Super-resolution ultrasonic imaging method and system
The ultrasonic contrast agent signal is processed through multi-scale convolutional spatiotemporal neural network, which solves the imaging distortion problem of ultrasonic imaging under high-concentration microbubbles, and realizes high-precision reconstruction of blood vessel structure and blood flow parameter maps, which is suitable for animals, humans and other living biological tissues.
Patent Information
- Application Number
- CN202510478890.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-16
- Publication Date
- 2025-07-18
AI Technical Summary
Under high concentration microbubbles, traditional ultrasound positioning microscopy is difficult to achieve high-precision reconstruction of microvascular structure and blood flow parameter maps, resulting in reduced imaging distortion and clinical diagnostic accuracy.
Multi-scale convolutional spatiotemporal neural network is used to process data containing ultrasonic contrast agent signals. Combined with the encoder module, recurrent neural network and decoder module, multi-scale features are extracted and fused to generate high-dimensional feature maps.
Under high-concentration microbubble conditions, high-quality vascular structure and blood flow parameter map reconstruction in a shorter time is achieved, trajectory tracking ability and imaging accuracy are improved, and are suitable for various living biological tissues.
Smart Images

Figure CN120339438A_ABST
Abstract
Description
Technical Field
[0001] This application belongs to the technical field of image processing, and specifically relates to a super-resolution ultrasound imaging method and system. Background Art
[0002] Ultrasound localization microscopy is a super-resolution imaging technology that breaks through the diffraction limit. It relies on the high-precision localization and tracking of intravascular microbubble contrast agents to reconstruct super-resolution images of the microvascular system. Compared with traditional ultrasound imaging, ultrasound localization microscopy achieves a resolution much higher than the ultrasound wavelength through the independent scattering characteristics of microbubbles, enabling the visualization of submillimeter- or even micrometer-scale vascular structures.
[0003] However, in practical applications, its imaging performance is affected by the microbubble concentration. A lower concentration of microbubbles helps improve the localization accuracy because separated microbubbles can be individually identified and accurately located, thus ensuring the super-resolution of the image. However, the injection of low-concentration microbubbles results in a smaller number of microbubbles entering the field of view in each time frame, leading to a slower perfusion process of the entire microvascular network and requiring a longer acquisition time to obtain a complete super-resolution angiogram. This time delay not only reduces the practicality of ultrasound localization microscopy in dynamic blood flow imaging but also poses challenges to in vivo imaging, especially in clinical applications where physiological conditions change rapidly, such as cardiovascular disease assessment.
[0004] On the other hand, increasing the microbubble injection concentration can accelerate the perfusion of the microvasculature, enabling more vascular systems to be visualized in a shorter time, thereby shortening the data acquisition window and improving the imaging efficiency. However, high-concentration microbubbles are prone to overlap and aggregation in ultrasound images, making it difficult for traditional microbubble localization algorithms to distinguish individual microbubbles, which in turn affects the accuracy of super-resolution imaging. This microbubble signal aliasing can introduce artifacts or incorrect vascular structures, causing the reconstructed super-resolution angiogram to be distorted and even potentially misestimating the blood flow velocity, affecting the accuracy of clinical diagnosis.
[0005] Therefore, how to ensure the precise localization of ultrasound localization microscopy under high-concentration microbubble conditions and achieve high-quality reconstruction of super-resolution velocity trajectory maps is a key challenge in the current field of super-resolution research.
[0006] Content of the Application
[0007] To overcome the above defects of the prior art, this application proposes a super-resolution ultrasound imaging method and system, which solves the problem of distortion in the reconstruction of super-resolution vascular structure diagrams and blood flow velocity diagrams under high-concentration microbubble conditions compared with traditional super-resolution vascular blood flow diagrams.
[0008] Specifically, it is achieved through the following technical solutions:
[0009] A super-resolution ultrasonic imaging method, which is applied to the vascular blood flow structure imaging of animals, humans, and other living biological tissues, includes:
[0010] Obtaining data containing ultrasonic contrast agent signals of the blood vessels of a living biological tissue;
[0011] Using a machine learning method to process the data containing ultrasonic contrast agent signals to obtain a structural and blood flow parameter map of the blood vessels of a living biological tissue.
