Blood vessel ultrasonic image three-dimensional reconstruction method, device and equipment and storage medium

Abstract

The invention discloses a blood vessel ultrasonic image three-dimensional reconstruction method, device and equipment and a storage medium, and relates to the technical field of image processing, and the method comprises the steps: collecting an ultrasonic image of a blood vessel in real time; acquiring space pose information of the ultrasonic probe; segmenting the ultrasonic image based on a U2-Net model, and extracting to obtain an ultrasonic blood vessel image; according to coordinates and space pose information of the calibration key points in an ultrasonic image coordinate system, an electromagnetic positioning coordinate system and a world coordinate system, a transformation matrix from the ultrasonic image coordinate system to an electromagnetic sensor space, a transformation matrix from the electromagnetic sensor space to a magnetic field space and a transformation matrix from the magnetic field space to the world coordinate system are obtained through calculation; converting each pixel point of the blood vessel in the ultrasonic image into a navigation coordinate system based on a geometric transformation algorithm to obtain a converted image; and carrying out real-time rendering on the obtained multi-frame conversion image in the world coordinate system. According to the invention, the three-dimensional reconstruction precision and speed of the ultrasonic image of the blood vessel are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of ultrasonic image processing, and in particular to a method, device, equipment and storage medium for three-dimensional reconstruction of ultrasonic images of blood vessels. Background Art

[0002] In recent years, ultrasound imaging has been widely used in cardiovascular interventional surgeries, especially in intravascular stent implantation, cardiac catheter intervention and other surgeries due to its non-invasive, real-time and convenient features. The real-time and high-frequency acquisition of ultrasound images requires efficient image processing and analysis technology to assist doctors in making accurate decisions. However, the traditional three-dimensional reconstruction method of ultrasound images of blood vessels faces multiple challenges: First, ultrasound images are greatly interfered by noise and have low segmentation accuracy, especially in dynamically changing vascular structures, resulting in low three-dimensional reconstruction accuracy. Secondly, real-time processing of ultrasound images requires fast spatial reconstruction, but traditional methods rely on CPU calculations, which are slow and difficult to meet real-time requirements.

[0003] Therefore, the applicant has developed a method, device, equipment and storage medium for three-dimensional reconstruction of ultrasound images of blood vessels to solve the above problems. Summary of the invention

[0004] The present invention provides a method, device, equipment and storage medium for three-dimensional reconstruction of ultrasonic images of blood vessels, so as to solve the problems of low precision and slow speed of the existing three-dimensional reconstruction methods of ultrasonic images of blood vessels.

[0005] The present invention achieves the above-mentioned purpose through the following technical solutions:

[0006] In a first aspect, the present invention provides a method for three-dimensional reconstruction of an ultrasonic image of a blood vessel, comprising:

[0007] Real-time acquisition of ultrasound images of blood vessels;

[0008] Acquiring spatial position information of the ultrasound probe through an electromagnetic sensor, wherein the spatial position information includes the position and rotation quaternion of the ultrasound probe in space;

[0009] The ultrasound image is segmented based on the U2-Net model to extract the ultrasound vascular image, and the U2-Net model is accelerated inference using the TensorRT framework during the segmentation process;

[0010] According to the coordinates of the calibration key points in the ultrasound image coordinate system, the electromagnetic positioning coordinate system, and the world coordinate system, as well as the spatial posture information, a conversion matrix from the ultrasound image coordinate system to the electromagnetic sensor space, a conversion matrix from the electromagnetic sensor space to the magnetic field space, and a conversion matrix from the magnetic field space to the world coordinate system are calculated, wherein the calibration key points are key points calibrated in the ultrasound image according to the ultrasound calibration algorithm;

[0011] According to the conversion matrix from the ultrasound image coordinate system to the electromagnetic sensor space, the conversion matrix from the electromagnetic sensor space to the magnetic field space, and the conversion matrix from the magnetic field space to the world coordinate system, each pixel point of the blood vessel in the ultrasound image is converted into the navigation coordinate system based on a geometric transformation algorithm to obtain a converted image;

[0012] The obtained multiple frames of the converted images are rendered in real time in the world coordinate system to obtain a real-time three-dimensional reconstruction result.

