Central vein risk hierarchical evaluation system and method based on data fusion

Abstract

The invention provides a central vein risk hierarchical evaluation system and method based on data fusion. The method comprises the following steps: acquiring a fixed image and a dynamic image; performing interpolation transformation on the dynamic image and calculating a rigid registration value of the fixed image and the transformed dynamic image; performing rigid registration on the fixed image and the dynamic image to obtain an initial registration fixed image and an initial registration dynamic image; constructing an image pyramid and carrying out deformable registration to obtain a target registration fixed image and a target registration dynamic image; based on the target registration fixed image and the target registration dynamic image, respectively extracting anatomical information features and functional information features; and constructing a target three-dimensional anatomical model based on the anatomical information features and the functional information features, performing risk hierarchical assessment on the central vein, and outputting a risk hierarchical assessment result. According to the method, two kinds of non-contrast enhanced sequence image fusion are adopted to fully consider association between anatomical information and functional information, and the reliability of a central vein risk assessment result is improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent data processing, and in particular to a central venous risk stratification assessment system and method based on data fusion. Background Technology

[0002] Central veins are crucial pathways for blood circulation in the human body, primarily including the brachiocephalic vein, subclavian vein, and superior vena cava. The patency of central veins is a critical factor in addressing various clinical conditions, especially during hemodialysis, where narrowing or occlusion can directly lead to dialysis access failure. Therefore, improving the reliability of central venous risk stratification assessment results is a technical issue that requires further research.

[0003] Currently, existing technologies primarily employ digital subtraction angiography (DSA), computed tomography (CT) angiography, color Doppler ultrasound, and traditional magnetic resonance angiography (MRI) for central venous risk stratification. While DSA is considered the "gold standard" for diagnosing anatomical stenosis, its invasiveness, radiation exposure, and inability to quantify hemodynamic parameters lead to unreliable risk assessment results. Furthermore, while CT angiography is non-invasive, it involves radiation and requires contrast agent injection, limiting its accuracy. Color Doppler ultrasound is non-invasive and radiation-free, but its acoustic window limitation restricts its visualization of deep central veins, resulting in unreliable risk assessments. Traditional MRI requires gadolinium contrast agents and, for tortuous or variable-velocity central veins, is prone to signal loss and exaggeration of stenosis, impacting the accuracy of risk assessments. Therefore, existing technologies lack integrated analysis of anatomical and functional information, resulting in unreliable risk assessment results for central veins. Summary of the Invention

[0004] To address the aforementioned issues, this invention proposes a central venous risk stratification assessment system and method based on data fusion. This system utilizes the fusion of two non-contrast enhanced image sequences to fully consider the correlation between anatomical and functional information, thereby improving the reliability of central venous risk assessment results.

[0005] To achieve the above objectives, embodiments of the present invention provide a central venous risk stratification assessment system based on data fusion, comprising: an image acquisition module for acquiring fixed and dynamic images within a preset anatomical range; a mutual information calculation module for interpolating and transforming the dynamic images to obtain transformed dynamic images and calculating rigid registration mutual information values ​​between the fixed and transformed dynamic images; a rigid registration module for rigidly registering the fixed and dynamic images based on the rigid registration mutual information values ​​to obtain initially registered fixed and initially registered dynamic images; and a deformable registration module for using a preset image sequence processing algorithm and the initially registered fixed image... The system uses an initial registration dynamic image and an image pyramid to obtain an image pyramid. Deformable registration is then performed on the image pyramid to obtain a target registration fixed image and a target registration dynamic image. A first feature extraction module is used to obtain anatomical information features based on a preset anatomical feature extraction algorithm and the target registration fixed image. A second feature extraction module is used to obtain functional information features based on a preset functional feature extraction algorithm and the target registration dynamic image. A risk stratification assessment module is used to fuse anatomical and functional information features to construct a target three-dimensional anatomical model. Based on the target three-dimensional anatomical model, a risk stratification assessment of the central vein is performed, and the risk stratification assessment results are output.

[0006] This invention proposes a central venous risk stratification assessment system based on data fusion. An image acquisition module acquires corresponding fixed and dynamic images within a preset anatomical range. A mutual information calculation module performs rigid registration on the fixed and dynamic images, followed by deformable registration on the initially registered fixed and dynamic images, providing a reliable data foundation for subsequent feature extraction. Then, a first and second feature extraction module extracts anatomical and functional information features from the target registered fixed and dynamic images, respectively. Finally, a risk stratification assessment module fuses these anatomical and functional features to perform risk stratification assessment of the central venous system, yielding the assessment results. Thus, by integrating image acquisition, registration, feature extraction, and fusion assessment, the system performs a unified analysis of anatomical and functional information features, ultimately providing a comprehensive assessment from morphology to function. This system achieves improved reliability of central venous risk assessment results by fusing two non-contrast-enhanced image sequences to fully consider the correlation between anatomical and functional information.

[0007] Furthermore, the image acquisition module is used to acquire fixed and dynamic images within a preset anatomical range, including: performing three-dimensional sequence scanning within the preset anatomical range to obtain fixed images; performing four-dimensional sequence scanning within the preset anatomical range to obtain time-resolved three-dimensional imaging data; and performing velocity encoding on the time-resolved three-dimensional imaging data to obtain dynamic images.

[0008] In the above scheme, three-dimensional sequence scanning, four-dimensional sequence scanning and velocity encoding techniques are used to obtain two types of non-contrast-enhanced image sequences: fixed images and dynamic images. By using these two types of image sequences in combination, accurate anatomical and functional information features can be obtained simultaneously without relying on contrast agents and radiation. This solves the problem of poor reliability of risk assessment results for central veins due to the lack of consideration of anatomical and functional information features in existing technologies, and helps to improve the reliability of central vein risk assessment results.

[0009] Furthermore, the mutual information calculation module is used to perform interpolation transformation on the dynamic image to obtain the transformed dynamic image and calculate the rigid registration mutual information value between the fixed image and the transformed dynamic image. This includes: calculating transformation interpolation based on the dynamic image, a preset rotation matrix, and a preset translation vector; performing interpolation transformation on the dynamic image based on the transformation interpolation to obtain the transformed dynamic image; identifying the overlapping region between the fixed image and the transformed dynamic image, and obtaining the grayscale value of the fixed image and the grayscale value of the transformed dynamic image at each pixel position in the overlapping region; constructing a joint histogram based on the grayscale values ​​of the fixed image and the transformed dynamic image, and calculating a joint probability distribution based on the joint histogram; generating a probability distribution of grayscale values ​​for the fixed image and the transformed dynamic image based on the joint probability distribution; calculating the information entropy of the fixed image based on the grayscale value probability distribution of the fixed image; calculating the information entropy of the dynamic image based on the grayscale value probability distribution of the transformed dynamic image; and calculating the rigid registration mutual information value between the fixed image and the transformed dynamic image based on the information entropy of the fixed image and the information entropy of the dynamic image.

[0010] In the above scheme, mutual information is used as a similarity measure for multimodal image registration. It does not rely on the linear relationship between gray levels between images and can effectively handle non-contrast enhancement sequence images with two different imaging mechanisms. The mutual information value is then used as the target for subsequent registration optimization, providing an objective and reliable data foundation for subsequent data processing. This ensures the initial accurate alignment of fixed and dynamic images in the overall spatial position, which helps to improve the reliability of central venous risk assessment results.

[0011] Furthermore, the rigid registration module is used to perform rigid registration on a fixed image and a dynamic image based on the rigid registration mutual information value to obtain an initial registered fixed image and an initial registered dynamic image. This includes: constructing a mutual information value gradient vector based on the transform interpolation and the rigid registration mutual information value; updating the transform interpolation along the gradient direction; and recalculating the rigid registration mutual information value of the fixed image and the transformed dynamic image whenever the transform interpolation is updated to obtain an optimized rigid registration mutual information value; if the optimized rigid registration mutual information value meets the preset registration requirements, the initial registered fixed image and the initial registered dynamic image are obtained.

[0012] In the above scheme, a mutual information value gradient vector is constructed and the transformation parameters are iteratively updated along the gradient direction to make the mutual information value meet the preset registration requirements, thereby realizing automatic and accurate registration between images. By transforming the complex spatial alignment problem into a solvable numerical optimization problem, the reliability of registration is improved. The rigid registration provides a good initial alignment state for subsequent deformable registration, avoiding the problem of poor reliability of subsequent registration due to excessive initial position deviation, and effectively improving the reliability of central venous risk assessment results.

