Multi-modal medical image registration method and system, computer equipment and medium

Abstract

The invention provides a multi-modal medical image registration method and system, computer equipment and a medium, and belongs to the technical field of medical image processing. The method comprises the following steps: firstly, pre-processing a paired multi-modal medical image, and then converting a source modal image into a composite image of target modal gray scale distribution by using a structure-preserving modal conversion network; network parameters are optimized by constructing positive and negative sample pairs and combining a comparison loss function with gray level consistency loss and structure retention loss to form a composite loss function. And finally, inputting the synthesized image and the fixed image into an unsupervised registration network, and outputting a deformation field to obtain a registered medical image. By adopting the multi-modal medical image registration method and system, the computer equipment and the medium, the registration precision can be effectively improved, the robustness and discrimination capability of the model to a local region can be enhanced, the accurate alignment of key anatomical structures can be ensured, and the clinical application prospect is good.

Description

Technical Field

[0001] The present invention relates to the technical field of medical image processing, and in particular to a multimodal medical image registration method, system, computer equipment and medium. Background Art

[0002] Medical image registration is a key technology for clinical diagnosis and treatment planning, especially in multimodal scenarios (such as CT and MRI fusion). Its goal is to achieve complementary integration of information from different imaging modalities through spatial alignment. However, multimodal images exhibit significant grayscale distribution inconsistencies and structural differences due to differences in imaging principles. The grayscale similarity metrics relied on by traditional registration methods fail when there are significant grayscale differences between modalities. This causes the optimization process to easily fall into local optimality, leading to problems such as structural mismatch and blurred boundaries, which severely restricts registration accuracy and clinical applicability.

[0003] Existing multimodal medical image registration methods primarily include those based on traditional optimization, supervised deep learning, unsupervised learning, and modality conversion strategies, but all have significant shortcomings. Traditional optimization methods rely on handcrafted features, resulting in poor registration performance and high computational complexity when modality grayscale differences are large. Supervised deep learning methods rely on labels of real or simulated deformation fields, but real deformation fields are difficult to obtain, and there is a gap between simulated samples and real conditions, which limits model generalization performance. The similarity metric used in unsupervised learning methods fails when grayscale differences are significant, making structural mismatches more likely. Modality conversion-based strategies suffer from structural information distortion, blurred boundaries, and complex training, making it particularly difficult to maintain structural integrity in fine-grained anatomical regions.

[0004] While existing research attempts to mitigate modality discrepancies through modality conversion or feature alignment, two major shortcomings remain: First, existing modality conversion methods fail to effectively preserve structural boundaries and fine-grained texture information. The converted images lack structural representation and cannot provide accurate supervision for registration. Second, most methods neglect the registration accuracy of local regions, especially small-scale anatomical structures, resulting in unstable matching in key areas, limiting their clinical application value. Therefore, improving the structural consistency and alignment accuracy of multimodal image registration requires addressing two key issues: interference from modality grayscale differences and robustness of local region registration. Summary of the Invention

[0005] The purpose of the present invention is to provide a multimodal medical image registration method, system, computer equipment and medium, which can effectively solve the problem of decreased registration accuracy caused by grayscale distribution differences in multimodal medical image registration, ensure the dual consistency of the converted image at the grayscale and structural levels, and effectively improve the accuracy of image positioning and the robustness of registration.

[0006] To achieve the above object, the present invention provides a multimodal medical image registration method, comprising the following steps: Step S1: Acquire and preprocess paired multimodal medical images to obtain preprocessed source modality images and target modality images; Step S2: input the preprocessed source modality image into the structure-preserving modality conversion network, and output a synthetic image with the same grayscale distribution as the preprocessed target modality image; Based on the source modality image processed by the structure-preserving modality conversion network and the preprocessed target modality image, the corresponding image blocks are extracted to construct positive sample pairs, and multi-source negative sample pairs are generated through cross-position selection and noise perturbation. The contrast loss is calculated based on normalized cosine similarity, and a composite loss function is formed with grayscale consistency loss and structure preservation loss to optimize the parameters of the structure-preserving modal conversion network. Step S3: Input the synthetic image and the fixed image into the unsupervised registration network, extract multi-scale features through the cascaded encoder-decoder structure, and output the deformation field after fusion; Step S4: outputting a registered medical image based on the deformation field and the synthesized image.

