A digital subtraction angiography method, device, computer equipment and medium
By employing a variable calibration model optimized through alternating temporal background synthesis and sparse constraints, the problem of preserving vascular structures and suppressing background artifacts in digital subtraction angiography was solved, achieving efficient and accurate image correction results that meet clinical needs.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- ZHEJIANG NORMAL UNIV
- Filing Date
- 2026-05-11
- Publication Date
- 2026-07-31
AI Technical Summary
Existing technologies struggle to balance vascular structure preservation and background artifact suppression in digital subtraction angiography, resulting in low computational efficiency, the introduction of block artifacts or false features, and an inability to meet clinical requirements for the accuracy, efficiency, and realism of image correction.
A variable calibration model is constructed by employing a temporal background synthesis and sparse constraint alternating optimization strategy. The deformation field gradient is dynamically adjusted through an adaptive regularization term, and iterative solutions are obtained by combining a deep learning network to generate high-quality pseudo-mask images and accurately strip vascular structures, thus avoiding artifacts.
It significantly improves artifact suppression capabilities, ensures the integrity and clarity of vascular anatomy, meets clinical requirements for image authenticity, stability, and real-time performance, and improves the accuracy and efficiency of image correction.
Smart Images
Figure CN122492490A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of medical image processing, specifically relating to a method for registering digital subtraction angiography images. Background Technology
[0002] Digital subtraction angiography (DSA) is the gold standard for vascular imaging and interventional therapy guidance. It eliminates background information by subtracting the contrast image after contrast agent injection into the blood vessel from the mask image before contrast, thus highlighting the morphology of the blood vessel and providing visualization support for the diagnosis of vascular lesions and the planning of interventional procedures.
[0003] The diagnostic value of DSA images heavily relies on the precise spatial alignment of the contrast image and the mask image. However, physiological movements such as breathing, heartbeat, and gastrointestinal motility can easily cause inter-frame spatial misalignment, producing motion artifacts that can obscure lesion areas and even mislead clinical decisions. Existing correction techniques for DSA motion artifacts are mainly divided into traditional methods and deep learning methods, both of which have limitations.
[0004] For example, early parametric models such as affine transformation and thin-plate spline interpolation had limited expressive power and could not compensate for complex non-rigid deformations. Furthermore, their reliance on iterative optimization led to low computational efficiency, and block matching methods were prone to introducing block artifacts, disrupting the spatial continuity of the image. Later, deep learning methods were adopted, divided into generative and registration methods. Generative methods (such as U-net and Pix2pix networks) suffer from data dependency issues, easily resulting in missing vascular structures or false features in complex motion scenes. Registration methods (such as VoxelMorph) rely on pre-estimation of blood vessels, and the generated dense deformation field easily distorts blood vessels, leading to blurred blood vessels.
[0005] Therefore, existing technologies cannot simultaneously preserve vascular structure and suppress background artifacts. Some methods are prone to introducing secondary artifacts or lack real-time performance, failing to meet clinical requirements for the accuracy, efficiency, and realism of DSA image correction. Summary of the Invention
[0006] To address the problem that existing technologies struggle to balance vascular structure preservation and background artifact suppression, this invention provides a registration method, apparatus, computer equipment, and medium for digital subtraction angiography.
[0007] To achieve the above objectives, the present invention provides the following technical solution: A digital subtraction angiography method, the method comprising: Acquire sequences of contrast images and pre-contrast mask images; Based on the contrast image and the pre-contrast mask image sequence, image block matching and reconstruction are performed along the time dimension to synthesize a pseudo-mask image that matches the structure of the contrast image; A variable alignment model is constructed, wherein the objective function of the variable alignment model includes a similarity metric for aligning the background of the angiography image with the pseudo-mask image after distortion by the deformation field, a regularization term for constraining the sparsity of vascular information, and an adaptive regularization term for applying spatially differentiated smoothing constraints to the deformation field; wherein the adaptive regularization term constructs a spatial weight map based on the pixel differences between the background of the angiography image and the pseudo-mask image to dynamically adjust the penalty intensity of the deformation field gradient according to the differences in image regions; The variable calibration model is iteratively solved, and the vascular information and deformation field are estimated alternately during the solution process. The adaptive regularization term guides the generation of a deformation field that adapts to the non-uniform motion characteristics during each deformation field estimation. When the final stage is reached, the final deformation field is output, and vascular information is extracted based on the final background image and the angiography image to obtain a digital subtraction angiography image with motion artifacts removed.