[0013] Furthermore, the deformable registration module is used to obtain an image pyramid based on a preset image sequence processing algorithm, an initial registration fixed image, and an initial registration dynamic image, and to perform deformable registration on the image pyramid to obtain a target registration fixed image and a target registration dynamic image. This includes: traversing the initial registration fixed image and the initial registration dynamic image, and performing smooth sampling sequentially according to the image resolution to generate the image pyramid; calculating the mutual information value of the initial registration fixed image and the initial registration dynamic image, and constructing an optimization objective function based on a preset regularization penalty term; traversing the image pyramid according to a preset direction, and performing deformable registration on each layer of the image pyramid until each layer of the image pyramid has completed deformable registration until the optimization objective function meets the preset optimization requirements, thereby obtaining the target registration fixed image and the target registration dynamic image. The process of deformable registration, which involves traversing the image pyramid in a preset direction and performing deformable registration on each layer of the image pyramid until each layer is deformably registered and the objective function meets the preset optimization requirements, to obtain a fixed image and a dynamic image of target registration, includes the following steps: covering the image corresponding to each layer of the image pyramid with a control point grid; calculating the normalized coordinates of any coordinate point in the local grid cell for any coordinate point in the image; determining the function value of the coordinate point in the normalized coordinates based on the control point index value of the control point grid, and using the function value as the influence weight of the control point; calculating the deformation value of the coordinate point based on the influence weight and coordinates of the control point, and performing deformable registration based on the deformation value.

[0014] In the above scheme, an image pyramid and a regularization penalty term are introduced for deformable registration. An image pyramid is constructed, and a layer-by-layer optimization strategy is used to quickly capture overall deformation at a large scale, followed by optimization of local details at a fine scale. This effectively avoids getting trapped in local optima. Complex image deformation is modeled as a displacement problem of sparse control points, and a regularization penalty term is used to prevent non-physical excessive deformation or folding, ensuring that the optimization objective function meets preset optimization requirements. This yields both a fixed-target registration image and a dynamic-target registration image. Therefore, based on the good initial alignment provided by rigid registration, deformable registration of the image provides a reliable data foundation for subsequent feature extraction, helping to improve the reliability of central venous risk assessment results.

[0015] Furthermore, the first feature extraction module is used to obtain anatomical information features based on a preset anatomical feature extraction algorithm and a target registration fixed image, including: extracting thrombus region features and vessel outer wall contour features from the target registration fixed image based on the preset anatomical feature extraction algorithm, and constructing corresponding thrombus 3D models and 3D vessel outer wall models according to a preset 3D reconstruction algorithm; generating several cross-sections based on the vessel orientation of the 3D vessel outer wall model; calculating the unobstructed lumen area based on the several cross-sections, and determining the stenosis location based on the unobstructed lumen area; determining the thrombus boundary based on the thrombus 3D model at the stenosis location, and determining the stenosis length at the stenosis location based on the thrombus boundary; obtaining vessel wall morphological features based on the 3D vessel outer wall model at the stenosis location; calculating the absolute stenosis degree based on the thrombus 3D model and the 3D vessel outer wall model at the stenosis location; and outputting the stenosis location, stenosis length, vessel wall morphological features, and absolute stenosis degree as anatomical information features.

[0016] In the above scheme, a preset anatomical feature extraction algorithm is used to extract thrombus region features and vessel outer wall contour features. Combined with a three-dimensional reconstruction algorithm, a three-dimensional model of the thrombus and a three-dimensional model of the vessel outer wall are constructed. Then, the stenosis morphology of the central vein is depicted. Subjective image interpretation is transformed into objective and repeatable quantitative data. The output anatomical information features can reflect the real impact of irregular thrombus lesions, providing a more reliable anatomical basis for subsequent risk assessment and helping to improve the reliability of central vein risk assessment results.

[0017] Furthermore, the second feature extraction module is used to obtain functional information features based on a preset functional feature extraction algorithm and a target registration dynamic image, including: obtaining the vessel wall surface and vessel cross-section based on the target registration dynamic image, the thrombus 3D model, and the 3D vessel outer wall model; calculating the gradient of blood flow velocity in the direction perpendicular to the vessel wall surface based on the vessel wall surface to obtain the wall shear stress; determining the oscillatory shear index based on the preset cardiac cycle and wall shear stress; solving for energy loss based on the vessel cross-section; and outputting the wall shear stress, oscillatory shear index, and energy loss as functional information features.

[0018] In the above scheme, key hemodynamic features are extracted from the registered blood flow data to quantify the mechanical effects of blood flow on the vessel wall, the degree of blood flow turbulence, and the mechanical energy loss caused by stenosis. Wall shear stress, oscillatory shear index, and energy loss are used as core indicators to assess the functional significance of stenosis, providing a reliable data foundation for subsequent in-depth analysis from morphological to functional assessment and effectively improving the reliability of central venous risk assessment results.

[0019] Furthermore, the risk stratification assessment module is used to integrate anatomical and functional information features to construct a target three-dimensional anatomical model, and to perform risk stratification assessment on the central vein based on the target three-dimensional anatomical model, outputting the risk stratification assessment results, including: reconstructing an initial three-dimensional anatomical model based on anatomical information features and mapping functional information features to the initial three-dimensional anatomical model to obtain the target three-dimensional anatomical model; based on the target three-dimensional anatomical model, if the anatomical information features are less than or equal to a preset first threshold, the output anatomical information features are considered low-risk; if the anatomical information features are greater than the preset first threshold and less than or equal to a preset second threshold, the output anatomical information features are considered medium-risk; if the anatomical information features are greater than the preset second threshold, the output anatomical information features are considered high-risk; if the functional information features are less than or equal to .... If a third threshold is set, the output functional information feature is considered low-risk. If the functional information feature is greater than the preset third threshold and less than or equal to the preset fourth threshold, the output functional information feature is considered medium-risk. If the functional information feature is greater than the preset fourth threshold, the output functional information feature is considered high-risk. If either the anatomical information feature or the functional information feature outputs high-risk, the risk stratification assessment result is high-risk. If the anatomical information feature outputs medium-risk, and the functional information feature outputs medium-risk and / or low-risk, the risk stratification assessment result is medium-risk. If the functional information feature outputs medium-risk, and the anatomical information feature outputs medium-risk and / or low-risk, the risk stratification assessment result is medium-risk. If both the anatomical information feature and the functional information feature outputs are low-risk, the risk stratification assessment result is low-risk.

[0020] In the above scheme, anatomical and functional information features extracted from registered fixed and dynamic images are fused, fully considering the correlation between anatomical and functional information features. Then, a rule-based risk stratification method is used to transform multidimensional heterogeneous data into intuitive clinical risk levels, and a quantitative correlation between anatomical and functional information features is established, avoiding the shortcomings of single-parameter assessment in existing technologies. This facilitates the use of two non-contrast-enhanced image sequences to fully consider the correlation between anatomical and functional information, improving the reliability of central venous risk assessment results.

[0021] This invention also provides a central venous risk stratification assessment method based on data fusion, comprising: acquiring fixed and dynamic images within a preset anatomical range; performing interpolation transformation on the dynamic images to obtain transformed dynamic images and calculating rigid registration mutual information values ​​between the fixed and transformed dynamic images; performing rigid registration on the fixed and dynamic images based on the rigid registration mutual information values ​​to obtain initial registration fixed and initial registration dynamic images; obtaining an image pyramid based on a preset image sequence processing algorithm, the initial registration fixed and initial registration dynamic images, and performing deformable registration on the image pyramid to obtain a target registration fixed and target registration dynamic images; obtaining anatomical information features based on a preset anatomical feature extraction algorithm and the target registration fixed image; obtaining functional information features based on a preset functional feature extraction algorithm and the target registration dynamic image; fusing the anatomical information features and functional information features to construct a target three-dimensional anatomical model, and performing risk stratification assessment of the central venous system based on the target three-dimensional anatomical model, and outputting the risk stratification assessment results.

