Template image determination method and device, equipment, medium and product

Abstract

The invention discloses a template image determination method and device, equipment, a medium and a product. The method comprises the following steps: determining a feature set and feature images corresponding to image features in the feature set; based on a hierarchy incremental principle, determining a hierarchy distance between the current hierarchy and the current hierarchy, selecting a feature cluster under the current hierarchy from the feature set according to the hierarchy distance, the current hierarchy distance being greater than the hierarchy distance under the previous hierarchy; determining an image combination under the current level based on the sub-template image under the previous level and the feature cluster under the current level; determining a sub-template image corresponding to the image combination; and if the feature set has the undistributed image features, returning to the step of determining the current hierarchy and the hierarchy distance under the current hierarchy until the feature set does not have the undistributed image features, and taking the sub-template image under the current hierarchy as a target template image. According to the embodiment of the invention, it can be ensured that the determined target template image has high accuracy.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular, to a method, apparatus, device, medium and product for determining a template image. Background Art

[0002] For the required template medical images in a target scenario, the prior art usually first obtains a large number of scanned images of target objects, then performs clustering analysis on the obtained scanned images, determines the sub-template images corresponding to each classification cluster, and then determines the target template image based on the sub-template images corresponding to each classification cluster.

[0003] When the distances between different classification clusters are far, it is impossible to ensure the accuracy of the target template image determined based on the sub-template images corresponding to each classification cluster. Summary of the Invention

[0004] The present invention provides a method, apparatus, device, medium and product for determining a template image to solve the problem that the prior template determination method cannot ensure the accuracy of the determined template image.

[0005] According to an aspect of the present invention, there is provided a method for determining a template image, including:

[0006] Determine a feature set corresponding to a scanned image set and feature images corresponding to each image feature in the feature set, where the scanned image set includes scanned images of multiple target objects, and the feature set includes image features in each of the scanned images that have an associated relationship with predetermined variable information;

[0007] Based on the principle of increasing levels, determine the current level and the level distance at the current level, and select the feature cluster at the current level from the feature set according to the level distance, where the current level distance is greater than the level distance at the previous level;

[0008] Determine the image combination at the current level by using the sub-template image at the previous level to replace the feature image belonging to the previous level in the feature cluster at the current level;

[0009] Process the image combination based on local popularity shrinkage to obtain a sub-template image corresponding to the image combination;

[0010] If there are unassigned image features in the feature set, return to the step of determining the current level and the level distance at the current level based on the principle of increasing levels until there are no unassigned image features in the feature set, and use the sub-template image at the current level as the target template image corresponding to the predetermined variable information.

[0011] According to another aspect of the present invention, there is provided a method for determining a template image, including:

[0012] Determine a feature set corresponding to a set of scanned images and feature images corresponding to each image feature in the feature set, the set of scanned images including scanned images of multiple target objects, and the feature set including image features in each of the scanned images that have an associated relationship with predetermined variable information;

[0013] Based on the principle of increasing levels, determine the current level and the level distance at the current level, and select a feature cluster at the current level from the feature set according to the level distance, where the current level distance is greater than the level distance at the previous level;

[0014] Determine an image combination at the current level by using the sub-template image at the previous level to replace the feature images belonging to the previous level in the feature cluster at the current level;

[0015] Process the image combination based on local popularity shrinkage to obtain a sub-template image corresponding to the image combination;

[0016] If there are unassigned image features in the feature set, return to the step of determining the current level and the level distance at the current level based on the principle of increasing levels until there are no unassigned image features in the feature set, and use the sub-template image at the current level as the target template image corresponding to the predetermined variable information.

[0017] According to another aspect of the present invention, there is provided an electronic device, the electronic device including:

[0018] At least one processor; and

[0019] A memory communicatively connected to the at least one processor; wherein,

[0020] The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the template image determination method according to any embodiment of the present invention.

[0021] According to another aspect of the present invention, there is provided a computer-readable storage medium, the computer-readable storage medium storing computer instructions for causing a processor to implement the template image determination method according to any embodiment of the present invention when executed.

[0022] According to another aspect of the present invention, there is provided a computer program product, which includes a computer program that, when executed by a processor, implements the template image determination method according to any one of the embodiments.

[0023] In the technical solution of the template image determination method provided by the embodiments of the present invention, since there is an association relationship between the image features and the predetermined variable information, and the feature image is determined based on the image features, there is also an association relationship between the feature image and the predetermined variable information; since the feature clusters at the current level are determined based on the current level distance, and the current level distance is greater than the level distance of the previous level, the feature clusters at the current level include all the feature images within the feature clusters at the previous level, so a sequence of feature clusters with a nested relationship can be obtained. Therefore, it is allowed to replace the image features belonging to the previous level in the feature clusters at the current level with the sub-template images at the previous level, and replace each image feature in the feature clusters at the current level with the corresponding feature image to obtain the image combination at the current level. After repeated iteration, the image combination at the highest level and the sub-template image of this image combination, that is, the target template image, can be obtained. The transmission of the sub-template image can significantly improve the accuracy of the sub-template image determined at the current level, thereby improving the accuracy of the target template image corresponding to the pre-variable information.

[0024] It should be understood that the content described in this part is not intended to identify the key or important features of the embodiments of the present invention, nor is it used to limit the scope of the present invention. Other features of the present invention will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0026] Figure 1 is a flowchart of the template image determination method provided by the embodiments of the present invention;

[0027] Figure 2 is a schematic diagram of the image feature extraction process provided by the embodiments of the present invention;

[0028] Figure 3 is a schematic diagram of hierarchical registration for the image combination provided by the embodiments of the present invention;

[0029] Figure 4 is a schematic diagram of the comparison result of the target template images generated by the embodiments of the present invention and other similar algorithms based on the HCP-D dataset;

[0030] Figure 5 It is a schematic diagram of the comparison result of the target template image generated by the embodiment of the present invention and other similar algorithms based on the OASIS dataset;

[0031] Figure 6 It is another flowchart of the template image determination method provided according to the embodiment of the present invention;

[0032] Figure 7 It is a schematic diagram of the determination process of the feature cluster and image combination provided according to the embodiment of the present invention;

[0033] Figure 8 It is another flowchart of the template image determination method provided according to the embodiment of the present invention;

[0034] Figure 9 It is a schematic diagram of the relationship between the deformation field and the inverse deformation field of each feature image provided according to the embodiment of the present invention;

[0035] Figure 10 It is a schematic diagram of the structure of the adaptive multi-level registration network provided according to the embodiment of the present invention;

[0036] Figure 11 It is a schematic diagram of the determination process of the sub-template image provided according to the embodiment of the present invention;

[0037] Figure 12 It is a schematic diagram of the structure of the template image determination device provided according to the embodiment of the present invention;

[0038] Figure 13 It is a schematic diagram of the structure of the electronic device for implementing the template image determination method of the embodiment of the present invention. Detailed implementation manners

[0039] In order to enable those skilled in the art of the present technology to better understand the solution of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without making creative efforts shall fall within the protection scope of the present invention.

