Four-dimensional ultrasonic image anisotropy enhancement method and device based on multi-scale tensor field guidance and storage medium

Abstract

The invention provides a four-dimensional ultrasonic image anisotropy enhancement method based on multi-scale tensor field guidance, and relates to the technical field of medical image processing. The anisotropy enhancement of the four-dimensional ultrasonic image is realized through the steps of space-time multi-resolution pyramid construction, low-resolution baseband component analysis and effective anatomical structure indication mask generation, tensor field nonlinear remapping, cross-scale tensor projection, tensor-oriented detail enhancement and image reconstruction. The problem of inaccurate structure direction prediction under strong noise is solved, maze-shaped artifacts generated during denoising are solved, the whole process is non-iterative, complex tensor feature value analysis is only carried out on a low-resolution layer, the calculation cost is reduced, broken vessel wall edges can be filled through the tensor remapping technology, and the structure direction prediction accuracy is improved. And a visual effect similar to long-time iterative diffusion is realized.

Description

Technical Field

[0001] This invention relates to the field of medical image processing technology, specifically to a method, device, and storage medium for anisotropy enhancement of four-dimensional ultrasound images guided by a multi-scale tensor field. Background Technology

[0002] Four-dimensional ultrasound imaging technology can capture the dynamic three-dimensional structure of human organs at a high frame rate, but due to the phase interference principle of ultrasound imaging, severe speckle noise is inevitably present in the images. This noise is multiplicative, tissue-structure related, and manifests as random flickering on the time axis, severely interfering with doctors' observation of fine structures (such as endocardial boundaries and valve movement). Traditional image denoising methods, such as Gaussian smoothing, blur image edges. To preserve edges while denoising, existing advanced techniques typically employ anisotropic diffusion or coherent enhancement diffusion (CED). These methods rely on a "structure tensor" to perceive the local geometric orientation of the image. However, when directly applied to four-dimensional ultrasound, existing techniques have two fatal flaws: First, the calculation of the structure tensor depends on the image gradient, and the strong speckle noise in the original ultrasound image leads to extremely chaotic gradient calculations. To obtain an accurate tensor, pre-smoothing of the image is usually required, but this leads to the loss of fine structures; second, the traditional diffusion equation (PDE) requires multiple iterative evolutions. With large time steps, the tensor field calculated at the initial moment quickly becomes invalid and cannot match the evolved image structure, resulting in step-like artifacts or structural breaks at the edges. If the step size is reduced and the number of iterations is increased, the computational load will increase exponentially, which cannot meet the real-time requirements of ultrasound equipment (usually requiring >20 fps), causing a lag in data iteration updates.

[0003] Therefore, there is an urgent need for a non-iterative four-dimensional ultrasonic processing method that can stably extract structural information from a strong noise background and perform high-quality enhancement in real time.

[0004] It should be noted that the above content is only used to help understand the technical solution of the present invention, and does not represent an admission that the above content is prior art. Summary of the Invention

[0005] The main objective of this invention is to provide a method, device, and storage medium for anisotropy enhancement of four-dimensional ultrasound images based on multi-scale tensor field guidance, aiming to solve the technical problems of chaotic gradient calculation, loss of fine structure, and lagging data iteration updates in the prior art when calculating structural tensors.

[0006] To achieve the above objectives, this invention provides a method for anisotropic enhancement of four-dimensional ultrasound images guided by a multi-scale tensor field. This method includes the following steps: Constructing a spatiotemporal multi-resolution pyramid: The collected discrete four-dimensional ultrasound volume data sequence is downsampled using a predefined spatial interpolation operator to obtain the low-resolution baseband component, and the high-resolution detail component of the discrete four-dimensional ultrasound volume data sequence is calculated. Low-resolution baseband component analysis and generation of effective anatomical structure indicator masks: The four-dimensional gradient vector of the low-resolution baseband component is calculated by a separable spatiotemporal smoothing differential operator to construct the original local structure tensor of the low-resolution baseband component; the eigenvalue features of the original local structure tensor are estimated by matrix trace operation, the local structure energy and the anisotropy of the local structure are calculated, and a sparse mask indicating the effective anatomical structure is generated based on the local structure energy and the anisotropy of the local structure. Tensor field nonlinear remapping: Based on sparse masks and the anisotropy of local structures, the original local structure tensor is sharpened through a nonlinear mapping function to generate a low-resolution control tensor field. The control tensor field exhibits enhanced anisotropy in the structure region and is suppressed in the background region. Cross-scale tensor projection: The low-resolution control tensor field and sparse mask are mapped to the spatial coordinate system of the high-resolution detail components by spatial interpolation operators to obtain the guiding tensor field and high-resolution mask. Tensor-guided detail enhancement: Based on a high-resolution mask, the high-resolution detail components are partitioned: in the background region, isotropic Gaussian smoothing is applied to the high-resolution detail components, and in the structural region, a set of four-dimensional basis filters are controlled by a guiding tensor field for weighted synthesis to obtain the enhanced high-resolution detail components. Image reconstruction: After upsampling the low-resolution baseband components using a spatial interpolation operator, the images are superimposed with the enhanced high-resolution detail components to synthesize an enhanced four-dimensional ultrasound image.

[0007] Optionally, in the step of constructing the original local structure tensor of the low-resolution baseband components, the formula for calculating the original local structure tensor is:

[0008] in For the original local structure tensor, Four-dimensional spacetime coordinates, An isotropic Gaussian smoothing kernel, These are low-resolution baseband components. This is a four-dimensional gradient vector. The individual components are obtained by converting the low-resolution baseband components. It is obtained by performing separable convolution with the corresponding dimensional difference kernel and the smoothing kernel of the orthogonal dimension. This is the matrix transpose operation.