[0012] In a specific embodiment, the step of using a machine learning method to process the data containing ultrasonic contrast agent signals to obtain a structural and blood flow parameter map of the blood vessels of a living biological tissue includes:
[0013] Using a multi-scale convolutional spatio-temporal neural network composed of an encoder module with multi-scale convolution, a recurrent neural network module, and a decoder module to process the data containing ultrasonic contrast agent signals to obtain a structural and blood flow parameter map of the blood vessels of a living biological tissue.
[0014] In a specific embodiment, the encoder module with multi-scale convolution includes a max pooling layer and a first convolutional layer, and the first convolutional layer includes multiple parallel convolutional kernels;
[0015] The steps for the encoder module with multi-scale convolution to process the data containing ultrasonic contrast agent signals are as follows:
[0016] Inputting the data containing ultrasonic contrast agent signals into the max pooling layer to obtain the pooled features of the data containing ultrasonic contrast agent signals;
[0017] Inputting the pooled features into the first convolutional layer to extract multi-scale features from different scales through multiple convolutional kernels;
[0018] Performing fusion processing on the extracted multi-scale features to obtain high-dimensional features.
[0019] In a specific embodiment, obtaining the data containing ultrasonic contrast agent signals of the blood vessels of a living biological tissue includes obtaining simulation data of the blood vessels of a living biological tissue, where the simulation data of the blood vessels of a living biological tissue is used for input into the multi-scale convolutional spatio-temporal neural network for training;
[0020] The process of obtaining the simulation data of the blood vessels of a living biological tissue is as follows:
[0021] Based on the traditional super-resolution vascular blood flow map of the blood vessels of a living biological tissue, performing processing to obtain the vascular structure map and blood flow velocity information of the blood vessels of a living biological tissue;
[0022] Put microbubble ultrasound contrast agent in the blood vessel structure diagram and initialize the microbubble amplitude and microbubble position to obtain the microbubble trajectory according to the blood flow velocity information;
[0023] Calculate the average velocity of each pixel point according to the microbubble trajectory and time to obtain the microbubble trajectory velocity map;
[0024] Send the microbubble position and the microbubble amplitude of each frame into the simulation model, obtain the corresponding ultrasonic RF signal and perform beam synthesis to obtain the ultrasonic frame sequence data containing the ultrasonic contrast agent signal;
[0025] Take the ultrasonic frame sequence data containing the ultrasonic contrast agent signal and the microbubble trajectory velocity map as simulation data.
[0026] In a specific embodiment, the process of obtaining the data containing the ultrasonic contrast agent signal of the blood vessels of the living biological tissue includes obtaining the in-vivo data of the blood vessels of the living biological tissue, wherein the in-vivo data of the blood vessels of the living biological tissue is used as an input to the multi-scale convolutional spatio-temporal neural network for prediction;
[0027] The process of obtaining the in-vivo data of the blood vessels of the living biological tissue is as follows:
[0028] Inject microbubble ultrasound contrast agent into the blood vessels of the living biological tissue, use an ultrasonic probe to obtain the data containing the ultrasonic contrast agent signal, and use SVD to filter out tissue clutter for processing to obtain the in-vivo data of the blood vessels of the living biological tissue.
[0029] In a specific embodiment, the data containing the ultrasonic contrast agent signal includes the simulation data containing the ultrasonic contrast agent signal and the in-vivo data containing the ultrasonic contrast agent signal;
[0030] The process of training and testing the data containing the ultrasonic contrast agent signal by using the multi-scale convolutional spatio-temporal neural network composed of an encoder module with multi-scale convolution, a recurrent neural network module, and a decoder module to obtain the structure and blood flow parameter map of the blood vessels of the living biological tissue includes:
[0031] Use the multi-scale convolutional spatio-temporal neural network composed of an encoder module with multi-scale convolution, a recurrent neural network module, and a decoder module to train the simulation data containing the ultrasonic contrast agent signal to obtain weights;
[0032] Optimize the neural network model based on the weights and the multi-scale convolutional spatio-temporal neural network;
[0033] Input the in-vivo data containing the ultrasonic contrast agent signal into the optimized neural network model for testing to obtain the structure and blood flow parameter map of the blood vessels of the living biological tissue.
[0034] In a specific embodiment, at least two of the multiple convolution kernels are used to extract features of different scales; and / or, the multiple convolution kernels include at least two combinations of 3×3 convolution kernels, 5×5 convolution kernels, and 7×7 convolution kernels.