[0040] It should be noted that the terms "first", "second", etc. in the description, claims and above-mentioned drawings of the present invention are used to distinguish similar objects, and do not necessarily have to be used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances, so that the embodiments of the present invention described here can be implemented in an order other than those illustrated or described here. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device that includes a series of steps or units does not have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0041] Figure 1 FIG. is a flowchart of a template image determination method provided by an embodiment of the present invention. This embodiment is applicable to the case where, based on a feature cluster with a nested relationship and the principle that the corresponding sub-template images of the image combination can be transmitted to the corresponding image combination at the next level, the image combinations at each level and the sub-template images corresponding to the image combinations at each level are gradually determined, so as to determine the target template image. This method can be executed by a template image determination device, which can be implemented in the form of hardware and / or software, and the template image determination device can be configured in the processor of an electronic device. As Figure 1 shown, the method includes:

[0042] S110. Determine a feature set corresponding to the scanned image set and feature images corresponding to each image feature in the feature set. The scanned image set includes scanned images of multiple target objects, and the feature set includes image features in each scanned image that have an associated relationship with predetermined variable information.

[0043] The scanned image refers to a clinical medical image of a target object, such as an MRI image (magnetic resonance image) or a CT image (computed tomography image), etc.

[0044] The scanned image set is a set of scanned images of all target objects in the target population. The target population is a population set that meets the requirements of predetermined variable information. For example, a male population aged between 40 and 50 years old. It should be noted that the number of the target population can be set according to the actual situation, and this embodiment will not elaborate here.

[0045] The predetermined variable information can be age information or disease state information; of course, it can also be other information that changes with independent variables such as time.

[0046] In one embodiment, the predetermined variable information is age, and the scanned image is a brain MRI image.

[0047] In one embodiment, the predetermined variable information is disease status information, such as moderate cerebellar atrophy, and the scanned image is a brain MRI image.

[0048] An image feature having an association relationship with the predetermined variable information refers to an image feature that changes as the predetermined variable information changes. For example, for a young child, as they grow older, their brain regions also develop accordingly.

[0049] In one embodiment, based on the existing image extraction algorithm, image features having an association relationship with the corresponding predetermined variable information are extracted from each scanned image. For the convenience of describing the technical solution, in this embodiment, all the image features extracted from all the scanned images obtained by a target object in one scan are regarded as a whole, which is simply referred to as image features.

[0050] In another embodiment, the scanned images of each target object are analyzed based on a pre-trained feature extraction model to obtain image features having an association relationship with the corresponding predetermined variable information in the scanned images. Among them, the pre-trained feature extraction model includes an encoder, a feature linear modulation layer, and a decoder; the feature linear modulation layer learns the association relationship between the predetermined variable information and the image features; the encoding layer is used to determine the image features having an association relationship with the predetermined variable information in the scanned image; the decoder is used to reconstruct the image features into the feature image.

[0051] Optionally, the feature extraction model is a convolutional autoencoder. During the training process, the feature extraction model inputs the scanned image and the predetermined variable information into the encoder. As Figure 2 shown, the goal of the encoder is to learn the association relationship between the predetermined variable information of the scanned image and the image features , and the goal of the decoder is to reconstruct the feature image with the image features. The feature linear modulation layer adaptively adds the predetermined variable information to each layer of the encoder by applying a radiometric transformation, and uses a four-layer multi-layer perception mechanism to generate scale information and displacement parameters. For example, 、 etc., and then projects the scale information and displacement parameters onto each layer of the encoder to generate the transformed image features to improve the accuracy of the image features.

[0052] The feature linear modulation layer can be expressed as: , where is the The feature map of a scanned image in the c-th network channel. In one embodiment, the encoder consists of three convolutional blocks, each convolutional block consisting of two convolutional layers with a stride of 3×3×3 and a convolutional layer with a stride of 2 of 3×3×3; at the end of the encoder, there is a convolutional layer with a stride of 1 of 3×3×3 and a fully connected layer to further integrate the high-level features of the latent representation. The decoder includes three transposed convolutional blocks and a 1×1×1 convolutional layer for receiving the output data of the three transposed convolutional blocks; wherein, each transposed convolutional block includes a transposed convolutional layer, two consecutive convolutional layers of 3×3×3, a LeakyReLU activation layer and a batch normalization layer connected in sequence, wherein the strides of the transposed convolutional layer and the convolutional layer are both 1, for upsampling the latent representation and restoring the image details; the 1×1×1 convolutional layer is used to output the feature image.

[0053] In one embodiment, the feature extraction model is trained using the OASIS dataset, the HCP-D dataset, and the enhanced images of both. In the test phase, they are directly called to extract the high-dimensional image features of the images. Network training is completed on a set electronic device using the PyTorch toolkit in Python, such as an NVIDIA TITAN Xp 12GB GPU server. The Adam optimizer is used to update the model parameters, and the learning rate is set to 0.001.

[0054] The pre-trained feature extraction model outputs both the image features collected by the encoder and the feature image determined by the decoder based on the image features to a set position.

[0055] S120. Based on the principle of hierarchical increment, determine the current level and the level distance at the current level, and select the feature cluster at the current level from the feature set according to the level distance, where the current level distance is greater than the level distance at the previous level.

[0056] The level distance can be understood as the maximum feature distance allowed at the corresponding level. That is to say, within the feature cluster at each level, the maximum feature distance allowed between two image features is the level distance at the corresponding level. Among them, the feature distance can reflect the similarity between the feature images corresponding to the two image features. Therefore, the closer the feature distance between the two image features, the higher the similarity between the feature images corresponding to the two image features, and vice versa, the lower the similarity.

[0057] In this embodiment, the level identifier of the current level is determined based on the level identifier of the previous level. For example, the level identifier of the previous level is n, and the level identifier of the current level is n + 1, where n is a positive integer.

[0058] In this embodiment, the level distance increases as the level increases.

[0059] In one embodiment, each level corresponds to a predetermined level distance. After the level identifier of the current level is determined, the predetermined level distance corresponding to the current level identifier is used as the level distance under the current level. This embodiment can simply and quickly determine the level distance under the current level.

[0060] In one embodiment, the level distance is determined based on a predetermined distance increment rule. Such as the arithmetic principle or the geometric principle. Exemplarily, the level distance of the previous level is r, and the level distance under the current level is 2r, or r + d.