[0013] Furthermore, the U2-Net model is accelerated in the segmentation process through the TensorRT framework, including:

[0014] The pth model trained by the U2-Net model is converted into an onnx model. The formula is as follows:

[0015] ;

[0016] Represents a PyTorch computation graph, which represents a directed computation graph, where V is a node in the computation graph, E is an edge in the computation graph, W is a weight tensor, M represents dynamic computation symbol mapping, and O represents computation graph optimization processing, which includes operator fusion, computation graph pruning, and computation flow rearrangement. Represents the conversion function that converts the PyTorch model to the ONNX model. ONNX computation graph representing the input;

[0017] Convert the onnx model to the trt model, the formula is as follows:

[0018] ;

[0019] in, represents the trt model, represents the ONNX model weight, O represents the computational graph optimization processing, F represents the operator fusion, S represents the scaling factor, C represents the compiler mapping function, and the scaling factor S is calculated as follows:

[0020] ;

[0021] Among them, b=n, which means weight quantization ONNX model weights Perform FPn quantization, where FPn represents the model accuracy and n can be 8, 16, or 32.

[0022] Furthermore, according to the coordinates of the calibrated key points in the ultrasound image coordinate system, the electromagnetic positioning coordinate system, and the world coordinate system, and the spatial posture information, a conversion matrix from the ultrasound image coordinate system to the electromagnetic sensor space, a conversion matrix from the electromagnetic sensor space to the magnetic field space, and a conversion matrix from the magnetic field space to the world coordinate system are calculated, including:

[0023] ;

[0024] ;

[0025] ;

[0026] ;

[0027]

[0028] N is the number of calibration key points determined in the ultrasound coordinate system and the electromagnetic positioning coordinate system obtained by calibration, where one calibration key point , is the coordinate of the key point in the ultrasound image coordinate system, is the coordinate of the key point in the electromagnetic positioning coordinate system, is the coordinate of the key point in the world coordinate system, TM represents the matrix required on the left side of the minimization formula, is the conversion matrix from ultrasound image coordinate system to electromagnetic sensor space, is the conversion matrix from electromagnetic sensor space to magnetic field space, is the transformation matrix from magnetic field space to world coordinate system, is the rotation quaternion, T represents the translation, i.e. the position of the ultrasound probe in space, and t x ,t y ,t z The table represents the position coordinates of the probe on the three coordinate axes, and R represents the rotation matrix.

[0029] Furthermore, according to the conversion matrix from the ultrasound image coordinate system to the electromagnetic sensor space, the conversion matrix from the electromagnetic sensor space to the magnetic field space, and the conversion matrix from the magnetic field space to the world coordinate system, each pixel point of the blood vessel in the ultrasound image is converted into the navigation coordinate system based on a geometric transformation algorithm to obtain a conversion image, including:

[0030] ;

[0031] in, is the coordinate of the pixel point in the converted image, (x, y) is the coordinate of the pixel point of the blood vessel in the ultrasound image coordinate system, S x , S yare the scaling factors of the x-axis and y-axis in the ultrasound image coordinate system, respectively.

[0032] In a second aspect, the present invention further provides a device for three-dimensional reconstruction of ultrasound images of blood vessels, comprising:

[0033] An acquisition module, the acquisition module is used to acquire an ultrasonic image of a blood vessel in real time;

[0034] An acquisition module, the acquisition module is used to acquire spatial posture information of the ultrasound probe through an electromagnetic sensor, the spatial posture information including the position and rotation quaternion of the ultrasound probe in space;

[0035] A segmentation module, wherein the segmentation module is used to segment the ultrasound image based on a U2-Net model to extract an ultrasound vascular image, and to accelerate the reasoning of the U2-Net model through a TensorRT framework during the segmentation process;

[0036] A calculation module, the calculation module is used to calculate the ultrasound-to-magnetic field space conversion matrix and the magnetic field-to-ultrasound space conversion matrix according to the spatial posture information through a calibration model and an algorithm based on feature points;

[0037] A conversion module, the conversion module is used to convert each pixel point in the ultrasonic vascular image into a navigation coordinate system based on a geometric transformation algorithm according to the ultrasonic-to-magnetic field space conversion matrix and the magnetic field-to-ultrasound space conversion matrix to obtain a conversion image;

[0038] A rendering module performs real-time rendering on the converted multiple-frame images in a world coordinate system to obtain a real-time three-dimensional reconstruction result.