[0022] This invention proposes a central venous risk stratification assessment method based on data fusion. Within a preset anatomical range, corresponding fixed and dynamic images are acquired. Rigid registration is then performed on the fixed and dynamic images, followed by deformable registration on the initially registered fixed and dynamic images, providing a reliable data foundation for subsequent feature extraction. Anatomical and functional information features are then extracted from the target registered fixed and dynamic images, respectively. These anatomical and functional features are then fused to perform risk stratification assessment of the central venous system, yielding the risk stratification assessment result. Thus, by integrating the entire process of image acquisition, registration, feature extraction, and fusion assessment, anatomical and functional information features are analyzed in an integrated manner. Finally, a comprehensive assessment from morphology to function is performed. Ultimately, this method utilizes the fusion of two non-contrast-enhanced image sequences to fully consider the correlation between anatomical and functional information, improving the reliability of central venous risk assessment results. Attached Figure Description

[0023] Figure 1 A schematic diagram of the module structure of a central venous risk stratification assessment system based on data fusion, provided in a certain embodiment of the present invention; Figure 2 A schematic diagram of a fixed image acquired by the image acquisition module of a central venous risk stratification assessment system based on data fusion, provided in a certain embodiment of the present invention. Figure 1 ; Figure 3 A schematic diagram of a fixed image acquired by the image acquisition module of a central venous risk stratification assessment system based on data fusion, provided in a certain embodiment of the present invention. Figure 2 ; Figure 4A schematic diagram of a dynamic image acquired by the image acquisition module of a data fusion-based central venous risk stratification assessment system according to a certain embodiment of the present invention; Figure 5 A flowchart illustrating the risk stratification logic of a risk stratification assessment module in a data fusion-based central venous risk stratification assessment system, as provided in a certain embodiment of the present invention; Figure 6 This is a flowchart illustrating the steps of a central venous risk stratification assessment method based on data fusion, provided in a certain embodiment of the present invention. Detailed Implementation

[0024] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] Example 1 See Figure 1 , Figure 1 This is a schematic diagram of the module structure of a central venous risk stratification assessment system based on data fusion, provided in a certain embodiment of the present invention; as shown below. Figure 1 As shown, this embodiment of the invention provides a central venous risk stratification assessment system based on data fusion, comprising: an image acquisition module 101, used to acquire fixed and dynamic images within a preset anatomical range; a mutual information calculation module 102, used to perform interpolation transformation on the dynamic image to obtain a transformed dynamic image and calculate the rigid registration mutual information value between the fixed image and the transformed dynamic image; a rigid registration module 103, used to perform rigid registration on the fixed image and the dynamic image based on the rigid registration mutual information value to obtain an initial registered fixed image and an initial registered dynamic image; and a deformable registration module 104, used to perform rigid registration on the fixed image and the initial registered fixed image based on a preset image sequence processing algorithm. The system combines an initial registration dynamic image with an image pyramid to obtain an image pyramid. Deformable registration is then performed on the image pyramid to obtain a target registration fixed image and a target registration dynamic image. A first feature extraction module 105 is used to obtain anatomical information features based on a preset anatomical feature extraction algorithm and the target registration fixed image. A second feature extraction module 106 is used to obtain functional information features based on a preset functional feature extraction algorithm and the target registration dynamic image. A risk stratification assessment module 107 is used to fuse anatomical information features and functional information features to construct a target three-dimensional anatomical model. Based on the target three-dimensional anatomical model, a risk stratification assessment of the central vein is performed, and the risk stratification assessment results are output.

[0026] One specific implementation involves first acquiring fixed and dynamic images within a predetermined anatomical region. In this embodiment, the predetermined anatomical region is represented as the target central venous region, the fixed images are represented as 3D-Vane sequence images, and the dynamic images are represented as 4D-Flow MRI sequence images. Then, a magnetic resonance scanner is used to jointly scan the target central venous region with non-contrast-enhanced 3D-Vane sequence images and 4D-Flow MRI sequence images, respectively acquiring 3D-Vane sequence images and 4D-Flow MRI sequence images. The 3D-Vane sequence images are high-resolution anatomical images, and the 4D-Flow MRI sequence images are time-resolved three-dimensional blood flow velocity data.

[0027] Next, three-dimensional spatial registration is performed on the 3D-Vane sequence images and the 4D-Flow MRI sequence images. The three-dimensional spatial registration includes rigid registration and deformable registration. The main execution process of rigid registration is as follows: First, the transformation interpolation of the current 4D-Flow MRI sequence image is determined. Then, the 4D-Flow MRI sequence image is initially transformed to obtain the transformed 4D-Flow MRI sequence image, i.e., the transformed dynamic image. Then, the mutual information value between the transformed 4D-Flow MRI sequence image and the 3D-Vane sequence image is calculated. By optimizing the mutual information value, the 4D-Flow MRI sequence image and the 3D-Vane sequence image achieve optimal registration. At this time, the 4D-Flow MRI sequence image and the 3D-Vane sequence image have been initially and accurately aligned in the overall spatial position.

[0028] Next, for the rigidly registered 3D-Vane and 4D-Flow MRI sequences, corresponding to the initial registration fixed image and the initial registration dynamic image, deformable registration is performed. The main process is as follows: Using the rigidly registered 3D-Vane and 4D-Flow MRI sequences as input, continuous Gaussian smoothing and downsampling generate a set of spatially pre-aligned image sequences with progressively decreasing resolution, represented as an image pyramid. In the image pyramid, deformable registration is performed on each layer of the image according to a certain processing direction, generally from the top to the bottom. Deformable registration involves overlaying a control point grid on the image and continuously adjusting the position of the control points through an optimization algorithm to calculate the final deformation of each coordinate point in the image, resulting in the target registration fixed image and the target registration dynamic image.

[0029] Then, for the registered 3D-Vane sequence images, i.e., the target registered fixed images, the anatomical features of the central vein are extracted using a preset anatomical feature extraction algorithm. The anatomical features include at least one of the following: absolute stenosis, stenosis location, stenosis length, and vessel wall morphology features. For the registered 4D-Flow MRI sequence data, i.e., the target registered dynamic images, the hemodynamic features of the central vein are extracted using a preset functional feature extraction algorithm. The hemodynamic features include at least one of the following: wall shear stress (WSS), oscillatory shear index (OSI), energy loss (EL), blood flow velocity, and flow rate. Both the preset anatomical feature extraction algorithm and the preset functional feature extraction algorithm can be implemented using existing mature technologies.

[0030] Finally, the extracted anatomical features and hemodynamic features are fused and analyzed, and a comprehensive evaluation report containing anatomical and functional information is generated based on the reconstructed three-dimensional model.

[0031] This invention proposes a central venous risk stratification assessment system based on data fusion. An image acquisition module acquires corresponding fixed and dynamic images within a preset anatomical range. A mutual information calculation module performs rigid registration on the fixed and dynamic images, followed by deformable registration on the initially registered fixed and dynamic images, providing a reliable data foundation for subsequent feature extraction. Then, a first and second feature extraction module extracts anatomical and functional information features from the target registered fixed and dynamic images, respectively. Finally, a risk stratification assessment module fuses these anatomical and functional features to perform risk stratification assessment of the central venous system, yielding the assessment results. Thus, by integrating image acquisition, registration, feature extraction, and fusion assessment, the system performs a unified analysis of anatomical and functional information features, ultimately providing a comprehensive assessment from morphology to function. This system achieves improved reliability of central venous risk assessment results by fusing two non-contrast-enhanced image sequences to fully consider the correlation between anatomical and functional information.

[0032] In a preferred embodiment, the image acquisition module is used to acquire fixed and dynamic images within a preset anatomical range, including: performing a three-dimensional sequence scan within the preset anatomical range to obtain fixed images; performing a four-dimensional sequence scan within the preset anatomical range to obtain time-resolved three-dimensional imaging data; and performing velocity encoding on the time-resolved three-dimensional imaging data to obtain dynamic images.

[0033] One preferred implementation method is described in [reference]. Figure 2 and Figure 3 , Figure 2 A schematic diagram of a fixed image acquired by the image acquisition module of a central venous risk stratification assessment system based on data fusion, provided in a certain embodiment of the present invention. Figure 1 ; Figure 3 A schematic diagram of a fixed image acquired by the image acquisition module of a central venous risk stratification assessment system based on data fusion, provided in a certain embodiment of the present invention. Figure 2 ;like Figure 2 and Figure 3 As shown, a magnetic resonance scanner equipped with a 3D-Vane sequence was used to scan and acquire aqueous phase, lipid phase, positive phase, and negative phase images as fixed images. Figure 2 This is an image of the aqueous phase. Figure 3 The images are lipid phase images; fixed images are used to depict the anatomical structure of the central vein, the location of thrombi, and the contour of the vessel wall. Generally, a 3.0T Philips Ingenia Elition X scanner can be used for scanning, and specific scanning parameters and positioning methods can be achieved using existing mature technologies.

[0034] See Figure 4 , Figure 4 A schematic diagram of dynamic images acquired by the image acquisition module of a central venous risk stratification assessment system based on data fusion, provided in a certain embodiment of the present invention; as shown. Figure 4 As shown, a 4D-Flow MRI sequence was used for scanning, covering the same anatomical range as the 3D-Vane sequence scan, to acquire time-resolved three-dimensional phase-contrast vascular imaging data. In post-processing, the 4D-PCA algorithm was enabled, and then a velocity encoding algorithm was used to velocity encode the time-resolved three-dimensional phase-contrast vascular imaging data. During the velocity encoding process, the expected flow velocity of the central vein can generally be set to 80-150 cm / s to cover the entire cardiac cycle, thus obtaining dynamic images.

[0035] In the above scheme, three-dimensional sequence scanning, four-dimensional sequence scanning and velocity encoding techniques are used to obtain two types of non-contrast-enhanced image sequences: fixed images and dynamic images. By using these two types of image sequences in combination, accurate anatomical and functional information features can be obtained simultaneously without relying on contrast agents and radiation. This solves the problem of poor reliability of risk assessment results for central veins due to the lack of consideration of anatomical and functional information features in existing technologies, and helps to improve the reliability of central vein risk assessment results.

