the training sample in the original training sample set is spatially transformed by using a spatial transformation method such as translation, rotation, and elastic transformation to generate a new training sample, i.e., an enhanced training sample set.

[0026] Step S2: extracting the contour of the heart by using the segmentation map of the end-diastolic cardiac image and the end-systolic cardiac image, and obtaining a registration point set corresponding to the contour of the heart image by using a farthest point sampling algorithm.

[0027] In the embodiment, the gradients in the X direction and the Y direction are calculated on the segmentation map of the cardiac image obtained in step S1, and the positions where the gradients are not zero constitute the contour of the heart. The point set on the contour of the heart is taken as a candidate point set, and a contour point is randomly selected as an initial point, and a final registration point set is obtained by using a farthest point sampling algorithm.

[0028] Step S3: constructing a variational neural network, inputting the end-diastolic cardiac image and the end-systolic cardiac image and the registration point set thereof for training to obtain network parameters for calculating the distribution of latent variables.

[0029] More specifically, the cardiac deformation field estimation model based on the registration point set enhancement comprises at least one Transformer layer for extracting image features of the end-diastolic cardiac image and the end-systolic cardiac image, two PRNet layers, one PRNet layer for extracting image correlation features between the image features output by the Transformer layer, and one PRNet layer for learning spatial correlation features between the registration point sets of the end-diastolic cardiac image and the end-systolic cardiac image, one multilayer perceptron for fusing the image correlation features and the spatial correlation features to infer the distribution parameters of the deformation parameters of the control points. A radial basis function is used to decode the latent variables to output the deformation field of the end-diastolic image, and the registered end-diastolic cardiac image is determined according to the deformation field.

[0030] Figure 2 The network structure diagram of the cardiac deformation field estimation model based on the registration point set enhancement provided by the embodiment is shown in the figure. Figure 2 ​As shown in the figure, the network structure includes three layers of Transformer layers, two layers of PRNet layers, a multi-layer perceptron, and a radial basis function. Among them, the three layers of Transformer layers are directly cascaded. Each layer of the Transformer layer contains a Transformer encoder and a Transformer decoder. The three layers of Transformer and one layer of PRNet are used to learn the image feature correlation of the image. One layer of PRNet is used to learn the spatial feature correlation of the registration point set. One layer of multi-layer perceptron is used to fuse the image correlation feature and the spatial correlation feature to output the latent variable distribution. The radial basis function is used to decode the latent variable to output the deformation field of the end-diastolic image.

[0031] Figure 2 The embodiment shows the structure diagram of the Transformer layer and the PRNet layer in the network structure of the cardiac deformation field estimation model based on the registration point set enhancement. As shown in the figure, Figure 2 The Transformer layer has a Transformer encoder and a Transformer decoder. The Transformer encoder layer is composed of a multi-head attention layer and a layer of multi-layer perceptron. The Transformer decoder is composed of two multi-head attention layers and a layer of multi-layer perceptron. The PRNet layer is composed of two fusion modules, two multi-layer perceptrons, two maximum pooling layers, and a correlation module.

[0032] Figure 3 The operation diagram of the fusion module of the PRNet module of the cardiac deformation field estimation model based on the registration point set enhancement provided by the embodiment of the application is shown in the figure. Figure 3 As shown in the figure, for any feature f i (i=1,…,n), it is spliced with all features g i (i=1,…,m) to obtain the fusion feature F i (i=1,…,n).

[0033] Figure 4 The operation diagram of the correlation module of the PRNet module of the cardiac deformation field estimation model based on the registration point set enhancement provided by the embodiment of the application is shown in the figure. Figure 4 As shown in the figure, for any feature f i (i=1,…,n), the correlation is calculated with all features g i (i=1,…,n). In the application, the vector dot product is used to measure the correlation, and the results of each vector dot product are spliced to obtain the correlation feature F i (i=1,…,n).

[0034] Step S4: constructing a radial basis dense deformation field based on the distribution of the latent variable.

[0035] More specifically, the radial basis function is used to decode the latent variable to output the deformation field of the end-diastole image. The formula of the deformation field in X direction is as follows:

[0036]

[0037] where v x is the position of pixel v in X direction, z i,x is the control point p i , ||v-p i || represents the Euclidean distance between pixel v and control point p i , and ψ(·) represents the compactly supported radial basis function with compact support radius r.

[0038] Step S5: Calculate the cross-correlation coefficient between the deformed end-diastole image and the end-systole image, the KL divergence of the prior of the latent variable and the variational distribution, the chamfer distance between the registration point set of the deformed end-diastole image and the registration point set of the end-systole image, and the Jacobian value of the deformation field, and then calculate the objective function of the network.

