A 3D hemodynamic atlas construction method for ultrasound examination

Abstract

The present application relates to the field of intelligent medicine, and particularly relates to a 3D hemodynamic atlas construction method for ultrasonic examination, the present application designs an optimized Unet model, the linear complexity of a state space model is used to model the global topological structure of a blood vessel, the continuity of the blood vessel and the spatial relationship between branches are captured, local features are extracted through a non-overlapping window, the spatial attention is combined to enhance the blood vessel edge response, dynamic receptive field adjustment is used to enhance the recognition and segmentation ability of different blood vessels, multi-directional scanning is used to model the spatial relationship of blood vessel branches, the problems of information loss and missegmentation caused by one-way scanning are avoided, and accurate three-dimensional blood vessel modeling is realized; the present application quantifies macroscopic and microscopic parameters and integrates them into the atlas, enriches the arterial information in the atlas, maps individual blood vessel geometry to the ATLAS coordinate system through B-spline registration, solves the parameter deviation caused by the angle variation of the aortic branch, increases the accuracy of registration, and improves the atlas quality.

Description

Technical Field

[0001] This invention belongs to the field of intelligent medicine, specifically a method for constructing 3D hemodynamic maps for ultrasound examination. Background Technology

[0002] The construction of 3D hemodynamic atlases primarily involves the fields of medical image processing and hemodynamic simulation. This technology can help the medical field better understand and assess blood flow in the human body, providing important information for disease diagnosis and treatment, and offering patients a more intuitive display of vascular blood flow. However, existing algorithms and technologies still have limitations in constructing 3D hemodynamic atlases, making it difficult to accurately construct 3D blood vessels from medical images, resulting in low-quality atlases based on dynamics. Summary of the Invention

[0003] To address the above issues and overcome the shortcomings of existing technologies, this invention provides a method for constructing 3D hemodynamic maps for ultrasound examinations. Addressing the limitations of existing technologies in accurately constructing 3D blood vessels from medical images and the low quality of hemodynamic-based map construction, this invention designs an optimized Unet model. It models the global topology of blood vessels using a state-space model (SSM) with linear complexity, capturing the continuity of blood vessels and the spatial relationships between branches, thus reducing computational complexity. Local features are extracted through non-overlapping windows, and spatial attention is used to enhance the response of blood vessel edges. Furthermore, dynamic receptive field adjustment adaptively adjusts the window size according to the blood vessel diameter, improving... This invention enhances compatibility with both large and small blood vessels, improves the ability to identify and segment different vessels, and explicitly models the spatial relationships of vascular branches through multi-directional scanning, avoiding information loss during unidirectional scanning. It also solves the problem of missegmentation at vascular intersections and bifurcations that is common in traditional techniques, ultimately achieving accurate 3D vascular segmentation and modeling. Furthermore, this invention simultaneously quantifies macroscopic (flow velocity, kinetic energy) and microscopic (wall shear stress, eddy current) parameters and integrates them into the 3D atlas, increasing the atlas information and making it more sensitive to the prediction of aneurysm rupture risk. By using B-spline registration to map the geometry of individual blood vessels to the ATLAS coordinate system, it solves the parameter deviation caused by variations in aortic branch angles, increases the accuracy of registration, and improves the quality of the atlas.

[0004] This invention provides a method for constructing 3D hemodynamic maps for ultrasound examination, specifically including the following steps:

[0005] Step S1: Data acquisition, acquiring three-dimensional ultrasound images of the areas to be mapped, including the blood flow velocity vector field;

[0006] Step S2: Data preprocessing, preprocess the 3D ultrasound image and slice it into 2D images, preprocess the 2D images and convert them into grayscale images, copy the grayscale images to the three channels and convert them into 3-channel images;

[0007] Furthermore, step S2 specifically includes the following steps:

[0008] Step S21: Remove labels from irrelevant structures and small objects in the 3D ultrasound image;

[0009] Step S22: Normalize the voxel intensity values ​​of the three-dimensional ultrasound image using window settings [400, 40];

[0010] Step S23: Convert the three-dimensional ultrasound image into a 2D slice, i.e., a two-dimensional image;

[0011] Step S24: Resample all slices to a uniform spatial resolution of 1mm × 1mm;

[0012] Step S25: Retain non-zero slices and normalize the intensity of the two-dimensional image;

[0013] Step S26: Adjust the processed 2D image to 1024×1024 pixels;

[0014] Step S27: Convert the two-dimensional image to a grayscale image, and convert it into a three-channel image by copying the grayscale image to three channels (1024×1024×3);

[0015] Step S3: Segmentation based on the optimized Unet model. The three-channel image is segmented using the optimized Unet model to obtain a blood vessel segmentation map.

