Pyramid cyclic deformation medical image registration method, device, equipment and medium

Abstract

The present application relates to a pyramid cyclic deformation medical image registration method, device, equipment and medium, wherein the method comprises: obtaining a fixed medical image and a mobile medical image to be registered and preprocessing them to obtain an initial fixed medical image and an initial mobile medical image; extracting features from the initial fixed medical image and the initial mobile medical image through an encoder of a target pyramid cyclic deformation model to obtain a fixed medical feature map and a mobile medical feature map; performing feature fusion and image deformation on the fourth-layer fixed medical feature map and the mobile medical feature map to obtain an initial cyclic fusion feature map and a first target deformation field; performing cyclic convolution processing based on the initial cyclic fusion feature map, the first three layers of fixed medical feature maps and the mobile medical feature maps to output a second target deformation field, and performing deformation processing on the mobile image through the second target deformation field to obtain a target registration result. The present invention improves the efficiency and accuracy of medical image registration.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a pyramid cyclic deformation medical image registration method, device, equipment and medium. Background Art

[0002] Deformable image registration has always been an important research direction in the field of medical imaging, and is crucial for preoperative planning, surgical navigation, disease diagnosis and follow-up. For a given moving image and fixed image, deformable registration crops the moving image to match the fixed image by predicting a physically possible nonlinear transformation. In this way, information from unidirectional / multimodal images can be fused into a common coordinate system to assist doctors in diagnosis and treatment. Although a variety of algorithms have been developed, accurate and efficient registration remains challenging, especially in the case of large deformations and high computational requirements for voxel data.

[0003] In recent years, deep neural network-based registration techniques have become a powerful standard for large-scale medical image registration due to their fast inference time and accuracy that is generally superior to traditional iterative methods. Deep registration models use the entire training set as the optimization target to train network parameters during the training phase, rather than using a pair of images. After training, these deep registration models typically take only a few seconds to infer the deformation field of an image pair. However, due to the iterative optimization during the deformation estimation process, most traditional registration methods can be computationally intensive and time-consuming, and have difficulty handling complex volumetric deformations, which are often encountered in medical image registration tasks. Although various advanced registration models have been proposed, accurate and efficient deformable registration remains challenging, especially for handling large volumetric deformations. Therefore, there is an urgent need for a medical image registration method to improve the efficiency and accuracy of medical image registration. Summary of the invention

[0004] The purpose of the embodiments of the present application is to propose a pyramid cyclic deformation medical image registration method, device, equipment and medium to improve the efficiency and accuracy of medical image registration.

[0005] In order to solve the above technical problems, the present application provides a pyramid cyclic deformation medical image registration method, including:

[0006] Acquire a fixed medical image and a mobile medical image to be registered;

[0007] Preprocessing the fixed medical image and the moving medical image respectively to obtain an initial fixed medical image and an initial moving medical image;

[0008] The encoder of the target pyramid cyclic deformation model performs feature extraction on the initial fixed medical image and the initial mobile medical image respectively to obtain a fixed medical feature map and a mobile medical feature map, wherein the fixed medical feature map and the mobile medical feature map both include four layers of feature maps;

[0009] Performing feature fusion and image deformation processing on the fourth-layer fixed medical feature map and the fourth-layer mobile medical feature map through the decoder of the target pyramid cyclic deformation model to obtain an initial cyclic fusion feature map and a first target deformation field;

[0010] Based on the initial circular fusion feature map, the first three layers of fixed medical feature maps and the first three layers of mobile medical feature maps, circular convolution processing is performed in their respective corresponding decoders to output a second target deformation field, and the mobile feature map is deformed by the second target deformation field to obtain a target alignment result.

[0011] In order to solve the above technical problems, the present application provides a medical image registration device with pyramid cyclic deformation, comprising:

[0012] A medical image acquisition unit, used for acquiring a fixed medical image and a mobile medical image to be registered;

[0013] An image preprocessing unit, used to preprocess the fixed medical image and the moving medical image respectively to obtain an initial fixed medical image and an initial moving medical image;

[0014] A feature extraction unit, configured to perform feature extraction on the initial fixed medical image and the initial mobile medical image respectively through an encoder of a target pyramid cyclic deformation model to obtain a fixed medical feature map and a mobile medical feature map, wherein the fixed medical feature map and the mobile medical feature map both include four-layer feature maps;

[0015] An image deformation processing unit, used for performing feature fusion and image deformation processing on the fourth-layer fixed medical feature map and the fourth-layer mobile medical feature map through a decoder of the target pyramid cyclic deformation model to obtain an initial cyclic fusion feature map and a first target deformation field;

[0016] The registration result generating unit is used to perform circular convolution processing in the respective corresponding decoders based on the initial circular fusion feature map, the first three layers of fixed medical feature maps and the first three layers of mobile medical feature maps to output a second target deformation field, and deform the mobile feature map through the second target deformation field to obtain a target registration result.

[0017] In order to solve the above technical problems, a technical solution adopted by the present invention is: providing a computer device, including one or more processors; a memory for storing one or more programs, so that the one or more processors implement the medical image registration method of pyramid cyclic deformation described in any one of the above.

[0018] In order to solve the above technical problems, a technical solution adopted by the present invention is: a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the medical image registration method of pyramid cyclic deformation described in any one of the above is implemented.