[0008] Optionally, based on the contrast image and the pre-contrast mask image sequence, image patch matching and reconstruction are performed along the time dimension to synthesize a pseudo-mask image that matches the structure of the contrast image, including: Each frame of the contrast image and the pre-contrast mask image sequence is decomposed into overlapping image blocks; For each spatial location in the contrast image, a matching image block is searched in the pre-contrast mask image sequence along the time dimension, and the optimal matching time index is determined by minimizing the Euclidean distance. A weighted average strategy is used to fuse and reconstruct the pixel values of all optimal matching image blocks covering the same pixel, generating the pseudo-mask image that is spatially continuous and similar in structure to the imaging image.
[0009] Optionally, the objective function formula of the variable allocation standard model is: ; in, For imaging images, For deformation field, For vascular information, For the deformation field gradient, , For regularization parameters, The image is the result of distortion of the pseudomask by a deformation field.
[0010] Optionally, in the deformation field registration step, the previous background target is registered with the pseudo-mask image, and the current deformation field is predicted using a deep learning registration network with a self-attention mechanism. ; in, For the TransMorph backbone network, It is a pseudo-mask image. For the first Background target and initial value for stage estimation .
[0011] Optionally, in the vascular information estimation step, the current deformation field is fixed, the objective function is simplified to a univariate optimization problem about vascular information, and a closed-form solution is obtained using a soft thresholding operator. This soft thresholding operator is applied to the residual image between the angiographic image and the deformed pseudo-mask image to extract sparse vascular information; thus, the first... Vascular information in the next iteration : ; in, This is the residual image after the k-th iteration. It is a symbolic function.
[0012] Optionally, the background image is updated according to the following formula: ; in, This is the background image updated after the k-th iteration. For imaging images, For the first The vascular information obtained in the next iteration.
[0013] Optionally, in the K-round alternating optimization process, constructing a spatial weight map based on the pixel difference between the background image and the pseudo-mask image, and applying an adaptive smoothing constraint to the gradient of the deformation field using the spatial weight map includes: Calculate the absolute value of the pixel-by-pixel residual between the background image and the pseudo-mask image in the current iteration stage; The negative exponential function is used to map the absolute value of the residuals into a spatial weight map, so that regions with large residuals correspond to smaller weights, and regions with small residuals correspond to larger weights. The regularization loss is calculated by element-wise multiplying the spatial weight map with the gradients of each spatial dimension of the deformation field, so as to dynamically adjust the smoothing constraint strength of different regions during training or iteration; wherein, the calculation formula of the spatial weight map is: ; in, For spatial weighting, The background image estimated for the k-th stage. It is a pseudo-mask image. To control the scaling factor for the weights' sensitivity to the residuals, It is a Gaussian smoothing kernel.
[0014] A digital subtraction angiography device, the device comprising: The acquisition module is used to acquire sequences of contrast images and pre-contrast mask images; The synthesis module is used to perform image patch matching and reconstruction along the time dimension based on the contrast image and the pre-contrast mask image sequence, and synthesize a pseudo-mask image that matches the structure of the contrast image; A construction module is used to construct a variable alignment model. The objective function of the variable alignment model includes a similarity metric for aligning the background of the angiography image with the pseudo-mask image after distortion by the deformation field, a regularization term for constraining the sparsity of vascular information, and an adaptive regularization term for applying spatially differentiated smoothing constraints to the deformation field. The adaptive regularization term constructs a spatial weight map based on the pixel differences between the background of the angiography image and the pseudo-mask image to dynamically adjust the penalty intensity of the deformation field gradient according to the differences in image regions. The iterative module is used to iteratively solve the variable calibration model, alternately estimating vascular information and deformation field during the solution process, and guiding the generation of deformation field adapted to non-uniform motion characteristics through the adaptive regularization term during each deformation field estimation. The output module is used to output the final deformation field when the final stage is reached, and extract vascular information based on the final background image and the angiography image to obtain a digital subtraction angiography image with motion artifacts removed.
[0015] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the aforementioned digital subtraction angiography method.