[0039] In a third aspect, the present invention further provides a device for three-dimensional reconstruction of ultrasound images of blood vessels, comprising:

[0040] Memory for storing computer programs;

[0041] A processor is used to implement the steps of the method for three-dimensional reconstruction of an ultrasonic image of a blood vessel when executing the computer program.

[0042] The present invention also discloses a storage medium, which is a readable storage medium. A computer program is stored on the storage medium. When the computer program is executed by a processor, the steps of a method for three-dimensional reconstruction of an ultrasonic image of a blood vessel are implemented.

[0043] The beneficial effects of the present invention are:

[0044] The present invention proposes a method for three-dimensional reconstruction of ultrasound images of blood vessels to solve the problems of low accuracy and slow speed of existing methods for three-dimensional reconstruction of ultrasound images of blood vessels. The present invention collects ultrasound images of blood vessels in real time and accelerates the reasoning of the U2-Net model, using its highly parallelized computing characteristics to significantly improve the real-time performance and computing accuracy in the ultrasound image segmentation process. Subsequently, the electromagnetic sensor is combined to obtain the precise spatial position information of the ultrasound probe, and the local coordinate system of the ultrasound image is docked with the global navigation system coordinate system with high precision through a geometric transformation algorithm, thereby improving the accuracy and speed of three-dimensional reconstruction of ultrasound images of blood vessels. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 The present invention provides a flow chart of a method for three-dimensional reconstruction of ultrasound images of blood vessels. DETAILED DESCRIPTION

[0046] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0047] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0048] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, further definition and explanation thereof is not required in subsequent drawings.

[0049] The specific implementation modes of the present invention are described in detail below in conjunction with the accompanying drawings.

[0050] like Figure 1 As shown, the present invention provides a method for three-dimensional reconstruction of an ultrasonic image of a blood vessel, comprising:

[0051] S1: Real-time acquisition of ultrasound images of blood vessels;

[0052] The ultrasound image of the blood vessel is collected in real time by a video acquisition card. The two-dimensional ultrasound image collected by a sensor (such as an ultrasound probe) can be expressed as I(x, y, t), where (x, y) is the spatial coordinate of the image and t is the time dimension. Specifically, at each time t, the acquisition card obtains the ultrasound image from the sensor and converts it into digital image data through signal processing.

[0053] S2: Acquire spatial posture information of the ultrasound probe through an electromagnetic sensor, where the spatial posture information includes the position and rotation quaternion of the ultrasound probe in space;

[0054] Combined with electromagnetic sensors to obtain accurate spatial position information of the ultrasound probe, the electromagnetic sensor provides the position and rotation quaternion of the ultrasound probe, where , T represents the translation (i.e. the position of the ultrasound probe in space), A quaternion is used to describe rotation.

[0055] S3: segmenting the ultrasound image based on the U2-Net model to extract the ultrasound vascular image, and accelerating the inference of the U2-Net model through the TensorRT framework during the segmentation process;

[0056] The collected ultrasound images are segmented using the U2-Net model, which can accurately extract the vascular structure. At the same time, the U2-Net model is accelerated through the TensorRT framework, and its highly parallel computing characteristics are used to significantly improve the real-time performance and computing accuracy of the ultrasound image segmentation process. The complex segmentation tasks of the U2-Net model are specifically optimized through the TensorRT framework, including adaptive operator fusion, weight quantization, cache optimization, and CUDA C preprocessing to reduce CPU-GPU data copy overhead, improve inference throughput, make inference more efficient, and reduce video memory usage.

[0057] Specifically, S3 includes:

[0058] First, the PTH model trained by the U2-Net model is converted into an ONNX model. The formula is as follows:

[0059] ;

[0060] Represents a PyTorch computation graph, which represents a directed computation graph, where V is a node in the computation graph, E is an edge in the computation graph, W is a weight tensor, M represents dynamic computation symbol mapping, and O represents computation graph optimization processing, which includes operator fusion, computation graph pruning, and computation flow rearrangement. Represents the conversion function that converts the PyTorch model to the ONNX model, thereby ensuring the consistency of the computational logic. During the conversion process, O is responsible for computational graph optimization, including operator fusion, which integrates multiple mergeable operators into an efficient operator; computational graph pruning, which reduces redundant calculations by removing invalid computing nodes; and computational flow reordering, which optimizes the execution order of the computational graph to improve reasoning efficiency and resource utilization.