[0035] In a specific embodiment, the fusing the extracted multiple scale features to obtain high-dimensional features includes:
[0036] Concatenating and fusing the extracted multiple scale features to obtain the concatenated multi-scale features;
[0037] Inputting the concatenated multi-scale features into the second convolutional layer of the encoder module with multi-scale convolution for more advanced feature extraction;
[0038] Inputting the multi-scale features after feature extraction into a batch normalization layer for batch normalization and performing a non-linear transformation through an activation function layer to obtain high-dimensional features.
[0039] In a specific embodiment, the recurrent neural network module includes a long short-term memory network module or a bidirectional recurrent neural network module.
[0040] A super-resolution ultrasound imaging system applied to the vascular blood flow structure imaging of animals, humans, and other living biological tissues, includes:
[0041] An acquisition module for acquiring data containing ultrasound contrast agent signals of the blood vessels of a living biological tissue;
[0042] A processing module for processing the data containing ultrasound contrast agent signals by using a machine learning method to obtain a structural and blood flow parameter map of the blood vessels of a living biological tissue.
[0043] This application has at least the following beneficial effects:
[0044] This application proposes a super-resolution ultrasound imaging method and system applied to the vascular blood flow structure imaging of animals, humans, and other living biological tissues. The method includes: acquiring data containing ultrasound contrast agent signals of the blood vessels of a living biological tissue; processing the data containing ultrasound contrast agent signals by using a machine learning method to obtain a structural and blood flow parameter map of the blood vessels of a living biological tissue. This method not only solves the problem of distortion of the reconstructed super-resolution vascular structural and blood flow parameter maps under high-concentration microbubble conditions, but also maintains a high-quality trajectory tracking ability within a shorter acquisition time and at a higher microbubble concentration, obtaining higher-quality structural and blood flow parameter maps. Moreover, this method has a wide range of applications and can be applied to the blood vessels of various different living biological tissues. Description of the Drawings
[0045] To more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the accompanying drawings required for the description of the embodiments. Obviously, the accompanying drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can also be obtained based on these drawings.
[0046] Figure 1 It is the first flowchart of the super-resolution ultrasonic imaging method of the present application;
[0047] Figure 2 It is the second flowchart of the super-resolution ultrasonic imaging method of the present application;
[0048] Figure 3 It is the third flowchart of the super-resolution ultrasonic imaging method of the present application;
[0049] Figure 4 It is the fourth flowchart of the super-resolution ultrasonic imaging method of the present application;
[0050] Figure 5 It is the fifth flowchart of the super-resolution ultrasonic imaging method of the present application;
[0051] Figure 6 It is the overall flowchart of the super-resolution ultrasonic imaging method of the present application;
[0052] Figure 7 It is the module schematic diagram of the super-resolution ultrasonic imaging system of the present application. Detailed implementation manners
[0053] In the following, various embodiments of the present application will be described more comprehensively. The present application can have various embodiments, and adjustments and changes can be made therein. However, it should be understood that there is no intention to limit the various embodiments of the present application to the specific embodiments disclosed herein, but the present application should be understood to cover all adjustments, equivalents, and / or alternative solutions that fall within the spirit and scope of the various embodiments of the present application.
[0054] In the following, the term "comprising" or "may comprise" that can be used in various embodiments of the present application indicates the presence of the disclosed functions, operations, or elements, and does not limit the addition of one or more functions, operations, or elements. In addition, as used in various embodiments of the present application, the terms "comprising", "having" and their cognates are only intended to indicate the presence of specific features, numbers, steps, operations, elements, components, or combinations of the foregoing items, and should not be understood as precluding the presence or adding the possibility of one or more other features, numbers, steps, operations, elements, components, or combinations of the foregoing items.
[0055] Embodiment 1
[0056] As shown Figure 1 in the figure, a super-resolution ultrasound imaging method is applied to the vascular blood flow structure imaging of animals, humans, and other living biological tissues, including:
[0057] Obtaining data containing ultrasound contrast agent signals of the blood vessels of living biological tissues;
[0058] Using machine learning methods to process the data containing ultrasound contrast agent signals to obtain the structure and blood flow parameter maps of the blood vessels of living biological tissues.