[0009] Optionally, in the steps of estimating the eigenvalues ​​of the original local structure tensor using matrix trace operations and calculating the local structure energy and the anisotropy degree of the local structure, the formula for calculating the anisotropy degree of the local structure is:

[0010] in, For each degree of heterogeneity, It is a function with maximum value. For the original local structure tensor traces, The square of the original local structure tensor is... traces, This is a minimum value, representing the principal eigenvalue of the original local structure tensor. and the secondary eigenvalues ​​of the original local structure tensor The difference With principal eigenvalues and secondary eigenvalues The sum of These are the eigenvalues ​​of the original local structure tensor. The anisotropy degree of the local structure.

[0011] Optionally, in the step of generating the low-resolution control tensor field, the calculation formula for the control tensor field is as follows:

[0012]

[0013] in As an intermediate variable, To control the tensor field, For sparse masks, It is the identity matrix. For even powers, This is the gain coefficient; non-principal eigenvalues ​​are suppressed through even-power operations. The corresponding feature vector components are obtained and processed through a sparse mask. In unstructured regions, the control tensor field is set to zero.

[0014] Optionally, in the step of obtaining the enhanced high-resolution detail components, the enhanced high-resolution detail components are synthesized using a hybrid filtering strategy, the calculation formula of which is:

[0015] in, For high-resolution detail components, For the enhanced high-resolution detail components, For high-resolution masks, For isotropic Gaussian smoothing or soft thresholding noise reduction, For the preset first A four-dimensional spatiotemporal anisotropic basis filter For guiding tensor fields in dual tensors The weighting coefficients obtained by projection upwards, weighting coefficients The calculation formula is:

[0016] in, To guide the tensor field, It is a dual tensor.

[0017] Optionally, in the step of generating a sparse mask indicating the effective anatomical structure based on the local structural energy and the anisotropy of the local structure, the calculation formula for the sparse mask is:

[0018] in, It is a standard sigmoid activation function. For local structural energy, Energy threshold For anisotropy threshold, β is the discriminant gain coefficient for the local structure energy, and β is the discriminant gain coefficient for the anisotropy of the local structure, where the local structure energy... The calculation formula is .

[0019] Optionally, a four-dimensional gradient vector The calculation formula is ; For each dimension d∈{x,y,z,t}, compute the gradient component of the d-th dimension separately. gradient components The spatial approximation method using the Riesz transform operator is employed, specifically as follows: For four-dimensional gradient vectors The first in Construct a four-dimensional separable convolution kernel using gradient components in each dimension. ; Four-dimensional separable convolution kernel By the Difference kernels in each dimension With the four-dimensional gradient vector The other three and the first Smoothing kernel in each orthogonal dimension The tensor product is constructed using a four-dimensional separable convolution kernel. For low-resolution baseband components Perform convolution to obtain the gradient component of the d-th dimension. The gradient component of the d-th dimension The calculation formula is: ; The difference kernel The difference kernel is selected from either the Scharr operator or the Farid-Simoncelli operator, which are optimized for rotation invariance. Configured to extract the first Edge information in each dimension; The smoothing kernel The smoothing kernel is selected from either the Scharr operator or the Farid-Simoncelli operator, which are optimized for rotation invariance. Configured to suppress the first High-frequency noise in orthogonal directions in each dimension is used to ensure that the generated structural tensor has isotropic response characteristics.

[0020] Furthermore, to achieve the above objectives, this invention also proposes a four-dimensional ultrasound image anisotropy enhancement device based on multi-scale tensor field guidance, comprising: Data acquisition module: used to acquire discrete four-dimensional ultrasound body data sequences and construct a spatiotemporal multi-resolution pyramid from the discrete four-dimensional ultrasound body data sequences; Data Augmentation Module: Used to process spatiotemporal multiresolution pyramids constructed from discrete four-dimensional ultrasound body data sequences, configured to perform low-resolution baseband component analysis and generation of masks indicating effective anatomical structures, tensor field nonlinear remapping, cross-scale tensor projection, and tensor-guided detail enhancement steps; Data reconstruction module: used to synthesize enhanced four-dimensional ultrasound images; Data output module: Used to output enhanced four-dimensional ultrasound images.

[0021] Furthermore, to achieve the above objectives, the present invention also proposes a computer-readable storage medium storing at least one instruction, at least one program, code set, or instruction set, wherein the at least one instruction, at least one program, code set, or instruction set is loaded and executed by a processor to implement a method for anisotropic enhancement of four-dimensional ultrasound images based on multi-scale tensor field guidance.