[0016] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the aforementioned digital subtraction angiography method.
[0017] The digital subtraction angiography registration method provided by this invention has the following beneficial effects: This invention effectively solves the common problem of existing technologies struggling to simultaneously preserve vascular structure and suppress background artifacts by constructing a cascaded optimization framework of "synthesis-registration". First, it utilizes a pseudomask that highly matches the temporal background synthesis with the angiographic image structure to overcome background mismatch caused by large-scale non-rigid motion. Second, it introduces a sparse constraint alternating optimization strategy to decouple vascular information estimation from deformation field registration iterations, accurately stripping vascular structures during registration and avoiding blurring or distortion of vascular details due to background alignment operations. Simultaneously, it employs a residual-aware adaptive regularization mechanism to dynamically adjust the deformation field smoothing constraints based on image region differences, effectively correcting motion artifacts in large deformation areas while preventing topological distortion in well-aligned areas. The overall framework iteratively refines the deformation field in an end-to-end manner, balancing registration accuracy and processing efficiency. Ultimately, it significantly improves artifact suppression capabilities while ensuring the integrity and clarity of vascular anatomy, meeting the comprehensive clinical requirements for image realism, stability, and real-time performance. Attached Figure Description
[0018] To more clearly illustrate the embodiments and design schemes of the present invention, the accompanying drawings required for this embodiment will be briefly described below. The drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0019] Figure 1 This is a schematic flowchart of a digital subtraction angiography method provided by the present invention according to an exemplary embodiment.
[0020] Figure 2 This is a schematic diagram of an optimized cascade framework for "synthesis-registration" provided by the present invention according to an exemplary embodiment.
[0021] Figure 3 This is an example diagram of a registration model calculation provided by the present invention according to an exemplary embodiment.
[0022] Figure 4 This is a block diagram of a digital subtraction angiography device provided by the present invention according to an exemplary embodiment. Detailed Implementation
[0023] To enable those skilled in the art to better understand and implement the technical solutions of the present invention, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention and should not be construed as limiting the scope of protection of the present invention.
[0024] The technical solutions provided by the various embodiments of the present invention will be described in detail below with reference to the accompanying drawings.
[0025] First, this invention provides a digital subtraction angiography method, specifically as follows: Figure 1 As shown, it includes the following steps: S101. Obtain the sequence of contrast images and pre-contrast mask images.
[0026] S102. Based on the contrast image and the pre-contrast mask image sequence, perform image block matching and reconstruction along the time dimension to synthesize a pseudo-mask image that matches the structure of the contrast image.
[0027] In this step, the contrast image and each frame in the precontrast mask image sequence are decomposed into overlapping image blocks. For each image block corresponding to a spatial location in the contrast image, a matching image block is searched in the precontrast mask image sequence along the time dimension, and the optimal matching time index is determined by minimizing the Euclidean distance. A weighted average strategy is used to fuse and reconstruct the pixel values of all optimal matching image blocks covering the same pixel to generate a spatially continuous pseudo-mask image that is structurally similar to the contrast image.
[0028] S103. Construct a variable alignment model. The objective function of the variable alignment model includes a similarity metric for aligning the background of the angiographic image with the pseudomask image after distortion by the deformation field, a regularization term for constraining the sparsity of vascular information, and an adaptive regularization term for applying spatially differentiated smoothing constraints to the deformation field.
[0029] The adaptive regularization term constructs a spatial weight map based on the pixel differences between the background of the imaging image and the pseudo-mask image, so as to dynamically adjust the penalty intensity of the deformation field gradient according to the differences in image regions.
[0030] The objective function formula is: ; in, For imaging images, For deformation field, For vascular information, For the deformation field gradient, , For regularization parameters, The image is the result of distortion of the pseudomask by a deformation field.
[0031] S104. Iteratively solve the variable calibration model, alternately estimating vascular information and deformation field during the solution process, and guide the generation of deformation field that adapts to non-uniform motion characteristics through the adaptive regularization term during each deformation field estimation.
[0032] In the deformation field registration step, the target background from the previous round is registered with the pseudo-mask image, and a deep learning registration network with a self-attention mechanism is used to predict the current deformation field. : ; in, For the TransMorph backbone network, It is a pseudo-mask image. For the first Background target and initial value for stage estimation .