[0019] The embodiment of the present invention provides a pyramid cyclic deformation medical image registration method, device, equipment and medium. The method includes: obtaining a fixed medical image and a mobile medical image to be registered; preprocessing the fixed medical image and the mobile medical image respectively to obtain an initial fixed medical image and an initial mobile medical image; extracting features from the initial fixed medical image and the initial mobile medical image respectively through an encoder of a target pyramid cyclic deformation model to obtain a fixed medical feature map and a mobile medical feature map, wherein the fixed medical feature map and the mobile medical feature map both include four layers of feature maps; performing feature fusion and image deformation processing on the fourth layer of fixed medical feature map and the fourth layer of mobile medical feature map through a decoder of the target pyramid cyclic deformation model to obtain an initial cyclic fusion feature map and a first target deformation field; performing cyclic convolution processing in the decoders corresponding to the initial cyclic fusion feature map, the first three layers of fixed medical feature map and the first three layers of mobile medical feature map to output a second target deformation field, and deforming the mobile feature map through the second target deformation field to obtain a target registration result. The embodiment of the present invention separately models the features of the mobile medical image and the fixed medical image, and then performs circular convolution processing and image deformation processing on the mobile medical image and the fixed medical image in the decoder, and generates a coarse-to-fine and step-by-step circular prediction deformation field, and finally generates a target registration image based on the predicted deformation field, which is beneficial to improving the efficiency and accuracy of medical image registration. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the scheme in the present application, a brief introduction is given below to the drawings required for use in the description of the embodiments of the present application. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0021] Figure 1 It is a flowchart of the implementation process of the pyramid cyclic deformation medical image registration method provided in the embodiment of the present application;

[0022] Figure 2 It is a schematic diagram of the overall structure of the network provided in the embodiment of the present application;

[0023] Figure 3 is a schematic diagram of the encoder structure provided in an embodiment of the present application;

[0024] Figure 4 is a schematic diagram of the decoder structure provided in an embodiment of the present application;

[0025] Figure 5 It is a flowchart for implementing a sub-process in the pyramid cyclic deformation medical image registration method provided in an embodiment of the present application;

[0026] Figure 6 It is a flowchart for implementing a sub-process in the pyramid cyclic deformation medical image registration method provided in an embodiment of the present application;

[0027] Figure 7 It is a flowchart for implementing a sub-process in the pyramid cyclic deformation medical image registration method provided in an embodiment of the present application;

[0028] Figure 8 It is a flowchart for implementing a sub-process in the pyramid cyclic deformation medical image registration method provided in an embodiment of the present application;

[0029] Fig. 9 It is a flowchart for implementing a sub-process in the pyramid cyclic deformation medical image registration method provided in an embodiment of the present application;

[0030] Fig.10 It is a flowchart for implementing a sub-process in the pyramid cyclic deformation medical image registration method provided in an embodiment of the present application;

[0031] Fig.11 Schematic diagram of a medical image registration device with pyramid cyclic deformation provided in an embodiment of the present application;

[0032] Fig.12 It is a schematic diagram of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0033] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by technicians in the technical field of the present application; the terms used in the specification of the application herein are only for the purpose of describing specific embodiments and are not intended to limit the present application; the terms "including" and "having" and any variations thereof in the specification and claims of the present application and the above-mentioned drawings are intended to cover non-exclusive inclusions. The terms "first", "second", etc. in the specification and claims of the present application or the above-mentioned drawings are used to distinguish different objects, not to describe a specific order.

[0034] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0035] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the accompanying drawings.

[0036] The present invention is described in detail below with reference to the accompanying drawings and embodiments.

[0037] It should be noted that the pyramid cyclic deformation medical image registration method provided in the embodiment of the present application is generally executed by a server, and accordingly, the pyramid cyclic deformation medical image registration device is generally configured in the server.

[0038] See also Figures 1 to 4 , Figure 1 A specific implementation of the pyramid cyclic deformation medical image registration method is shown. Figure 2 It is a schematic diagram of the overall structure of the network provided in the embodiment of the present application; Figure 3 is a schematic diagram of the encoder structure provided in an embodiment of the present application; Figure 4 It is a schematic diagram of the decoder structure provided in an embodiment of the present application.

[0039] It should be noted that if there are substantially the same results, the method of the present invention is not limited to Figure 1 The process sequence shown is limited to the following steps:

[0040] S1: Obtain a fixed medical image and a mobile medical image to be registered.

[0041] In the embodiment of the present application, the medical image may be a brain magnetic resonance image and an abdominal CT image, or other medical images. Specifically, the medical image is acquired, and two images are randomly selected from the medical images of the same modality as the fixed medical image and the mobile medical image. In addition, the fixed medical image and the mobile medical image may be specified from the medical images.

[0042] S2: Preprocessing the fixed medical image and the moving medical image respectively to obtain an initial fixed medical image and an initial moving medical image.

[0043] In the embodiment of the present application, the fixed medical image and the mobile medical image are cropped to obtain the cropped fixed medical image and the mobile medical image, and the cropped fixed medical image and the mobile medical image are normalized so that the gray value of the image is between 0 and 1, thereby obtaining the initial fixed medical image I f and the initial mobile medical image I m .

[0044] S3: extracting features from the initial fixed medical image and the initial moving medical image respectively through an encoder of the target pyramid cyclic deformation model to obtain a fixed medical feature map and a moving medical feature map.

[0045] Wherein, the fixed medical feature map and the mobile medical feature map both include four layers of feature maps.

[0046] The embodiment of the present application designs a pyramid cycle structure, and the pyramid network is a U-net structure as a whole. First, the dual encoders extract features from the initial fixed medical image and the initial moving medical image respectively to obtain a fixed medical feature map and a moving medical feature map, and then the pyramid cycle decoder obtains a series of deformation subfields, and finally all deformation subfields are cascaded to obtain the final deformation field deformation moving image, and obtain the target registration image.

[0047] See also Figure 5 , Figure 5 A specific implementation of step S3 is shown, which is described in detail as follows:

[0048] S31: Inputting the initial fixed medical image and the initial moving medical image into the target pyramid cyclic deformation model.

[0049] S32: The hierarchical features of the initial fixed medical image and the initial mobile medical image are extracted respectively through the four-layer convolution module of the non-shared weight residual convolution encoder, and the extracted hierarchical features are down-sampled through the last three layers of the convolution module to obtain the fixed medical feature map and the mobile medical feature map.

[0050] In an embodiment of the present application, the target pyramid cyclic deformation model is a U-net structure, which is a pyramid cyclic deformation model after model training. The encoder in the embodiment of the present application mainly adopts a residual network design, which is a powerful feature extraction backbone network. The encoder adopts a dual-branch strategy that does not share weights. Such a strategy can encode the initial fixed medical image and the initial mobile medical image separately to improve performance. Furthermore, the encoder uses a 4-layer convolution module to extract hierarchical features, and the number of channels in each layer is doubled. The initial fixed medical image and the initial mobile medical image are input into their respective encoders, and the feature map M is generated after convolution, instance normalization and leakage ReLU units. 1 and F 1 Except for the first convolution block in the encoder, there is no downsampling. The following three layers first downsample the feature map by a convolution with a step size of 2, and enter the residual module after each layer of downsampling. The final output feature map size is (1 / 8) of the original 3 , generating a fixed medical feature map {M 1 ,M 2 ,M 3 ,M 4} and mobile medical feature map {F 1 ,F 2 ,F 3 ,F 4}.