[0016] Step S31: Image segmentation and linear projection. Input a three-channel image and segment it into 4×4 non-overlapping blocks. Obtain feature maps by linearly projecting the non-overlapping blocks.

[0017] Step S32: Encoder encoding, using a visual state space module based on receptive field optimization to encode the feature map, obtaining the initial encoding result;

[0018] Step S321: Use the Visual State Space Model (VSSM) based on residual optimization to globally model the feature map and obtain global features:

[0019] ;

[0020] In the formula, Represents the visual state space model. Represents normalization operation, For learnable scaling factor, Representative feature map, adjust residual connection strength, For global features;

[0021] Step S322: Unfold the feature map into a one-dimensional sequence in four directions, perform Mamba processing on each sequence, and then merge them into a 2D feature map.

[0022] ;

[0023] In the formula, For feature expansion operations, Represents the direction of development. This represents a merge operation. Represents 2D features;

[0024] Step S323: Extract the receptive field through a non-overlapping window and combine it with spatial attention to perform local feature enhancement on the 2D features, resulting in enhanced local features:

[0025] ;

[0026] In the formula, This represents a grouped convolution with a kernel size of o*o. This represents average pooling. Represents standardized processing. and Represents the activation function. This is a local enhancement feature;

[0027] Step S324: Dynamic feature fusion, dynamically fusing global features and local enhanced features to obtain the initial coding result;

[0028] Step S33: Downsampling, downsampling the primary encoding result to obtain the primary downsampling result;

[0029] Step S34: Multi-layer downsampling coding, repeat steps S32 to S33 three times to obtain downsampling codes;

[0030] Step S35: Multi-layer upsampling encoding. The above-mentioned visual state space module based on receptive field optimization and Patch Expanding are used to perform 4-layer upsampling on the downsampling encoding to obtain the decoding features.

[0031] Step S36: Restore the decoded features to the original three-channel image size through a linear layer and output the blood vessel segmentation map;

[0032] Step S4: 3D reconstruction, stitching together the blood vessel segmentation maps of all two-dimensional images to obtain a 3D blood vessel segmentation map;

[0033] Step S5: Reconstruct the blood flow velocity field, reconstruct the blood flow velocity vector field to obtain hemodynamic parameters and blood flow path:

[0034] Step S51: Doppler angle correction, combined with the geometric parameters of the three-dimensional ultrasound probe, to reconstruct the blood flow velocity vector field:

[0035] ;

[0036] In the formula, The nodal velocities representing the blood flow velocity vector field. The direction vector of blood flow. The direction vector of the sound beam. The included angle is estimated using 3D probe attitude data. , This represents the speed after recovery;

[0037] Step S52: Use radial basis function interpolation to fill in the missing values ​​of the blood flow velocity vector field;

[0038] Step S53: Calculate hemodynamic parameters, including the wall shear stress (WSS) and vorticity on the vessel surface.

[0039] , ;

[0040] In the formula, Represents blood viscosity. Represents the tangential velocity in the blood flow velocity vector field. Represents wall shear stress. The spatial gradient representing the blood flow velocity vector field;

[0041] Step S54: Blood flow path construction, generate streamlines by integrating along the velocity field to describe the blood flow path;

[0042] Step S6: Map hemodynamic parameters and blood flow paths onto the surface of the 3D vessel segmentation map to generate a 3D hemodynamic atlas, wherein the hemodynamic parameters are 3D mapped based on B-splines:

[0043] ;

[0044] In the formula, , and Represents the B-spline basis functions. Represents control points. The coordinates of the control points in the 3D vessel segmentation map represent the corresponding three-dimensional coordinates. These control points are the locations where 3D mapping is required. , and These represent the wall shear stress (WSS) on the blood vessel surface, the recovered velocity, and the vorticity, respectively. This represents the result of mapping hemodynamic parameters such as wall shear stress (WSS), velocity, and vorticity onto a 3D vessel segmentation map. This represents the three-dimensional coordinates of the corresponding point in the hemodynamic map.