[0022] This invention provides a method, device, and storage medium for anisotropy enhancement of four-dimensional ultrasound images based on multi-scale tensor fields. The invention calculates the structural tensor at the low-resolution baseband component level after downsampling. Because the low-resolution baseband component has a high signal-to-noise ratio, the tensor field calculated at this level exhibits extremely high topological stability. Subsequently, a nonlinear tensor remapping mechanism is introduced to artificially enhance the anisotropy of the tensor through mathematical means, simulating the structure after multiple iterative diffusions. Finally, the simulated structural tensor field after multiple iterative diffusions is projected back to the high-resolution level, enabling directional filtering of high-frequency details. This invention offers stability, solving the problem of inaccurate structural orientation prediction under strong noise and eliminating the "maze-like" artifacts generated during denoising. This invention is real-time, as the entire process is non-iterative, and complex tensor eigenvalue analysis is performed only at the low-resolution layer (with a data volume of 1 / 4 or 1 / 8 of the original image), reducing computational costs. This invention provides high image quality; through tensor remapping technology, it can fill in the edges of ruptured blood vessel walls, achieving a visual effect similar to long-term iterative diffusion. Attached Figure Description

[0023] Figure 1 This illustration shows a schematic diagram of the steps of a four-dimensional ultrasound image anisotropy enhancement method based on a multi-scale tensor field guided by an exemplary embodiment of this application. Figure 2 This illustration shows a schematic diagram of the overall processing flow of a four-dimensional ultrasound image anisotropy enhancement method based on a multi-scale tensor field guided by an exemplary embodiment of this application; Figure 3 This illustration shows a schematic diagram of multi-scale pyramid decomposition and tensor guidance logic provided in an exemplary embodiment of this application; Figure 4 This illustration shows a schematic diagram of the geometric principle of tensor field nonlinear remapping provided in an exemplary embodiment of this application; Figure 5 This illustration shows a schematic diagram of a four-dimensional ultrasound image anisotropy enhancement device based on a multi-scale tensor field guided by an exemplary embodiment of this application. Figure 6 This illustration shows a structural block diagram of a computer device provided in an exemplary embodiment of this application. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of this invention clearer, the following description will be provided in conjunction with the appendix. Figures 1-6 The embodiments of this application will be described in further detail.

[0025] In this article, "multiple" refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. The character " / " generally indicates that the preceding and following related objects have an "or" relationship.

[0026] It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the invention.

[0027] To transfer data between different resolution levels, a spatial interpolation operator is first defined. Spatial interpolation operator The calculation formula is as follows:

[0028] For input data, For any point in the target high-resolution grid, Its corresponding coordinates in the source low-resolution grid are ,make for The integer coordinates of the top left corner, with the decimal part as... , For the target scaling factor, when Spatiotemporal interpolation operators Corresponding downsampling, spatial interpolation operator Perform the decimation operation after low-pass filtering; when Spatiotemporal interpolation operators Corresponding to upsampling, spatial interpolation operator Perform the interpolation operation.

[0029] like Figures 1-2 As shown, Figure 1 This illustration shows a schematic diagram of the steps of a four-dimensional ultrasound image anisotropy enhancement method based on multi-scale tensor field guidance, provided in an exemplary embodiment of this application. Figure 1 The method for enhancing the anisotropy of four-dimensional ultrasound images based on multi-scale tensor field guidance includes the following steps: S100, Constructing a Spatiotemporal Multi-Resolution Pyramid: such as Figure 3 As shown, Figure 3 This illustration shows a schematic diagram of multi-scale pyramid decomposition and tensor guidance logic provided in an exemplary embodiment of this application, constructing a spatiotemporal multi-resolution pyramid as follows: The acquired discrete four-dimensional ultrasound volume data sequence is downsampled using a predefined spatial interpolation operator to obtain the low-resolution baseband component, and the high-resolution detail component of the discrete four-dimensional ultrasound volume data sequence is calculated.

[0030] Acquiring four-dimensional ultrasound data sequences using ultrasound equipment Through an isotropic Gaussian smoothing kernel The acquired four-dimensional ultrasound body data sequence Spatial smoothing is achieved through spatial interpolation operators. (The target scale factor s is set to 0.5) The spatially smoothed four-dimensional ultrasound body data sequence is downsampled to obtain the low-resolution baseband component. Low-resolution baseband components The data sequence of four-dimensional ultrasound body was preserved. The primary temporal resolution is achieved by reducing spatial resolution to suppress speckle noise; subsequently, spatial interpolation operators are used. (Target scale factor s is set to 2) for low-resolution baseband components Upsampling is performed, and residuals are calculated to obtain high-resolution detail components. High-resolution detail component The calculation formula is: At this point, most of the high-frequency speckle noise is isolated. In the middle, the main anatomical structures and movement information are preserved in middle.

[0031] S200, Low-resolution baseband component analysis and generation of effective anatomical structure mask: The four-dimensional gradient vector of the low-resolution baseband component is calculated by a separable spatiotemporal smoothing differential operator to construct the original local structure tensor of the low-resolution baseband component; the eigenvalue features of the original local structure tensor are estimated by matrix trace operation, the local structure energy and the anisotropy of the local structure are calculated, and a sparse mask indicating the effective anatomical structure is generated based on the local structure energy and the anisotropy of the local structure.

[0032] For low-resolution baseband components The following operations are performed at each level to extract robust structural features: Riesz gradient calculation: To obtain a rotationally invariant four-dimensional gradient estimate while avoiding the significant overhead of frequency domain computation, this invention employs a separable spatiotemporal smoothing differential operator to approximate the first-order Riesz transform. The Riesz transform behaves as a phase-shifting operation in the frequency domain and corresponds to a high-dimensional generalization of the Hilbert transform in the spatial domain. This applies to bandpass signals (i.e., low-resolution baseband components). Under the premise of applying a combination of smoothing and difference operators, the first-order Riesz transform is approximated. The specific operation is as follows: Define a four-dimensional gradient vector For each dimension Four-dimensional gradient vector No. Riesz components in each dimension The calculation formula is four-dimensional convolution:

[0033] in, A four-dimensional convolution kernel. It is separable, a four-dimensional convolution kernel. Difference kernel of one dimension Tensor product and smoothing kernel in other three dimensions The tensor product is formed.