[0051] S4: Performing feature fusion and image deformation processing on the fourth-layer fixed medical feature map and the fourth-layer mobile medical feature map through the decoder of the target pyramid cyclic deformation model to obtain an initial cyclic fusion feature map and a first target deformation field.

[0052] In the present application example, v s ,φ s ,φ s ,s=1,2,…,n 4 +n 3 +n 2 +n 1 Represent the velocity field, residual field and deformation field at the sth level of the cycle, n 1 、n 2 、n 3 、n 4 They represent the number of loop layers from shallow to deep layers in the decoder. The first convolution module is CConv, which represents a two-level convolution → instance normalization → leaky ReLU module; the first convolution module is used to generate a fusion feature map iC lrepresents the fused feature map generated by level l CConv, where l = 1, 2, 3, 4 corresponds to the level of the encoder. The second convolution module is RegBlock, which represents the convolution module that generates the residual field, where Reg convolution is weighted N(0,1e -5 ) is sampled from it, and the 3-channel convolution layer with the bias initialized to 0 is used to generate the velocity field, and the ss layer is used to ensure differential homeomorphism. The third convolution module is DConv, which represents a layer of convolution → instance normalization → leaky ReLU, which is used to reduce the number of channels of the fused feature map to generate D l , l = 1,2,3,4, so as to cycle.

[0053] See also Figure 6 , Figure 6 A specific implementation of step S4 is shown, which is described in detail as follows:

[0054] S41: splicing the fourth-layer fixed medical feature map and the fourth-layer mobile medical feature map to obtain a spliced ​​feature map, and performing feature fusion on the spliced ​​feature map through the first convolution module of the outermost decoder in the target pyramid cyclic deformation model to obtain an initial fused feature map.

[0055] S42: Inputting the initial fusion feature map into the second convolution module for convolution processing to generate a first deformation field, and performing image deformation processing on the fourth layer mobile medical feature map according to the first deformation field to obtain a first deformation feature map.

[0056] In the embodiment of the present application, since the coarsest layer of the decoder part initially has only the fourth layer of fixed medical feature map M 4 With the fourth layer mobile medical feature map F 4 , and no feature map is fused. Therefore, in order to unify the number of channels, only the first convolution module is used in this layer. Specifically, the fourth layer fixes the medical feature map M 4 and the fourth layer mobile medical feature map F 4 The spliced ​​feature map is then convolved by the first convolution module to obtain the initial fusion feature map C. 4 (D 4 ), and then the initial fusion feature map C 4 (D 4 ) is input to the second convolution module for convolution processing to generate the first deformation field φ 1 , and according to the first deformation field φ 1 For the fourth layer mobile medical feature map F 4 The image is deformed to obtain the first deformation feature map, where the deformation action is completed by the spatial transformation network. It is expressed as follows:

[0057]

[0058] S43: Performing circular convolution processing based on the initial fused feature map, the fourth-layer fixed medical feature map and the first deformation feature map to generate the initial circular fused feature map and the first target deformation field.

[0059] See also Figure 7 , Figure 7 A specific implementation of step S43 is shown, which is described in detail as follows:

[0060] S431: splicing the initial fusion feature map, the fourth-layer fixed medical feature map and the first deformation feature map to obtain a first spliced ​​feature map.

[0061] S432: Perform convolution processing on the first spliced ​​feature map through the first convolution module to obtain a second fused feature map.

[0062] S433: Input the second fused feature map into the second convolution module to generate a first deformation sub-field, and fuse the first deformation sub-field with the first deformation field to obtain a first deformation total field.

[0063] S434: Perform image deformation processing on the first deformation feature map based on the first deformation total field to obtain the first cyclic deformation feature map, and determine whether the first preset number of cycles is reached. If not, re-execute the circular convolution processing, and output the deformation sub-field during each circular convolution processing, and fuse the deformation sub-field with the deformation total field at the end of the previous cycle to perform image deformation processing, until the first preset number of cycles is reached, and the initial circular fusion feature map and the first target deformation field are obtained.

[0064] In the embodiment of the present application, the initial fusion feature map, the fourth layer fixed medical feature map and the first deformation feature map are first spliced ​​to obtain a first spliced ​​feature map, and then the first spliced ​​feature map is convoluted by the first convolution module to obtain a second fusion feature map C. 4 , each time the second fusion feature map C 4 The first deformation subfield is generated by inputting the second fused feature map into the second convolution module, and the first deformation subfield is fused with the first deformation field to obtain the first deformation total field, and the deformation subfield generated subsequently is fused with the deformation total field. Then, the first deformation feature map is subjected to image deformation processing in the first deformation total field to obtain a deformed image, and then the process returns to step S431, and steps S431 to S434 are re-executed to perform circular convolution processing until the number of cycles reaches the first preset number of cycles, and the initial circular fusion feature map and the first target deformation field are obtained. It is expressed by the following formula:

[0065]

[0066]

[0067]

[0068] in, represents the deformation field φ obtained previously s-1 Deformation and movement feature map M l , the deformation action is completed by the spatial transformation network, Represents the fusion formula of the deformation field. This loop is performed by taking the output C of CConv 4 , the deformed moving feature map is spliced ​​with the fixed feature map, and then C is obtained through CCBlock 4 , each time C 4 The deformation subfield is obtained through RegBlock and then fused with the total deformation field. Through this cycle, the network can always pass on the high-level semantics of the previous layer. Since the number of input channels of the initial convolution module and the subsequent convolution modules in the coarsest layer is different, the weights of the initial convolution module and the subsequent ones are not shared in this layer, but the deformation solution convolution (RegConv) is shared. In the subsequent l-1 layer, both modules share weights.