[0033] In the vascular information estimation step, the current deformation field is fixed, and the objective function is simplified to a univariate optimization problem concerning vascular information. A closed-form solution is obtained using a soft thresholding operator, which is applied to the residual image between the angiographic image and the deformed pseudomask image to extract sparse vascular signals. The formula for solving for the vascular information is as follows: ; in, The vascular information estimated in the k-th iteration. The residual image for the k-th iteration. It is a symbolic function.
[0034] Based on the above deformation field update and vascular information estimation, the target background update follows the formula: ; in, The target background is updated after the k-th iteration. For imaging images, This represents the vascular information obtained in the k-th iteration.
[0035] In addition, this step also requires calculating the absolute value of the pixel-wise residual between the target background and the pseudo-mask image in the current iteration stage; mapping this absolute value of the residual to a spatial weight map using a negative exponential function, so that regions with large residuals correspond to smaller weights, and regions with small residuals correspond to larger weights; calculating the regularization loss by element-wise multiplying this spatial weight map with the gradients of each spatial dimension of the deformation field, so as to dynamically adjust the smoothing constraint strength of different regions during training or iteration; the formula for calculating the spatial weight map is: ; in, For spatial weighting, The background image estimated for the k-th stage. It is a pseudo-mask image. To control the scaling factor for the weights' sensitivity to the residuals, The smoothing kernel is 5×5 Gaussian with a standard deviation σ=1.0.
[0036] S105. When the final stage is reached, the final deformation field is output, and the vascular information is extracted based on the final background image and the angiography image to obtain a digital subtraction angiography image with motion artifacts removed.
[0037] By employing the above method and constructing a cascaded optimization framework of "synthesis-registration," the common problem of existing technologies struggling to simultaneously preserve vascular structure and suppress background artifacts is effectively solved. First, a pseudomask highly matching the angiographic image structure is generated using temporal background synthesis, overcoming background mismatch caused by large-scale non-rigid motion and reducing artifact generation at its source. Second, a sparse constraint alternating optimization strategy is introduced to decouple vascular information estimation from deformation field registration iterations. During registration, vascular structures are precisely stripped and protected, avoiding blurring or distortion of vascular details due to over-registration. Simultaneously, a residual-aware adaptive regularization mechanism is used to dynamically adjust the deformation field smoothing constraints based on image region differences. This effectively corrects motion artifacts in large deformation regions while preventing topological distortion in well-aligned regions, thus eliminating the introduction of secondary artifacts. The overall framework iteratively refines the deformation field in an end-to-end manner, balancing registration accuracy and processing efficiency. Ultimately, it significantly improves artifact suppression capabilities while ensuring the integrity and clarity of vascular anatomy, meeting the comprehensive clinical requirements for image realism, stability, and real-time performance.
[0038] To illustrate the steps of the above method, the present invention further provides a mechanistic explanation of the key steps.
[0039] This invention proposes a cascaded optimization framework for DSA image registration that integrates temporal template matching and adaptive deformation field estimation. Through collaborative optimization in the "synthesis-registration" two-stage process, it achieves motion artifact removal and vascular structure preservation. Simultaneously, it introduces a residual-aware adaptive regularization strategy to adapt to non-uniform motion scenarios. Specifically, it includes the following key steps:
[0040] 1. Temporal background synthesis to generate high-quality pseudomasks.
[0041] Utilizing the temporal redundancy of DSA sequences, the contrast image and pre-contrast mask image sequences are decomposed into overlapping image blocks. An optimal mask image block is matched for each contrast image block along the temporal dimension. A pseudo-mask highly similar to the contrast image structure is reconstructed through weighted averaging. This avoids block artifacts and provides high-quality input for subsequent registration. The optimal time index is achieved by minimizing... The norm is determined by the formula:
[0042] ; in, For imaging images In spatial location The block, For the first A mask frame at position The block, This is the optimal time index.
[0043] 2. Construct a sparse-guided recurrent registration network (SGR-Net) to iteratively solve for vascular information and deformation field.
[0044] Based on TransMorph as the registration backbone network, a fusion network is constructed. The sparse-constrained variable registration model decomposes the model into two sub-problems: vascular information estimation and deformation field registration, through an alternating optimization strategy. Iterative refinement of the deformation field achieves decoupling of blood vessels from the background and avoids blurring of vascular structures during registration.