[0039] More specifically, the objective function of the network is as follows:

[0040]

[0041] where F and M are the end-systole image and the end-diastole image, respectively; z = {z1, …, z n} is the latent deformation parameter; P M = {P M1 , …, P Mm} and P F = {P F1 , …, P Fm} are the registration point sets of the end-diastole and end-systole cardiac images, respectively; n is the number of control points, m is the number of points in the registration point set, φ is the deformation field, λ is a constant coefficient, p(z) is the prior distribution of the latent deformation parameter z, q β (z|F,M) is the variational distribution of the latent deformation parameter z; LCC(M,F,φ) is the cross-correlation coefficient between the deformed end-diastole image M and the end-systole image F; is the expectation under the variational distribution q β (z|F,M) of the latent deformation parameter z; KL[q β (z|F,M)||p(z)] is the KL divergence between the variational distribution and the prior distribution p(z) of the latent deformation parameter z; L chamfer (P M ,P FL (F, M, φ) is the chamfer distance between the deformed point set of the end diastolic image and the point set of the end systolic image, L jacobian J (φ) is the Jacobian of the deformation field.

[0042] More specifically, the formula of the cross-correlation coefficient between the deformed end diastolic image and the end systolic image is as follows:

[0043]

[0044] where Ω is the image domain, M φ is the deformed image of the end diastolic image M by the deformation field φ, A is the local neighborhood centered at (x, y), is the average intensity value of F (x, y) in the neighborhood A, is the average intensity value of M φ (x, y) in the neighborhood A.

[0045] More specifically, the expectation is calculated by using the Monte Carlo approximation:

[0046]

[0047] where φ i is the latent deformation parameter z β sampled from the q i (z|F, M) distribution,

[0048] More specifically, the formula of the KL divergence between the prior of the latent variable and the variational distribution is as follows:

[0049]

[0050] where the prior distribution p (z) of the latent deformation parameter z is a Gaussian distribution with mean 0 and covariance matrix I, n is the number of control points, the variational distribution q β (z|F, M) is the chain product of n Gaussian distributions, where the parameters μ i and of the variational distribution are the mean and variance of the latent deformation parameter z i , respectively.

[0051] More specifically, the formula of the chamfer distance between the deformed point set of the end diastolic image and the point set of the end systolic image is as follows:

[0052]

[0053] where P is a set of registration points of the end-diastolic image M P is a set of registration points of the end-diastolic image

[0054] More specifically, the formula for calculating the Jacobian value of the deformation field is as follows:

[0055]

[0056] where Ω is the image domain, |J φ (v)| is the determinant of the Jacobian matrix.

[0057] Step S6: Determine whether the objective function converges.

[0058] Step S7: When the objective function does not converge, calculate the gradient of the network parameters using the objective function, perform gradient backpropagation, and update the parameters. When the objective function converges, obtain the trained cardiac deformation field estimation model.

[0059] More specifically, assume that the i-th network parameter is w i , calculate Then the network parameter w i is adjusted to w i ′ = w i + γΔw i , where γ is the learning rate, which can be automatically determined according to the adaptive stochastic gradient descent algorithm.

[0060] Embodiment 2

[0061] The embodiment of the present application provides a cardiac deformation field estimation system based on a set of registration points, as shown in the figure, comprising: Figure 5

[0062] The receiving module 1 is used for receiving end-diastolic and end-systolic cardiac images and their segmentation maps;

[0063] The sampling module 2 is used for extracting the cardiac contours of the end-diastolic and end-systolic cardiac images, and obtaining the corresponding registration point set of the cardiac images using the farthest point sampling algorithm;

[0064] The processing module 3 is used for processing the end-diastolic and end-systolic cardiac images and their registration point sets using the trained variational neural network, and outputting the latent variable distribution. The variational neural network comprises at least one Transformer layer, two PRNet layers and one multilayer perceptron;

[0065] The determining module 4 is used for decoding the latent variable by the radial basis function to output the deformation field of the end-diastolic image, and determining the registration end-diastolic cardiac image according to the deformation field.

[0066] The optional module further comprises: ​

[0067] The construction module 5 is configured to construct a variational neural network of a registration point set enhanced cardiac deformation field estimation model.

[0068] The training module 6 is configured to train the variational neural network of the registration point set enhanced cardiac deformation field estimation model by using the end-diastolic cardiac image and the end-systolic cardiac image and their segmentation maps, to output the registered end-diastolic cardiac image.