[0069] S5: Based on the initial circular fusion feature map, the first three layers of fixed medical feature maps and the first three layers of mobile medical feature maps, circular convolution processing is performed in their respective corresponding decoders to output a second target deformation field, and the mobile feature map is deformed by the second target deformation field to obtain a target alignment result.

[0070] See also Figure 8 , Figure 8 A specific implementation of step S5 is shown, which is described in detail as follows:

[0071] S51: In any layer of the inner three-layer decoder, the deformation field and fusion feature map input by the previous layer of decoder are obtained.

[0072] In the embodiment of the present application, the same processing process is adopted in any layer decoder of the inner three layers of decoders, so in the embodiment of the present application, only the specific embodiment of any layer decoder is described.

[0073] S52: multiplying the deformation field input by the previous layer decoder to obtain a multiplied deformation field.

[0074] S53: Perform trilinear interpolation upsampling on the deformation feature map input by the previous layer decoder to obtain an initial basic feature map.

[0075] In the embodiment of the present application, after the above steps are subjected to circular convolution, the initial circular fusion feature map and the first target deformation field are obtained. In the embodiment of the present application, the first target deformation field φ k Multiply it by 2 to get the multiplied deformation field, and then convert the initial deformation feature map C l+1 The multiplied deformation field is subjected to trilinear interpolation upsampling to obtain a sampled deformation feature map and a sampled deformation field, and finally the sampled deformation feature map is subjected to image deformation processing by the sampled deformation field to obtain a second deformed image. The upsampling formula is as follows:

[0076] D l =upConv(C l+1 ),φ k = up(2·φ k );

[0077] Among them, up(.) represents trilinear interpolation upsampling, upConv(.) represents 3D transposed convolution, instance normalization and leaky ReLU, and the number of output channels is the same as the deformed image M l (F l ) to allow for looping.

[0078] S54: deforming the mobile medical feature map input by the layer decoder through the multiplied deformation field to obtain a second deformation feature map.

[0079] S55: Performing circular convolution processing in the decoder of this layer based on the second deformation feature map, the multiplied deformation field and the initial basic feature map to obtain the deformation field and fusion feature map output by the decoder of this layer.

[0080] See also Fig. 9 , Fig. 9 A specific implementation of step S55 is shown, which is described in detail as follows:

[0081] S551: splicing the second deformation feature map, the sampled deformation field and the sampled deformation feature map to obtain a second spliced ​​feature map.

[0082] S552: Obtain a third fused feature map by performing convolution processing on the second spliced ​​feature map on the first convolution module.

[0083] S553: ​​Perform convolution processing on the third fused feature map through a third convolution module so that the number of channels of the third fused feature map is consistent with that of the mobile feature map input by the decoder of this layer, and obtain a basic feature map.

[0084] S554: Input the basic feature map into the second convolution module to generate a second deformation subfield, and fuse the second deformation subfield with the sampled deformation field to obtain an initial second deformation total field.

[0085] S555: Perform image deformation processing on the second deformation feature map based on the initial second deformation total field to obtain the first cyclic deformation feature map of the decoder of this layer, and determine whether the second preset number of cycles has been reached. If not, re-execute the circular convolution processing, and output the deformation sub-field during each circular convolution processing, and fuse the deformation sub-field with the deformation total field at the end of the previous cycle to perform image deformation processing, until the second preset number of cycles is reached, and the deformation field and fused feature map output by the decoder of this layer are obtained.

[0086] In an embodiment of the present application, the second deformation feature map, the sampled deformation field and the sampled deformation feature map are spliced ​​to obtain a second spliced ​​feature map, and then the second spliced ​​feature map is convolved by the first convolution module to obtain a third fused feature map, and then the third fused feature map is convolved by the third convolution module to make the third fused feature map consistent with the number of channels of the mobile feature map input by the decoder of this layer, to obtain a basic feature map, and then the basic feature map is input into the second convolution module to generate a second deformation subfield, and the second deformation subfield is fused with the sampled deformation field to obtain an initial second deformation total field, and the second deformation feature map is subjected to image deformation processing based on the initial second deformation total field, and then returns to step S541, and re-executes steps S541 to S545, so that a circular convolution process is performed until the number of cycles reaches the second preset number of cycles. It should be noted that the first preset number of cycles and the second preset number of cycles are set according to actual conditions and are not limited here. The circular convolution process is as follows:

[0087]

[0088]

[0089]

[0090] D l =DConv(C l );

[0091] Among them, C l The number of channels of DConv is consistent with the number of input channels of CConv. The function of DConv is to reduce the number of channels so that it is consistent with the mobile feature map or fixed feature map at this level. Figure 1 This loop is done by taking the output D of the third convolution module DConv l, the deformed mobile feature map is concatenated with the fixed feature map, and then C is obtained through the first convolution module CConvBlock l , the obtained D l The number of channels is reduced by the third convolution module DConv to be consistent with the number of channels of the mobile feature map and the fixed feature map of this level, and the cycle is entered again. l The deformation subfield is obtained through the second convolution module RegBlock and then fused with the deformation total field. These modules can always share weights in the loop, which also allows high-level semantics to be transferred between layers. Finally, the decoder will output a semantically rich second target deformation field to deform the moving image on the first layer.

[0092] Furthermore, in order to balance efficiency and performance, the embodiment of the present application adopts an unbalanced loop strategy. Because looping at full resolution is very computationally intensive, in order to compensate for the accuracy, the present application loops once more at the penultimate layer and does not loop at the final full resolution, and finally sets it to n 1 ,n 2 ,n 3 ,n 4 =1,3,2,2.

[0093] S56: When all decoder circular convolution processes are completed, the second target deformation field is obtained, and the mobile feature map is deformed by the second target deformation field to obtain a target registration result.

[0094] See also Fig.10 , Fig.10 A specific implementation method before step S3 is shown, which is described in detail as follows:

[0095] S31: Collect a medical image dataset, and preprocess the medical image dataset to obtain an initial medical image dataset.

[0096] S32: Randomly select two images from the initial medical image dataset of the same modality as a fixed image and a moving image, and construct a training data sample based on the fixed image and the moving image.