[0059] This method not only solves the problem of distortion in reconstructing the structure and blood flow parameter maps of super-resolution blood vessels under high-concentration microbubble conditions, but also maintains a high-quality trajectory tracking ability within a shorter acquisition time and at a higher microbubble concentration, obtaining higher-quality structure and blood flow parameter maps. Moreover, this method has a wide range of applications and can be applied to the blood vessels of various different living biological tissues.
[0060] Using machine learning methods to process the data containing ultrasound contrast agent signals to obtain the structure and blood flow parameter maps of the blood vessels of living biological tissues includes using a multi-scale convolutional spatio-temporal neural network composed of an encoder module with multi-scale convolutions, a recurrent neural network module, and a decoder module to process the data containing ultrasound contrast agent signals to obtain the structure and blood flow parameter maps of the blood vessels of living biological tissues.
[0061] Among them, the encoder module with multi-scale convolutions effectively captures the local (such as the position of microbubble ultrasound contrast agents) and global (such as the morphology of blood vessel branches) spatial information of the blood vessel structure through hierarchical convolution and feature fusion, avoiding the loss of details or the neglect of global structures that may be caused by single-scale feature extraction, and improving the spatial resolution and anatomical accuracy of the structure and blood flow parameter maps.
[0062] Using a recurrent neural network module to process high-dimensional feature sequences can capture the temporal dynamic changes of microbubble ultrasound contrast agent signals (such as flow velocity fluctuations, aggregation and dispersion of microbubble ultrasound contrast agents), significantly improving the calculation accuracy of blood vessel hemodynamic parameters (such as shear stress, blood flow resistance), and providing richer dynamic information for the study of blood vessel physiology and pathology.
[0063] The decoder module gradually restores the low-resolution feature map to the original image size through upsampling operations, while retaining the key information in the high-dimensional features, avoiding edge blurring and artifacts caused by direct interpolation, and finally generating high-resolution structure and blood flow parameter maps that match the original blood vessel structure, providing a more intuitive tool for vascular function evaluation in clinical diagnosis and treatment.
[0064] It can be seen that the present application uses a multi-scale convolutional spatio-temporal neural network composed of an encoder module with multi-scale convolutions, a recurrent neural network module, and a decoder module to solve the problems of reconstructing super-resolution vascular structure diagrams and distorted structural and blood flow parameter diagrams under high-concentration microbubble conditions, so as to maintain a high-quality trajectory tracking ability within a shorter acquisition time and at a higher concentration of microbubble ultrasound contrast agents, ensuring that the processed structural and blood flow parameter diagrams have high-quality structural and velocity information. Moreover, the super-resolution ultrasound imaging method of the present application is applied to the vascular blood flow structure imaging of living biological tissues such as animals and humans, such as the chorioallantoic membrane of chicken embryos, mouse cerebral blood vessels, human cerebral blood vessels, etc., and can be applicable to more complex vascular dynamics environments, making ultrasound super-resolution vascular imaging more valuable in application.
[0065] Among them, the blood flow parameter diagram contains parameters such as blood flow diameter, flow velocity, and flow direction.
[0066] As Figure 2 shown, the encoder module with multi-scale convolutions includes a max-pooling layer and a first convolutional layer, and the first convolutional layer includes multiple parallel convolutional kernels;
[0067] The steps for the encoder module with multi-scale convolutions to process the data containing ultrasound contrast agent signals are as follows:
[0068] Input the data containing ultrasound contrast agent signals into the max-pooling layer to obtain the pooled features of the data containing ultrasound contrast agent signals;
[0069] Input the pooled features into the first convolutional layer, and extract multiple-scale features from different scales through multiple convolutional kernels;
[0070] Fuse the extracted multiple-scale features to obtain high-dimensional features.
[0071] The present application processes the data containing ultrasound contrast agent signals through the max-pooling layer to obtain pooled features, thereby reducing the spatial resolution of the data, and combining the first convolutional layer to perform multi-scale feature extraction on the pooled features to obtain multiple-scale features, enabling the network to simultaneously extract local and global information, thereby enhancing the perception ability of different-scale blood vessels and microbubble motion patterns; finally, fuse the multiple-scale features to obtain high-dimensional features. Compared with the original data, the high-dimensional features are more representative and discriminative, and also reduce the computational amount, so as to be more efficiently processed by the subsequent recurrent neural network module and decoder module.