[0045] (1) Core objective function of the registration model: ; in, For deformation field, For vascular information, For the deformation field gradient, , For regularization parameters, The image is the result of distortion of the pseudomask by a deformation field.
[0046] (2) Deformation field update: fix vascular information The problem is transformed into background registration, and the TransMorph network is used to directly predict the first... Deformation field of the next iteration : ; in, For the TransMorph backbone network, For the first Background target and initial value for stage estimation .
[0047] (3) Vascular information estimation: fixed deformation field By solving the closed-form solution using the soft threshold operator, the first... Vascular information in the next iteration : ; in, For residual images, It is a symbolic function.
[0048] (4) Target background update: based on the deformation field predicted in the current iteration With vascular information To obtain a cleaner background as the background target for the next stage: .
[0049] 3. Iterative refinement and total loss optimization are performed to output an artifact-free subtraction image.
[0050] Expand the alternating optimization process Each stage is a cascaded process. After each stage completes "deformation field registration → blood vessel estimation → background update", the background target is updated. By integrating the similarity loss (MSE) and adaptive regularization loss to construct the total loss function, end-to-end training is achieved. ; The final output is .
[0051] 4. Residual sensing adaptive regularization dynamically adjusts deformation field smoothing constraints.
[0052] Based on the current estimation background With fake masks Constructing a spatial weight map from pixel-by-pixel residuals The smoothing constraint strength of the deformation field is dynamically adjusted: the constraint is relaxed in significant areas with large residuals to ensure matching accuracy, and the constraint is strengthened in well-aligned areas with small residuals to avoid topological distortion.
[0053] (1) Formula for calculating spatial weight map: ; in, To control the scaling factor for the weights' sensitivity to the residuals, It is a Gaussian smoothing kernel.
[0054] (2) Adaptive regularization loss formula: ; in, For element-wise multiplication, For the deformation field in the first The gradient of each spatial dimension.
[0055] Based on the above mechanism explanation, the present invention also provides an embodiment.
[0056] Step 1: Time background compositing.
[0057] The core of this step is to utilize the temporal redundancy of the DSA sequence to synthesize a pseudomask that is highly similar in structure to the contrast image. This provides high-quality input for subsequent registration and avoids block artifacts.
[0058] 1.1 Image Patch Decomposition: Input DSA image sequence, including pre-contrast image sequence. and single imaging image The imaging images and each pre-contrast image All are decomposed into overlapping patches, with a patch size set to [size missing]. The sliding step size of adjacent blocks is set to ( To ensure overlapping coverage. For contrast images In spatial location Extract target block For each pre-contrast image In the same spatial location Extract the corresponding block at the location .
[0059] 1.2 Time-dimension block matching for each target block Traversing the pre-contrast image sequence in the time dimension By minimizing Norm finds the optimal matching block and determines the optimal time index. The calculation formula is as follows: ; in, express Norm, Indicates the first Zhang's pre-contrast image is in position The block at that location.
[0060] 1.3 The pseudomask is reconstructed using a weighted average strategy. Due to block overlap, each pixel is contained within multiple optimal matching blocks. The average of the corresponding pixel values across all the optimal blocks covering that pixel is used to obtain the pixel's position within the optimal matching blocks. The final value in the composition process. This strategy ensures a smooth transition of block boundaries and effectively avoids block artifacts. The algorithm for the synthesis process is summarized below:
[0061] Algorithm 1: Temporal background synthesis algorithm.
[0062] ① Input pre-contrast sequence Imaging images Set the block size and sliding step size; ② and all Perform block decomposition to obtain and ( (for spatial location index) ③ For each Solving for the optimal time index Record the corresponding optimal block ; ④ For each pixel in the image, calculate all the pixels that cover that pixel. The average pixel value is used to generate a pseudo mask. ; ⑤ Output The input mask image is used in the registration stage.
[0063] Step 2: Construction and discretization of the sparsely guided recurrent registration network (SGR-Net).
[0064] This step is based on the TransMorph network to build a sparsely guided cyclic registration framework. By separating blood vessel and background information and iteratively refining the deformation field, the integrity of the blood vessel structure is ensured.