[0069] The adjusting module 7 is configured to calculate a target function by using a cross-correlation coefficient between the deformed end-diastolic image and the end-systolic image, a KL divergence of a prior of a latent variable and a variational distribution, a chamfer distance between a registration point set of the deformed end-diastolic image and a registration point set of the end-systolic image, and a Jacobian value of the deformation field, and update parameters of the variational neural network of the registration point set enhanced cardiac deformation field estimation model according to the target function, to output the trained variational neural network.

[0070] It is worth noting that the present embodiment and the embodiment 1 are in a relationship of corresponding devices and methods, and the content described in the embodiment 1 is also applicable to the present embodiment, and thus the present embodiment will not be repeated.

[0071] Embodiment 3

[0072] The embodiment of the present application provides a computer device, such as Figure 6As shown, it comprises at least one processor 401, such as a CPU (Central Processing Unit), at least one communication interface 403, a memory 404, and at least one communication bus 402. The communication bus 402 is used to realize the connection and communication between the components. The communication interface 403 can include a display, a keyboard, and can also include a standard wired interface and a wireless interface. The memory 404 can be a volatile RAM (Random Access Memory) or a non-volatile memory such as at least one disk memory. The memory 404 can also be at least one storage device located away from the aforementioned processor 401. The processor 401 can execute the method for estimating a cardiac deformation field based on a set of registration points in embodiment 1 or the system for estimating a cardiac deformation field based on a set of registration points in embodiment 2. The memory 404 stores a set of program codes, and the processor 401 calls the program codes stored in the memory 404 to execute the method for estimating a cardiac deformation field based on a set of registration points in embodiment 1 or the system for estimating a cardiac deformation field based on a set of registration points in embodiment 2. The communication bus 402 can be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. The communication bus 402 can be divided into an address bus, a data bus, and a control bus. For ease of representation, Figure 6 Only one line is used in the figure, but it does not mean that there is only one bus or only one type of bus.

[0073] The memory 404 can include a volatile memory such as a RAM (Random-Access Memory) and can also include a non-volatile memory such as a flash memory, a hard disk drive (HDD), or a solid-state drive (SSD). The memory 404 can also include a combination of the above-mentioned types of memories.

[0074] The processor 401 can be a central processing unit (CPU), a network processor (NP), or a combination of the CPU and the NP.

[0075] The processor 401 can further include a hardware chip. The hardware chip can be an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof. The PLD can be a complex programmable logic device (CPLD), a field-programmable gate array (FPGA), a generic array logic (GAL), or any combination thereof.

[0076] Optionally, the memory 404 is further configured to store program instructions. The processor 401 can invoke the program instructions to implement the method of the cardiac deformation field estimation model based on the registration point set enhancement in Embodiment 1 or the system of the cardiac deformation field estimation model based on the registration point set enhancement in Embodiment 2.

[0077] The embodiments of the present application further provide a computer-readable storage medium having computer executable instructions stored thereon. The computer executable instructions can execute the method of the cardiac deformation field estimation model based on the registration point set enhancement in Embodiment 1 or the system of the cardiac deformation field estimation model based on the registration point set enhancement in Embodiment 2. The storage medium can be a disk, an optical disk, a read-only memory (ROM), a random access memory (RAM), a flash memory, a hard disk drive (HDD), a solid-state drive (SSD), or the like. The storage medium can further include a combination of the above-mentioned storage media.

[0078] In several embodiments provided in the present application, it should be understood that the disclosed apparatus and method can be implemented in other manners. For example, the described apparatus embodiments are merely schematic. For example, the division of the modules is merely a logical function division. There can be another division manner for the actual implementation, for example, multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the displayed or discussed mutual couplings or direct couplings or communication connections between different modules can be indirect couplings or communication connections through some interfaces, and electrical, mechanical or other forms.

[0079] The modules illustrated as separated components can or can not be physically separated, and the components illustrated as modules can or can not be physical modules, i.e., can be located in one place, or can be distributed on multiple network modules. Some or all of the modules can be selected according to actual needs to achieve the purpose of the embodiments.

[0080] In addition, each functional module in each embodiment of the present application can be integrated into a processing module, or each module can be physically present alone, or two or more modules can be integrated into one module. The integrated module can be realized in the form of hardware or in the form of a software functional module.

[0081] When the integrated module is realized in the form of a software functional module and sold or used as an independent product, it can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application essentially or the part that contributes to the prior art or the whole or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a number of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the embodiments of the present application.

[0082] It should be noted that, for the foregoing method embodiments, in order to facilitate description, they are all described as a series of action combinations, but those skilled in the art should know that the present application is not limited to the order of the actions described, because according to the present application, some steps can be performed in other orders or at the same time. Secondly, those skilled in the art should know that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily all the necessary actions of the present application.