[0061] After the feature set is determined, the distances between pairwise image features in the feature set are determined, and the corresponding feature clusters are selected from the feature set based on the level distance under the current level. The distance between each image feature in the feature cluster and at least one other image feature in the feature cluster is less than the level distance.

[0062] It can be understood that as the level distance increases, the number of image features in the feature cluster also increases. Therefore, in this embodiment, the number of image features included in the feature cluster under the current level is greater than or equal to the number of image features included in the feature cluster under the previous level, that is, the feature cluster under the previous level is a subset of the feature cluster under the current level. In this way, a nested feature cluster is formed.

[0063] In one embodiment, from the perspective of topological data analysis, a multi-scale progressive grouping strategy based on persistent homology is designed to obtain sequence grouping in an integrated manner. Assume that N scanned images are all located in the same manifold M, and the topological structure of their distribution can be described by a graph based on feature similarity.

[0064] The feature distance between pairwise image features is determined by the following formula:

[0065] ;

[0066] where is the image feature labeled i, is the image feature labeled j; Pearson(·) is the Pearson correlation coefficient of the image feature, and its value range is [0, 1]. The more similar two scanned images are, the smaller the feature distance between the image features corresponding to the two scanned images.

[0067] The Vietoris-Rips complex (Rips complex) is used to establish the connectivity of the simplicial complex. Therefore, the Rips complex can be equivalently represented as a Rips subgraph (feature cluster) at scale r > 0 , where the subgraph (feature cluster) is defined as:

[0068] ;

[0069] Among them, is a feature cluster at the r scale. Based on determine the image combination formed by the feature clusters at each hierarchical structure h i below. Use the Rips filtering scheme to extend the Rips subgraph to multi-scale grouping to generate a nested subgraph (feature cluster) combination.

[0070] Specifically, based on the feature similarity that persists in the image at multiple scales, gradually cluster the feature images. During the clustering process, set multiple levels of predetermined hierarchical distances according to the critical tolerance of the feature distance, and generate a multi-scale subgraph (feature cluster) combination by gradually expanding through the predetermined hierarchical distances, which is represented as a series of nested subgraph (feature cluster) combinations:

[0071] ;

[0072] Among them, , .

[0073] S130. Replace the image features belonging to the previous level in the feature cluster at the current level with the sub-template image at the previous level, and replace each image feature in the feature cluster at the current level with the corresponding feature image to obtain the image combination at the current level.

[0074] The sub-template image at the previous level is substantially determined based on all the feature images in the corresponding image combination at the previous level. Therefore, the sub-template image at the previous level can be used to replace all the image features belonging to the previous level in the feature cluster at the current level, and then replace the new image features of this feature cluster compared with the corresponding feature cluster at the previous level with the corresponding feature images to obtain the image combination at the current level.

[0075] Since the sub-template image at the previous level is determined based on all the feature images in the corresponding image combination at the previous level and passed to the corresponding image combination at the current level, the image combinations at different levels ( ) can be considered to have a nested relationship, which can be specifically expressed as:

[0076] .

[0077] The image combination with a nested relationship can increase the correlation and tightness between image combinations at adjacent levels, making it possible to use the sub-template image corresponding to the image combination at the previous level as the feature image in the corresponding image combination at the current level and participate in the determination process of the sub-template image corresponding to the image combination at the current level; moreover, passing the sub-template image at the previous level to the corresponding image combination at the current level to participate in and guide the determination of the sub-template image corresponding to the image combination at the current level can improve the accuracy of determining the sub-template image corresponding to the image combination at the current level.

[0078] It should be noted that if level n is the highest level, then it only includes one image combination.

[0079] S140. Process the image combination based on local manifold shrinkage to obtain the sub-template image corresponding to the image combination.

[0080] Local manifold shrinkage can be understood as, during the image registration process, gradually registering multiple local images (image combinations) to gradually reduce the differences between the images in the multiple local images (image combinations) to complete the registration of the multiple local images (image combinations).

[0081] Specifically, in order to avoid the adverse effects of blurred average images or biased single reference images on group registration, the embodiment of the present invention performs hierarchical group registration within the image combination based on local manifold shrinkage. As Figure 3 shown, assume that the image combination contains n feature images, which are located on the local image manifold m(I) and are represented as an undirected weighted graph. Among them, each feature image is defined as a vertex in the graph, and the edge weight is determined by the distance between pairwise feature images. Considering simple and fast calculations, the distance is defined as the gray difference between two images, and the gray difference can be determined by the following formula:

[0082] ;

[0083] where P is the total number of voxels in the feature image, is the gray value of the th voxel in image ( represents the th iteration), and group registration is performed within each image combination.

[0084] In one embodiment, the sub-template image corresponding to the image combination is determined according to the deformation fields of the feature images in the image combination to improve the accuracy of the sub-template image corresponding to the image combination.

[0085] Since the sub-template image corresponding to the image combination at the current level is determined based on local popularity shrinkage, the sub-template image at the current level is located at the center of the image combination at the current level. By analogy, the sub-template images at each level are located at the center of the image combination at each level. Therefore, the sub-template image at the previous level is located at the center of the image combination at the previous level. Then, when the sub-template image at the previous level is passed to the current level, it is located at the center of all the feature images corresponding to the corresponding feature cluster at the previous level.

[0086] Since all the feature images are divided into a sequence of image combinations with nested relationships, the determination of the sub-template images from the low level to the high level is completed under the guidance of the sub-template images and the real individual images at the previous level, achieving the technical effect of avoiding the propagation of registration errors between levels and improving the accuracy of the sub-template images at each level.

[0087] In one embodiment, the sub-template image at the previous level is passed to the current level through the following formula as a feature image at the current level:

[0088] ;

[0089] By passing the sub-template image at the previous level to the corresponding image combination at the current level as a feature image of the corresponding image combination at the current level, the image registration at the current level is guided.

[0090] S150. If there are unassigned image features in the feature set, return to the step of determining the current level and the level distance at the current level based on the principle of increasing levels until there are no unassigned image features in the feature set, and use the sub-template image at the current level as the target template image corresponding to the predetermined variable information.

[0091] After the sub-template image at the current level is determined, if there are no unassigned image features in the feature set, it means that the current level is the highest level. Therefore, use the sub-template image at the current level as the target template image corresponding to the predetermined variable information. If there are unassigned image features in the feature set, it means that the current level is not the highest level. Therefore, return to S120.