[0065] 2.1 Registration Model Establishment: A variable registration model incorporating sparse constraints is established, with the core objective of aligning pseudo-masks. Background components of contrast image F ( , (For vascular information), the objective function of the model is as follows: , in, Represents the deformation field. The gradient of the deformation field; Indicates vascular information, for Sparsity constraints characterize the spatial sparsity of blood vessels. The deformation field smoothing regularization parameter, For sparse constraint parameters, This represents the image after the pseudomask has been deformed by the deformation field. To ensure data fidelity, the deformed dummy mask is aligned with the angiographic image after devascularization.
[0066] 2.2 Alternating optimization strategy.
[0067] Deformation field registration ( -step) Fix the estimated vascular information from the previous round. Solve the deformation field The problem then transforms into a registration problem between background images: ; Using TransMorph network As a registration backbone network, its self-attention mechanism is used to capture long-range dependencies and predict deformation fields. The calculation formula is as follows: in, The initial value is the estimated background for the previous iteration. .
[0068] Vascular information estimation (S-step) fixed deformation field Solving for vascular information ,correspond The proximal operator for regularization is solved using the soft-thresholding operator, and the closed-form solution is as follows: , in, For the current residual image, It is a symbolic function.
[0069] 2.3 Iterative Refinement: The alternating optimization process is expanded into... Each cascade stage (in this embodiment) (balancing accuracy and efficiency), such as Figure 2 As shown, all stages share network parameters to ensure parameter efficiency. Each stage executes " The process is "Iteration step → S iteration step → background update", and the background update formula is as follows:
[0070] ; Through iterative refinement Gradually approaching the real background, guide the network to focus on background alignment and avoid blurring of vascular structures.
[0071] Step 3: Residual-aware adaptive regularization.
[0072] This step improves the local registration rationality by dynamically adjusting the regularization constraint of the deformation field to adapt to the non-uniform motion characteristics of DSA images.
[0073] 3.1 Spatial weight graph construction based on the current estimated background With fake masks Pixel-level residuals are used to construct a spatial weight map. The calculation formula is as follows: ; in, For spatial weighting, The background image estimated for the k-th stage. It is a pseudo-mask image. To control the scaling factor for the weights' sensitivity to the residuals, Gaussian smoothing kernel, standard deviation This formula assigns smaller weights to regions with larger residuals. Regions with smaller residuals correspond to larger weights. .
[0074] 3.2 Adaptive regularization loss calculation.
[0075] Weight graph The Hadamard product is performed with the deformation field gradient, and the fixed-parameter regularization term is replaced. The adaptive regularization loss formula is as follows: ; in, It represents the Hadamah accumulation. The deformation field is represented in the first... The gradient of each spatial dimension.
[0076] Step 4: Overall training and iterative solution.
[0077] 4.1 Definition of Total Loss Function: Integrating similarity loss and adaptive regularization loss, a total loss function is constructed for end-to-end training: ; in, Mean squared error (MSE) loss is used to quantify background alignment accuracy; The weights are used for regularization loss. and The first The final estimation of the background and deformation field for the stage.
[0078] This invention employs a "synthesis-registration" cascade framework to first address the background mismatch caused by large-scale motion, and then achieves fine registration through sparse guidance and adaptive regularization, effectively balancing artifact suppression and vascular structure preservation. Experimental results show that the PSNR and image entropy values of blood vessels in DSA images processed by this method are significantly superior to existing mainstream methods, providing high-quality image support for clinical interventional therapy. Figure 3 The figure illustrates a computational example of the "synthesis-registration" model proposed in this invention, including a background image, an angiographic image, a direct subtraction image, a subtraction image obtained using existing registration methods, and a subtraction image obtained using the method proposed in this invention. As can be seen from the figure, compared to the processing results of existing technologies, the subtraction image obtained by the method of this invention has significantly fewer artifacts while maintaining clear visibility of blood vessels, thus better preserving the structure of the blood vessels.