[0097] S33: Training the pyramid cyclic deformation model based on a preset loss function and training data samples to obtain the target pyramid cyclic deformation model.

[0098] In the embodiment of the present application, before implementing step S3, the embodiment of the present application needs to train the pyramid cyclic deformation model to obtain a target pyramid cyclic deformation model. In the embodiment of the present application, a medical image data set (such as a brain magnetic resonance image and an abdominal CT image, etc.) is collected, and the medical image data set is subjected to preprocessing operations such as cropping and image normalization, and the grayscale value is between 0 and 1, and then two images are randomly selected from the initial medical image data set of the same modality as a fixed image and a moving image, and a training data sample is constructed based on the fixed image and the moving image, and finally the pyramid cyclic deformation model is trained based on a preset loss function and the training data sample to obtain a target pyramid cyclic deformation model.

[0099] Among them, the preset loss function is:

[0100]

[0101] Among them, L total is the total loss, λ is the weight of the regularization term, I f is the fixed image, I m is the moving image, φ is the deformation field, L reg is the residual loss, if the single-modality medical image matching criterion L sim is the normalized mutual loss, if the multimodal medical image matching criterion L sim is the local mutual information loss.

[0102] Furthermore, during the model training process, the Adam optimizer and the learning rate decay strategy of the embodiment of the present application are as follows:

[0103]

[0104] Among them, lr m Indicates the learning rate of the mth round, lr init It represents the learning rate of the initial round. The number of learning rounds depends on whether the loss decreases. The training can be stopped when it converges.

[0105] Furthermore, in the embodiment of the present application, the ss layer in the convolution layer of the deformation field uses a static velocity field to ensure differential homeomorphism properties, such as differentiability, reversibility and topology preservation, which can make the generated deformation field more realistic and reduce the local folding and flipping of the deformation image. The deformation field is defined as an ordinary differential equation about time t:

[0106]

[0107] When t = 0, φ (0) is the identity transformation, φ (1) This is the deformation field required by the embodiment of the present application. Further, the embodiment of the present application adopts the scaling and square method to integrate φ(t) From 0 to 1, specifically, it is obtained by the following cycle:

[0108]

[0109] Among them, the output v of the convolution solution is calculated for each level of deformation s Zoom to the initial After the above formula is repeated t times, the residual deformation field of this level is finally obtained.

[0110] In an embodiment of the present application, a fixed medical image and a mobile medical image to be registered are obtained; the fixed medical image and the mobile medical image are preprocessed respectively to obtain an initial fixed medical image and an initial mobile medical image; feature extraction is performed on the initial fixed medical image and the initial mobile medical image respectively through an encoder of a target pyramid cyclic deformation model to obtain a fixed medical feature map and a mobile medical feature map, wherein the fixed medical feature map and the mobile medical feature map both include four layers of feature maps; feature fusion and image deformation processing are performed on the fourth layer of fixed medical feature map and the fourth layer of mobile medical feature map through a decoder of the target pyramid cyclic deformation model to obtain an initial cyclic fusion feature map and a first target deformation field; based on the initial cyclic fusion feature map, the first three layers of fixed medical feature map and the first three layers of mobile medical feature map, cyclic convolution processing is performed in the respective corresponding decoders to output a second target deformation field, and the mobile feature map is deformed through the second target deformation field to obtain a target registration result.

[0111] The embodiment of the present invention uses a residual convolution encoder with non-shared weights to separately model the features of the moving image and the fixed image, and then uses a convolutional network in the decoder to solve the deformation subfield and continuously fuse the deformation subfield. The embodiment of the present invention uses a step-by-step recursive strategy with high-level semantic integration to predict the deformation field from coarse to fine and step-by-step loop, while ensuring the rationality of the differential homeomorphism of the deformation field. At the same time, due to the recursive strategy, the embodiment of the present invention can effectively realize deformable registration without the need for a separate affine pre-registration, which is conducive to improving the efficiency and accuracy of medical image registration.

[0112] Furthermore, the embodiments of the present application can be used to register two-dimensional images, three-dimensional images, multi-modal medical images, and multi-modal natural images, while some patents may limit the scope of application, such as being applicable only to two-dimensional images or three-dimensional images.

[0113] Please refer to Fig.11 , as a response to the above Figure 1 The present application provides an embodiment of a medical image registration device for pyramid cyclic deformation, and the device embodiment is similar to Figure 1Corresponding to the method embodiment shown, the device can be specifically applied to various electronic devices.

[0114] like Fig.11 As shown, the pyramid cyclic deformation medical image registration device of this embodiment includes: a medical image acquisition unit 61, an image preprocessing unit 62, a feature extraction unit 63, an image deformation processing unit 64 and a registration result generation unit 65, wherein:

[0115] The medical image acquisition unit 61 is used to acquire the fixed medical image and the moving medical image to be registered.

[0116] The image preprocessing unit 62 is used to preprocess the fixed medical image and the moving medical image respectively to obtain an initial fixed medical image and an initial moving medical image.

[0117] The feature extraction unit 63 is used to perform feature extraction on the initial fixed medical image and the initial mobile medical image respectively through the encoder of the target pyramid cyclic deformation model to obtain a fixed medical feature map and a mobile medical feature map, wherein the fixed medical feature map and the mobile medical feature map both include four-layer feature maps.

[0118] The image deformation processing unit 64 is used to perform feature fusion and image deformation processing on the fourth layer fixed medical feature map and the fourth layer mobile medical feature map through the decoder of the target pyramid cyclic deformation model to obtain an initial cyclic fusion feature map and a first target deformation field.

[0119] The registration result generating unit 65 is used to perform circular convolution processing in the respective corresponding decoders based on the initial circular fusion feature map, the first three layers of fixed medical feature maps and the first three layers of mobile medical feature maps to output a second target deformation field, and deform the mobile feature map through the second target deformation field to obtain a target registration result.

[0120] Further, the feature extraction unit 63 includes:

[0121] An image input unit, which inputs the initial fixed medical image and the initial moving medical image into the target pyramid cyclic deformation model;

[0122] The hierarchical feature extraction unit uses a four-layer convolution module of a residual convolution encoder with non-shared weights to extract hierarchical features from the initial fixed medical image and the initial moving medical image respectively, and downsamples the extracted hierarchical features through the last three layers of the convolution module to obtain the fixed medical feature map and the moving medical feature map.