[0007] Preferably, in step S1, the preprocessing includes: cropping, denoising and grayscale normalization.

[0008] Preferably, in step S2, the structure-preserving modal transfer network is based on the DenseUNet architecture, an Early Transition module is introduced in the encoding path, and a three-dimensional linear upsampling module combined with a post-convolution module is used in the decoding path.

[0009] Preferably, in step S2, the process of constructing the positive sample pair is as follows: From the pre-processed source modality image and the preprocessed target modality image Extract the image blocks at the same position and ; represents the source modality image patch, represents the target modality image patch; The source modality image is converted to Mapped to pseudo target modality image patch , construct positive sample pairs ,in represents the mode conversion network.

[0010] Preferably, in step S2, cross-position selection of negative samples includes selecting from the target modality image Other image blocks that do not correspond to the same position And the noise perturbation includes the target modality image block Add mixed disturbance to generate noisy negative samples ,in, represents the Gaussian noise term, represents the symbolic function, Represents contrast loss , is the target modality image patch.

[0011] Preferably, in step S2, the composite loss function is as follows: ; in, represents the compound loss, represents the grayscale consistency loss, represents the loss of structural preservation, represents a composite image, represents the weight, represents the gradient operator (implemented using Sobel filtering), Indicates the number of blocks for each pair of images, Indicates the number of negative samples corresponding to each image block, Indicates the sequence number of the image block that is currently a positive sample, Indicates the pseudo target modality image block and the The feature similarity between negative samples across positions, represents the temperature parameter, Indicates the The feature similarity of the noisy negative samples, It represents the normalized cosine similarity, which is calculated as follows: ; in, represents the inner product of the pseudo target modality image block and the target modality image block, express norm.

[0012] Preferably, in step S3, the loss function of the unsupervised registration network is It consists of similarity loss and smoothness loss, and the calculation formula is as follows: ; in, represents the hyperparameter that balances the similarity term and the regularization term, represents the image domain, represents the gradient of the deformation field, represents the normalized cross-correlation, Deformation field Apply to images , Represents the composite operation of functions, Indicates a fixed image.

[0013] The present invention also provides a multimodal medical image registration system, comprising: A data acquisition and preprocessing module is used to acquire and preprocess paired multimodal medical images to obtain preprocessed source modality images and target modality images; The structure-preserving modality conversion module is built on the DenseUNet architecture and includes an early transition module in the encoding path and a 3D linear upsampling combined with a post-convolution module in the decoding path. The structure-preserving modality conversion module receives the preprocessed source modality image and outputs a composite image with the same grayscale distribution as the target modality. Contrastive learning optimization module, including: Positive sample construction unit: extracts image blocks with corresponding spatial positions from the source modality image processed by the structure-preserving modality conversion network and the preprocessed target modality image to construct positive sample pairs; Multi-source negative sample generation unit: generates negative sample pairs through cross-position selection and noise perturbation; Composite loss calculation unit: Calculates contrast loss based on normalized cosine similarity, combines grayscale consistency loss and structure preservation loss to form a composite loss function, and optimizes the parameters of the modal conversion module; The unsupervised registration module uses a cascaded encoder-decoder structure to receive the synthetic image and the fixed image, and outputs the deformation field after multi-scale feature fusion; The registration image generation module outputs the registered medical image based on the deformation field and the synthetic image.

[0014] The present invention also provides a computer device, comprising a memory and a processor, wherein the memory is used to store instructions, and the processor is used to execute the instructions to implement the multimodal medical image registration method as described above.

[0015] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the multimodal medical image registration method as described above.