[0079] As can be seen from the above, this invention proposes a novel DSA image registration and artifact removal model based on "synthesis-registration" cascade optimization and adaptive deformation field. This model innovatively integrates a temporal background synthesis strategy, a sparse-guided cyclic registration network, and a residual-aware adaptive regularization mechanism. Specifically, it synthesizes high-quality pseudo-pre-angiography images through image patch matching and overlap fusion mechanisms, solving the background mismatch problem caused by large-scale motion; it introduces L1 sparse constraints to decouple blood vessel and background information, avoiding damage to blood vessel structures during registration. Furthermore, it employs a dynamic regularization parameter adjustment strategy to adapt to non-uniform motion scenarios, and uses deep learning to replace traditional iterative optimization, balancing registration accuracy and computational efficiency, significantly simplifying the solution of DSA image correction problems under complex non-rigid deformation. The model and algorithm proposed in this invention not only theoretically fill the gap in the adaptability of existing technologies to complex motion scenarios, but also demonstrate strong stability and wide applicability in practical clinical applications.
[0080] Secondly, the present invention also provides a digital subtraction angiography device, such as... Figure 4 As shown, it includes: The acquisition module 201 is used to acquire the sequence of contrast images and pre-contrast mask images.
[0081] The synthesis module 202 is used to perform image patch matching and reconstruction along the time dimension based on the contrast image and the pre-contrast mask image sequence, and synthesize a pseudo mask image that matches the structure of the contrast image.
[0082] The construction module 203 is used to construct a variable alignment model. The objective function of the variable alignment model includes a similarity metric for aligning the background of the angiographic image with the pseudo-mask image after distortion by the deformation field, a regularization term for constraining the sparsity of vascular information, and an adaptive regularization term for applying spatially differentiated smoothing constraints to the deformation field. The adaptive regularization term constructs a spatial weight map based on the pixel differences between the background of the angiographic image and the pseudo-mask image to dynamically adjust the penalty intensity of the deformation field gradient according to the differences in image regions.
[0083] The iteration module 204 is used to iteratively solve the variable calibration model. During the solution process, the vascular information and deformation field are estimated alternately. The adaptive regularization term guides the generation of deformation field that adapts to the non-uniform motion characteristics during each deformation field estimation.
[0084] The output module 205 is used to output the final deformation field when the final stage is reached, and extract vascular information based on the final background image and the angiography image to obtain a digital subtraction angiography image with motion artifacts removed.
[0085] The present invention also provides a computer-readable storage medium storing a computer program that can be used to execute the above-described... Figure 1 The steps of a digital subtraction angiography method are provided.
[0086] This invention also provides a computer device. At the hardware level, the computer device includes a processor, an internal bus, a network interface, memory, and non-volatile memory, and may also include other hardware required for various operations. The processor reads the corresponding computer program from the non-volatile memory into memory and then executes it to achieve the above-mentioned functions. Figure 1 The steps of a digital subtraction angiography method are provided.
[0087] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0088] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, as well as combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0089] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.
[0090] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0091] It should be noted that the specific embodiments described above enable those skilled in the art to more fully understand the present invention, but do not limit the present invention in any way. Therefore, although the present invention has been described in detail in this specification, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the present invention; and all technical solutions and improvements that do not depart from the spirit and scope of the present invention are covered within the protection scope of the patent of the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.
Claims
1. A digital subtraction angiography method, characterized in that, The method includes: Acquire sequences of contrast images and pre-contrast mask images; Based on the contrast image and the pre-contrast mask image sequence, image block matching and reconstruction are performed along the time dimension to synthesize a pseudo-mask image that matches the structure of the contrast image; A variable alignment model is constructed, wherein the objective function of the variable alignment model includes a similarity metric for aligning the background of the angiography image with the pseudo-mask image after distortion by the deformation field, a regularization term for constraining the sparsity of vascular information, and an adaptive regularization term for applying spatially differentiated smoothing constraints to the deformation field; wherein the adaptive regularization term constructs a spatial weight map based on the pixel differences between the background of the angiography image and the pseudo-mask image to dynamically adjust the penalty intensity of the deformation field gradient according to the differences in image regions; The variable calibration model is iteratively solved, and the vascular information and deformation field are estimated alternately during the solution process. The adaptive regularization term guides the generation of a deformation field that adapts to the non-uniform motion characteristics during each deformation field estimation. When the final stage is reached, the final deformation field is output, and vascular information is extracted based on the final background image and the angiography image to obtain a digital subtraction angiography image with motion artifacts removed.