[0092] Figure 4 This is the comparison result between the target template image corresponding to the HCP-D dataset provided by the embodiment of the present invention and the target template image corresponding to the HCP-D dataset determined by other methods; Figure 5The comparison result between the target template image corresponding to the OASIS dataset provided by the embodiments of the present invention and the target template image corresponding to the OASIS dataset determined by other methods. Obviously, the target template image corresponding to the HCP-D dataset determined by the embodiments of the present invention has clearer anatomical structures compared to the target template image corresponding to the HCP-D dataset constructed by other methods. For example Figure 4 the axial plane, sagittal plane, and coronal plane from top to bottom in Figure 5 and the axial plane, sagittal plane, and coronal plane from top to bottom in

[0093] In summary, the method described in the embodiments of the present invention has a large capacity for accommodating large-scale image groups, strong adaptability to highly heterogeneous groups among individuals, and can generate clearer and unbiased central template images, that is, target template images, in a relatively short time.

[0094] For the technical solution of the template image determination method provided by the embodiments of the present invention, since there is an association relationship between the image features and the predetermined variable information, and the feature image is determined based on the image features, there is also an association relationship between the feature image and the predetermined variable information; since the feature clusters at the current level are determined based on the current level distance, and the current level distance is greater than the level distance of the previous level, the feature clusters at the current level include all the feature images within the feature clusters at the previous level, so a nested sequence of feature clusters can be obtained. Therefore, it is allowed to replace the image features belonging to the previous level in the feature clusters at the current level with the sub-template images at the previous level, and replace each image feature in the feature clusters at the current level with the corresponding feature image to obtain the image combination at the current level. After repeated iteration, the image combination at the highest level and the sub-template image of this image combination, that is, the target template image, can be obtained. The transmission of the sub-template image can significantly improve the accuracy of the sub-template image determined at the current level, thereby improving the accuracy of the target template image corresponding to the pre-variable information.

[0095] Figure 6 Another flowchart of the template image determination method provided by the embodiments of the present invention. This embodiment is used to refine the image combination determination step in the above embodiment, as Figure 6 shown, the method includes:

[0096] S210. Determine the feature set corresponding to the set of scanned images and the feature images corresponding to each image feature in the feature set. The set of scanned images includes the scanned images of multiple target objects, and the feature set includes the image features in each scanned image that have an association relationship with the predetermined variable information.

[0097] S2201. Based on the principle of increasing levels, determine the current level and the level distance at the current level, where the level distance at the current level is greater than the level distance at the previous level.

[0098] S2202. Select at least one feature cluster at the current level from the feature set according to the level distance.

[0099] Among at least one feature cluster at the current level, the feature distance between each image feature belonging to the same feature cluster and at least one other image feature in the feature cluster is less than or equal to the predetermined level distance (threshold) at the current level; the feature distance between image features belonging to different feature clusters is greater than the predetermined level distance (threshold) at the current level. That is to say, compared with the similarity between image features within the same feature cluster, the similarity between image features belonging to different feature clusters is lower.

[0100] As Figure 7 shown, the first level includes a first feature cluster and a second feature cluster, where the first feature cluster includes image features , image features and image features , and the second feature cluster includes image features , image features , image features and image features . The feature distance between image features and image features , and the feature distance between image features and image features are both less than the predetermined level distance at the current level; the feature distance between image features and image features , the feature distance between image features and image features , the feature distance between image features and image features as well as the feature distance between image features and image features are all less than the predetermined level distance at the current level. The image features , image features and image features in the first feature cluster and the image features in the second feature cluster have a feature distance greater than the predetermined level distance (threshold) at the current level.

[0101] As Figure 7 can be seen, each feature cluster can be represented as a connected graph, the image features can be understood as vertices, and the line connecting two image features and less than the predetermined level distance (threshold) is the edge.

[0102] S230. Replace the image features belonging to the previous level in the corresponding feature clusters at the current level with at least one sub-template image from the previous level, and replace the remaining image features in the at least one feature cluster at the current level with corresponding feature images, to obtain at least one image combination at the current level.

[0103] In this embodiment, different feature clusters at each level correspond to different image combinations, so as to improve the similarity between pairwise feature images in each image combination. Since one image combination corresponds to one sub-template image, each level will correspond to one or more sub-template images.

[0104] Exemplarily, there are two image combinations at the previous level, which are the A1 image combination and the A2 image combination respectively. There are three feature clusters at the current level, and the three feature clusters include the B1 feature cluster, the B2 feature cluster, and the B3 feature cluster. Among them, the B1 feature cluster includes the image features corresponding to the respective feature images in the A1 image combination, the B2 feature cluster includes the image features corresponding to the respective feature images in the A2 image combination, and the B3 feature cluster is a newly added feature cluster relative to the previous level. In this way, there are three image combinations at the current level, and the three image combinations respectively include the C1 image combination corresponding to the B1 feature cluster, the C2 image combination corresponding to the B2 feature cluster, and the C3 image combination corresponding to the B3 feature cluster. Specifically, the C1 image combination includes the sub-template image corresponding to the A1 image combination and one or more target feature images. The one or more target feature images are the feature images corresponding to the remaining image features after removing all the image features associated with the A1 image combination from the B1 feature cluster. Among them, the distance between the sub-template image corresponding to the A1 image combination and each of the one or more target feature images is equal to the distance between the center of the A1 image combination and each of the one or more target feature images; the C2 image combination includes the sub-template image corresponding to the A2 image combination and one or more target feature images. The one or more target feature images are the feature images corresponding to the remaining image features after removing all the image features associated with the A2 image combination from the B2 feature cluster; in the C2 image combination, the distance between the sub-template image corresponding to the A2 image combination and each of the one or more target feature images is equal to the distance between the center of the A2 image combination and each of the one or more target feature images; the C3 image combination includes the feature images corresponding to the respective image features in the B3 feature cluster.

[0105] Exemplarily, as Figure 7 shown, there are a first feature cluster and a second feature cluster at the first level. The first feature cluster includes the image feature , the image feature , and the image feature , and the second feature cluster includes the image feature , Image Features , Image Features and Image Features . There are two image combinations in the first level. The two image combinations are the A1 image combination including feature image , feature image and feature image , and the A2 image combination including feature image , feature image , feature image and feature image . Among them, the identification subscripts of the corresponding image features and feature images are the same. For example, the feature image of image feature is , the sub-template image corresponding to the A1 image combination is , and the sub-template image corresponding to the A2 image combination is . There is one feature cluster in the second level. The feature cluster includes all the image features in the first feature cluster, all the image features in the second feature cluster, and image feature . Correspondingly, there is one image combination in the second level. The image combination includes sub-template image , sub-template image and feature image . It should be noted that Figure 7 in which S means passing the image features of the previous feature cluster to the feature cluster of the next level, or the feature cluster of the next level includes the feature cluster of the previous level; R means passing the sub-template image corresponding to the image combination at the previous level to the corresponding image combination at the next level, or when the current level is the highest level, taking the sub-template image at the current level as the target template image .