[0083] In the above embodiments, the description of each embodiment has its own focus, and the parts not described in detail in a certain embodiment can be referred to the relevant description of other embodiments.

[0084] The above is the description of the method, system, device and medium for estimating the cardiac deformation field based on the registration point set provided by the application. For those skilled in the art, according to the idea of the embodiments of the application, the specific implementation and application range will be changed. In conclusion, the content of the specification should not be understood as a limitation of the application.

Claims

1. A method for cardiac deformation field estimation based on a set of registration points, characterized in that, The method comprises the following steps: receiving diastolic and systolic heart images and their segmentation maps; extracting the contours of the heart using the segmentation maps of the diastolic and systolic heart images, and obtaining a set of registration points corresponding to the heart contours of the heart images by using a farthest point sampling algorithm; inputting the diastolic and systolic heart images and the set of registration points into a pre-trained variational neural network to obtain a distribution of latent variables; wherein the variational neural network comprises at least one Transformer layer for extracting image features of the diastolic and systolic heart images, the Transformer layer comprising an encoder and a decoder; two PRNet layers, one of which is used to extract image correlation features between the image features output by the Transformer layer, and the other of which is used to learn spatial correlation features between the sets of registration points of the diastolic and systolic heart images, the PRNet layer comprising a fusion module, a multilayer perceptron, a max-pooling layer and a correlation module; and one multilayer perceptron for fusing the image correlation features and the spatial correlation features to infer the distribution parameters of the deformation parameters of the control points; inputting the distribution of latent variables into a pre-constructed radial basis function to decode and obtain a deformation field of the diastolic image, and performing deformation according to the deformation field to obtain a deformed diastolic image.

2. The landmark set-based cardiac deformation field estimation method of claim 1, wherein, The step of extracting the contours of the heart using the segmentation maps of the diastolic and systolic heart images comprises calculating the gradients in the X and Y directions of the segmentation maps of the heart images, and the positions with non-zero gradients constitute the contours of the heart.

3. The landmark set-based cardiac deformation field estimation method of claim 1, wherein, The step of obtaining a set of registration points corresponding to the heart contours of the heart images by using a farthest point sampling algorithm comprises taking the set of points on the contours of the heart as a candidate point set, randomly selecting an initial point on the contour as an initial point, and obtaining a final set of registration points by using a farthest point sampling algorithm.

4. The landmark set-based cardiac deformation field estimation method of claim 1, wherein, The Transformer layer has an encoder and a decoder, the diastolic heart image takes a plurality of image blocks centered on the control points, the number of image blocks is the same as the number of control points, the systolic heart image takes a plurality of image blocks centered on the control points, the number of image blocks is the same as the number of control points, and the image blocks are input into the encoder layer and the decoding layer of the Transformer, respectively.

5. The landmark set-based cardiac deformation field estimation method of claim 1, wherein, The PRNet layer comprises a fusion module, a multilayer perceptron, a max-pooling layer and a correlation module; the fusion module completes the splicing of features, the multilayer perceptron and the max-pooling layer learn the features, and the correlation module measures the correlation between the features.

6. A cardiac deformation field estimation system based on a set of registration points, characterized in that, The method comprises the following steps: a receiving module for receiving diastolic and systolic heart images and their segmentation maps; a point sampling module for extracting the contours of the heart in the diastolic and systolic heart images, and obtaining a set of registration points corresponding to the heart images by using a farthest point sampling algorithm; The processing module inputs the end-diastolic and end-systolic heart images and the set of registration points into a pre-trained variational neural network to obtain a distribution of latent variables; wherein the variational neural network comprises: at least one layer of a Transformer layer for extracting image features of the end-diastolic and end-systolic heart images, the Transformer layer comprising an encoder and a decoder; two layers of a PRNet layer, one layer of the PRNet layer being used to extract image correlation features between the image features output by the Transformer layer, and the other layer of the PRNet layer being used to learn spatial correlation features between the set of registration points of the end-diastolic and end-systolic heart images, the PRNet layer comprising a fusion module, a multilayer perceptron, a max-pooling layer and a correlation module; and one layer of a multilayer perceptron for fusing the image correlation features and the spatial correlation features to infer distribution parameters of deformation parameters of the control points. The determining module is configured to decode the latent variables by using a radial basis function to output a deformation field of the end-diastolic image, and determine the end-diastolic heart image after registration according to the deformation field. 7.An electronic device comprising: A memory and a processor, wherein a computer program is stored in the memory and can be run on the processor, and when the processor executes the computer program, the method in any one of claims 1 to 5 is implemented.

8. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the method in any one of claims 1 to 5.

Images