[0061] Then convert the onnx model into a trt model , the formula is as follows:

[0062] ;

[0063] in, The ONNX computational graph representing the input consists of multiple computational nodes (operators) and is processed by computational graph optimization O, including optimization strategies such as computational flow reordering, operator fusion, and computational graph pruning to improve inference efficiency. The optimized computational graph is subjected to operator fusion F to merge multiple independent operators into more efficient operators, further reducing computational overhead. Subsequently, weight quantization ONNX model weights Perform FP16 quantization, where the scaling factor S is calculated as:

[0064] ;

[0065] Where b=16 (if FP16 quantization), the quantized calculation graph passes through the compiler mapping function C, converts the optimized ONNX calculation graph into an efficient TensorRT calculation engine, and finally generates , which is the TensorRT engine for efficient reasoning. This process combines computational optimization, weight quantization, and hardware acceleration, allowing the model to have faster reasoning speed, lower video memory usage, and better energy efficiency when running on NVIDIA GPUs.

[0066] S4: According to the coordinates of the calibration key points in the ultrasound image coordinate system, the electromagnetic positioning coordinate system, and the world coordinate system and the spatial posture information, a conversion matrix from the ultrasound image coordinate system to the electromagnetic sensor space, a conversion matrix from the electromagnetic sensor space to the magnetic field space, and a conversion matrix from the magnetic field space to the world coordinate system are calculated, wherein the calibration key points are key points calibrated in the ultrasound image according to the ultrasound calibration algorithm, including:

[0067] ;

[0068] ;

[0069] ;

[0070] ;

[0071] ;

[0072] N is the number of calibration key points determined in the ultrasound coordinate system and the electromagnetic positioning coordinate system obtained by calibration, where one calibration key point , is the coordinate of the key point in the ultrasound image coordinate system, is the coordinate of the key point in the electromagnetic positioning coordinate system, is the coordinate of the key point in the world coordinate system, TM represents the matrix required on the left side of the minimization formula, is the conversion matrix from ultrasound image coordinate system to electromagnetic sensor space, is the conversion matrix from electromagnetic sensor space to magnetic field space, is the transformation matrix from magnetic field space to world coordinate system, is the rotation quaternion, T represents the translation, i.e. the position of the ultrasound probe in space, and t x ,t y ,t z The subtable represents the position coordinates of the probe on the three coordinate axes, and R represents the rotation matrix;

[0073] The ultrasonic calibration algorithm can adopt one of the single point (line) method, the two-dimensional plane alignment method and the Freehand method. For example, the Freehand method uses a model with several parallel thin lines that can be imaged in ultrasound to implement ultrasound probe calibration. This method makes the image scanning plane of the ultrasound probe intersect with these parallel lines. The model can be an open rectangular block composed of five plexiglass plates. The model contains several copper wires, and the two ends of each copper wire are fixed on the corresponding holes on the front wall and the back wall of the model. All copper wires form multiple N-shaped structures, and ultrasound scanning will form 3 bright spots, among which the middle bright spot is the key point.

[0074] S5: According to the conversion matrix from the ultrasound image coordinate system to the electromagnetic sensor space, the conversion matrix from the electromagnetic sensor space to the magnetic field space, and the conversion matrix from the magnetic field space to the world coordinate system, each pixel point of the blood vessel in the ultrasound image is converted into the navigation coordinate system based on a geometric transformation algorithm to obtain a converted image, including:

[0075] ;

[0076] in, is the coordinate of the pixel point in the converted image, (x, y) is the coordinate of the pixel point of the blood vessel in the ultrasound image coordinate system, S x , S y are the scaling factors of the x-axis and y-axis in the ultrasound image coordinate system, respectively.

[0077] S6: Render the obtained multi-frame converted images in real time in the world coordinate system to obtain real-time three-dimensional reconstruction results, thereby realizing spatial mapping between ultrasound data and the navigation system, and optimizing volume rendering of ultrasound slices in the world coordinate system, thereby realizing rapid three-dimensional reconstruction of two-dimensional ultrasound.