[0123] Furthermore, the image deformation processing unit 64 includes:

[0124] an initial fused feature map generating unit, configured to splice the fourth-layer fixed medical feature map and the fourth-layer mobile medical feature map to obtain a spliced ​​feature map, and to perform feature fusion on the spliced ​​feature map through the first convolution module of the outermost decoder in the target pyramid cyclic deformation model to obtain an initial fused feature map;

[0125] A first deformation feature map generating unit is used to input the initial fusion feature map into the second convolution module for convolution processing to generate a first deformation field, and perform image deformation processing on the fourth layer mobile medical feature map according to the first deformation field to obtain a first deformation feature map;

[0126] The first circular convolution processing unit is used to perform circular convolution processing based on the initial fused feature map, the fourth layer fixed medical feature map and the first deformation feature map to generate the initial circular fused feature map and the first target deformation field.

[0127] Further, the first circular convolution processing unit includes:

[0128] A first splicing feature map generating unit, configured to splice the initial fusion feature map, the fourth-layer fixed medical feature map and the first deformation feature map to obtain a first splicing feature map;

[0129] A second fused feature map generating unit, configured to perform convolution processing on the first concatenated feature map through the first convolution module to obtain a second fused feature map;

[0130] A first deformation total field generating unit, configured to input the second fused feature map into the second convolution module to generate a first deformation subfield, and fuse the first deformation subfield with the first deformation field to obtain a first deformation total field;

[0131] The first target deformation field generating unit is used to perform image deformation processing on the first deformation feature map based on the first deformation total field to obtain the first cyclic deformation feature map, and determine whether the first preset number of cycles is reached. If not, the circular convolution processing is re-executed, and a deformation sub-field is output during each circular convolution processing, and the deformation sub-field is fused with the deformation total field at the end of the previous cycle to perform image deformation processing, until the first preset number of cycles is reached, and the initial circular fusion feature map and the first target deformation field are obtained.

[0132] Furthermore, the registration result generating unit 65 includes:

[0133] A feature map acquisition unit, used to acquire, in any layer of the inner three-layer decoder, the deformation field and fused feature map input by the previous layer of decoder;

[0134] The deformation field multiplication unit is used to multiply the deformation field input by the previous layer decoder to obtain the multiplied deformation field

[0135] An upsampling processing unit, used to perform trilinear interpolation upsampling on the deformation feature map input by the previous layer decoder to obtain an initial basic feature map;

[0136] A second deformation feature map generating unit, configured to perform deformation processing on the mobile medical feature map input by the layer decoder through the multiplied deformation field to obtain a second deformation feature map;

[0137] A second circular convolution processing unit, configured to perform circular convolution processing in the decoder of the layer based on the second deformation feature map, the multiplied deformation field and the initial basic feature map, so as to obtain a deformation field and a fused feature map output by the decoder of the layer;

[0138] The deformation processing unit is used to obtain the second target deformation field when all decoder circular convolution processes are completed, and to perform deformation processing on the moving feature map through the second target deformation field to obtain a target alignment result.

[0139] Further, the second circular convolution processing unit includes:

[0140] A second splicing feature map generating unit, used for splicing the second deformation feature map, the sampled deformation field and the sampled deformation feature map to obtain a second splicing feature map;

[0141] A third fused feature map generating unit, configured to obtain a third fused feature map by performing convolution processing on the second spliced ​​feature map;

[0142] A basic feature map generating unit, configured to obtain a third fused feature map by performing convolution processing on the first convolution module with the second concatenated feature map;

[0143] A basic feature map generating unit, configured to perform convolution processing on the third fused feature map through a third convolution module so that the number of channels of the third fused feature map is consistent with the number of channels of the mobile feature map input by the decoder of this layer, thereby obtaining a basic feature map;

[0144] an initial second deformation total field generating unit, configured to input the basic feature map into the second convolution module to generate a second deformation subfield, and fuse the second deformation subfield with the sampled deformation field to obtain an initial second deformation total field;

[0145] A loop processing unit is used to perform image deformation processing on the second deformation feature map based on the initial second deformation total field, obtain the first loop deformation feature map of the decoder of this layer, and determine whether the second preset number of cycles is reached. If not, re-execute the loop convolution processing, and output the deformation sub-field during each loop convolution processing, and fuse the deformation sub-field with the deformation total field at the end of the previous cycle to perform image deformation processing, until the second preset number of cycles is reached, and the deformation field and fused feature map output by the decoder of this layer are obtained.

[0146] Furthermore, the medical image acquisition unit 61 further includes:

[0147] A medical image data set acquisition unit, used for acquiring a medical image data set and preprocessing the medical image data set to obtain an initial medical image data set;

[0148] A training data sample construction unit, configured to randomly select two images from the initial medical image data set of the same modality as a fixed image and a moving image, and to construct a training data sample based on the fixed image and the moving image;

[0149] A model training unit, used for training the pyramid cyclic deformation model based on a preset loss function and training data samples to obtain the target pyramid cyclic deformation model;

[0150] Among them, the preset loss function is:

[0151] L total =L sim (I f ,I m φ)+λL reg (φ);

[0152] Among them, L total is the total loss, λ is the weight of the regularization term, I f is the fixed image, I m is the moving image, φ is the deformation field, L reg is the residual loss, if the single-modality medical image matching criterion L sim is the normalized mutual loss, if the multimodal medical image matching criterion L sim is the local mutual information loss.

[0153] To solve the above technical problems, the present application also provides a computer device. Fig.12 , Fig.12 This is a basic structural block diagram of the computer device in this embodiment.

[0154] The computer device 7 includes a memory 71, a processor 72, and a network interface 73 that are interconnected through a system bus. It should be noted that the figure only shows a computer device 7 having three components: a memory 71, a processor 72, and a network interface 73, but it should be understood that it is not required to implement all the components shown, and more or fewer components can be implemented instead. Among them, those skilled in the art can understand that the computer device here is a device that can automatically perform numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes but is not limited to microprocessors, application specific integrated circuits (ASIC), programmable gate arrays (FPGA), digital signal processors (DSP), embedded devices, etc.