[0072] Among them, at least two of the multiple convolutional kernels are used to extract features of different scales; and / or, the multiple convolutional kernels include at least two combinations of 3×3 convolutional kernels, 5×5 convolutional kernels, and 7×7 convolutional kernels. The present application provides multiple processing paths through multiple parallel convolutional kernels, and at least two convolutional kernels are used to extract features of different scales, which can be at least two combinations of 3×3 convolutional kernels, 5×5 convolutional kernels, and 7×7 convolutional kernels, or combinations of other convolutional kernels of different scales. This design can effectively integrate local and global information, further enhance the ability to analyze vascular structures and flow velocity information of different scales, and improve the imaging quality and the accuracy of flow velocity estimation.
[0073] In one embodiment, as Figure 6 shown, the first convolutional layer includes four convolutional kernels, namely Block k3 (3×3 convolutional kernel), Block k3 (3×3 convolutional kernel), Block k5 (5×5 convolutional kernel), and Block k7 (7×7 convolutional kernel). Among them, the 3×3 convolutional kernel is suitable for extracting local detail information, which helps to enhance the network's ability to distinguish smaller microbubbles; the 5×5 convolutional kernel is used to capture medium-scale microbubble flow information; the 7×7 convolutional kernel can perceive the spatial morphology of larger-scale microbubbles, which helps to completely restore the vascular network structure. Since microbubbles of different scales have different effects on ultrasonic imaging, this design of multiple convolutional kernels of different scales can effectively integrate local and global information and improve the imaging quality and the accuracy of flow velocity estimation.
[0074] As Figure 3 shown, obtaining data containing ultrasonic contrast agent signals of blood vessels in a living biological tissue includes obtaining simulation data of blood vessels in a living biological tissue, where the simulation data containing ultrasonic contrast agent signals is used to be input into a multi-scale convolutional spatio-temporal neural network for training;
[0075] The process of obtaining simulation data of blood vessels in a living biological tissue is as follows:
[0076] Based on the traditional super-resolution vascular blood flow map of blood vessels in a living biological tissue, process it to obtain the vascular structure map and blood flow velocity information of blood vessels in a living biological tissue;
[0077] Put microbubble ultrasonic contrast agent in the vascular structure map and initialize the microbubble amplitude and microbubble position to obtain microbubble trajectories according to the blood flow velocity information;
[0078] Calculate the average velocity of each pixel point according to the microbubble trajectories and time to obtain a microbubble trajectory velocity map;
[0079] Send the microbubble position and the microbubble amplitude of each frame into the simulation model, obtain the corresponding ultrasonic RF signal and perform beamforming to obtain ultrasonic frame sequence data containing ultrasonic contrast agent signals;
[0080] The ultrasonic frame sequence data containing ultrasonic contrast agent signals and the microbubble trajectory velocity map are used as simulation data.
[0081] The simulation data containing ultrasonic contrast agent signals obtained by the above method is closer to real data. This data records the flow of microbubble ultrasonic contrast agents in blood vessels and contains strong tissue signals and blood flow signals simultaneously.
[0082] In one embodiment, the simulation model is the Field-II simulation model. According to actual needs, the simulation model of the present application can adjust the number of frames and control the length of the microbubble ultrasonic contrast agent trajectory formed during the simulation stage to obtain a preset number of multi-frame images.
[0083] Obtaining data containing ultrasonic contrast agent signals of the blood vessels of a living biological tissue includes obtaining in-vivo data of the blood vessels of the living biological tissue. Among them, the in-vivo data containing ultrasonic contrast agent signals is used as input to a multi-scale convolutional spatio-temporal neural network for prediction;
[0084] The process of obtaining the in-vivo data of the blood vessels of a living biological tissue is as follows:
[0085] Inject microbubble ultrasonic contrast agents into the blood vessels of the living biological tissue, use an ultrasonic probe to obtain data containing ultrasonic contrast agent signals, and perform processing using SVD to filter out tissue clutter to obtain the in-vivo data of the blood vessels of the living biological tissue.
[0086] In order to better extract the in-vivo data containing ultrasonic contrast agent signals in the present application, an ultrasonic probe is used to obtain data containing ultrasonic contrast agent signals, and processing is performed using SVD to filter out tissue clutter to ensure that only the microbubble ultrasonic contrast agent information (the position and amplitude of microbubbles) of each frame of image is retained, so as to facilitate better prediction.