[0016] Therefore, the present invention adopts the above-mentioned multimodal medical image registration method, system, computer device and medium, and the beneficial technical effects are as follows: (1) In terms of network design, based on the DenseUNet backbone architecture, an early transition module is introduced in the encoding path, and a three-dimensional linear upsampling module combined with a post-convolution module is used in the decoding path to achieve direct grayscale mapping from the source modality to the target modality, effectively avoiding the error accumulation caused by the intermediate modality. At the same time, the representation of the anatomical region is enhanced through the structure preservation module. A composite loss function is proposed by combining the grayscale consistency loss, structure preservation loss and ratio loss function, further enhancing the ability to preserve key anatomical structures during modality conversion. This design ensures the dual consistency of the converted image at the grayscale and structural levels, providing a high-quality strength foundation for subsequent single-modality image registration.

[0017] (2) The image patch-level contrast learning mechanism is introduced to fully explore the spatial correspondence between the source modality and the target modality in the local area. By constructing image patch pairs with the same spatial position as positive samples and different spatial positions as negative samples, the network can focus on the spatial consistency modeling of key structures, thereby significantly improving the semantic alignment ability of local areas and the cross-modal structure recognition ability. In the registration task, this structural alignment-based perception feature can effectively improve the model's positioning accuracy and registration robustness for key parts (such as lesions or cerebellum).

[0018] (3) A multi-source negative sample strategy is introduced into the contrastive learning mechanism. By combining position perturbation and noise perturbation, the negative sample construction form is enriched, and the discriminability and generalization ability of contrast features are improved. Among them, position perturbation changes the spatial semantic relationship of image patches, and noise perturbation simulates the intensity uncertainty commonly seen in modal changes. The two work together to enhance the network's robust modeling ability for structural alignment and effectively suppress the interference of modal artifacts and noise on structure preservation and alignment accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 Flowchart for modality conversion and contrastive learning; Figure 2 This is the registration flow chart; Figure 3 Qualitative visualization comparison of the same-modal registration model; Figure 4 Comparison of multimodal registration model effects. DETAILED DESCRIPTION

[0020] The technical solution of the present invention is further described below with reference to the accompanying drawings and embodiments.

[0021] Unless otherwise defined, technical or scientific terms used in the present invention shall have the same meaning as commonly understood by one of ordinary skill in the art to which the present invention belongs.

[0022] Example 1 The present invention provides a medical image registration method, comprising the following steps: Step S1: Data preparation and preprocessing: Acquire and preprocess paired multimodal medical images that have been aligned or roughly registered to obtain preprocessed source modality images and target modality images.

[0023] The preprocessing steps include: Image cropping: remove irrelevant areas in the image and retain the area of interest.

[0024] Denoising: Use filtering or noise reduction methods to remove noise interference.

[0025] Grayscale normalization: linear normalization is used to map image pixel values to a uniform range , to reduce the grayscale differences between modalities.

[0026] Step S2: Joint optimization of modality conversion and contrastive learning: Step S21: Input the preprocessed source modality image into the structure-preserving modality conversion network, and output a synthetic image with the same grayscale distribution as the preprocessed target modality image.

[0027] like Figure 1 As shown in the figure, the structure-preserving modality transfer network is based on the DenseUNet architecture. Specifically, it first constructs an end-to-end feature learning framework (DenseUNet architecture) through convolution, pooling, and upsampling operations. Then, it introduces an early transition module in the encoding path to expand the receptive field through early downsampling, thereby assisting the model in establishing a global grayscale mapping between modalities. Finally, in the decoding path, it uses a three-dimensional linear upsampling module combined with a post-convolution module to avoid the checkerboard artifacts introduced by traditional deconvolution operations, improving the smoothness of edge areas and the restoration of details.

[0028] The input of the structure-preserving modal transfer network is the preprocessed source modality image , output a synthetic image that is consistent with the grayscale distribution of the preprocessed target modality image ,Right now: ; in, represents the mode conversion network.

[0029] Step S22: Based on the preprocessed source modality image and target modality image, spatially corresponding image blocks are extracted to construct positive sample pairs, and multi-source negative sample pairs are generated through cross-position selection and noise perturbation.

[0030] The construction process of positive sample pairs is as follows: From the pre-processed source modality image and the preprocessed target modality image Extract the image blocks at the same position and ; represents the source modality image patch, represents the target modality image patch; The source modality image is converted to Mapped to pseudo target modality image patch , construct positive sample pairs .