[0036] In a preferred embodiment, the mutual information calculation module is used to perform interpolation transformation on the dynamic image to obtain a transformed dynamic image and calculate the rigid registration mutual information value between the fixed image and the transformed dynamic image. This includes: calculating a transformation interpolation based on the dynamic image, a preset rotation matrix, and a preset translation vector; performing an interpolation transformation on the dynamic image based on the transformation interpolation to obtain the transformed dynamic image; identifying the overlapping region between the fixed image and the transformed dynamic image, and obtaining the grayscale value of the fixed image and the grayscale value of the transformed dynamic image at each pixel position in the overlapping region; constructing a joint histogram based on the grayscale values ​​of the fixed image and the transformed dynamic image, and calculating a joint probability distribution based on the joint histogram; generating a probability distribution of grayscale values ​​for the fixed image and the transformed dynamic image based on the joint probability distribution; calculating the information entropy of the fixed image based on the grayscale value probability distribution of the fixed image; calculating the information entropy of the dynamic image based on the grayscale value probability distribution of the transformed dynamic image; and calculating the rigid registration mutual information value between the fixed image and the transformed dynamic image based on the information entropy of the fixed image and the information entropy of the dynamic image.

[0037] One preferred implementation method involves a registration process that finds an optimal spatial transformation model to maximize the anatomical similarity between the transformed dynamic image and the fixed image. First, a rigid transformation is performed. This rigid transformation corrects for differences caused by overall patient positional movement and rotation between the two scans. It's worth noting that the rigid transformation assumes the object itself does not deform, only undergoing overall translation and rotation. Generally, a six-degree-of-freedom three-dimensional rigid transformation model can be used. The specific implementation process is as follows: The transformation of a spatial point from a dynamic image to a fixed image can be expressed as: T_rigid(x) = R * x + t, where T_rigid(x) represents the transformation interpolation, x is the coordinate of a point in the dynamic image, R is a 3x3 rotation matrix (i.e., a preset rotation matrix), and t is a 3x1 translation vector (i.e., a preset translation vector).

[0038] In this embodiment, we take a fixed image, i.e., a 3D-Vane sequence image, denoted as F, which is generally a lipid phase, and a dynamic image, i.e., a 4D-Flow MRI sequence image, denoted as M, as an example for explanation. For the transformed dynamic image, i.e., the dynamic image after interpolation transformation, denoted as M_T, for each pixel position in the overlapping area of ​​the fixed image F and the transformed dynamic image M_T, we take the gray value of the fixed image F as a and the gray value of the transformed dynamic image M_T as b, and construct a gray-level pair, denoted as (a, b). Then, we use this gray-level pair (a, b) to construct a two-dimensional joint histogram H. The size of the histogram is N×M, where N and M are the gray levels of the fixed image and the dynamic image, respectively. The value of H(a, b) is the number of times the gray-level pair (a, b) appears. Then, the joint histogram H is normalized to obtain the joint probability distribution, expressed as P_FM(a, b) = H(a, b) / N_total, where N_total is the total number of pixels in the overlapping region. From the joint probability distribution, the probability distribution of the grayscale value of each image is derived. The edge probability distribution P_F(a) of the fixed image F is obtained by summing each row of the joint probability distribution P_FM: P_F(a) = Σ_b P_FM(a, b); the edge probability distribution P_M(b) of the transformed dynamic image M_T is obtained by summing each column of the joint probability distribution P_FM: P_M(b) = Σ_a P_FM(a, b). Finally, based on the edge probability distribution P_F(a) of the fixed image F and the edge probability distribution P_M(b) of the transformed dynamic image M_T, the information entropy of the fixed image is calculated, expressed as H(F) = ... -Σ_aP_F(a)·log(P_F(a)); calculate the dynamic image information entropy, expressed as H(M)= -Σ_b P_M(b)·log(P_M(b)); and calculate the joint entropy of the two, expressed as H(F,M) = -Σ_aΣ_b P_FM(a,b)·log(P_FM(a,b)); finally, based on the fixed image information entropy, dynamic image information entropy, and joint entropy, calculate the rigid registration mutual information value of the fixed image and the transformed dynamic image, expressed as MI(F; M) = H(F) + H(M) - H(F,M), where H(F) + H(M) is the total uncertainty when the two images are independent; H(F,M) is the total uncertainty after the two images are correlated; the difference MI represents the uncertainty saved due to the correlation between the two images, that is, the amount of information they share. More specifically, the mutual information value MI(T) under the current spatial transformation T can be expressed as: MI(T)=Σ{a,b}P_{F,M_T}(a,b)·log_2(P_{F,M_T}(a,b) / (P_F(a)·P_{M_T}(b))); In the formula, T represents the spatial transformation parameters in the current iteration (such as rotation, translation, and displacement of deformation control points); F is the fixed image, i.e., the 3D-Vane sequence image; M_T is the moving image, i.e., the image obtained after resampling the 4D-Flow MRI sequence image through the current spatial transformation T; a is a gray value in the fixed image F; b is a gray value in the transformed moving image M_T; P_F(a) is the edge probability of the gray value a appearing in the fixed image F; P_{M_T}(b) is the edge probability of the gray value b appearing in the transformed moving image M_T; P_{F,M_T}(a,b) is the joint probability of the gray pair (a,b) appearing together in the overlapping area of ​​the image; Σ_{a,b} is the summation over all possible gray pairs (a,b).

[0039] In the above scheme, mutual information is used as a similarity measure for multimodal image registration. It does not rely on the linear relationship between gray levels between images and can effectively handle non-contrast enhancement sequence images with two different imaging mechanisms. The mutual information value is then used as the target for subsequent registration optimization, providing an objective and reliable data foundation for subsequent data processing. This ensures the initial accurate alignment of fixed and dynamic images in the overall spatial position, which helps to improve the reliability of central venous risk assessment results.

[0040] In a preferred embodiment, the rigid registration module is used to perform rigid registration on a fixed image and a dynamic image based on rigid registration mutual information values ​​to obtain an initial registered fixed image and an initial registered dynamic image. The module includes: constructing a mutual information value gradient vector based on transform interpolation and rigid registration mutual information values; updating the transform interpolation along the gradient direction; and recalculating the rigid registration mutual information values ​​of the fixed image and the transformed dynamic image each time the transform interpolation is updated to obtain optimized rigid registration mutual information values; if the optimized rigid registration mutual information values ​​meet preset registration requirements, the initial registered fixed image and the initial registered dynamic image are obtained.

[0041] One preferred implementation involves fine-tuning the parameters of the spatial transformation T using an optimization algorithm. For each parameter adjustment, a new mutual information value is recalculated. The goal of the optimization algorithm is to find the set of spatial transformation interpolations that maximizes the mutual information MI(F, M), denoted as T_optimal. The loop terminates when the value of MI(F, M) no longer increases significantly or reaches the maximum number of iterations, at which point the two images are considered to have achieved optimal registration. The specific execution process can be explained as follows: In this embodiment, the gradient descent method is used. The core of the gradient descent method is to advance along the direction of the fastest rise of the function value, with small steps, gradually approaching the maximum value. Specifically, at the beginning of the algorithm, an initial set of change interpolation T_0 is needed, which can usually be set as a unit transformation with zero rotation and zero translation. Then, the mutual information value gradient vector ∇MI(T) is calculated based on the current change interpolation T. Each component of the mutual information value gradient vector ∇MI(T) represents the rate of change of mutual information in the corresponding parameter dimension. Specifically, a small perturbation Δθ is applied to each parameter θᵢ in turn, and the change in mutual information ΔMI is calculated. Then, ΔMI / Δθ is used to approximate the partial derivative in the direction of the parameter, expressed as: ∇MI(T) = [∂MI / ∂θ1, ∂MI / ∂θ2,..., ∂MI / ∂θ n ]; In the formula, θᵢ is the transformation parameter; The independent process of applying perturbation to each parameter θᵢ is as follows: keeping all other parameters θⱼ (j≠i) unchanged, only the parameter θᵢ is increased by a small positive value Δθ, resulting in a new set of transformation parameters T. + =(θ1,...,θᵢ+Δθ,...,θ n A positive perturbation is applied, and then the mutual information value MI(T) under the current interpolation value T and the perturbation-adjusted interpolation value T are calculated respectively. + Mutual information value MI(T) + According to the definition of partial derivatives, the approximate calculation using the first-order forward difference formula is: ∂MI / ∂θᵢ ≈ (MI(T) + The calculated value is the approximate value of the i-th component ∂MI / ∂θᵢ in the gradient vector ∇MI(T). Repeating this process for each transformation parameter θᵢ in order from i to n, we obtain n values. Arranging these values ​​sequentially into a vector is represented as: [(MI( )-MI(T)) / Δθ,(MI( )-MI(T)) / Δθ,...,(MI( )-MI(T)) / Δθ]; This numerical vector is the discrete approximation of the gradient ∇MI(T) as defined in mathematics.