[0106] In one embodiment, it is determined whether the number of feature images included in each of the at least one image combination is within a predetermined number range; if the number of feature images included in any one of the at least one image combination is greater than the upper boundary of the predetermined number range, the hierarchical distance at the current level is reduced, and the step of selecting at least one feature cluster at the current level from the feature set according to the hierarchical distance is returned; if the number of feature images included in any one of the at least one image combination is less than the lower boundary of the predetermined number range, and the feature set includes unassigned image features, the hierarchical distance at the current level is increased, and the step of selecting at least one feature cluster at the current level from the feature set according to the hierarchical distance is returned; if the number of image combinations at the current level is 1, and the number of feature images included in the image combination at the current level is less than the lower boundary of the predetermined number range, and at the same time the feature set does not include unassigned image features, the image combination at the current level is taken as the image combination at the highest level.

[0107] Specifically, the predetermined number range can be expressed as:

[0108] ;

[0109] Wherein, represents the number of feature images included in the image combination labeled m, and are respectively used to control the minimum and maximum values of the number of feature images included in the corresponding image combination. It can be understood that in the case where the target image combination includes the sub-template image corresponding to the corresponding image combination at the previous level, this sub-template image is also regarded as a feature image of the target image combination.

[0110] By controlling the number of feature images in each image combination within a predetermined number range, the accuracy of the sub-template images corresponding to each image combination can be ensured, thereby improving the accuracy of the target template image.

[0111] S240. Based on local popularity shrinkage, process each of the at least one image combination at the current level to obtain the sub-template images corresponding to each of the at least one image combination at the current level.

[0112] After determining the at least one image combination at the current level, determine the sub-template images corresponding to each of the at least one image combination respectively.

[0113] S250. If there are unallocated image features in the feature set, return the step of determining the current level and the hierarchical distance at the current level based on the principle of increasing levels until there are no unallocated image features in the feature set, and use the sub-template image at the current level as the target template image corresponding to the predetermined variable information.

[0114] If the current level includes one feature cluster and there are no unallocated image features in the feature set (see Figure 7 the last level in), at this time the current level is the highest level; if the current level includes at least two feature clusters and there are no unallocated image features in the feature set, then transfer the sub-template images corresponding to the respective image combinations of the at least two feature clusters to the next level as the feature images in the only image combination at the next level, and the next level is the highest level. Determine the sub-template image based on the only image combination and use the sub-template image as the target registration template.

[0115] In one embodiment, after determining at least one sub-template image at the current level, adjust the spatial positions of the respective feature images in the corresponding image combinations at all previous levels based on the at least one sub-template image at the current level, so that the respective feature images in the corresponding image combinations at all previous levels and the sub-template image at the current level are in the same coordinate space, achieving the technical effects of completing the spatial registration of all registered images based on the local image registration result and improving the capacity of the method of the present invention to accommodate an image group (image combination).

[0116] Exemplarily, the first level includes a first image combination and a second image combination, the second level includes a third image combination and a fourth image combination, wherein the third image combination includes the sub-template image corresponding to the first image combination, and the fourth image combination includes the sub-template image corresponding to the second image combination; the third level includes a fifth image combination and a sixth image combination, wherein the fifth image combination includes the sub-template image corresponding to the third image combination, and the sixth image combination includes the sub-template image corresponding to the third image combination. Then, based on the sub-template image corresponding to the third image combination, the spatial positions of the respective feature images in the first image combination can be adjusted so that the respective feature images in the first image combination and the sub-template image corresponding to the third image combination are in the same coordinate space. Based on the fifth image combination, the spatial positions of the respective feature images in the third image combination and the first image combination can be adjusted in sequence so that the respective feature images in the third image combination and the first image combination and the sub-template image corresponding to the fifth image combination are in the same coordinate space.

[0117] In one embodiment, the spatial positions of the respective feature images in the corresponding image combinations at all previous levels are corrected based on the sub-template image through the following formula:

[0118] ;

[0119] in, is the current level, express arrive The feature images in the corresponding feature combinations in all levels, Indicates the current level The deformation field from each feature image in to the group center (the sub-template image at the current level).

[0120] Based on the foregoing embodiments, the embodiments of the present invention determine at least one feature cluster at each level based on the feature distance between each image feature and the predetermined level distance at each level, and then use at least one sub-template image at the previous level to replace the image features belonging to the previous level in the corresponding feature cluster at the current level, and replace the remaining image features in the at least one feature cluster at the current level with the corresponding feature image, to obtain at least one image combination at the current level, and then determine the sub-template images corresponding to the at least one image combination, thereby improving the universality of the image registration method and the accuracy of the determined target template image.

[0121] Figure 8 This is a flow chart of a template image determination method provided in an embodiment of the present invention. This embodiment is used to refine the sub-template image determination step in the above embodiment. Figure 8 As shown, the method includes:

[0122] S310, determining a feature set corresponding to a scanned image set and a feature image corresponding to each image feature in the feature set, wherein the scanned image set includes scanned images of multiple target objects, and the feature set includes image features in each scanned image that are associated with predetermined variable information.

[0123] S320, based on the principle of increasing levels, determine the current level and the level distance under the current level, select the feature cluster under the current level from the feature set according to the level distance, and the current level distance is greater than the level distance under the previous level.

[0124] S330, determining an image combination at the current level by replacing a feature image belonging to the previous level in a feature cluster at the current level with a sub-template image at the previous level.

[0125] S3401. For the image combination at the current level, determine the deformation field corresponding to each feature image.

[0126] In one embodiment, for the image combination at the current level, an inverse deformation field between each of the feature images and other feature images is determined; a deformation field weight between each of the feature images and other feature images is determined according to the distance between each of the feature images and other feature images; an inverse deformation field weighted sum between each of the feature images and other feature images is determined according to the deformation field weight between each of the feature images and other feature images and the inverse deformation field between each of the feature images and other feature images; and the inverse of the weighted sum is used as the deformation field corresponding to each of the feature images.

[0127] Specifically, the inverse deformation field is the inverse of the corresponding deformation field. Figure 9 The relationship between the deformation field and the inverse deformation field between the feature images is shown. For each feature image in the image combination, the image distance between the current feature image and other feature images is determined. and the deformation field weight corresponding to the image distance.

[0128] In one embodiment, the value corresponding to the Gaussian function with a standard deviation of the image distance is used as the deformation field weight corresponding to the image distance. This embodiment can simply and quickly determine the required deformation field weight.

[0129] Specifically, the Gaussian function is:

[0130] ;

[0131] where is the standard deviation.