[0031] The construction process of multi-source negative sample pairs is as follows: Cross-position negative samples: Cross-position selection involves selecting negative samples from the target modality image. Other image blocks that do not correspond to the same position , used to simulate the situation of mismatched anatomical regions; Noisy negative samples: Noisy perturbations include target modality image patches Add mixed disturbance to generate noisy negative samples ,in, represents the Gaussian noise term, represents the symbolic function, represents contrast loss.

[0032] Step S23: Calculate the contrast loss based on the normalized cosine similarity, and form a composite loss function with the grayscale consistency loss and the structure preservation loss to optimize the parameters of the structure-preserving modal conversion network.

[0033] The composite loss function is as follows: ; in, represents the compound loss, represents the grayscale consistency loss, represents the loss of structural preservation, represents a composite image, represents the weight, represents the gradient operator, Indicates the number of blocks for each pair of images, Indicates the number of negative samples corresponding to each image block, Indicates the sequence number of the image block that is currently a positive sample, Indicates the pseudo target modality image block and the The feature similarity between negative samples across positions, represents the temperature parameter, Indicates the The feature similarity of the noisy negative samples, It represents the normalized cosine similarity, which is calculated as follows: ; in, represents the inner product of the output image block and the target modality image block, express norm.

[0034] Step S3: Input the composite image and the fixed image into the unsupervised registration network. The registration process is as follows: Figure 2 shown.

[0035] The registration network is designed based on UNet. and fixed images After splicing the channel dimension as input, the cascaded encoder and decoder structure extracts multi-scale features and generates a high-quality deformation field. The encoder part consists of 5 layers, each layer uses a 32-channel feature map and performs 1 / 2, 1 / 4, 1 / 8, and 1 / 16 scale downsampling in turn; the decoder uses a symmetrical structure to gradually upsample to the original resolution and finally outputs the final deformation field. , Indicates the height of the corresponding image, Indicates the width of the corresponding image, represents the depth of the corresponding image, and 3 represents the displacement component.

[0036] Loss function of unsupervised registration network It consists of similarity loss and smoothness loss, and the calculation formula is as follows: ; in, represents the hyperparameter that balances the similarity term and the regularization term, represents the image domain, represents the gradient of the deformation field, represents the normalized cross-correlation, Deformation field Apply to images , Represents the composite operation of functions, Indicates a fixed image.

[0037] Step S4: outputting a registered medical image based on the deformation field and the synthesized image.

[0038] The present invention will be further described below through specific examples.

[0039] This example introduces this network into several typical registration frameworks and conducts systematic comparative experiments. The experiments cover both homomodal registration frameworks (Voxelmorph, LapIRN) and multimodal registration frameworks (MIND, multiGranICON, and MambaMorph). The DICE coefficient (DSC) and 95% Hausdorff distance (HD95) are used as evaluation metrics to comprehensively assess the registration performance of each method.

[0040] In the homomodal registration experiments, the modality conversion network was integrated into the Voxelmorph and LapIRN frameworks (labeled as Voxelmorph-T and LapIRN-T, respectively) and evaluated on standard CT-MR datasets.

[0041] Table 1 summarizes the quantitative evaluation results of each method. The results show that, after adding the modality conversion module, Voxelmorph-T's DICE improved from 68.88 to 74.04, and its HD95 decreased from 3.15 to 3.02; LapIRN-T's DICE improved from 70.81 to 75.86, and its HD95 decreased from 2.76 to 2.54. These results demonstrate that the proposed modality conversion network can effectively alleviate the adverse effects of cross-modal grayscale discrepancies on the same-modality registration model, significantly improving the alignment performance of anatomical structures.

[0042] Table 1 Quantitative evaluation results of the homomodal registration task ;

[0043] To further evaluate the structural consistency of each method from a visual perspective, Figure 3 Qualitative results of various models in the CT-to-MR registration task are presented. Each image group in the figure shows the fixed image (MR), the original moving image (CT), and the registered image, with three types of contours superimposed: red represents anatomical structures in the fixed image, pink represents structures in the original moving image, and orange represents structures in the registered image. By observing the degree of overlap between the red and orange contours, the consistency of the registered anatomical structures can be intuitively assessed. The method incorporating the modality conversion module demonstrates higher contour overlap in key areas, further validating its ability to improve registration accuracy.