[0034] by Components of direction For example, its convolution kernel Defined as:

[0035] Similarly, And so on.

[0036] Preferred, specific difference kernel and smooth kernel The Scharr operator or the Farid-Simoncelli 5-point derivative operator, which has optimal rotation invariance, are selected as follows: Differential kernel The preferred core is a 3-point core [0.5, 0, -0.5] or a 5-point optimized core [0.1096, 0.2766, 0, -0.2766, -0.1096].

[0037] Smooth kernel The preferred cores are 3-point cores [0.25, 0.5, 0.25] or 5-point optimized cores [0.0377, 0.2492, 0.4263, 0.2492, 0.0377].

[0038] By performing difference in the differentiation direction and smoothing in the orthogonal direction, the interference of high-frequency noise on the gradient direction is effectively suppressed, and the isotropicity of gradient calculation is significantly improved. Compared with the simple central difference [1,0, -1], the above approximate implementation given in this patent can more accurately capture the normal direction of continuous structures such as blood vessel walls in ultrasound images, so that the generated structural tensor is closer to the result of the first-order Riesz transform.

[0039] Original tensor construction: Calculate the four-dimensional gradient vector The outer product is then subjected to local Gaussian smoothing to obtain the original local structure tensor. The original local structure tensor The calculation formula is as follows:

[0040] in, An isotropic Gaussian smoothing kernel is used to integrate the four-dimensional gradient vector. The neighborhood information of the outer product smooths out local fluctuations in all directions.

[0041] Anisotropy analysis and sparse mask generation: To efficiently compute and distinguish effective anatomical structures from background noise on a GPU, this invention utilizes the original local structure tensor. traces To quickly estimate feature values ​​and generate a sparse mask based on them, the specific operations are as follows: (1) Fast eigenvalue estimation method: Utilizing the properties of matrix trace and matrix square trace, the approximate original local structure tensor can be analyzed without iteration. Main eigenvalues and secondary eigenvalues .

[0042] First, calculate the original local structure tensor. First-order trace and the original local structure tensor second-order trace Original local structure tensor First-order trace The calculation formula is as follows:

[0043] in, For the original local structure tensor No. Line number Column elements; original local structure tensor second-order trace The calculation formula is as follows:

[0044] in, For the original local structure tensor No. Line number The elements of the column.

[0045] Suppose that the local structure is mainly composed of the local structure tensor. Main eigenvalues and secondary eigenvalues Decision (i.e., setting) (This is approximately true when describing edge or wall structures), then the local structure tensor... Main eigenvalues and secondary eigenvalues The following relationships exist:

[0046]

[0047] Based on algebraic relations The principal eigenvalues ​​can be derived. and secondary eigenvalues eigenvalue difference between With eigenvalues ​​and values The estimation formula is:

[0048]

[0049] This leads to the anisotropy degree. Quick calculation formula:

[0050] in To prevent local minima where the denominator is zero, the anisotropy degree is obtained through a fast estimation method using eigenvalues. The method for quickly calculating formulas involves only matrix multiplication and addition, making it ideal for parallel acceleration via GPUs.

[0051] (2) Sparse mask generate Based on local structure energy and the above anisotropy degree Sparse masks are generated through dual threshold control. sparse mask The calculation formula is:

[0052] in This is a standard sigmoid activation function used to map input values ​​to... The interval is used to achieve soft threshold switching. Standard sigmoid activation function. Defined as:

[0053] Energy threshold, energy threshold Used to distinguish signal regions from background regions, when local structural energy Greater than hour, tending towards 1; For local structural energy Energy discrimination gain coefficient, energy discrimination gain coefficient Controlling the steepness of energy selection, The larger the value, the sparser the mask. The transition is sharper at the boundary between the signal area and the background area; The smaller the value, the sparser the mask. The smoother the transition at the boundary between the signal area and the background area; Anisotropy threshold is used to distinguish oriented structures (such as blood vessel walls) from random textures (such as speckle noise). The anisotropy degree of a local structure... Greater than hour, Tend to 1, Local structure discrimination gain coefficients Sensitivity to control structure selection The larger the value, the stricter the algorithm's requirement for directional consistency. Strict algorithms only retain extremely significant linear or planar structures for directional consistency. This is for exponential operations.

[0054] The above calculation of sparse masks , The region indicates the boundaries of high-energy and well-oriented anatomical structures in the image, and expensive filtering operations are performed only in this region, thus significantly reducing computational costs.

[0055] S300, Tensor Field Nonlinear Remapping: such as Figure 4 As shown, Figure 4 The diagram illustrates the geometric principle of tensor field nonlinear remapping provided in an exemplary embodiment of this application. Tensor field nonlinear remapping is based on sparse masks and the anisotropy of local structures. The original local structure tensor is sharpened by a nonlinear mapping function to generate a low-resolution control tensor field. The control tensor field exhibits enhanced anisotropy in the structure region and is suppressed in the background region.

[0056] right Perform nonlinear mapping to generate a control tensor field. To strengthen the structure, only in sparse masks Enhance areas with higher values:

[0057]

[0058] in Even numbers (e.g.) Setting it to 8) enables the control of the tensor field. It exhibits strong anisotropy (flattened ellipsoid) in the structural region, while it is suppressed to zero tensor in the background region.

[0059] S400, cross-scale tensor projection: The low-resolution control tensor field and sparse mask are mapped to the spatial coordinate system of the high-resolution detail components through spatial interpolation operators to obtain the guiding tensor field and high-resolution mask.