[0045] The beneficial results achieved by the present invention using the above solution are as follows:

[0046] (1) This invention designs an optimized Unet model, which models the global topology of blood vessels with the linear complexity of the state space model (SSM), captures the continuity of blood vessels and the spatial relationship between branches, reduces computational complexity, extracts local features through non-overlapping windows, enhances the response of blood vessel edges by combining spatial attention, and improves the compatibility of thick and thin blood vessels by dynamically adjusting the receptive field according to the diameter of blood vessels, enhances the ability to identify and segment different blood vessels, explicitly models the spatial relationship of blood vessel branches by multi-directional scanning, avoids information loss by unidirectional scanning, solves the problem of missegmentation at blood vessel intersections and bifurcations in traditional technology, and finally achieves accurate 3D blood vessel segmentation modeling;

[0047] (2) This invention increases the information in the 3D map by simultaneously quantifying macroscopic (flow velocity, kinetic energy) and microscopic parameters (wall shear stress, eddy current) and integrating them into the 3D map, making it more sensitive to the prediction of aneurysm rupture risk. By using B-spline registration to map the geometry of individual blood vessels to the ATLAS coordinate system, the parameter deviation caused by the variation of aortic branch angle is solved, the accuracy of registration is increased, and the quality of the map is improved. Attached Figure Description

[0048] Figure 1 A flowchart of a method for constructing 3D hemodynamic maps for ultrasound examination provided by the present invention;

[0049] Figure 2 This invention provides a data processing flowchart for a visual state space module based on receptive field optimization.

[0050] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof. Detailed Implementation

[0051] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0052] Example 1, see Figure 1 A method for constructing 3D hemodynamic maps for ultrasound examination, specifically including the following steps:

[0053] Step S1: Acquire three-dimensional ultrasound images of the areas where the atlas needs to be constructed, including the blood flow velocity vector field;

[0054] Step S2: Preprocess the three-dimensional ultrasound image and slice it into a two-dimensional image. Preprocess the two-dimensional image and convert it into a grayscale image. Copy the grayscale image to three channels to convert it into a three-channel image.

[0055] Step S3: Use the optimized Unet model to segment the three-channel image to obtain a blood vessel segmentation map;

[0056] Step S4: Stitch together the blood vessel segmentation images from all two-dimensional images to obtain a 3D blood vessel segmentation image;

[0057] Step S5: Reconstruct the blood flow velocity vector field to obtain hemodynamic parameters and blood flow path;

[0058] Step S6: Map hemodynamic parameters and blood flow paths onto the surface of the 3D vessel segmentation map to generate a 3D hemodynamic atlas, wherein the hemodynamic parameters are 3D mapped based on B-splines:

[0059] ;

[0060] In the formula, , and Represents the B-spline basis functions. Represents control points. The coordinates of the control points in the 3D vessel segmentation map represent the corresponding three-dimensional coordinates. These control points are the locations where 3D mapping is required. , and These represent the wall shear stress (WSS) on the blood vessel surface, the recovered velocity, and the vorticity, respectively. This represents the result of mapping hemodynamic parameters such as wall shear stress (WSS), velocity, and vorticity onto a 3D vessel segmentation map. This represents the three-dimensional coordinates of the corresponding point in the hemodynamic map.

[0061] Example 2, this example is based on the above example, step S3 specifically includes the following steps:

[0062] Step S31: Input a three-channel image, divide it into 4×4 non-overlapping blocks, and obtain the feature map by linear projection of the non-overlapping blocks;

[0063] Step S32: Encoder encoding, using a visual state space module based on receptive field optimization to encode the feature map, obtaining the initial encoding result;

[0064] Step S33: Downsample the primary encoding result to obtain the primary downsampled result;

[0065] Step S34: Multi-layer downsampling coding, repeat steps S32 to S33 three times to obtain downsampling codes;

[0066] Step S35: Multi-layer upsampling encoding. The above-mentioned visual state space module based on receptive field optimization and Patch Expanding are used to perform 4-layer upsampling on the downsampling encoding to obtain the decoding features.

[0067] Step S36: Restore the decoded features to the original three-channel image size through a linear layer and output the blood vessel segmentation map;

[0068] Applying the above scheme to cerebral blood vessel segmentation:

[0069] Background: Early diagnosis of intracranial aneurysms relies on high-precision segmentation of cerebral blood vessels, but MRA (magnetic resonance angiography) data often suffers from low contrast and small vessel breaks.