[0044] In multimodal registration experiments, the proposed method was compared with existing advanced multimodal registration models (MIND, multiGranICON, and MambaMorph) (see Table 2) and evaluated uniformly on the same CT-MR dataset. Because these methods inherently handle modality differences, the proposed method did not integrate additional modules to maximize the inherent performance of each method. The results demonstrated that the proposed method outperformed the comparison models in both DICE and HD95 metrics, demonstrating its superior robustness and accuracy in maintaining anatomical consistency and addressing modality grayscale differences, making it suitable for multimodal registration scenarios with significant grayscale distribution differences.

[0045] Table 2 Quantitative evaluation results of multimodal registration task ;

[0046] Figure 4 The qualitative visualization results of each method under the multimodal registration task are further shown. Figure 3 Similarly, the overlap between the red and orange outlines in the registered image can be used to reflect the effectiveness of structural alignment. It can be seen that the proposed method exhibits better structural alignment capabilities in key anatomical regions, demonstrating its superior performance in multimodal registration tasks.

[0047] Example 2 A multimodal medical image registration system, comprising: A data acquisition and preprocessing module is used to acquire and preprocess paired multimodal medical images to obtain preprocessed source modality images and target modality images; The structure-preserving modality conversion module is built on the DenseUNet architecture and includes an early transition module in the encoding path and a 3D linear upsampling combined with a post-convolution module in the decoding path. The structure-preserving modality conversion module receives the preprocessed source modality image and outputs a composite image with the same grayscale distribution as the target modality. Contrastive learning optimization module, including: Positive sample construction unit: extracts image blocks with corresponding spatial positions from the source modality image processed by the structure-preserving modality conversion network and the preprocessed target modality image to construct positive sample pairs; Multi-source negative sample generation unit: generates negative sample pairs through cross-position selection and noise perturbation; Composite loss calculation unit: Calculates contrast loss based on normalized cosine similarity, combines grayscale consistency loss and structure preservation loss to form a composite loss function, and optimizes the parameters of the modal conversion module; The unsupervised registration module uses a cascaded encoder-decoder structure to receive the synthetic image and the fixed image, and outputs the deformation field after multi-scale feature fusion; The registration image generation module outputs the registered medical image based on the deformation field and the synthetic image.

[0048] If the above functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes various media that can store program code, such as USB flash drives, mobile hard drives, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical disks.

[0049] The logic and / or steps represented in the flowcharts or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing the logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (e.g., a computer-based system, a system including a processor, or other system that can fetch and execute instructions from an instruction execution system, apparatus, or device). For purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0050] More specific examples (a non-exhaustive list) of computer-readable media include the following: an electrical connection with one or more wires (electronic devices), a portable computer disk cartridge (magnetic devices), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disc read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, and then editing, interpreting, or processing in another suitable manner as necessary, and then storing it in a computer memory.

[0051] It is worth noting that the contents not elaborated in detail in the present invention are all prior art and are well known to those skilled in the art.

[0052] Therefore, the present invention adopts the above-mentioned multimodal medical image registration method, system, computer equipment and medium, which can effectively improve the registration accuracy, enhance the robustness and discrimination ability of the model for local areas, ensure the accurate alignment of key anatomical structures, and has good clinical application prospects.

[0053] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit the same. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solutions of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A multimodal medical image registration method, characterized in that: The following steps are involved: Step S1: Acquire and preprocess paired multimodal medical images to obtain preprocessed source modality images and target modality images; Step S2: input the preprocessed source modality image into the structure-preserving modality conversion network, and output a synthetic image with the same grayscale distribution as the preprocessed target modality image; Based on the source modality image processed by the structure-preserving modality conversion network and the preprocessed target modality image, the corresponding image blocks are extracted to construct positive sample pairs, and multi-source negative sample pairs are generated through cross-position selection and noise perturbation. The contrast loss is calculated based on normalized cosine similarity, and a composite loss function is formed with grayscale consistency loss and structure preservation loss to optimize the parameters of the structure-preserving modal conversion network. The structure-preserving modal transfer network is based on the DenseUNet architecture, which introduces an early transition module in the encoding path and adopts a 3D linear upsampling module combined with a post-convolution module in the decoding path. Step S3: Input the synthetic image and the fixed image into the unsupervised registration network, extract multi-scale features through the cascaded encoder-decoder structure, and output the deformation field after fusion; Step S4: outputting a registered medical image based on the deformation field and the synthesized image.