[0078] The embodiment of the present invention further provides a device for three-dimensional reconstruction of an ultrasonic image of a blood vessel, comprising:

[0079] An acquisition module, the acquisition module is used to acquire an ultrasonic image of a blood vessel in real time;

[0080] An acquisition module, the acquisition module is used to acquire spatial posture information of the ultrasound probe through an electromagnetic sensor, the spatial posture information including the position and rotation quaternion of the ultrasound probe in space;

[0081] A segmentation module, wherein the segmentation module is used to segment the ultrasound image based on a U2-Net model to extract an ultrasound vascular image, and to accelerate the reasoning of the U2-Net model through a TensorRT framework during the segmentation process;

[0082] A calculation module, the calculation module is used to calculate the ultrasound-to-magnetic field space conversion matrix and the magnetic field-to-ultrasound space conversion matrix according to the spatial posture information through a calibration model and an algorithm based on feature points;

[0083] A conversion module, the conversion module is used to convert each pixel point in the ultrasonic vascular image into a navigation coordinate system based on a geometric transformation algorithm according to the ultrasonic-to-magnetic field space conversion matrix and the magnetic field-to-ultrasound space conversion matrix to obtain a conversion image;

[0084] A rendering module performs real-time rendering on the converted multiple-frame images in a world coordinate system to obtain a real-time three-dimensional reconstruction result.

[0085] The embodiment of the present invention further provides a device for three-dimensional reconstruction of an ultrasonic image of a blood vessel, comprising:

[0086] Memory for storing computer programs;

[0087] A processor is used to implement the steps of the method for three-dimensional reconstruction of an ultrasonic image of a blood vessel when executing the computer program.

[0088] An embodiment of the present invention further provides a storage medium, which is a readable storage medium and stores a computer program. When the computer program is executed by a processor, the steps of the method for three-dimensional reconstruction of an ultrasonic image of a blood vessel are implemented.

[0089] A three-dimensional reconstruction method for ultrasonic images of blood vessels proposed by the present invention solves the problems of low accuracy and slow speed in the existing three-dimensional reconstruction methods for ultrasonic images of blood vessels. By collecting ultrasonic images of blood vessels in real time and accelerating the inference of the U2-Net model, the real-time performance and calculation accuracy in the ultrasonic image segmentation process are significantly improved by using its highly parallel computing characteristics. Subsequently, the precise spatial pose information of the ultrasonic probe is obtained by combining an electromagnetic sensor, and the local coordinate system of the ultrasonic image is accurately docked with the global navigation system coordinate system through a geometric transformation algorithm, improving the three-dimensional reconstruction accuracy and speed of the ultrasonic images of blood vessels.

[0090] The above are only the preferred embodiments of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the technical principle of the present invention, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present invention.

Claims

1. A method for three-dimensional reconstruction of an ultrasonic image of a blood vessel, characterized in that: include: Real-time acquisition of ultrasound images of blood vessels; Acquiring spatial position information of the ultrasound probe through an electromagnetic sensor, wherein the spatial position information includes the position and rotation quaternion of the ultrasound probe in space; The ultrasound image is segmented based on the U2-Net model to extract the ultrasound vascular image, and the U2-Net model is accelerated inference using the TensorRT framework during the segmentation process; According to the coordinates of the calibration key points in the ultrasound image coordinate system, the electromagnetic positioning coordinate system, and the world coordinate system, as well as the spatial posture information, a conversion matrix from the ultrasound image coordinate system to the electromagnetic sensor space, a conversion matrix from the electromagnetic sensor space to the magnetic field space, and a conversion matrix from the magnetic field space to the world coordinate system are calculated, wherein the calibration key points are key points calibrated in the ultrasound image according to the ultrasound calibration algorithm; According to the conversion matrix from the ultrasound image coordinate system to the electromagnetic sensor space, the conversion matrix from the electromagnetic sensor space to the magnetic field space, and the conversion matrix from the magnetic field space to the world coordinate system, each pixel point of the blood vessel in the ultrasound image is converted into the navigation coordinate system based on a geometric transformation algorithm to obtain a converted image; The obtained multiple frames of the converted images are rendered in real time in the world coordinate system to obtain a real-time three-dimensional reconstruction result.