[0070] 3D Improvement: A 3D Improvement Design for a Visual State Space Module Based on Receptive Field Optimization

[0071] Extending 2D four-directional scanning to six-directional (bidirectional x / y / z axis) 3D sequence scanning models the spatial continuity of blood vessels.

[0072] 3D grouped convolution (kernel size 3×3×3) is used to extract local vascular features, and attention map calculation is changed to 3D average pooling;

[0073] Network structure:

[0074] Input: 3D MRA volume data (256×256×128);

[0075] Encoder: 4-stage downsampling (3D Patch Merging), number of channels per stage [96,192,384,768];

[0076] Skip connection: The 3D features of the encoder are spliced ​​with the corresponding layers of the decoder to preserve the spatial information of blood vessel branches.

[0077] Example 3, see Figure 2 This embodiment is based on the above embodiment. Step S32 specifically includes the following steps:

[0078] Step S321: Use the Visual State Space Model (VSSM) based on residual optimization to globally model the feature map and obtain global features:

[0079] ;

[0080] In the formula, Represents the visual state space model. Represents normalization operation, For learnable scaling factor, Representative feature map, adjust residual connection strength, For global features;

[0081] Step S322: Unfold the feature map into a one-dimensional sequence in four directions, perform Mamba processing on each sequence, and then merge them into a 2D feature map.

[0082] ;

[0083] In the formula, For feature expansion operations, Represents the direction of development. This represents a merge operation. Represents 2D features;

[0084] Step S323: Extract the receptive field through a non-overlapping window and combine it with spatial attention to perform local feature enhancement on the 2D features, resulting in enhanced local features:

[0085] ;

[0086] In the formula, This represents a grouped convolution with a kernel size of o*o. This represents average pooling. Represents standardized processing. and Represents the activation function. This is a local enhancement feature;

[0087] Step S324: Dynamically fuse global features and local enhanced features to obtain the initial coding result;

[0088] The above approach was applied to coronary CTA segmentation (quantification of stenosis).

[0089] Task Background: In coronary CTA, calcified plaques and vascular stenosis cause local gray-scale abrupt changes, which are prone to oversegmentation using traditional methods;

[0090] (1) Dynamic receptive field adjustment:

[0091] Variable kernel grouping convolution: Adaptively selects the convolution kernel size (1×1×1 to 5×5×5) based on the vessel radius; small kernels are used to preserve details in narrow vessels, while large kernels are used to suppress noise in wide vessels.

[0092] Original form:

[0093] ;

[0094] optimization:

[0095] ; This represents the local blood vessel diameter, which needs to be standardized before calculation to fit the log function.

[0096] (2) Multimodal input fusion:

[0097] Dual-path encoder: processes raw CTA data and calcification enhancement maps (generated via threshold segmentation) in parallel, fusing features within a visual state space module based on receptive field optimization.

[0098] Example 4: Based on the above examples, this example constructs an ultrasound-based 3D aortic hemodynamic atlas for early identification of vulnerable plaques and prediction of acute events (such as plaque rupture).

[0099] Data collection

[0100] Equipment: Using a 3D ultrasound (TEE) probe, 5MHz frequency, scanning from the ascending aorta to the descending aorta;

[0101] Parameters: Acquire B-mode and Doppler data for 10 consecutive cardiac cycles, spatial resolution 0.4×0.4×0.5mm³, frame rate 25Hz;

[0102] Preprocessing and segmentation

[0103] Filtering: The Perona-Malik model (κ=0.1) is used to suppress blood flow speckle noise while preserving the vessel wall structure;

[0104] Segmentation: The aortic lumen is initialized based on the optimized Unet model (step S3) (Dice=0.94), and the calcified plaque boundaries are corrected by level set evolution (50 iterations).

[0105] Hemodynamic calculations

[0106] Key parameters:

[0107] Low shear stress zone: Mark vulnerable areas of the plaque;

[0108] Oscillatory shear index (OSI): Calculates the time-varying shear direction change; OSI > 0.3 indicates inflammation risk.

[0109] Blood flow retention time: Through particle tracking simulation (blood flow velocity vector field), a retention time >300ms indicates a tendency for thrombosis.

[0110] Map Construction and Validation

[0111] Registration: Data from 50 patients were registered to the standard aortic ATLAS (B-spline control point spacing 10mm).