[0132] In one embodiment, the product of the reciprocal of the image distance and a set coefficient is used as the deformation field weight corresponding to the image distance. This embodiment can simply and quickly determine the required deformation field weight.

[0133] The deformation field between the current feature image and other feature images can be expressed as:

[0134] ;

[0135] where is the inverse deformation field, and the inverse of the above inverse deformation field weighted sum is used as the deformation field of the current feature image.

[0136] In one embodiment, for the image combination at the current level, based on a pre-trained adaptive multi-level registration network, the deformation fields of each of the feature images with respect to other feature images are determined (see Figure 10); Determine the deformation fields corresponding to each of the feature images according to the deformation fields between each of the feature images and other feature images; wherein, the loss function used in the training of the adaptive multi-level registration network includes a velocity field constraint term; the velocity field constraint term includes the difference between the velocity field corresponding to the deformation field in the label and the velocity field corresponding to the predicted deformation field.

[0137] Based on the pre-trained adaptive multi-level registration network, adopting a multi-resolution strategy and always learning the deformation field at the whole-brain image level, it can maintain the continuity of the deformation field and achieve fast and high-performance registration of scanned images with different deformation lengths.

[0138] To ensure the reversibility and topological invariance of the deformation field, the adaptive multi-level registration model in this embodiment uses the velocity field to determine the network parameters so that the model has the property of diffeomorphism. Among them, the deformation field is defined as follows:

[0139] ;

[0140] Among them, " " is a composite operator, representing the combined operation of functions, used to describe the continuous action of multiple transformations. Specifically, it means to first perform deformation through , and then perform deformation through ; is the identity transformation, t is the time, integrate the velocity field v within the unit time, and use the scaling and squaring operation with a time step T = 7 to obtain the final deformation field . The diffeomorphic adaptive multi-level registration model has an adaptive ability for large or complex deformations, can ensure fast and accurate registration within the image combination, and avoid the accumulation of global registration errors.

[0141] In one embodiment, the adaptive multi-level registration model is trained using the OASIS dataset and the HCP-D dataset. In the test stage, directly call them to estimate the deformation field for subsequent hierarchical group registration. The network training is completed on the set electronic device using the PyTorch toolkit in Python, such as on an NVIDIA TITAN Xp 12GB GPU server. For the adaptive multi-level registration network, use the Adam optimizer to update the model parameters, set the learning rate to 0.001, and the network is trained for 15 rounds to converge. It has the same parameters as the feature extraction network except that the weight parameter of the regularization loss is set to 2.5 to obtain the optimal result. Save the training result with the best performance on the validation set and use it for testing. After the model passes the test, it can be used.

[0142] S3402. Deform each feature image according to the deformation field corresponding to each feature image to update the image combination.

[0143] Deform each feature image according to the deformation field corresponding to each feature image in the image combination, so that each feature image moves towards the center of the image combination, and an updated image combination is obtained. It can be understood that, compared with the image combination, the similarity between the feature images in the updated image combination is higher.

[0144] S3403. If the sum of the distances between pairwise feature images in the updated image combination does not meet the set distance condition, then return to the step of determining the deformation field corresponding to each feature image for the image combination at the current level.

[0145] If the sum of the distances between pairwise feature images in the updated image combination is greater than the set distance, then return to S3401.

[0146] S3304. If the sum of the distances between pairwise image features in the updated image combination meets the set distance condition, take the mean value of all the feature images in the updated image combination as the sub-template image at the current level.

[0147] As Figure 11 shown, in the absence of a reference prior, after several iterations, the total distance between all the feature images in the image combination is minimized. At this time, it is determined that each feature image has been gradually moved to the center of the image combination. When all the feature images in the image combination are very close on the manifold of the image combination, therefore, take the mean value of all the feature images in the image combination at this time as the sub-template image ( ). Among them, the sum of the distances can be determined based on the following formula:

[0148] ;

[0149] Among them, and represent the feature images registered at the th iteration in the image combination, that is, . Among them, represents the feature image registered at the th registration.

[0150] The sub-template image corresponding to the image combination can be determined by the following formula:

[0151] .

[0152] S350. If the current level is not the highest level, then return to the step of determining the current level based on the principle of increasing levels until the current level is the highest level, and take the sub-template image as the target template image corresponding to the predetermined variable information.

[0153] In the technical solution provided by the embodiment of the present invention, since the deformation fields corresponding to the respective feature images are based on the deformation field information between each feature and all other images in the image combination, in the case where the image combination is not an image combination at the first level, the movement of each feature image in the image combination involves the participation of the corresponding sub-template image at the previous level included therein. Therefore, in this embodiment, the accuracy of the movement of each feature image in the image combination is relatively high, and the accuracy of the sub-template image determined based on the movement results of the feature images is also relatively high.

[0154] Figure 12 It is a schematic structural diagram of a template image determination device provided by an embodiment of the present invention. As Figure 12 shown, the device includes:

[0155] A feature module 41, configured to determine a feature set corresponding to a set of scanned images and feature images corresponding to each image feature in the feature set, the set of scanned images including scanned images of multiple target objects, and the feature set including image features in each of the scanned images that have an associated relationship with predetermined variable information;

[0156] A feature cluster determination module 42, configured to determine a current level and a level distance at the current level based on the principle of increasing levels, and select a feature cluster at the current level from the feature set according to the level distance, where the current level distance is greater than the level distance at the previous level;

[0157] An image combination module 43, configured to determine an image combination at the current level by replacing the feature images belonging to the previous level in the feature cluster at the current level with the sub-template images at the previous level;

[0158] A template image module 44, configured to process the image combination based on local popular contraction to obtain a sub-template image corresponding to the image combination;

[0159] An iteration module 45, configured to, if there are unallocated image features in the feature set, return to the step of determining the current level and the level distance at the current level based on the principle of increasing levels until there are no unallocated image features in the feature set, and use the sub-template image at the current level as the target template image corresponding to the predetermined variable information.

[0160] In one embodiment, the feature cluster module 42 is specifically configured to:

[0161] Select at least one feature cluster at the current level from the feature set according to the level distance;

[0162] The image combination module 43 is specifically configured to:

[0163] Determine at least one image combination at the current level by replacing, respectively, at least one sub-template image at the previous level with the image features at the previous level belonging to the corresponding feature cluster at the current level;

[0164] The template image module 44 is specifically configured to:

[0165] Process at least one image combination at the current level respectively based on local popularity shrinkage to obtain sub-template images respectively corresponding to at least one image combination at the current level.