[0060] Using spatial interpolation operators Low-resolution control tensor field Mapped to high-resolution detail components The guiding tensor field is obtained in the coordinate space where it is located. Guiding the tensor field The calculation formula is:

[0061] in, It is a low-resolution coordinate system. These are high-resolution coordinates. This is due to the control tensor field. The guiding tensor field is calculated and nonlinearly sharpened in a low-noise environment. It can provide smooth and continuous structural orientation guidance for high-resolution layers, and repair edge breaks caused by speckle noise.

[0062] S500, Tensor-guided detail enhancement: Based on a high-resolution mask, high-resolution detail components are partitioned: in the background region, isotropic Gaussian smoothing is applied to the high-resolution detail components, and in the structural region, a set of four-dimensional basis filters are controlled by a guiding tensor field for weighted synthesis to obtain the enhanced high-resolution detail components.

[0063] Using guided tensor fields and high-resolution masks For high-resolution detail components Processing is performed, including high-resolution masks. Through spatial interpolation operators For sparse masks Upsampling yields a high-resolution mask. The calculation formula is: The specific operating steps are as follows: Pre-design a set of total A four-dimensional spatiotemporal anisotropic basis filter To process four-dimensional data, these four-dimensional spatiotemporal anisotropic basis filters must be 4D Gabor filters or 4D Gaussian second-order derivative filters that include a time dimension. For example: Smoothing occurs spatially along a specific direction (such as the course of blood vessels) and temporally along a motion trajectory. The number of four-dimensional spatiotemporal anisotropic basis filters. Based on the symmetry of four-dimensional space (it is recommended that M≥4, for example...), Or adopt a simplified version Spatial orientation + time axis processing).

[0064] Calculate the weights of each four-dimensional spatiotemporal anisotropic basis filter. ,in It is related to the four-dimensional spatiotemporal anisotropic basis filter The dual tensor corresponding to the direction.

[0065] To reduce computational costs, based on high-resolution masks The image is divided into structural regions and background regions, and these regions are processed separately to obtain enhanced high-resolution detail components. Enhanced high-resolution detail components The calculation formula is:

[0066] The background area is a high-resolution mask. For background areas, a low-cost IsoSmooth operation can be performed, which can choose isotropic Gaussian smoothing or simple soft thresholding to remove background speckles.

[0067] The structural region is a high-resolution mask. The structural regions undergo expensive, tensor-guided basis filter weighted synthesis, which is enhanced with high fidelity along the anatomical structure and direction of motion.

[0068] S600, Image Reconstruction: After upsampling the low-resolution baseband components using a spatial interpolation operator, the images are superimposed with the enhanced high-resolution detail components to synthesize an enhanced four-dimensional ultrasound image.

[0069] Through spatial interpolation operators The original low-resolution baseband components Upsampling, and enhanced high-resolution detail components Add them together to create the final four-dimensional ultrasound image:

[0070] Through the above steps, this invention effectively solves the contradiction between "structure preservation" and "noise suppression" in four-dimensional ultrasound image denoising while ensuring real-time performance.

[0071] This invention provides a method, device, and storage medium for anisotropy enhancement of four-dimensional ultrasound images based on multi-scale tensor fields. The invention calculates the structural tensor at the low-resolution baseband component level after downsampling. Because the low-resolution baseband component has a high signal-to-noise ratio, the tensor field calculated at this level exhibits extremely high topological stability. Subsequently, a nonlinear tensor remapping mechanism is introduced to artificially enhance the anisotropy of the tensor through mathematical means, simulating the structure after multiple iterative diffusions. Finally, the simulated structural tensor field after multiple iterative diffusions is projected back to the high-resolution level, enabling directional filtering of high-frequency details. This invention offers stability, solving the problem of inaccurate structural orientation prediction under strong noise and eliminating the "maze-like" artifacts generated during denoising. This invention is real-time, as the entire process is non-iterative, and complex tensor eigenvalue analysis is performed only at the low-resolution layer (with a data volume of 1 / 4 or 1 / 8 of the original image), reducing computational costs. This invention provides high image quality; through tensor remapping technology, it can fill in the edges of ruptured blood vessel walls, achieving a visual effect similar to long-term iterative diffusion.

[0072] Furthermore, in the step of constructing the original local structure tensor of the low-resolution baseband components, the formula for calculating the original local structure tensor is:

[0073] in For the original local structure tensor, Four-dimensional spacetime coordinates, An isotropic Gaussian smoothing kernel, These are low-resolution baseband components. This is a four-dimensional gradient vector. The components are obtained by... It is obtained by performing separable convolution with the corresponding dimensional difference kernel and the smoothing kernel of the orthogonal dimension. This is the matrix transpose operation.

[0074] Furthermore, in the steps of estimating the eigenvalues ​​of the original local structure tensor using matrix trace operations and calculating the local structure energy and the anisotropy degree of the local structure, the formula for calculating the anisotropy degree of the local structure is:

[0075] in, For each degree of heterogeneity, It is a function with maximum value. For the original local structure tensor traces, The square of the original local structure tensor is... traces, This is a minimum value, representing the principal eigenvalue of the original local structure tensor. and the secondary eigenvalues ​​of the original local structure tensor The difference With principal eigenvalues and secondary eigenvalues The sum of These are the eigenvalues ​​of the original local structure tensor. The anisotropy degree of the local structure.

[0076] Furthermore, in the step of generating the low-resolution control tensor field, the calculation formula for the control tensor field is as follows:

[0077]

[0078] in As an intermediate variable, To control the tensor field, For sparse masks, It is the identity matrix. For even powers, This is the gain coefficient; non-principal eigenvalues ​​are suppressed through even-power operations. The corresponding feature vector components are obtained and processed through a sparse mask. In unstructured regions, the control tensor field is set to zero.