[0087] As Figure 4 shown, the data containing ultrasonic contrast agent signals includes simulation data containing ultrasonic contrast agent signals and in-vivo data containing ultrasonic contrast agent signals;
[0088] Use a multi-scale convolutional spatio-temporal neural network composed of an encoder module with multi-scale convolution, a recurrent neural network module, and a decoder module to train and test the data containing ultrasonic contrast agent signals to obtain the structure and blood flow parameter map of the blood vessels of a living biological tissue, including:
[0089] Use a multi-scale convolutional spatio-temporal neural network composed of an encoder module with multi-scale convolution, a recurrent neural network module, and a decoder module to train the simulation data containing ultrasonic contrast agent signals to obtain weights;
[0090] Optimize the neural network model based on the weights and the multi-scale convolutional spatio-temporal neural network;
[0091] Input the in-vivo data containing ultrasound contrast agent signals into an optimized neural network model for testing to obtain the structural and blood flow parameter maps of the blood vessels in living biological tissues.
[0092] As Figure 5 shown, fuse the extracted multi-scale features to obtain high-dimensional features including:
[0093] Stitch and fuse the extracted multi-scale features to obtain the stitched multi-scale features;
[0094] Input the stitched multi-scale features into the second convolutional layer of the encoder module with multi-scale convolutions for more advanced feature extraction;
[0095] Input the multi-scale features after feature extraction into a batch normalization layer for batch normalization and perform non-linear transformation through an activation function layer to obtain high-dimensional features.
[0096] In this application, first stitch and fuse the extracted multi-scale features to obtain the stitched multi-scale features to retain the complete multi-scale information. This fusion strategy enables the network to more comprehensively understand the blood vessel structure and avoid information loss caused by single-scale features. Further, perform feature fusion and processing through the first convolutional layer, batch normalization layer, and activation function layer to enhance the non-linear representation ability of the network, improve the robustness of the model, and accelerate the training process.
[0097] Among them, the recurrent neural network module includes a long short-term memory network module or a bidirectional recurrent neural network module. Processing the high-dimensional feature sequence through the long short-term memory network module or the bidirectional recurrent neural network module can capture the temporal dynamic changes of the microbubble ultrasound contrast agent signals (such as flow velocity fluctuations, aggregation and dispersion of microbubble ultrasound contrast agents), significantly improving the calculation accuracy of blood vessel hemodynamic parameters (such as shear stress, blood flow resistance) and providing richer dynamic information for vascular physiology and pathology research.
[0098] Embodiment 2
[0099] As Figure 6 shown, based on the super-resolution ultrasound imaging method in Embodiment 1, this application also proposes a super-resolution ultrasound imaging system, which is applied to the blood vessel structures of different animals and the blood vessel blood flow structure imaging of animals, humans, and other living biological tissues, including:
[0100] An acquisition module for acquiring data containing ultrasound contrast agent signals of the blood vessels in living biological tissues;
[0101] A processing module for processing the data containing ultrasound contrast agent signals using machine learning methods to obtain the structural and blood flow parameter maps of the blood vessels in living biological tissues.
[0102] Through the cooperation between the acquisition module and the processing module, this application solves the problem of distortion in reconstructing the super-resolution vascular structure diagram and blood flow parameter diagram under high-concentration microbubble conditions, so as to maintain a high-quality trajectory tracking ability within a shorter acquisition time and at a higher microbubble concentration, ensuring that the processed structure and blood flow parameter diagram have high-quality structure and velocity information. Moreover, the super-resolution ultrasound imaging system of this application is not limited to the chorioallantoic membrane of chicken embryos, mouse cerebral blood vessels, or human cerebral blood vessels. This system can also be applied to other more complex vascular dynamics environments, making ultrasound super-resolution vascular imaging more valuable in applications.
[0103] Among them, using machine learning methods to process data containing ultrasound contrast agent signals to obtain the structure and blood flow parameter diagram of blood vessels in living biological tissues includes using a multi-scale convolutional spatio-temporal neural network composed of an encoder module with multi-scale convolutions, a recurrent neural network module, and a decoder module to process data containing ultrasound contrast agent signals to obtain the structure and blood flow parameter diagram of blood vessels in living biological tissues.