[0155] Computer devices can be computing devices such as desktop computers, notebooks, PDAs, and cloud servers. Computer devices can interact with users through keyboards, mice, remote controls, touch pads, or voice control devices.

[0156] The memory 71 includes at least one type of readable storage medium, and the readable storage medium includes flash memory, hard disk, multimedia card, card-type memory (for example, SD or DX memory, etc.), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, disk, optical disk, etc. In some embodiments, the memory 71 can be an internal storage unit of the computer device 7, such as a hard disk or memory of the computer device 7. In other embodiments, the memory 71 can also be an external storage device of the computer device 7, such as a plug-in hard disk equipped on the computer device 7, a smart memory card (SmartMedia Card, SMC), a secure digital (Secure Digital, SD) card, a flash card, etc. Of course, the memory 71 can also include both the internal storage unit of the computer device 7 and its external storage device. In this embodiment, the memory 71 is generally used to store the operating system and various application software installed on the computer device 7, such as the program code of the medical image registration method of pyramid cyclic deformation, etc. In addition, the memory 71 can also be used to temporarily store various types of data that have been output or are to be output.

[0157] The processor 72 may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chips in some embodiments. The processor 72 is generally used to control the overall operation of the computer device 7. In this embodiment, the processor 72 is used to run the program code stored in the memory 71 or process data, such as running the program code of the above-mentioned pyramid cyclic deformation medical image registration method to implement various embodiments of the pyramid cyclic deformation medical image registration method.

[0158] The network interface 73 may include a wireless network interface or a wired network interface, and the network interface 73 is generally used to establish a communication connection between the computer device 7 and other electronic devices.

[0159] The present application also provides another embodiment, namely, providing a computer-readable storage medium, which stores a computer program, and the computer program can be executed by at least one processor to enable the at least one processor to perform the steps of a pyramid cyclic deformation medical image registration method as described above.

[0160] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, a disk, or an optical disk), and includes a number of instructions for a terminal device (which can be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) to execute the methods of each embodiment of the present application.

[0161] Obviously, the embodiments described above are only some embodiments of the present application, rather than all embodiments. The preferred embodiments of the present application are given in the accompanying drawings, but they do not limit the patent scope of the present application. The present application can be implemented in many different forms. On the contrary, the purpose of providing these embodiments is to make the understanding of the disclosure of the present application more thorough and comprehensive. Although the present application is described in detail with reference to the aforementioned embodiments, for those skilled in the art, it is still possible to modify the technical solutions recorded in the aforementioned specific implementation methods, or to perform equivalent replacement of some of the technical features therein. Any equivalent structure made using the contents of the specification and drawings of this application, directly or indirectly used in other related technical fields, is similarly within the scope of patent protection of this application.

Claims

1. A pyramid cyclic deformation medical image registration method, It is characterized in that include: Acquire a fixed medical image and a mobile medical image to be registered; Preprocessing the fixed medical image and the moving medical image respectively to obtain an initial fixed medical image and an initial moving medical image; The encoder of the target pyramid cyclic deformation model performs feature extraction on the initial fixed medical image and the initial mobile medical image respectively to obtain a fixed medical feature map and a mobile medical feature map, wherein the fixed medical feature map and the mobile medical feature map both include four layers of feature maps; Performing feature fusion and image deformation processing on the fourth-layer fixed medical feature map and the fourth-layer mobile medical feature map through the decoder of the target pyramid cyclic deformation model to obtain an initial cyclic fusion feature map and a first target deformation field; Based on the initial circular fusion feature map, the first three layers of fixed medical feature maps and the first three layers of mobile medical feature maps, circular convolution processing is performed in the respective corresponding decoders to output a second target deformation field, and the mobile medical feature map is deformed by the second target deformation field to obtain a target registration result; The method of performing circular convolution processing based on the initial circular fusion feature map, the first three layers of fixed medical feature maps, and the first three layers of mobile medical feature maps in their respective corresponding decoders to output a second target deformation field, and performing deformation processing on the mobile medical feature map through the second target deformation field to obtain a target registration result includes: In any decoder of the inner three layers, the deformation field and fused feature map input by the previous decoder are obtained; The deformation field input by the previous layer decoder is multiplied to obtain a multiplied deformation field; Perform trilinear interpolation upsampling on the deformation feature map input by the previous layer decoder to obtain the initial basic feature map; The mobile medical feature map input by the layer decoder is deformed by using the multiplied deformation field to obtain a second deformation feature map; Performing a circular convolution process in the decoder of the layer based on the second deformation feature map, the multiplied deformation field and the initial basic feature map to obtain a deformation field and a fused feature map output by the decoder of the layer; When all decoder circular convolution processes are completed, the second target deformation field is obtained, and the mobile medical feature map is deformed by the second target deformation field to obtain a target registration result.

2. The medical image registration method of pyramid cyclic deformation according to claim 1, It is characterized in that The encoder of the target pyramid cyclic deformation model extracts features from the initial fixed medical image and the initial moving medical image respectively to obtain a fixed medical feature map and a moving medical feature map, including: Inputting the initial fixed medical image and the initial moving medical image into the target pyramid cyclic deformation model; The four-layer convolution module of the non-shared weight residual convolution encoder is used to extract hierarchical features from the initial fixed medical image and the initial mobile medical image respectively, and the extracted hierarchical features are downsampled through the last three layers of the convolution module to obtain the fixed medical feature map and the mobile medical feature map.

3. The medical image registration method of pyramid cyclic deformation according to claim 1, It is characterized in that The decoder of the target pyramid cyclic deformation model performs feature fusion and image deformation processing on the fourth layer fixed medical feature map and the fourth layer mobile medical feature map to obtain an initial cyclic fusion feature map and a first target deformation field, including: The fourth layer of fixed medical feature map and the fourth layer of mobile medical feature map are spliced ​​to obtain a spliced ​​feature map, and the spliced ​​feature map is subjected to feature fusion through the first convolution module of the outermost decoder in the target pyramid cyclic deformation model to obtain an initial fused feature map; Inputting the initial fusion feature map into the second convolution module for convolution processing to generate a first deformation field, and performing image deformation processing on the fourth layer mobile medical feature map according to the first deformation field to obtain a first deformation feature map; Circular convolution processing is performed based on the initial fused feature map, the fourth-layer fixed medical feature map, and the first deformation feature map to generate the initial circular fused feature map and the first target deformation field.