[0042] Next, in this embodiment, the specific process of updating the transformation interpolation along the gradient direction is as follows: the transformation interpolation is updated according to the calculated gradient direction, expressed as: T_new=T_old+λ*∇MI(T_old); where λ is the learning rate; whenever the transformation interpolation T is adjusted, the rigid registration mutual information value is recalculated to obtain the optimized rigid registration mutual information value for subsequent calculations.

[0043] In this embodiment, the preset registration requirement can be expressed as a convergence condition for determining whether the gradient descent algorithm meets the stopping condition. Generally, it can be set as follows: the magnitude of the gradient vector (i.e., ||∇MI(T)||) is less than a preset threshold, indicating that a near-extreme point has been reached. The improvement of the mutual information value in several consecutive iterations is less than a threshold, or the preset maximum number of iterations has been reached. If any convergence condition is met, the current optimal parameter T_optimal is output; otherwise, the next round of iteration is started with the new parameter T_new. Thus, the rigid registration of the fixed image and the dynamic image is completed, and the initially registered fixed image and the initially registered dynamic image are obtained.

[0044] In the above scheme, a mutual information value gradient vector is constructed and the transformation parameters are iteratively updated along the gradient direction to make the mutual information value meet the preset registration requirements, thereby realizing automatic and accurate registration between images. By transforming the complex spatial alignment problem into a solvable numerical optimization problem, the reliability of registration is improved. The rigid registration provides a good initial alignment state for subsequent deformable registration, avoiding the problem of poor reliability of subsequent registration due to excessive initial position deviation, and effectively improving the reliability of central venous risk assessment results.

[0045] In a preferred embodiment, a deformable registration module is used to obtain an image pyramid based on a preset image sequence processing algorithm, an initial registration fixed image, and an initial registration dynamic image, and to perform deformable registration on the image pyramid to obtain a target registration fixed image and a target registration dynamic image. The module includes: traversing the initial registration fixed image and the initial registration dynamic image, and sequentially performing smooth sampling according to the image resolution to generate the image pyramid; calculating the mutual information value of the initial registration fixed image and the initial registration dynamic image, and constructing an optimization objective function based on a preset regularization penalty term; traversing the image pyramid according to a preset direction, and performing deformable registration on each layer of the image pyramid until each layer of the image pyramid has completed deformable registration until the optimization objective function meets the preset optimization requirements, thereby obtaining the target registration fixed image and the target registration dynamic image. The process of deformable registration, which involves traversing the image pyramid in a preset direction and performing deformable registration on each layer of the image pyramid until each layer is deformably registered and the objective function meets the preset optimization requirements, to obtain a fixed image and a dynamic image of target registration, includes the following steps: covering the image corresponding to each layer of the image pyramid with a control point grid; calculating the normalized coordinates of any coordinate point in the local grid cell for any coordinate point in the image; determining the function value of the coordinate point in the normalized coordinates based on the control point index value of the control point grid, and using the function value as the influence weight of the control point; calculating the deformation value of the coordinate point based on the influence weight and coordinates of the control point, and performing deformable registration based on the deformation value.

[0046] One preferred implementation involves taking the initially registered fixed image and the initially registered dynamic image after rigid registration as input. A preset image sequence processing algorithm employs multi-resolution strategies such as L-BFGS-B and other optimization algorithms. Through continuous Gaussian smoothing and downsampling, an image pyramid is constructed. A typical image pyramid usually contains 3 to 4 levels, such as L0, L1, and L2, where L0 is the highest resolution (original image), and L2 or L3 is the lowest resolution. The image pyramid generation process is as follows: starting from the highest resolution image, L0, Gaussian filtering is first performed to smooth noise and details. The smoothed image is then downsampled in three dimensions. The downsampling process can involve taking one point every other pixel to obtain the next layer image with half the resolution, i.e., L1. This process is repeated to generate the entire pyramid.

[0047] Then, using the same mutual information calculation method, the mutual information values ​​of the initial registered fixed image and the initial registered dynamic image are obtained. An optimization objective function is constructed using a data fidelity term, a regularization penalty term, and a balancing weight. The data fidelity term is the negative value of the mutual information, the regularization penalty term is the bending energy term, and the balancing weight is used to adjust the relative importance between these two terms. In this embodiment, the constructed optimization objective function introduces bending energy regularization and sets appropriate weights to suppress excessive oscillations and local abrupt changes in the deformation field, prevent non-physical tissue folding, ensure the overall continuity and smoothness of the deformation field, make it more consistent with the mechanical behavior of biological tissues, and improve the stability of algorithm convergence, avoiding falling into non-optimal solutions caused by noise or local details.

[0048] Furthermore, optimization is performed stepwise from coarse to fine according to a preset direction, which is the direction from the top of the image pyramid to the bottom. Specifically, each layer of the image pyramid is iteratively optimized using an optimization algorithm. In the low-resolution layer, the displacement of sparse control points is adjusted to maximize mutual information. When convergence occurs in the low-resolution layer, the optimized deformation field is upsampled to match the control point grid density with the next higher-resolution layer. This upsampled deformation field serves as the initial estimate for L1-level registration optimization. In the high-resolution layer, the deformation field passed from the upper layer is used for initialization, and optimization is performed on this basis. This process is repeated until optimization is completed in the highest-resolution original image layer, i.e., the optimization objective function meets the preset optimization requirements, and the final high-precision deformation field is output, resulting in the target registration fixed image and the target registration dynamic image. In this embodiment, the deformable registration process is as follows: A B-spline free deformation model is used for interpretation. This model generates a smooth and continuous local deformation field by overlaying a control point grid on the image and moving the control points. By optimizing the displacement of the control points, the deformation field of the entire image can be generated, expressed as: T_nonrigid(x, y, z) = Σ Σ Σ φ_i(u) φ_j(v) φ_k(w) * C_{i, j, k}; In the formula, (x, y, z) are the coordinates of a point on the image, φ is the B-spline basis function, C_{i, j, k} are the coordinates of the control points, i, j, k are the control point indices, set to integers, used to uniquely identify which control point in the control point grid covering the 3D image is in the x, y, z directions. For example, φ_i(u) refers to the B-spline basis function with index i in the x direction; u, v, w are the local normalized coordinates, which are the normalized coordinates of any point (x, y, z) in the image within its local B-spline grid cell. During calculation, the grid cell in which the point is located is first determined based on its global coordinates (x, y, z), and then its position within that grid cell is mapped to the range [0, 1]. The three coordinate values ​​u, v, and w within the local grid cell represent the relative position of the point. φ_i(u) represents the function value of the B-spline basis function with index i in the x-direction at the local coordinate u of the point (x, y, z). This value represents the influence weight of the control point C_{i, j, k} on the current point in the x-direction. Similarly, φ_j(v) and φ_k(w) represent the influence weights in the y and z directions, respectively. The triple summation ΣΣΣ represents the calculation of the total deformation displacement of the point (x, y, z), which requires traversing all control points in its local neighborhood that can influence it, usually 4x4x4. For each control point index combination (i, j, k), its comprehensive weight φ_i(u) * φ_j(v) * φ_k(w) is calculated, multiplied by the displacement C_{i, j, k} of the control point, and finally all products are summed to obtain the final deformation value of the point, and deformable registration is performed.

[0049] In the above scheme, an image pyramid and a regularization penalty term are introduced for deformable registration. An image pyramid is constructed, and a layer-by-layer optimization strategy is used to quickly capture overall deformation at a large scale, followed by optimization of local details at a fine scale. This effectively avoids getting trapped in local optima. Complex image deformation is modeled as a displacement problem of sparse control points, and a regularization penalty term is used to prevent non-physical excessive deformation or folding, ensuring that the optimization objective function meets preset optimization requirements. This yields both a fixed-target registration image and a dynamic-target registration image. Therefore, based on the good initial alignment provided by rigid registration, deformable registration of the image provides a reliable data foundation for subsequent feature extraction, helping to improve the reliability of central venous risk assessment results.

[0050] In a preferred embodiment, the first feature extraction module is used to obtain anatomical information features based on a preset anatomical feature extraction algorithm and a target registration fixed image, including: extracting thrombus region features and vessel outer wall contour features from the target registration fixed image based on the preset anatomical feature extraction algorithm, and constructing corresponding thrombus 3D models and 3D vessel outer wall models according to a preset 3D reconstruction algorithm; generating several cross-sections based on the vessel orientation of the 3D vessel outer wall model; calculating the unobstructed lumen area based on the several cross-sections, and determining the stenosis location based on the unobstructed lumen area; determining the thrombus boundary based on the thrombus 3D model at the stenosis location, and determining the stenosis length at the stenosis location based on the thrombus boundary; obtaining vessel wall morphological features based on the 3D vessel outer wall model at the stenosis location; calculating the absolute stenosis degree based on the thrombus 3D model and the 3D vessel outer wall model at the stenosis location; and outputting the stenosis location, stenosis length, vessel wall morphological features, and absolute stenosis degree as anatomical information features.