2. The method for three-dimensional reconstruction of an ultrasonic image of a blood vessel according to claim 1, characterized in that: During the segmentation process, the U2-Net model is accelerated through the TensorRT framework, including: The pth model trained by the U2-Net model is converted into an onnx model. The formula is as follows: ； Represents a PyTorch computation graph, which represents a directed computation graph, where V is a node in the computation graph, E is an edge in the computation graph, W is a weight tensor, M represents dynamic computation symbol mapping, and O represents computation graph optimization processing, which includes operator fusion, computation graph pruning, and computation flow rearrangement. Represents the conversion function that converts the PyTorch model to the ONNX model. ONNX computation graph representing the input; Convert the onnx model to the trt model, the formula is as follows: ； in, represents the trt model, represents the ONNX model weight, O represents the computational graph optimization processing, F represents the operator fusion, S represents the scaling factor, C represents the compiler mapping function, and the scaling factor S is calculated as follows: ； Among them, b=n, which means weight quantization ONNX model weights Perform FPn quantization, where FPn represents the model accuracy.

3. The method for three-dimensional reconstruction of a blood vessel ultrasound image according to claim 1, characterized in that: According to the coordinates of the calibration key points in the ultrasound image coordinate system, the electromagnetic positioning coordinate system, and the world coordinate system, as well as the spatial posture information, the conversion matrix from the ultrasound image coordinate system to the electromagnetic sensor space, the conversion matrix from the electromagnetic sensor space to the magnetic field space, and the conversion matrix from the magnetic field space to the world coordinate system are calculated, including: ； ； ； ； ； N is the number of calibration key points determined in the ultrasound coordinate system and the electromagnetic positioning coordinate system obtained by calibration, where one calibration key point , is the coordinate of the key point in the ultrasound image coordinate system, is the coordinate of the key point in the electromagnetic positioning coordinate system, is the coordinate of the key point in the world coordinate system, TM represents the matrix required on the left side of the minimization formula, is the conversion matrix from ultrasound image coordinate system to electromagnetic sensor space, is the conversion matrix from electromagnetic sensor space to magnetic field space, is the transformation matrix from magnetic field space to world coordinate system, is the rotation quaternion, T represents the translation, i.e. the position of the ultrasound probe in space, and t x ,t y ,t z The table represents the position coordinates of the probe on the three coordinate axes, and R represents the rotation matrix.

4. The method for three-dimensional reconstruction of ultrasound images of blood vessels according to claim 3, characterized in that: According to the conversion matrix from the ultrasound image coordinate system to the electromagnetic sensor space, the conversion matrix from the electromagnetic sensor space to the magnetic field space, and the conversion matrix from the magnetic field space to the world coordinate system, each pixel point of the blood vessel in the ultrasound image is converted into the navigation coordinate system based on a geometric transformation algorithm to obtain a conversion image, including: ， in, is the coordinate of the pixel point in the converted image, (x, y) is the coordinate of the pixel point of the blood vessel in the ultrasound image coordinate system, S x , S y are the scaling factors of the x-axis and y-axis in the ultrasound image coordinate system, respectively.

5. A device for three-dimensional reconstruction of ultrasound images of blood vessels, characterized in that: include: An acquisition module, the acquisition module is used to acquire an ultrasonic image of a blood vessel in real time; An acquisition module, the acquisition module is used to acquire spatial posture information of the ultrasound probe through an electromagnetic sensor, the spatial posture information including the position and rotation quaternion of the ultrasound probe in space; A segmentation module, wherein the segmentation module is used to segment the ultrasound image based on a U2-Net model to extract an ultrasound vascular image, and to accelerate the reasoning of the U2-Net model through a TensorRT framework during the segmentation process; A calculation module, the calculation module is used to calculate the ultrasound-to-magnetic field space conversion matrix and the magnetic field-to-ultrasound space conversion matrix according to the spatial posture information through a calibration model and an algorithm based on feature points; A conversion module, the conversion module is used to convert each pixel point in the ultrasonic vascular image into a navigation coordinate system based on a geometric transformation algorithm according to the ultrasonic to magnetic field space conversion matrix and the magnetic field to ultrasonic space conversion matrix to obtain a conversion image; A rendering module performs real-time rendering on the converted multiple-frame images in a world coordinate system to obtain a real-time three-dimensional reconstruction result.

6. A three-dimensional reconstruction device for ultrasonic images of blood vessels, characterized in that: include: Memory for storing computer programs; A processor is used to implement the steps of the method for three-dimensional reconstruction of an ultrasonic image of a blood vessel as claimed in any one of claims 1 to 4 when executing the computer program.

7. A storage medium, characterized in that: The storage medium is a readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, the steps of the method for three-dimensional reconstruction of an ultrasonic image of a blood vessel as claimed in any one of claims 1 to 4 are implemented.

Images