2. A multimodal medical image registration method according to claim 1, characterized in that: In step S1, preprocessing includes: cropping, denoising and grayscale normalization.

3. The multimodal medical image registration method according to claim 1, wherein: In step S2, the construction process of positive sample pairs is as follows: From the pre-processed source modality image and the preprocessed target modality image Extract the image blocks at the same position and ; represents the source modality image patch, represents the target modality image patch; The source modality image is converted to Mapped to pseudo target modality image patch , construct positive sample pairs ,in represents the mode conversion network.

4. A multimodal medical image registration method according to claim 3, characterized in that: In step S2, cross-position selection of negative samples includes selecting the image with the same position as the target modality image. Other image blocks that do not correspond to the same position , the noise perturbation includes the target modality image patch Add mixed disturbance to generate noisy negative samples ,in, represents the Gaussian noise term, represents the symbolic function, represents contrast loss.

5. A multimodal medical image registration method according to claim 4, characterized in that: In step S2, the composite loss function is as follows: ； in, represents the compound loss, represents the grayscale consistency loss, represents the loss of structural preservation, represents a composite image, represents the weight, represents the gradient operator, Indicates the number of blocks for each pair of images, Indicates the number of negative samples corresponding to each image block, Indicates the sequence number of the image block that is currently a positive sample, Indicates the pseudo target modality image block and the The feature similarity between negative samples across positions, represents the temperature parameter, Indicates the The feature similarity of the noisy negative samples, It represents the normalized cosine similarity, which is calculated as follows: ； in, represents the inner product of the pseudo target modality image block and the target modality image block, express norm.

6. A multimodal medical image registration method according to claim 5, characterized in that: In step S3, the loss function of the unsupervised registration network is It consists of similarity loss and smoothness loss, and the calculation formula is as follows: ； in, represents the hyperparameter that balances the similarity term and the regularization term, represents the image domain, represents the gradient of the deformation field, represents the normalized cross-correlation, Deformation field Apply to images , Represents the composite operation of functions, Indicates a fixed image.

7. A multimodal medical image registration system, characterized in that: include: A data acquisition and preprocessing module is used to acquire and preprocess paired multimodal medical images to obtain preprocessed source modality images and target modality images; A structure-preserving modal transfer module, built on the DenseUNet architecture, consisting of an early transition module in the encoding path and a 3D linear upsampling combined with post-convolution module in the decoding path; The structure-preserving modality conversion module receives the preprocessed source modality image and outputs a synthetic image with the same grayscale distribution as the target modality; Contrastive learning optimization module, including: Positive sample construction unit: extracts image blocks with corresponding spatial positions from the source modality image processed by the structure-preserving modality conversion network and the preprocessed target modality image to construct positive sample pairs; Multi-source negative sample generation unit: generates negative sample pairs through cross-position selection and noise perturbation; Composite loss calculation unit: Calculates contrast loss based on normalized cosine similarity, combines grayscale consistency loss and structure preservation loss to form a composite loss function, and optimizes the parameters of the modal conversion module; The unsupervised registration module uses a cascaded encoder-decoder structure to receive the synthetic image and the fixed image, and outputs the deformation field after multi-scale feature fusion; The registration image generation module outputs the registered medical image based on the deformation field and the synthetic image.

8. A computer device, characterized in that: The system comprises a memory and a processor, wherein the memory is used to store instructions, and the processor is used to execute the instructions to implement the multimodal medical image registration method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the multimodal medical image registration method according to any one of claims 1 to 6 is implemented.

Images