[0079] Furthermore, in the step of obtaining the enhanced high-resolution detail components, the enhanced high-resolution detail components are synthesized using a hybrid filtering strategy, the calculation formula of which is:

[0080] in, For high-resolution detail components, For the enhanced high-resolution detail components, For high-resolution masks, For isotropic Gaussian smoothing or soft thresholding noise reduction, For the preset first A four-dimensional spatiotemporal anisotropic basis filter For guiding tensor fields in dual tensors The weighting coefficients obtained by projection upwards, weighting coefficients The calculation formula is:

[0081] in, To guide the tensor field, It is a dual tensor.

[0082] Furthermore, in the step of generating a sparse mask indicating the effective anatomical structure based on the local structural energy and the anisotropy of the local structure, the calculation formula for the sparse mask is:

[0083] in, It is a standard sigmoid activation function. For local structural energy, Energy threshold For anisotropy threshold, β is the discriminant gain coefficient for the local structure energy, and β is the discriminant gain coefficient for the anisotropy of the local structure, where the local structure energy... The calculation formula is .

[0084] Furthermore, the four-dimensional gradient vector The calculation formula is ; For each dimension Calculate the first one separately gradient components in each dimension gradient components The spatial approximation method using the Riesz transform operator is employed, specifically as follows: For four-dimensional gradient vectors The first in Construct a four-dimensional separable convolution kernel using gradient components in each dimension. ; Four-dimensional separable convolution kernel By the Difference kernels in each dimension With the four-dimensional gradient vector The other three and the first Smoothing kernel in each orthogonal dimension The tensor product is constructed using a four-dimensional separable convolution kernel. For low-resolution baseband components Perform convolution to obtain the gradient component of the d-th dimension. The gradient component of the d-th dimension The calculation formula is: ; The difference kernel The difference kernel is selected from either the Scharr operator or the Farid-Simoncelli operator, which are optimized for rotation invariance. Configured to extract the first Edge information in each dimension; The smoothing kernel The smoothing kernel is selected from either the Scharr operator or the Farid-Simoncelli operator, which are optimized for rotation invariance. Configured to suppress the first High-frequency noise in orthogonal directions in each dimension is used to ensure that the generated structural tensor has isotropic response characteristics.

[0085] To achieve the above objectives, this invention also proposes a four-dimensional ultrasound image anisotropy enhancement device based on multi-scale tensor field guidance, such as... Figure 5 As shown, Figure 5 This illustration shows a schematic diagram of a four-dimensional ultrasound image anisotropy enhancement device based on multi-scale tensor field guidance, provided in an exemplary embodiment of this application. The four-dimensional ultrasound image anisotropy enhancement device based on multi-scale tensor field guidance includes: Data acquisition module 810: used to acquire discrete four-dimensional ultrasound body data sequences and construct a spatiotemporal multi-resolution pyramid on the discrete four-dimensional ultrasound body data sequences; Data Augmentation Module 820: Used to process spatiotemporal multiresolution pyramids constructed from discrete four-dimensional ultrasound body data sequences, configured to perform low-resolution baseband component analysis and generation of masks indicating effective anatomical structures, tensor field nonlinear remapping, cross-scale tensor projection, and tensor-guided detail enhancement steps; Data reconstruction module 830: used to synthesize enhanced four-dimensional ultrasound images; Data output module 840: Used to output enhanced four-dimensional ultrasound images.

[0086] Furthermore, to achieve the above objectives, the present invention also proposes a computer-readable storage medium storing at least one instruction, at least one program, code set, or instruction set, wherein the at least one instruction, at least one program, code set, or instruction set is loaded and executed by a processor to implement a method for anisotropic enhancement of four-dimensional ultrasound images based on multi-scale tensor field guidance.

[0087] It should be noted that the anisotropic enhancement device for four-dimensional ultrasound images based on multi-scale tensor field guidance provided in this application embodiment is only an example of the above-mentioned division of functional modules / functional units. In practical applications, the above functions can be assigned to different functional modules / functional units as needed, that is, the internal structure of a four-dimensional ultrasound image anisotropic enhancement device based on multi-scale tensor field guidance can be divided into different functional modules / functional units to complete all or part of the functions described above.

[0088] Figure 6This illustration shows a structural block diagram of a computer device provided in an exemplary embodiment of this application. The computer device can be a desktop computer, a laptop computer, a handheld computer, or a cloud server, etc. The computer device may include, but is not limited to, a processor and memory. The processor and memory can be connected via a bus or other means. The processor can be a Central Processing Unit (CPU). The processor can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs) or other programmable logic devices, graphics processing units (GPUs), embedded neural network processing units (NPUs) or other dedicated deep learning coprocessors, discrete gate or transistor logic devices, discrete hardware components, or combinations of the above types of chips.

[0089] The processor may include one or more processing cores, such as a quad-core processor or an octa-core processor. The processor may be implemented using at least one hardware form of DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), or PLA (Programmable Logic Array). The processor may also include a main processor and coprocessors. The main processor, also known as the CPU (Central Processing Unit), is used to process data in the wake-up state; the coprocessor is a low-power processor used to process data in the standby state. In some embodiments, the processor may integrate a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content required to be displayed on the screen. In some embodiments, the processor may also include an AI (Artificial Intelligence) processor, which is used to handle computational operations related to machine learning.