[0051] One preferred implementation involves inputting a registered target image and extracting anatomical information such as the location, length, and morphological features of the vessel wall through an automated or semi-automated image processing workflow. Specifically, based on aqueous images, threshold segmentation and region growing algorithms are used to extract the thrombus region. A continuous 3D thrombus model is generated using 3D reconstruction technology. Based on lipid and forward / backward phase images, edge detection and region growing algorithms are used to extract the vessel wall contour, establishing a complete 3D vessel wall model. The 3D reconstruction process is achievable with existing technology and will not be elaborated here. Then, a series of cross-sections are generated along the vessel's direction. The area of ​​the patent lumen and the total area of ​​the vessel wall are calculated on each cross-section. The cross-section with the smallest patent lumen area is identified as the stenosis location, and its 3D spatial coordinates are recorded. Next, the proximal and distal boundaries of the thrombus are marked on the 3D thrombus model. The boundary points are projected onto the vessel centerline, and the 3D path length along the centerline between the two projection points is calculated to obtain the stenosis length. Finally, the extraction of vascular wall morphological features includes: vascular wall thickness, vascular contour regularity, vascular wall component characteristics, and stenosis morphology classification. Specifically, vascular wall thickness is measured by the vertical distance between the outer and inner walls of the vessel in the stenotic region; vascular contour regularity is assessed by calculating changes in vascular wall curvature to evaluate contour smoothness; vascular wall component characteristics are analyzed by examining the signal features of both positive and negative phase images to identify abnormal changes such as calcification and fibrosis; and stenosis morphology classification is based on thrombus distribution patterns, classifying stenosis into types such as centripetal and eccentric. Absolute stenosis is also calculated: absolute stenosis = (thrombus diameter at stenosis / outer diameter of the vessel at stenosis) × 100%.

[0052] Ultimately, the location, length, morphological features of the vessel wall, and absolute stenosis are output as anatomical information features.

[0053] In the above scheme, a preset anatomical feature extraction algorithm is used to extract thrombus region features and vessel outer wall contour features. Combined with a three-dimensional reconstruction algorithm, a three-dimensional model of the thrombus and a three-dimensional model of the vessel outer wall are constructed. Then, the stenosis morphology of the central vein is depicted. Subjective image interpretation is transformed into objective and repeatable quantitative data. The output anatomical information features can reflect the real impact of irregular thrombus lesions, providing a more reliable anatomical basis for subsequent risk assessment and helping to improve the reliability of central vein risk assessment results.

[0054] In a preferred embodiment, the second feature extraction module is used to obtain functional information features based on a preset functional feature extraction algorithm and a target registration dynamic image, including: obtaining the vessel wall surface and vessel cross-section based on the target registration dynamic image, the thrombus 3D model, and the 3D vessel outer wall model; calculating the gradient of blood flow velocity in the direction perpendicular to the vessel wall surface based on the vessel wall surface to obtain the wall shear stress; determining the oscillatory shear index based on the preset cardiac cycle and the wall shear stress; solving for energy loss based on the vessel cross-section; and outputting the wall shear stress, oscillatory shear index, and energy loss as functional information features.

[0055] One preferred implementation method involves using the registered target dynamic image as input to extract hemodynamic parameters of the target region, including but not limited to: Wall Shear Stress (WSS): the tangential force per unit area of ​​the vessel wall; Oscillatory Shear Index (OSI): a measure of the degree of change of the WSS direction over time; Energy Loss (EL): the loss of mechanical energy of blood flow due to stenosis, eddies, etc.; blood flow velocity, flow rate, and streamline diagram, etc. Specifically, the target registration dynamic image contains three-dimensional velocity field data with time series, and the three velocity components are represented as: V_x(x,y,z,t), V_y(x,y,z,t), and V_z(x,y,z,t), where V_x(x,y,z,t) represents the velocity in the left-right direction; V_y(x,y,z,t) represents the velocity in the front-back direction; and V_z(x,y,z,t) represents the velocity in the head-to-toe direction. The temporal information t covers multiple time points throughout the cardiac cycle.

[0056] In this embodiment, the vessel wall surface is used to calculate WSS and OSI. The vessel wall surface is defined by the vessel inner wall contour extracted from the registered 3D-Vane image using an image segmentation algorithm, i.e., defined by a three-dimensional vessel outer wall model. The vessel cross-section or volume of interest is used to calculate energy loss, flow rate, and average velocity. The vessel cross-section or volume of interest is manually drawn by the user on the registered image or automatically generated by the system at a specified location, typically at the most severe stenosis or the distal end of the stenosis. Then, to ensure the quality of the original velocity data, two existing and mature methods, Gaussian filtering and median filtering, are used for noise filtering; the specific process is not detailed here. The specific calculation process is as follows: Wall Shear Stress (WSS) is characterized as the drag force generated by blood flow on the vessel wall. On the surface of the vessel wall, the gradient of blood flow velocity in the direction perpendicular to the wall is calculated, expressed as WSS = μ* (∂V_tangential / ∂n); where μ is the viscosity of blood, usually taken as 0.004 Pa·s; ∂V_tangential / ∂n is the spatial rate of change of blood flow velocity parallel to the vessel wall in the direction perpendicular to the vessel wall; that is, the gradient is calculated at each vertex of the vessel wall grid to obtain the WSS vector that changes with time. The Oscillatory Shear Index (OSI) is used to quantify the degree of turbulence in the direction of the WSS (vascular septum) during the cardiac cycle. A higher OSI value indicates a more drastic change in blood flow direction, which is closely related to endothelial cell damage. It is obtained by comparing the difference between the direction of the WSS vector and its time-averaged vector, and is expressed as OSI = 0.5 * (1 - |∫WSS dt| / ∫|WSS| dt); where ∫ represents the integral over one cardiac cycle, OSI is a dimensionless number between 0 and 0.5, where 0 indicates that the WSS direction is constant, 0.5 indicates that the forward and reverse WSS completely cancel each other out, and the time-averaged WSS is the average value over one cardiac cycle. Energy loss (EL) is used to assess the hemodynamic significance of stenosis. Based on Bernoulli's principle, it is indirectly derived by calculating the pressure difference across the region of interest, or directly solved by calculating the viscous energy dissipation function within the region. The expression is: EL = μ*∫∫∫_Ω2 * [ (∂V_x / ∂x)²+ (∂V_y / ∂y)²+ (∂V_z / ∂z)²] + (∂V_x / ∂y +∂V_y / ∂x)²+ (∂V_x / ∂z +∂V_z / ∂x)²+ (∂V_y / ∂z) +∂V_z / ∂y)²dΩ; where Ω is the cross-section of the blood vessel or the volume of interest. The integration is performed within the cross-section of the blood vessel or the volume of interest, that is, the function is calculated and integrated based on all spatial gradients of the velocity field, and finally the energy loss value in mW (milliwatts) is obtained.Blood flow velocity and flow rate are calculated at a specified vascular cross-section. Specifically, flow rate is expressed as Q(t) = ∫∫_A V_perpendicular(x,y,t) dA, which is the average flow rate obtained by integrating the velocity component perpendicular to the cross-section at each point and interpreting it as the average value over one period. Flow velocity is typically expressed as the peak velocity or spatial average velocity at the reported cross-section.

[0057] Finally, the wall shear stress, oscillatory shear index, and energy loss are output as functional information features.

[0058] In the above scheme, key hemodynamic features are extracted from the registered blood flow data to quantify the mechanical effects of blood flow on the vessel wall, the degree of blood flow turbulence, and the mechanical energy loss caused by stenosis. Wall shear stress, oscillatory shear index, and energy loss are used as core indicators to assess the functional significance of stenosis, providing a reliable data foundation for subsequent in-depth analysis from morphological to functional assessment and effectively improving the reliability of central venous risk assessment results.

[0059] A preferred embodiment includes a risk stratification assessment module, used to fuse anatomical information features and functional information features to construct a target three-dimensional anatomical model, and to perform risk stratification assessment on the central vein based on the target three-dimensional anatomical model, outputting the risk stratification assessment results, including: reconstructing an initial three-dimensional anatomical model based on anatomical information features, and mapping functional information features to the initial three-dimensional anatomical model to obtain the target three-dimensional anatomical model; based on the target three-dimensional anatomical model, if the anatomical information features are less than or equal to a preset first threshold, then the anatomical information features are output as low risk; if the anatomical information features are greater than the preset first threshold and less than or equal to a preset second threshold, then the anatomical information features are output as medium risk; if the anatomical information features are greater than the preset second threshold, then the anatomical information features are output as high risk; if the functional information features are less than or equal to... If a third threshold is preset, the output functional information feature is considered low-risk; if the functional information feature is greater than the third threshold and less than or equal to the fourth threshold, the output functional information feature is considered medium-risk; if the functional information feature is greater than the fourth threshold, the output functional information feature is considered high-risk; if either the anatomical information feature or the functional information feature outputs high-risk, the risk stratification assessment result is high-risk; if the anatomical information feature outputs medium-risk, and the functional information feature outputs medium-risk and / or low-risk, the risk stratification assessment result is medium-risk; if the functional information feature outputs medium-risk, and the anatomical information feature outputs medium-risk and / or low-risk, the risk stratification assessment result is medium-risk; if both the anatomical information feature and the functional information feature outputs are low-risk, the risk stratification assessment result is low-risk.