[0104] The encoder module with multi-scale convolutions includes a max-pooling layer and a first convolutional layer, and the first convolutional layer includes multiple parallel convolutional kernels;
[0105] The processing module includes:
[0106] A pooling module for inputting data containing ultrasound contrast agent signals into the max-pooling layer to obtain the pooling features of the data containing ultrasound contrast agent signals;
[0107] A multi-scale convolution module for inputting the pooling features into the first convolutional layer to extract multiple-scale features from different scales through multiple convolutional kernels;
[0108] A fusion module for fusing the extracted multiple-scale features to obtain high-dimensional features.
[0109] Through the cooperation between the pooling module, the multi-scale convolution module, and the fusion module, this application can simultaneously extract local and global information, thereby enhancing the perception ability of blood vessels and microbubble motion patterns at different scales; and fusing the multiple-scale features to obtain high-dimensional features. Compared with the original data, this high-dimensional feature is more representative and discriminative, and also reduces the computational amount, facilitating more efficient processing in the subsequent process.
[0110] In various embodiments of this application, the expression "or" or "at least one of A or / and B" includes any combination or all combinations of the words listed simultaneously. For example, the expression "A or B" or "at least one of A or / and B" may include A, may include B, or may include both A and B.
[0111] Expressions (such as "first", "second", etc.) used in various embodiments of the present application may modify various components in the various embodiments, but do not limit the corresponding components. For example, the above expressions do not limit the order and / or importance of the components. The above expressions are only for the purpose of distinguishing one component from other components. For example, the first user device and the second user device indicate different user devices, although both are user devices. For example, without departing from the scope of the various embodiments of the present application, the first component may be referred to as the second component, and similarly, the second component may also be referred to as the first component.
[0112] It should be noted that in the present application, unless otherwise clearly specified and defined, terms such as "installed", "connected", "fixed", etc. should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium; it can be the communication inside two components. For those of ordinary skill in the art, the specific meanings of the above terms in the present application can be understood according to specific circumstances.
[0113] In the present application, those of ordinary skill in the art need to understand that the terms indicating orientation or positional relationship in the text are based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present application and simplifying the description, rather than indicating or implying that the device or component referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation to the present application.
[0114] The terms used in the various embodiments of the present application are only for the purpose of describing specific embodiments and are not intended to limit the various embodiments of the present application. As used herein, the singular form is also intended to include the plural form unless the context clearly indicates otherwise. Unless otherwise defined, all terms (including technical terms and scientific terms) used herein have the same meaning as commonly understood by those of ordinary skill in the art to which the various embodiments of the present application belong. The terms (such as those defined in a commonly used dictionary) will be interpreted as having the same meaning as the contextual meaning in the relevant technical field and will not be interpreted as having an idealized meaning or an overly formal meaning unless clearly defined in the various embodiments of the present application.
Claims
1. A super-resolution ultrasound imaging method, which is applied to the vascular blood flow structure imaging of animals, humans and other living biological tissues, and is characterized in that Comprising: Obtaining data containing ultrasonic contrast agent signals of the blood vessels of a living biological tissue; Processing the data containing ultrasonic contrast agent signals by using a machine learning method to obtain a structural and blood flow parameter map of the blood vessels of a living biological tissue.
2. The super-resolution ultrasonic imaging method according to claim 1, characterized in that The processing of the data containing ultrasonic contrast agent signals by using a machine learning method to obtain a structural and blood flow parameter map of the blood vessels of a living biological tissue includes: Processing the data containing ultrasonic contrast agent signals by using a multi-scale convolutional spatio-temporal neural network composed of an encoder module with multi-scale convolution, a recurrent neural network module, and a decoder module to obtain a structural and blood flow parameter map of the blood vessels of a living biological tissue.
3. The super-resolution ultrasound imaging method according to claim 2, wherein The encoder module with multi-scale convolution includes a max pooling layer and a first convolutional layer, and the first convolutional layer includes a plurality of parallel convolutional kernels; The steps for the encoder module with multi-scale convolution to process the data containing ultrasonic contrast agent signals are: Inputting the data containing ultrasonic contrast agent signals into the max pooling layer to obtain the pooling features of the data containing ultrasonic contrast agent signals; Inputting the pooling features into the first convolutional layer, and extracting multi-scale features from different scales through the plurality of convolutional kernels; Performing fusion processing on the extracted multi-scale features to obtain high-dimensional features.