[0166] In one embodiment, the image combination module 43 is configured to:

[0167] Determine whether the number of feature images included in each of the at least one image combination is within a predetermined number range;

[0168] If the number of feature images included in any one of the at least one image combination is greater than the upper boundary of the predetermined number range, reduce the level distance at the current level, and return to the step of selecting at least one feature cluster at the current level from the feature set according to the level distance;

[0169] If the number of feature images included in any one of the at least one image combination is less than the lower boundary of the predetermined number range, and the feature set includes unassigned image features, increase the level distance at the current level, and return to the step of selecting at least one feature cluster at the current level from the feature set according to the level distance;

[0170] If the number of image combinations at the current level is 1, and the number of feature images included in the image combination at the current level is less than the lower boundary of the predetermined number range, and at the same time the feature set does not include unassigned image features, then use the image combination at the current level as the image combination at the highest level.

[0171] In one example, the template image module 44 includes:

[0172] A deformation field unit, configured to determine a deformation field corresponding to each of the feature images for the image combination at the current level;

[0173] An update unit, configured to deform each of the feature images according to the deformation field corresponding to each of the feature images to update the image combination;

[0174] An iteration unit, configured to, if the sum of the distances between any two of the feature images in the updated image combination does not meet the set distance condition, return to the step of determining the deformation field corresponding to each of the feature images for the image combination at the current level;

[0175] A template image determination unit, configured to use the mean value of all feature images in the updated image combination as the sub-template image of the current level if the sum of the distances between pairwise image features in the updated image combination meets a set distance condition.

[0176] In one example, the deformation field unit is configured to:

[0177] Determine the inverse deformation field between each feature image and other feature images for the image combination at the current level;

[0178] Determine the deformation field weight between each feature image and other feature images according to the distance between each feature image and other feature images;

[0179] Determine the weighted sum of the inverse deformation fields between each feature image and other feature images according to the deformation field weights between each feature image and other feature images and the inverse deformation fields between each feature image and other feature images;

[0180] Use the inverse of the weighted sum as the deformation field corresponding to each feature image.

[0181] In one embodiment, the template image module 44 is configured to:

[0182] For the image combination at the current level, determine the deformation field between each feature image and other feature images respectively based on a pre-trained adaptive multi-level registration network;

[0183] Determine the deformation field corresponding to each feature image according to the deformation fields between each feature image and other feature images respectively;

[0184] Wherein, the loss function used in the training of the adaptive multi-level registration network includes a velocity field constraint term;

[0185] The velocity field constraint term includes the difference between the velocity field corresponding to the deformation field in the label and the velocity field corresponding to the predicted deformation field.

[0186] In one embodiment, the pre-variable state information is age information or disease state information.

[0187] In one embodiment, after the template image module 44, it further includes:

[0188] A reverse adjustment module, configured to adjust the spatial positions of all feature images at the previous level based on the sub-template image of the current level, so that all feature images at the previous level and the sub-template image of the current level are located in the same coordinate space.

[0189] In the technical solution of the template image determination device provided by the embodiments of the present invention, since there is an association relationship between the image features and the predetermined variable information, and the feature image is determined based on the image features, there is also an association relationship between the feature image and the predetermined variable information; since the feature clusters at the current level are determined based on the current level distance, and the current level distance is greater than the level distance of the previous level, the feature clusters at the current level include all the feature images within the feature clusters at the previous level, so a sequence of feature clusters with a nested relationship can be obtained. Therefore, it is allowed to replace the image features belonging to the previous level in the feature clusters at the current level with the sub-template images at the previous level, and replace each image feature in the feature clusters at the current level with the corresponding feature image to obtain the image combination at the current level. After repeated iteration, the image combination at the highest level and the sub-template image of this image combination, that is, the target template image, can be obtained. The transmission of the sub-template image can significantly improve the accuracy of the sub-template image determined at the current level, thereby improving the accuracy of the target template image corresponding to the pre-variable information.

[0190] The template image determination device provided by the embodiments of the present invention can execute the template image determination method provided by any embodiment of the present invention, and has the corresponding functional modules and beneficial effects for executing the method.

[0191] Figure 13 FIG. shows a schematic structural diagram of an electronic device 10 that can be used to implement the embodiments of the present invention. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smart phones, wearable devices (such as helmets, glasses, watches, etc.) and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present invention described and / or claimed herein.

[0192] As Figure 13 shown, the electronic device 10 includes at least one processor 11, and a memory communicatively connected to at least one processor 11, such as a read-only memory (ROM) 12, a random access memory (RAM) 13, etc. Among them, the memory stores a computer program executable by at least one processor. The processor 11 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 12 or the computer program loaded from the storage unit 18 into the random access memory (RAM) 13. In the RAM 13, various programs and data required for the operation of the electronic device 10 can also be stored. The processor 11, the ROM 12, and the RAM 13 are connected to each other through a bus 14. The input / output (I / O) interface 15 is also connected to the bus 14.

[0193] Multiple components in the electronic device 10 are connected to the I / O interface 15, including: an input unit 16, such as a keyboard, a mouse, etc.; an output unit 17, such as various types of displays, speakers, etc.; a storage unit 18, such as a magnetic disk, an optical disc, etc.; and a communication unit 19, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 19 allows the electronic device 10 to exchange information / data with other devices via a computer network such as the Internet and / or various telecommunication networks.

[0194] The processor 11 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the processor 11 include but are not limited to a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The processor 11 executes the various methods and processes described above, such as the template image determination method.

[0195] In some embodiments, the template image determination method can be implemented as a computer program, which is tangibly contained in a computer-readable storage medium, such as the storage unit 18. In some embodiments, part or all of the computer program can be loaded and / or installed onto the electronic device 10 via the ROM 12 and / or the communication unit 19. When the computer program is loaded into the RAM 13 and executed by the processor 11, one or more steps of the template image determination method described above can be executed. Alternatively, in other embodiments, the processor 11 can be configured to execute the template image determination method by any other suitable means (e.g., by means of firmware).

[0196] The various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special-purpose or general-purpose programmable processor, and can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit the data and instructions to the storage system, the at least one input device, and the at least one output device.

[0197] A computer program for implementing the method of the present invention can be written in any combination of one or more programming languages. These computer programs can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when the computer programs are executed by the processor, the functions / operations specified in the flowchart and / or block diagram are implemented. The computer programs can be executed entirely on the machine, partially on the machine, as a stand-alone software package partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0198] In the context of the present invention, a computer-readable storage medium can be a tangible medium that can contain or store a computer program for use by or in connection with an instruction execution system, apparatus, or device. The computer-readable storage medium can include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. Alternatively, the computer-readable storage medium can be a machine-readable signal medium. More specific examples of the machine-readable storage medium would include an electrical connection based on 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 foregoing.