[0090] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs, non-transitory computer-executable programs, and modules, such as the program instructions / modules corresponding to the methods in the above embodiments of this application. The processor executes various functional applications and data processing by running the non-transitory software programs, instructions, and modules stored in the memory, thereby implementing the methods in the above embodiments. The memory may include a program storage area and a data storage area, wherein the program storage area may store the operating system and at least one application program required for a function; the data storage area may store data created by the processor, etc. Furthermore, the memory may include high-speed random access memory and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, the memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0091] In some embodiments, the computer device may also optionally include: a peripheral device interface and at least one peripheral device. The processor, memory, and peripheral device interface can be connected via a bus or signal lines. Each peripheral device can be connected to the peripheral device interface via a bus, signal lines, or a circuit board. Specifically, the peripheral device includes at least one of: a radio frequency circuit, a display screen, and a keyboard.

[0092] Peripheral device interfaces can be used to connect at least one I / O (Input / Output) related peripheral device to the processor and memory. In some embodiments, the processor, memory, and peripheral device interface are integrated on the same chip or circuit board; in some other embodiments, any one or two of the processor, memory, and peripheral device interface can be implemented on separate chips or circuit boards, which is not limited in this embodiment.

[0093] The display screen is used to display the UI (User Interface). This UI can include graphics, text, icons, videos, and any combination thereof. When the display screen is a touch screen, it also has the ability to collect touch signals on or above the surface of the display. These touch signals can be input as control signals to a processor for processing. In this case, the display screen can also be used to provide virtual buttons and / or a virtual keyboard, also known as soft buttons and / or a soft keyboard. In some embodiments, there may be one display screen, located on the front panel of the computer device; in other embodiments, there may be at least two display screens, respectively located on different surfaces of the computer device or in a folded design; in still other embodiments, the display screen may be a flexible display screen, located on a curved or folded surface of the computer device. Furthermore, the display screen can be configured as a non-rectangular, irregular shape, i.e., a non-rectangular screen. The display screen can be made of materials such as LCD (Liquid Crystal Display) and OLED (Organic Light-Emitting Diode).

[0094] A power supply is used to power the various components in a computer device. The power supply can be alternating current (AC), direct current (DC), a disposable battery, or a rechargeable battery. When the power supply includes a rechargeable battery, the rechargeable battery can be a wired rechargeable battery or a wireless rechargeable battery. A wired rechargeable battery is charged via a wired connection, while a wireless rechargeable battery is charged via a wireless coil. The rechargeable battery can also be used to support fast charging technology.

[0095] Those skilled in the art will understand that the structure shown in this embodiment does not constitute a limitation on the computer device, and may include more or fewer components than shown, or combine certain components, or use different component arrangements.

[0096] This application also discloses a computer-readable storage medium. Specifically, the computer-readable storage medium is used to store a computer program, which, when executed by a processor, implements the methods described in the above-described method embodiments. Those skilled in the art will understand that implementing all or part of the processes in the methods described above can be accomplished by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments described above. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk drive (HDD), or solid-state drive (SSD), etc.; the storage medium can also include combinations of the above types of memory.

[0097] The above are merely preferred embodiments of the present invention and do not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A method for enhancing the anisotropy of four-dimensional ultrasound images based on multi-scale tensor field guidance, characterized in that, The proposed method for enhancing the anisotropy of four-dimensional ultrasound images based on multi-scale tensor fields includes the following steps: Constructing a spatiotemporal multi-resolution pyramid: Using a predefined spatial interpolation operator, the acquired discrete four-dimensional ultrasound volume data sequence is downsampled to obtain low-resolution baseband components, and the high-resolution detail components of the discrete four-dimensional ultrasound volume data sequence are calculated. Low-resolution baseband component analysis and generation of effective anatomical structure indicator mask: The four-dimensional gradient vector of the low-resolution baseband component is calculated using a separable spatiotemporal smoothing differential operator to construct the original local structure tensor of the low-resolution baseband component; the eigenvalue features of the original local structure tensor are estimated using matrix trace operation; the local structure energy and the anisotropy of the local structure are calculated; and a sparse mask indicating the effective anatomical structure is generated based on the local structure energy and the anisotropy of the local structure. Tensor field nonlinear remapping: Based on the sparse mask and the anisotropy of the local structure, the original local structure tensor is sharpened through a nonlinear mapping function to generate a low-resolution control tensor field. The control tensor field exhibits enhanced anisotropy in the structural region and is suppressed in the background region. Cross-scale tensor projection: The low-resolution control tensor field and the sparse mask are mapped to the spatial coordinate system of the high-resolution detail component by the spatial interpolation operator to obtain the guiding tensor field and the high-resolution mask. Tensor-guided detail enhancement: Based on the high-resolution mask, the high-resolution detail components are partitioned: the high-resolution detail components are subjected to isotropic Gaussian smoothing in the background region, and a set of four-dimensional basis filters are weighted and synthesized in the structural region by controlling a set of four-dimensional basis filters through the guiding tensor field to obtain the enhanced high-resolution detail components. Image reconstruction: After upsampling the low-resolution baseband components using a spatial interpolation operator, the images are superimposed with the enhanced high-resolution detail components to synthesize an enhanced four-dimensional ultrasound image.

2. The method for enhancing the anisotropy of four-dimensional ultrasound images based on multi-scale tensor field guidance according to claim 1, characterized in that: In the step of constructing the original local structure tensor of the low-resolution baseband component, the formula for calculating the original local structure tensor is: in For the original local structure tensor, Four-dimensional spacetime coordinates, An isotropic Gaussian smoothing kernel, The low-resolution baseband component. The four-dimensional gradient vector is a four-dimensional gradient vector. The individual components are obtained by converting the low-resolution baseband components. It is obtained by performing separable convolution with the corresponding dimensional difference kernel and the smoothing kernel of the orthogonal dimension. This is the matrix transpose operation.