4. The medical image registration method of pyramid cyclic deformation according to claim 3, It is characterized in that The performing circular convolution processing based on the initial fused feature map, the fourth-layer fixed medical feature map, and the first deformation feature map to generate the initial circular fused feature map and the first target deformation field includes: Splicing the initial fusion feature map, the fourth-layer fixed medical feature map, and the first deformation feature map to obtain a first spliced ​​feature map; Performing convolution processing on the first concatenated feature map through the first convolution module to obtain a second fused feature map; Inputting the second fused feature map into the second convolution module to generate a first deformation subfield, and fusing the first deformation subfield with the first deformation field to obtain a first deformation total field; Based on the first deformation total field, the first deformation feature map is subjected to image deformation processing to obtain the first cyclic deformation feature map, and it is determined whether the first preset number of cycles is reached. If not, the circular convolution processing is re-executed, and a deformation sub-field is output during each circular convolution processing, and the deformation sub-field is fused with the deformation total field at the end of the previous cycle to perform image deformation processing, until the first preset number of cycles is reached, and the initial circular fusion feature map and the first target deformation field are obtained.

5. The medical image registration method of pyramid cyclic deformation according to claim 4, It is characterized in that The method of performing a circular convolution process in the layer decoder based on the second deformation feature map, the multiplied deformation field and the sampled deformation feature map to obtain the deformation field and the fused feature map output by the layer decoder includes: splicing the second deformation feature map, the multiplied deformation field and the sampled deformation feature map to obtain a second spliced ​​feature map; Obtain a third fused feature map by performing convolution processing on the first convolution module with the second spliced ​​feature map; Performing convolution processing on the third fused feature map through a third convolution module so that the number of channels of the third fused feature map is consistent with the number of channels of the mobile medical feature map input by the decoder of this layer, thereby obtaining a basic feature map; Inputting the basic feature map into the second convolution module to generate a second deformation subfield, and fusing the second deformation subfield with the multiplied deformation field to obtain an initial second deformation total field; Based on the initial second deformation total field, the second deformation feature map is deformed, and the first cyclic deformation feature map of the decoder of this layer is obtained. It is determined whether the second preset number of cycles is reached. If not, the circular convolution process is re-executed, and a deformation sub-field is output during each circular convolution process. The deformation sub-field is fused with the deformation total field at the end of the previous cycle to perform image deformation processing until the second preset number of cycles is reached, and the deformation field and fused feature map output by the decoder of this layer are obtained.

6. The medical image registration method of pyramid cyclic deformation according to any one of claims 1 to 5, It is characterized in that Before extracting features from the initial fixed medical image and the initial moving medical image respectively through the encoder of the target pyramid cyclic deformation model to obtain a fixed medical feature map and a moving medical feature map, the method further includes: Collecting a medical image data set, and preprocessing the medical image data set to obtain an initial medical image data set; Randomly selecting two images from the initial medical image dataset of the same modality as a fixed image and a moving image, and constructing a training data sample based on the fixed image and the moving image; Training the pyramid cyclic deformation model based on a preset loss function and training data samples to obtain the target pyramid cyclic deformation model; Among them, the preset loss function is: ； in, is the total loss, is the weight of the regularization term, is the fixed image, For the moving image, is the deformation field, is the residual loss, if the single-modality medical image matching criterion To normalize mutual loss, if the multimodal medical image matching criterion is the local mutual information loss.

7. A medical image registration device with pyramid cyclic deformation, It is characterized in that include: A medical image acquisition unit, used for acquiring a fixed medical image and a mobile medical image to be registered; An image preprocessing unit, used to preprocess the fixed medical image and the moving medical image respectively to obtain an initial fixed medical image and an initial moving medical image; A feature extraction unit, configured to perform feature extraction on the initial fixed medical image and the initial mobile medical image respectively through an encoder of a target pyramid cyclic deformation model to obtain a fixed medical feature map and a mobile medical feature map, wherein the fixed medical feature map and the mobile medical feature map both include four-layer feature maps; An image deformation processing unit, used for performing feature fusion and image deformation processing on the fourth-layer fixed medical feature map and the fourth-layer mobile medical feature map through a decoder of the target pyramid cyclic deformation model to obtain an initial cyclic fusion feature map and a first target deformation field; A registration result generating unit, configured to perform a circular convolution process in the respective corresponding decoders based on the initial circular fusion feature map, the first three layers of fixed medical feature maps, and the first three layers of mobile medical feature maps to output a second target deformation field, and perform deformation processing on the mobile medical feature map through the second target deformation field to obtain a target registration result; The registration result generating unit comprises: A feature map acquisition unit, used to acquire, in any layer of the inner three-layer decoder, the deformation field and fused feature map input by the previous layer of decoder; A deformation field multiplication unit is used to multiply the deformation field input by the previous layer decoder to obtain a multiplied deformation field; An upsampling processing unit, used to perform trilinear interpolation upsampling on the deformation feature map input by the previous layer decoder to obtain an initial basic feature map; A second deformation feature map generating unit, configured to perform deformation processing on the mobile medical feature map input by the layer decoder through the multiplied deformation field to obtain a second deformation feature map; Performing a circular convolution process in the decoder of the layer based on the second deformation feature map, the multiplied deformation field and the initial basic feature map to obtain a deformation field and a fused feature map output by the decoder of the layer; The second circular convolution processing unit is used to obtain the second target deformation field when all decoder circular convolution processes are completed, and to perform deformation processing on the mobile medical feature map through the second target deformation field to obtain a target registration result.

8. A computer device, It is characterized in that The invention comprises a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the medical image registration method of pyramid cyclic deformation according to any one of claims 1 to 6 is implemented.

9. A computer-readable storage medium, It is characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the pyramid cyclic deformation medical image registration method according to any one of claims 1 to 6 is implemented.

Images