[0199] In order to provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or a trackball) through which the user can provide input to the electronic device. Other kinds of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0200] The systems and techniques described herein can be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer having a graphical user interface or a web browser through which a user can interact with an implementation of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system can be interconnected with each other by digital data communication in any form or medium (e.g., a communication network). Examples of communication networks include: local area network (LAN), wide area network (WAN), blockchain network, and the Internet.

[0201] A computing system can include a client and a server. The client and the server are generally remote from each other and typically interact through a communication network. The client-server relationship is created by computer programs that run on respective computers and have a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or a cloud host, which is a host product in the cloud computing service system, solving the defects of difficult management and weak business scalability existing in traditional physical hosts and VPS services.

[0202] An embodiment of the present invention also provides a computer program product, including a computer program which, when executed by a processor, implements the template image determination method provided in any embodiment of the present application.

[0203] In the process of implementing the computer program product, computer program code for performing the operations of the present invention can be written in one or more programming languages or combinations thereof. The programming languages include object-oriented programming languages such as Java, Smalltalk, C++, and also include conventional procedural programming languages such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, executed as a stand-alone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer can 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 can be connected to an external computer (e.g., by connecting through the Internet using an Internet service provider).

[0204] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in the present invention can be executed in parallel, sequentially, or in a different order, as long as the desired results of the technical solution of the present invention can be achieved, and no limitation is imposed herein.

[0205] The above specific embodiments do not constitute a limitation on the protection scope of the present invention. Those skilled in the art should understand that various modifications, combinations, sub - combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention shall be included within the protection scope of the present invention.

Claims

1. A template image determination method, characterized in that: include: Determine a feature set corresponding to a scanned image set and a feature image corresponding to each image feature in the feature set, wherein the scanned image set includes scanned images of a plurality of target objects, and the feature set includes image features in each of the scanned images that are associated with predetermined variable information; Based on the principle of increasing levels, determine the current level and the level distance under the current level, select the feature cluster under the current level from the feature set according to the level distance, and the current level distance is greater than the level distance under the previous level; Using the sub-template image at the previous level to replace the image features belonging to the previous level in the feature cluster at the current level, and replacing the remaining image features in the feature cluster at the current level with the corresponding feature image, to obtain the image combination at the current level; Processing the image combination based on local manifold contraction to obtain a sub-template image corresponding to the image combination; If there are unassigned image features in the feature set, return to the step of determining the current level and the distance between levels below the current level based on the principle of increasing levels until there are no unassigned image features in the feature set, and use the sub-template image under the current level as the target template image corresponding to the predetermined variable information.

2. The method according to claim 1, characterized in that The selecting a feature cluster at the current level from the feature set according to the level distance includes: Selecting at least one feature cluster at the current level from the feature set according to the level distance; The method of using the sub-template image at the previous level to replace the feature image belonging to the previous level in the feature cluster at the current level to determine the image combination at the current level includes: Using at least one sub-template image at the previous level to replace the image features belonging to the previous level in the corresponding feature cluster at the current level, and replacing the remaining image features in the at least one feature cluster at the current level with the corresponding feature image, to obtain at least one image combination at the current level; The step of processing the image combination based on local manifold shrinkage to obtain a sub-template image corresponding to the image combination includes: At least one image combination at the current level is processed respectively based on local popular contraction to obtain sub-template images corresponding to the at least one image combination at the current level.

3. The method according to claim 2, characterized in that After determining at least one image combination at the current level, the method further includes: Determining whether the number of feature images included in each of the at least one image combination is within a predetermined number range; If the number of feature images included in any image combination of the at least one image combination is greater than the upper boundary of the predetermined number range, reducing the level distance at the current level, and returning to the step of selecting at least one feature cluster at the current level from the feature set according to the level distance; If the number of feature images included in any image combination of the at least one image combination is less than the lower boundary of the predetermined number range, and the feature set includes unassigned image features, then increasing the level distance at the current level, and returning to the step of selecting at least one feature cluster at the current level from the feature set according to the level distance; If the number of image combinations at the current level is 1, and the number of feature images included in the image combination at the current level is less than the lower boundary of the predetermined number range, and the feature set does not include unassigned image features, the image combination at the current level is taken as the image combination at the highest level.

4. The method according to claim 1, characterized in that: The step of processing the image combination based on local manifold shrinkage to obtain a sub-template image corresponding to the image combination includes: For the image combination at the current level, determining a deformation field corresponding to each of the feature images; Deforming each of the feature images according to the deformation field corresponding to each of the feature images to update the image combination; If the sum of the distances between the two feature images in the updated image combination does not meet the set distance condition, returning to the step of determining the deformation field corresponding to each feature image for the image combination at the current level; If the sum of the distances between the image features in each pair in the updated image combination meets the set distance condition, the average of all feature images in the updated image combination is used as the sub-template image of the current level.

5. The method according to claim 4, characterized in that The step of determining the deformation field corresponding to each of the feature images for the image combination at the current level includes: For the image combination at the current level, determining an inverse deformation field between each of the feature images and other feature images; Determine the deformation field weight between each of the feature images and other feature images according to the distance between each of the feature images and other feature images; Determine a weighted sum of inverse deformation fields between each of the feature images and other feature images according to the deformation field weights between each of the feature images and other feature images and the inverse deformation fields between each of the feature images and other feature images; The inverse of the weighted sum is used as the deformation field corresponding to each of the feature images.

6. The method according to claim 5, characterized in that The step of determining the deformation field corresponding to each of the feature images for the image combination at the current level includes: For the image combination at the current level, determining the deformation field of each of the feature images and other feature images respectively based on a pre-trained adaptive multi-level registration network; Determine the deformation field corresponding to each of the characteristic images according to the deformation fields of each of the characteristic images and other characteristic images respectively; Wherein, the loss function used in the training of the adaptive multi-level registration network includes a velocity field constraint term; The velocity field constraint term includes the difference between the velocity field corresponding to the deformation field in the label and the velocity field corresponding to the predicted deformation field.

7. The method according to claim 2, characterized in that After the image combination is processed based on local manifold shrinkage to obtain a sub-template image corresponding to the image combination, the method further includes: Based on at least one sub-template image of the current level, the spatial position of each feature image in all corresponding image combinations at the previous level is adjusted so that each feature image in all corresponding image combinations at the previous level and the sub-template image of the current level are located in the same coordinate space.

8. An electronic device, characterized in that: The electronic device comprises: at least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the template image determination method described in any one of claims 1-7.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the template image determination method described in any one of claims 1-7 when executed.

10. A computer program product, characterized in that The computer program product comprises a computer program, and when the computer program is executed by a processor, the computer program implements the template image determination method according to any one of claims 1 to 7.

Images