[0112] Statistical graphs: Generate probability distribution heatmaps for low WSS and high OSI;

[0113] Clinical output:

[0114] Risk Mapping: Visualizing 3D "High-Risk Red Zones" (Low WSS + High OSI + Long Stay Time) to guide intervention decisions.

[0115] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0116] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

[0117] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.

Claims

1. A 3D hemodynamic mapping method for ultrasound examination, characterized in that: Specifically comprising the following steps: Step S1: collect the three-dimensional ultrasound image of the required mapping site, including the blood flow velocity vector field; Step S2: pre-process the three-dimensional ultrasound image and slice it into two-dimensional images, pre-process the two-dimensional images and convert them into grayscale images, and copy the grayscale images to three channels to convert them into three-channel images; Step S3: use an optimized Unet model to segment the three-channel images to obtain a blood vessel segmentation map, specifically comprising the following steps: Step S31: input the three-channel image, divide it into 4x4 non-overlapping blocks, and obtain a feature map by linear projection of the non-overlapping blocks; Step S32: encoder coding, using a receptive field optimization-based visual state space module to encode the feature map to obtain a primary encoding result, specifically comprising the following steps: Step S321: use a residual optimization-based visual state space model VSSM to globally model the feature map to obtain global features: ； wherein represents a visual state space model, represents a normalization operation, is a learnable scaling factor, represents a feature map, adjusting the residual connection strength, is a global feature; Step S322: expand the feature map into a one-dimensional sequence in four directions, and perform Mamba processing on each direction, and then merge them into 2D features: ； wherein is a feature expansion operation, represents the direction of expansion, represents a merge operation, represents a 2D feature; Step S323: extract the receptive field through a non-overlapping window and combine it with spatial attention to enhance the local features of the 2D features to obtain locally enhanced features: ； In the formula, represents a grouped convolution with a convolution kernel size of o*o, represents an average pooling process, represents a normalization process, and represents an activation function, is a local enhanced feature; Step S324: dynamically fuse the global features and the locally enhanced features to obtain the primary encoding result; Step S33: downsample the primary encoding result to obtain a primary downsampled result; Step S34: multi-layer downsample coding, repeat steps S32 to S33 three times to obtain a downsampled coding; Step S35: multi-layer upsample coding, use the above-mentioned receptive field optimization-based visual state space module and Patch Expanding to perform 4-layer upsample on the downsampled coding to obtain decoding features; Step S36: restore the decoding features to the original three-channel image size through a linear layer, and output the blood vessel segmentation map; Step S4: splice the blood vessel segmentation maps of all two-dimensional images to obtain a 3D blood vessel segmentation map; Step S5: reconstruct the blood flow velocity vector field to obtain blood flow dynamics parameters and blood flow paths, specifically comprising the following steps: Step S51: Doppler angle correction, combined with the geometric parameters of the three-dimensional ultrasound probe, to restore the blood flow velocity vector field: ； wherein the nodal velocity representing the blood flow velocity vector field, is the blood flow direction vector, is the acoustic beam direction vector, is the included angle, wherein is estimated by the three-dimensional probe pose data represents the velocity after the recovery; Step S52: use radial basis function interpolation to fill in the missing values of the blood flow velocity vector field; Step S53: blood flow dynamics parameter calculation, calculate the wall shear stress WSS and vorticity on the surface of the blood vessel: ， ； wherein represents the blood viscosity, represents the tangential velocity in the blood flow velocity vector field, represents the wall shear stress, represents the spatial gradient of the blood flow velocity vector field, represents the vorticity; Step S54: blood flow path construction, integrate along the velocity field to generate streamlines to describe the blood flow path; Step S6: map the blood flow dynamics parameters and blood flow paths to the surface of the 3D blood vessel segmentation map to generate a 3D blood flow dynamics map, wherein the blood flow dynamics parameters are 3D mapped based on B-splines: ； In the formula, , , represents a B-spline basis function, represents a control point, represents a three-dimensional coordinate corresponding to the control point in the 3D blood vessel segmentation graph, and the control point is a point position required to be mapped in 3D, , , respectively represent a wall shear stress WSS of a blood vessel surface, a velocity after recovery, and a vorticity, represents a result of mapping the wall shear stress WSS, the velocity, and the vorticity of the hemodynamic parameter to the 3D blood vessel segmentation graph, that is, represents a three-dimensional coordinate of a corresponding point position in the hemodynamic atlas.

Images