3. The method for enhancing the anisotropy of four-dimensional ultrasound images based on multi-scale tensor field guidance according to claim 1, characterized in that: In the steps of estimating the eigenvalues ​​of the original local structure tensor using matrix trace operations and calculating the local structure energy and the anisotropy of the local structure, the formula for calculating the anisotropy of the local structure is as follows: in, For the aforementioned anisotropy, It is a function with maximum value. For the original local structure tensor traces, The square of the original local structure tensor is... traces, The principal eigenvalue of the original local structure tensor is a minimum value. and the secondary eigenvalues ​​of the original local structure tensor The difference With the principal eigenvalue and the secondary eigenvalues The sum of The eigenvalues ​​of the original local structure tensor are the features. The anisotropy degree of the local structure is denoted as .

4. The method for enhancing the anisotropy of four-dimensional ultrasound images based on multi-scale tensor field guidance according to claim 3, characterized in that: In the step of generating the low-resolution control tensor field, the calculation formula for the control tensor field is as follows: in As an intermediate variable, For the control tensor field, For the sparse mask, It is the identity matrix. For even powers, The gain coefficient is used to suppress non-principal eigenvalues ​​through even-power operations. The corresponding feature vector vector components, and through the sparse mask Set the control tensor field to zero in the unstructured region.

5. The method for enhancing the anisotropy of four-dimensional ultrasound images based on multi-scale tensor field guidance according to claim 1, characterized in that: In the step of obtaining the enhanced high-resolution detail components, the enhanced high-resolution detail components are synthesized using a hybrid filtering strategy, the calculation formula of which is: in, For high-resolution detail components, For the enhanced high-resolution detail components, For the high-resolution mask, For the isotropic Gaussian smoothing or soft thresholding noise reduction, For the preset first A four-dimensional spatiotemporal anisotropic basis filter For guiding tensor fields in dual tensors The weighting coefficients obtained by projection, the weighting coefficients The calculation formula is: in, Let the guiding tensor field be defined. Let be the dual tensor.

6. The method for enhancing the anisotropy of four-dimensional ultrasound images based on multi-scale tensor field guidance according to claim 1, characterized in that: In the step of generating a sparse mask indicating the effective anatomical structure based on the local structure energy and the anisotropy of the local structure, the calculation formula for the sparse mask is: in, It is a standard sigmoid activation function. The local structural energy, Energy threshold For anisotropy threshold, The discrimination gain coefficient for the local structure energy is... The discriminant gain coefficient is the anisotropy degree of the local structure, wherein the local structure energy The calculation formula is .

7. The method for enhancing the anisotropy of four-dimensional ultrasound images based on multi-scale tensor field guidance according to claim 2, characterized in that: The four-dimensional gradient vector The calculation formula is ; For each dimension Calculate the first one separately gradient components in each dimension The gradient component The spatial approximation method using the Riesz transform operator is employed, specifically as follows: For the four-dimensional gradient vector The first in Construct a four-dimensional separable convolution kernel using gradient components in each dimension. ; The four-dimensional separable convolution kernel From the first Difference kernels in each dimension With the four-dimensional gradient vector The other three and the first Smoothing kernel in each orthogonal dimension The tensor product is constructed using the four-dimensional separable convolution kernel. For the low-resolution baseband components Perform convolution to obtain the gradient component of the d-th dimension. The gradient component of the d-th dimension The calculation formula is: .

8. The method for enhancing the anisotropy of four-dimensional ultrasound images based on multi-scale tensor field guidance according to claim 7, characterized in that: The difference kernel The difference kernel is selected from either the Scharr operator or the Farid-Simoncelli operator, which are optimized for rotation invariance. Configured to extract the first Edge information in each dimension; The smoothing kernel The smoothing kernel is selected from either the Scharr operator or the Farid-Simoncelli operator, which are optimized for rotation invariance. Configured to suppress the first High-frequency noise in orthogonal directions in each dimension is used to ensure that the generated structural tensor has isotropic response characteristics.

9. A device for enhancing the anisotropy of four-dimensional ultrasound images based on multi-scale tensor field guidance, characterized in that, The aforementioned four-dimensional ultrasound image anisotropy enhancement device based on multi-scale tensor field guidance includes: Data acquisition module: used to acquire discrete four-dimensional ultrasound body data sequences and construct a spatiotemporal multi-resolution pyramid based on the discrete four-dimensional ultrasound body data sequences; Data augmentation module: used to process the spatiotemporal multiresolution pyramid constructed from the discrete four-dimensional ultrasound body data sequence, configured to perform low-resolution baseband component analysis and generation of effective anatomical structure mask, tensor field nonlinear remapping, cross-scale tensor projection, and tensor-guided detail enhancement steps; Data reconstruction module: used to synthesize enhanced four-dimensional ultrasound images; Data output module: Used to output enhanced four-dimensional ultrasound images.

10. A computer-readable storage medium, characterized in that, The readable storage medium stores at least one instruction, at least one program, code set, or instruction set, wherein the at least one instruction, the at least one program, the code set, or instruction set is loaded and executed by a processor to implement a method for enhancing the anisotropy of four-dimensional ultrasound images based on a multi-scale tensor field as described in any one of claims 1 to 7.

Images