4. The super-resolution ultrasound imaging method according to claim 2, characterized in that The obtaining of the data containing ultrasonic contrast agent signals of the blood vessels of a living biological tissue includes obtaining simulation data of the blood vessels of a living biological tissue, wherein the simulation data of the blood vessels of a living biological tissue is used for input into the multi-scale convolutional spatio-temporal neural network for training; The obtaining process of the simulation data of the blood vessels of a living biological tissue is: Processing based on a traditional super-resolution blood flow map of the blood vessels of a living biological tissue to obtain a blood vessel structure map and blood flow velocity information of the blood vessels of a living biological tissue; Placing microbubble ultrasonic contrast agent in the blood vessel structure map and initializing the microbubble amplitude and microbubble position to obtain microbubble trajectories according to the blood flow velocity information; Calculating the average velocity of each pixel point according to the microbubble trajectories and time to obtain a microbubble trajectory velocity map; Feeding the microbubble position and the microbubble amplitude of each frame into a simulation model, obtaining corresponding ultrasonic RF signals and performing beam synthesis to obtain ultrasonic frame sequence data containing ultrasonic contrast agent signals; Taking the ultrasonic frame sequence data containing ultrasonic contrast agent signals and the microbubble trajectory velocity map as simulation data.
5. The super-resolution ultrasound imaging method according to claim 2, wherein The obtaining of the data containing ultrasonic contrast agent signals of the blood vessels of a living biological tissue includes obtaining in-vivo data of the blood vessels of a living biological tissue, wherein the in-vivo data of the blood vessels of a living biological tissue is used for input into the multi-scale convolutional spatio-temporal neural network for prediction; The obtaining process of the in-vivo data of the blood vessels of a living biological tissue is: Injecting microbubble ultrasonic contrast agent into the blood vessels of a living biological tissue, using an ultrasonic probe to obtain data containing ultrasonic contrast agent signals, and performing processing by using SVD to filter out tissue clutter to obtain the in-vivo data of the blood vessels of a living biological tissue.
6. The super-resolution ultrasonic imaging method according to claim 1, wherein The data containing ultrasonic contrast agent signals includes simulation data containing ultrasonic contrast agent signals and in-vivo data containing ultrasonic contrast agent signals; Training and testing the data containing ultrasound contrast agent signals using a multi-scale convolutional spatio-temporal neural network composed of an encoder module with multi-scale convolution, a recurrent neural network module, and a decoder module to obtain the structure and blood flow parameter maps of blood vessels in living biological tissues includes: Training the simulation data containing ultrasound contrast agent signals using a multi-scale convolutional spatio-temporal neural network composed of an encoder module with multi-scale convolution, a recurrent neural network module, and a decoder module to obtain weights; Optimizing the neural network model based on the weights and the multi-scale convolutional spatio-temporal neural network; Inputting the in-vivo data containing ultrasound contrast agent signals into the optimized neural network model for testing to obtain the structure and blood flow parameter maps of blood vessels in living biological tissues.
7. The super-resolution ultrasound imaging method according to claim 3, wherein At least two of the multiple convolutional kernels are used to extract features of different scales; and / or, the multiple convolutional kernels include at least two combinations of 3×3 convolutional kernels, 5×5 convolutional kernels, and 7×7 convolutional kernels.
8. The super-resolution ultrasonic imaging method according to claim 3, wherein, The fusing process of the multiple extracted scale features to obtain high-dimensional features includes: Concatenating and fusing the multiple extracted scale features to obtain the concatenated multi-scale features; Inputting the concatenated multi-scale features into the second convolutional layer of the encoder module with multi-scale convolution for more advanced feature extraction; Inputting the multi-scale features after feature extraction into a batch normalization layer for batch normalization and performing non-linear transformation through an activation function layer to obtain high-dimensional features.
9. The super-resolution ultrasonic imaging method according to claim 2, wherein The recurrent neural network module includes a long short-term memory network module or a bidirectional recurrent neural network module.
10. A super-resolution ultrasound imaging system is applied to the vascular blood flow structure imaging of animals, humans, and other living biological tissues, and is characterized in that, Including: An acquisition module for acquiring data containing ultrasound contrast agent signals of blood vessels in living biological tissues; A processing module for processing the data containing ultrasound contrast agent signals using machine learning methods to obtain the structure and blood flow parameter maps of blood vessels in living biological tissues.