[0060] One preferred implementation method is described in [reference]. Figure 5 , Figure 5A risk stratification logic flowchart of a risk stratification assessment module in a data fusion-based central venous risk stratification assessment system, as provided in a certain embodiment of the present invention; Figure 5 As shown, the extracted anatomical and functional information features are fused and visualized on a unified three-dimensional model. For example, the initial three-dimensional anatomical model can be reconstructed using anatomical information features. Generally, the distribution of WSS or EL can be overlaid on the reconstructed vascular model in a color mapping manner to obtain the target three-dimensional anatomical model. A comprehensive evaluation system is constructed based on the target 3D anatomical model. If the anatomical information feature is less than or equal to a preset first threshold, the output anatomical information feature is low risk; if the anatomical information feature is greater than the preset first threshold and less than or equal to a preset second threshold, the output anatomical information feature is medium risk; if the anatomical information feature is greater than the preset second threshold, the output anatomical information feature is high risk; if the functional information feature is less than or equal to a preset third threshold, the output functional information feature is low risk; if the functional information feature is greater than the preset third threshold and less than or equal to a preset fourth threshold, the output functional information feature is medium risk; if the functional information feature is greater than the preset fourth threshold, the output functional information feature is high risk; if either the anatomical or functional information feature outputs high risk, the risk stratification assessment result is high risk; if the anatomical information feature outputs medium risk and the functional information feature outputs medium risk and / or low risk, the risk stratification assessment result is medium risk; if the functional information feature outputs medium risk and the anatomical information feature outputs medium risk and / or low risk, the risk stratification assessment result is medium risk; if both the anatomical and functional information feature outputs are low risk, the risk stratification assessment result is low risk.

[0061] A specific example is given. Taking the absolute stenosis degree S as an example, its risk stratification framework is as follows: Corresponding to mild stenosis, it is usually set below a relatively low threshold. The first preset threshold can be set as S ≤ 50%, and at this time, the output anatomical information feature is low risk; Corresponding to moderate stenosis, it is set between the above low-risk threshold and a higher threshold. The second preset threshold can be set as S ≤ 70%, that is, 50% < S ≤ 70%, and at this time, the output anatomical information feature is medium risk; Corresponding to severe stenosis, it is set above the above higher threshold, that is, S > 70%, and at this time, the output anatomical information feature is high risk. Similarly, taking the energy loss EL as an example, its risk stratification framework is as follows: Corresponding to mild blood flow disturbance, it is usually set below a low threshold. The third preset threshold can be set as EL ≤ 15 mW, and at this time, the output functional information feature is low risk; Corresponding to moderate blood flow disorder, it is set between the above low-risk threshold and a higher threshold. The fourth preset threshold can be set as 30 mW, that is, 15 mW < EL ≤ 30 mW, and at this time, the output functional information feature is medium risk; Corresponding to severe hemodynamic abnormality, it is set above the above higher threshold, that is, EL > 30 mW, and at this time, the output functional information feature is high risk. Then, according to the synergistic logic between the absolute stenosis degree S and the energy loss EL, the risk stratification assessment result is determined. When the absolute stenosis degree S is at "medium risk" and the energy loss EL is at "low risk", it indicates that although the current hemodynamic impact is relatively light, a clear moderate stenosis has occurred in the anatomical structure, and there is a risk of progression. At this time, the output risk stratification assessment result is medium risk. When the absolute stenosis degree S is at "low risk" and the energy loss EL is at "high risk", although the anatomical stenosis degree is not severe, it has caused significant hemodynamic disorder. At this time, the output risk stratification assessment result is high risk; When both the absolute stenosis degree S and the energy loss EL are at "high risk", it indicates that both the anatomical structure and the functional index confirm the existence of serious lesions. At this time, the output risk stratification assessment result is high risk; When both the absolute stenosis degree S and the energy loss EL are at "low risk", it indicates that the anatomical stenosis degree is not severe and the current hemodynamic impact is relatively light, and the output risk stratification assessment result is low risk.

[0062] In the above solution, the anatomical information features and functional information features extracted from the registered fixed image and dynamic image are fused, fully considering the association between the anatomical information features and functional information features. Then, a regularized risk stratification method is used to convert multi-dimensional heterogeneous data into intuitive clinical risk levels, and a quantitative association between the anatomical information features and functional information features is established, avoiding the defects of single parameter evaluation in the prior art. It is beneficial to realize the fusion of two non-contrast-enhanced sequence images to fully consider the association between anatomical information and functional information, and improve the reliability of the central venous risk assessment result.

[0063] Example 2 See Figure 6 , Figure 6 This is a flowchart illustrating the steps of a central venous risk stratification assessment method based on data fusion according to a certain embodiment of the present invention. As shown in Figure 6, this embodiment of the present invention proposes a central venous risk stratification assessment method based on data fusion, including steps 201 to 207, each step of which is as follows: Step 201: Acquire fixed and dynamic images within a preset anatomical range; Step 202: Perform interpolation transformation on the dynamic image to obtain the transformed dynamic image and calculate the rigid registration mutual information value between the fixed image and the transformed dynamic image; Step 203: Perform rigid registration on the fixed image and the dynamic image based on the rigid registration mutual information value to obtain the initially registered fixed image and the initially registered dynamic image; Step 204: Based on the preset image sequence processing algorithm, the initial registration fixed image and the initial registration dynamic image, an image pyramid is obtained, and deformable registration is performed on the image pyramid to obtain the target registration fixed image and the target registration dynamic image. Step 205: Obtain anatomical information features based on a preset anatomical feature extraction algorithm and a target registration fixed image; Step 206: Based on the preset functional feature extraction algorithm and the target registration dynamic image, obtain functional information features; Step 207: Integrate anatomical and functional information features to construct a target three-dimensional anatomical model, and perform risk stratification assessment of the central vein based on the target three-dimensional anatomical model, outputting the risk stratification assessment results.

[0064] This invention proposes a central venous risk stratification assessment method based on data fusion. Within a preset anatomical range, corresponding fixed and dynamic images are acquired. Rigid registration is then performed on the fixed and dynamic images, followed by deformable registration on the initially registered fixed and dynamic images, providing a reliable data foundation for subsequent feature extraction. Anatomical and functional information features are then extracted from the target registered fixed and dynamic images, respectively. These anatomical and functional features are then fused to perform risk stratification assessment of the central venous system, yielding the risk stratification assessment result. Thus, by integrating the entire process of image acquisition, registration, feature extraction, and fusion assessment, anatomical and functional information features are analyzed in an integrated manner. Finally, a comprehensive assessment from morphology to function is performed. Ultimately, this method utilizes the fusion of two non-contrast-enhanced image sequences to fully consider the correlation between anatomical and functional information, improving the reliability of central venous risk assessment results.

[0065] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

[0066] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. Furthermore, the described specific features, structures, materials, or characteristics may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of those different embodiments or examples.

[0067] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "a plurality of" means two or more, unless otherwise explicitly specified.

Claims

1. A central venous risk stratification assessment system based on data fusion, characterized in that, include: The image acquisition module is used to acquire fixed and dynamic images within a preset anatomical range; The mutual information calculation module is used to perform interpolation transformation on the dynamic image to obtain the transformed dynamic image and calculate the rigid registration mutual information value between the fixed image and the transformed dynamic image; A rigid registration module is used to perform rigid registration on the fixed image and the dynamic image based on the rigid registration mutual information value to obtain an initially registered fixed image and an initially registered dynamic image; The deformable registration module is used to obtain an image pyramid based on a preset image sequence processing algorithm, the initial registration fixed image and the initial registration dynamic image, and to perform deformable registration on the image pyramid to obtain a target registration fixed image and a target registration dynamic image. The first feature extraction module is used to obtain anatomical information features based on a preset anatomical feature extraction algorithm and the target registration fixed image; The second feature extraction module is used to obtain functional information features based on a preset functional feature extraction algorithm and the target registration dynamic image; The risk stratification assessment module is used to integrate the anatomical information features and the functional information features to construct a target three-dimensional anatomical model, and to perform risk stratification assessment on the central vein based on the target three-dimensional anatomical model, and output the risk stratification assessment results.