2. The method according to claim 1, characterized in that, Based on the contrast image and the pre-contrast mask image sequence, image patch matching and reconstruction are performed along the time dimension to synthesize a pseudo-mask image that matches the structure of the contrast image, including: Each frame of the contrast image and the pre-contrast mask image sequence is decomposed into overlapping image blocks; For each spatial location in the contrast image, a matching image block is searched in the pre-contrast mask image sequence along the time dimension, and the optimal matching time index is determined by minimizing the Euclidean distance. A weighted average strategy is used to fuse and reconstruct the pixel values of all optimal matching image blocks covering the same pixel, generating the pseudo-mask image that is spatially continuous and similar in structure to the imaging image.
3. The method according to claim 1, characterized in that, The objective function formula for the variable allocation standardization model is: ； in, For imaging images, For deformation field, For vascular information, For the deformation field gradient, , For regularization parameters, The image is the result of distortion of the pseudomask by a deformation field.
4. The method according to claim 1, characterized in that, In the deformation field registration step, the background target from the previous round is registered with the pseudo-mask image, and the current deformation field is predicted using a deep learning registration network with a self-attention mechanism. ； in, For the TransMorph backbone network, It is a pseudo-mask image. For the first Background target and initial value for stage estimation .
5. The method according to claim 3, characterized in that, In the vascular information estimation step, the current deformation field is fixed, and the objective function is simplified to a univariate optimization problem concerning vascular information. The vascular information is then solved using a soft thresholding operator, which is applied to the residual image between the angiographic image and the deformed pseudomask image to extract sparse vascular information. This yields the... Vascular information in the next iteration : ； in, This is the residual image after the k-th iteration. It is a symbolic function.
6. The method according to claim 1, characterized in that, The background image is updated according to the following formula: ； in, This is the background image updated after the k-th iteration. For imaging images, For the first The vascular information obtained in the next iteration.
7. The method according to claim 1, characterized in that, In the optimization process of alternating estimation, a spatial weight map is constructed based on the pixel difference between the background image and the pseudo-mask image. The spatial weight map is then used to apply a spatially differentiated smoothing constraint to the deformation field, including: Calculate the absolute value of the pixel-by-pixel residual between the background image and the pseudo-mask image in the current iteration stage; The negative exponential function is used to map the absolute value of the residuals into a spatial weight map, so that regions with large residuals correspond to smaller weights, and regions with small residuals correspond to larger weights. The regularization loss is calculated by element-wise multiplying the spatial weight map with the gradients of each spatial dimension of the deformation field, so as to dynamically adjust the smoothing constraint strength of different regions during training or iteration; wherein, the calculation formula of the spatial weight map is: ； in, For spatial weighting, The background image estimated for the k-th stage. It is a pseudo-mask image. To control the scaling factor for the weights' sensitivity to the residuals, It is a Gaussian smoothing kernel.
8. A digital subtraction angiography device, characterized in that, The device includes: The acquisition module is used to acquire sequences of contrast images and pre-contrast mask images; The synthesis module is used to perform image patch matching and reconstruction along the time dimension based on the contrast image and the pre-contrast mask image sequence, and synthesize a pseudo-mask image that matches the structure of the contrast image; A construction module is used to construct a variable alignment model. The objective function of the variable alignment model includes a similarity metric for aligning the background of the angiography image with the pseudo-mask image after distortion by the deformation field, a regularization term for constraining the sparsity of vascular information, and an adaptive regularization term for applying spatially differentiated smoothing constraints to the deformation field. The adaptive regularization term constructs a spatial weight map based on the pixel differences between the background of the angiography image and the pseudo-mask image to dynamically adjust the penalty intensity of the deformation field gradient according to the differences in image regions. The iterative module is used to iteratively solve the variable calibration model, alternately estimating vascular information and deformation field during the solution process, and guiding the generation of deformation field adapted to non-uniform motion characteristics through the adaptive regularization term during each deformation field estimation. The output module is used to output the final deformation field when the final stage is reached, and extract vascular information based on the final background image and the angiography image to obtain a digital subtraction angiography image with motion artifacts removed.
9. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, implements the method described in any one of claims 1 to 7.
10. A computer device, characterized in that, The method includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method described in any one of claims 1 to 7.