2. The central venous risk stratification assessment system based on data fusion as described in claim 1, characterized in that, The image acquisition module is used to acquire fixed and dynamic images within a preset anatomical range, including: Three-dimensional sequential scanning is performed within a pre-defined anatomical area to obtain fixed images; Four-dimensional sequence scanning was performed within a pre-defined anatomical area to obtain time-resolved three-dimensional imaging data; The time-resolved three-dimensional imaging data is velocity-encoded to obtain a dynamic image.

3. The central venous risk stratification assessment system based on data fusion as described in claim 1, characterized in that, The mutual information calculation module is used to perform interpolation transformation on the dynamic image to obtain a transformed dynamic image and to calculate the rigid registration mutual information value between the fixed image and the transformed dynamic image, including: Based on the dynamic image, the preset rotation matrix, and the preset translation vector, the transformation interpolation is calculated; The dynamic image is interpolated based on the aforementioned transformation interpolation to obtain a transformed dynamic image; Identify the overlapping region between the fixed image and the transformed dynamic image, and obtain the grayscale value of the fixed image and the grayscale value of the transformed dynamic image at each pixel position in the overlapping region; Based on the fixed image grayscale value and the transformed dynamic image grayscale value, a joint histogram is constructed and a joint probability distribution is calculated based on the joint histogram; Based on the joint probability distribution, a fixed image grayscale value probability distribution and a transformed dynamic image grayscale value probability distribution are generated; Calculate the information entropy of the fixed image based on the probability distribution of the grayscale values ​​of the fixed image; Calculate the dynamic image information entropy based on the probability distribution of grayscale values ​​in the transformed dynamic image; Based on the fixed image information entropy and the dynamic image information entropy, the rigid registration mutual information value between the fixed image and the transformed dynamic image is calculated.

4. The central venous risk stratification assessment system based on data fusion as described in claim 3, characterized in that, The rigid registration module is used to perform rigid registration on the fixed image and the dynamic image based on the rigid registration mutual information value, to obtain an initially registered fixed image and an initially registered dynamic image, including: Based on the transformation interpolation and the rigid registration mutual information value, a mutual information value gradient vector is constructed; The transformation interpolation is updated along the gradient direction. Whenever the transformation interpolation is updated, the rigid registration mutual information value of the fixed image and the transformed dynamic image is recalculated to obtain the optimized rigid registration mutual information value. If the optimized rigid registration mutual information value meets the preset registration requirements, then the initial registration fixed image and the initial registration dynamic image are obtained.

5. The central venous risk stratification assessment system based on data fusion as described in claim 1, characterized in that, The deformable registration module is used to obtain an image pyramid based on a preset image sequence processing algorithm, the initial registration fixed image, and the initial registration dynamic image, and to perform deformable registration on the image pyramid to obtain a target registration fixed image and a target registration dynamic image, including: The image pyramid is generated by iterating through the initial registration fixed image and the initial registration dynamic image and performing smooth sampling according to the image resolution. Calculate the mutual information value between the initial registered fixed image and the initial registered dynamic image, and construct an optimization objective function based on a preset regularization penalty term; Based on a preset direction, the image pyramid is traversed, and deformable registration is performed on each layer of the image pyramid until each layer of the image pyramid has completed deformable registration until the optimization objective function meets the preset optimization requirements, thereby obtaining a fixed target registration image and a dynamic target registration image.

6. The central venous risk stratification assessment system based on data fusion as described in claim 5, characterized in that, In the steps of traversing the image pyramid based on a preset direction, performing deformable registration on each layer of the image pyramid until each layer of the image pyramid has completed deformable registration until the optimization objective function meets the preset optimization requirements, and obtaining a fixed target registration image and a dynamic target registration image, the execution process of deformable registration includes: A grid of control points is overlaid on the image corresponding to each layer of the image pyramid. For any coordinate point on the image, calculate the normalized coordinates of the coordinate point within the local grid cell; Based on the control point index value of the control point grid, the function value of the coordinate point under the normalized coordinates is determined, and the function value is used as the influence weight of the control point. Based on the influence weight of the control point and the coordinates of the control point, the deformation value of the coordinate point is calculated, and deformable registration is performed based on the deformation value.

7. The central venous risk stratification assessment system based on data fusion as described in claim 1, characterized in that, The first feature extraction module is used to obtain anatomical information features based on a preset anatomical feature extraction algorithm and the target registration fixed image, including: Based on a preset anatomical feature extraction algorithm, the thrombus region features and vessel outer wall contour features of the target registration fixed image are extracted respectively, and the corresponding thrombus 3D model and 3D vessel outer wall model are constructed according to a preset 3D reconstruction algorithm. Several cross-sections are generated based on the blood vessel orientation of the aforementioned three-dimensional blood vessel outer wall model. The unobstructed lumen area is calculated based on several of the aforementioned cross sections, and the narrowing location is determined based on the unobstructed lumen area. The thrombus boundary is determined based on the three-dimensional model of the thrombus at the narrow location, and the stenosis length at the narrow location is determined based on the thrombus boundary. The morphological features of the vessel wall are obtained based on the three-dimensional vessel wall model at the narrow location; The absolute stenosis is calculated based on the three-dimensional model of the thrombus at the stenosis location and the three-dimensional model of the outer wall of the blood vessel at the stenosis location. The location, length, morphological features of the vessel wall, and absolute stenosis are output as anatomical information features.

8. The central venous risk stratification assessment system based on data fusion as described in claim 7, characterized in that, The second feature extraction module is used to obtain functional information features based on a preset functional feature extraction algorithm and the target registration dynamic image, including: Based on the target registration dynamic image, the thrombus 3D model and the 3D blood vessel outer wall model, the blood vessel wall surface and blood vessel cross-section are obtained; The gradient of blood flow velocity in the direction perpendicular to the wall surface is calculated based on the surface of the blood vessel wall to obtain the wall shear stress; The oscillatory shear index is determined based on the preset cardiac cycle and the wall shear stress. The energy loss is calculated based on the cross-section of the blood vessel. The wall shear stress, the oscillatory shear index, and the energy loss are output as functional information features.

9. A central venous risk stratification assessment system based on data fusion as described in claim 8, characterized in that, The risk stratification assessment module is used to fuse the anatomical information features and the functional information features to construct a target three-dimensional anatomical model, and to perform risk stratification assessment on the central vein based on the target three-dimensional anatomical model, outputting the risk stratification assessment results, including: An initial three-dimensional anatomical model is reconstructed based on the anatomical information features, and the functional information features are mapped onto the initial three-dimensional anatomical model to obtain the target three-dimensional anatomical model; Based on the target three-dimensional anatomical model, if the anatomical information features are less than or equal to a preset first threshold, the anatomical information features are output as low risk. If the anatomical information feature is greater than a preset first threshold and less than or equal to a preset second threshold, then the anatomical information feature is output as medium risk. If the anatomical information feature is greater than a preset second threshold, then the anatomical information feature is output as high risk; If the functional information feature is less than or equal to a preset third threshold, then the output functional information feature is low risk; If the functional information feature is greater than a preset third threshold and less than or equal to a preset fourth threshold, then the functional information feature is output as medium risk. If the functional information feature is greater than the preset fourth threshold, then the functional information feature is output as high risk; If either the anatomical information feature or the functional information feature outputs a high risk, then the risk stratification assessment result is high risk. If the anatomical information feature output is of medium risk, and the functional information feature output is of medium risk and / or low risk, then the risk stratification assessment result is medium risk. If the output of the functional information feature is medium risk, and the output of the anatomical information feature is medium risk and / or low risk, then the risk stratification assessment result is medium risk; If both the anatomical information feature and the functional information feature outputs are low risk, then the risk stratification assessment result is low risk.

10. A central venous risk stratification assessment method based on data fusion, characterized in that, The system applied to the central venous risk stratification assessment system based on data fusion as described in claims 1-9 includes: Acquire fixed and dynamic images within a preset anatomical area; The dynamic image is interpolated to obtain a transformed dynamic image, and the rigid registration mutual information value between the fixed image and the transformed dynamic image is calculated. Based on the rigid registration mutual information value, rigid registration is performed on the fixed image and the dynamic image to obtain an initially registered fixed image and an initially registered dynamic image; Based on a preset image sequence processing algorithm, the initial registration fixed image and the initial registration dynamic image, an image pyramid is obtained, and deformable registration is performed on the image pyramid to obtain a target registration fixed image and a target registration dynamic image. Based on the preset anatomical feature extraction algorithm and the target registration fixed image, anatomical information features are obtained; Based on the preset functional feature extraction algorithm and the target registration dynamic image, functional information features are obtained; By integrating the anatomical and functional information features, a target three-dimensional anatomical model is constructed, and a risk stratification assessment of the central vein is performed based on the target three-dimensional anatomical model, and the risk stratification assessment results are output.

Images