Single-array-element super-resolution ultrasonic imaging system based on deep learning

Abstract

The invention provides a single-array-element super-resolution ultrasonic imaging system based on deep learning, and the system is characterized in that the system comprises an ultrasonic collection module which comprises a single-array-element ultrasonic probe and is used for collecting a plurality of microbubble echo signals of an imaging target tissue at a single imaging angle; the system measurement matrix storage module is used for storing a corresponding system measurement matrix of the single array element ultrasonic probe at a single imaging angle; the microbubble reconstruction module is used for performing image reconstruction on each microbubble echo signal through a Moore-Penrose pseudo-inverse method according to the system measurement matrix to obtain a corresponding reconstructed microbubble image; the microbubble positioning module comprises a trained microbubble positioning network and is used for obtaining a corresponding microbubble positioning image according to the reconstructed microbubble image; and the image reconstruction module is used for carrying out superposition processing on all the microbubble positioning images to obtain a super-resolution ultrasonic microscopic image. In conclusion, the method can realize rapid imaging based on simple imaging equipment and has high imaging quality at the same time.

Description

Technical Field

[0001] The present invention relates to ultrasonic imaging technology, and in particular to a single-element super-resolution ultrasonic imaging system based on deep learning. Background Art

[0002] Ultrasound imaging technology is widely used in clinical diagnosis and treatment due to its advantages such as speed, safety, and low cost. However, due to the diffraction limit of ultrasound, its resolution has been limited to half the wavelength, making it impossible to image fine vascular structures within tissues. To overcome this spatial resolution limitation, ultrasound localization microscopy (ULM) was proposed. This method utilizes ultrasound contrast-enhanced microbubbles and a single-molecule localization strategy to achieve sub-diffraction-limited imaging, enabling ultrasound imaging to observe fine tissue structures.

[0003] Currently, three-dimensional ultrasound has been clinically proven to play an important role in the assessment of most medical problems. However, current ultrasound image acquisition methods generally achieve three-dimensional space scanning by translating and rotating linear array transducers. Although the imaging speed can be accelerated to a certain extent by optimizing the reconstruction algorithm, its data acquisition speed cannot be further improved due to the limitations of the imaging scanning method. By using area array transducers, the data acquisition speed of three-dimensional ultrasound imaging can be accelerated. However, according to the Nyquist theorem, the spatial resolution and imaging speed of ultrasound images are highly dependent on the area array ultrasound transducer, high data transmission rate, and the performance of integrated electronic devices. Therefore, the imaging cost of three-dimensional ultrasound imaging methods based on area arrays is high and the computational burden is heavy, which limits their widespread application in the biomedical field.

[0004] In addition, some studies have attempted to achieve compressed three-dimensional ultrasound imaging with the help of metaacoustic materials to assist single ultrasonic array elements. Specifically, this method uses metaacoustic materials to manufacture an acoustic phase plate to encode the single array element sound field, and uses the encoded field to image the target to be measured; by rotating the acoustic phase plate multiple times, the target is imaged with different encoded fields, thereby improving the image reconstruction quality. However, most existing single-element technologies rely on sufficient raw data, that is, raw data at multiple rotation angles. In order to collect sufficient data at various angles, the image acquisition time of this method is relatively long, which limits its three-dimensional imaging speed. Due to this limitation, although this method has good imaging capabilities for static three-dimensional targets, its imaging quality for flowing targets such as microbubbles is highly limited.

[0005] In summary, the existing ultrasound positioning microscopy technology still has problems such as slow imaging speed, low imaging quality, and high imaging hardware cost. Summary of the Invention

[0006] The present invention is made to solve the above problems, and its purpose is to provide a single-element super-resolution ultrasonic imaging system based on deep learning.

[0007] The present invention provides a single-element super-resolution ultrasound imaging system based on deep learning, which is used to obtain a super-resolution ultrasound microscopic image of an imaging target tissue. The system comprises: an ultrasound acquisition module, which includes a single-element ultrasound probe, and is used to acquire multiple microbubble echo signals of the imaging target tissue at a single imaging angle; a system measurement matrix storage module, which is used to store the system measurement matrix corresponding to the single-element ultrasound probe at a single imaging angle; a microbubble reconstruction module, which is used to reconstruct the image of each microbubble echo signal by the Moore-Penrose pseudo-inverse method according to the system measurement matrix, and obtain the corresponding reconstructed image. A microbubble image is constructed; a microbubble positioning module includes a trained microbubble positioning network, which is used to obtain a corresponding microbubble positioning image based on the reconstructed microbubble image; an image reconstruction module is used to superimpose all microbubble positioning images to obtain a super-resolution ultrasonic microscopy image, wherein the single-element ultrasonic probe includes a single-element ultrasonic transducer and a sound field phase encoding plate made of an acoustic conductive material, the sound field phase encoding plate has the same size as the single-element ultrasonic transducer and is arranged in front of the single-element ultrasonic transducer, and the surface of the sound field phase encoding plate is composed of a highly random grid, which causes the ultrasonic sound wave to be locally delayed through the grid to encode the ultrasonic sound field.

[0008] The single-element super-resolution ultrasonic imaging system based on deep learning provided by the present invention may also have the following features: wherein, the process of acquiring the system measurement matrix includes the following steps: step S1, according to the three-dimensional translation stage and the hydrophone, collecting and obtaining a plane sound field distribution parallel to the sound field phase encoding plate in the imaging area; step S2, performing a two-dimensional Fourier transform on the plane sound field distribution to obtain the corresponding plane angular spectrum; step S3, according to the angular spectrum method and the plane angular spectrum, calculating the sound field angular spectrum of each plane in the imaging area; step S4, performing a two-dimensional inverse Fourier transform on each sound field angular spectrum to obtain the corresponding sound field distribution; step S5, according to all sound field distributions, calculating the system measurement matrix by the reciprocity principle.

[0009] The single-element super-resolution ultrasound imaging system based on deep learning provided by the present invention may also have the following features: wherein, in step S1, the expression of the plane sound field distribution is: Where p(x,y,z0) is the plane sound field distribution, is the sound field amplitude at the coordinate (x, y, z0), and φ(x, y, z0) is the sound field phase at the coordinate (x, y, z0).

[0010] The single-element super-resolution ultrasound imaging system based on deep learning provided by the present invention may also have the following features: wherein, in step S3, the expression of the sound field angular spectrum is: k=(k x ,k y ,k z ), Where P(k x ,k y ,z) is the angular spectrum of the sound field at the coordinate (x, y, z), P(k x ,k y ,z0) is the angular spectrum of the sound field at the coordinate (x,y,z0), |k| is the wave number, c is the wave velocity, and ω is the angular frequency.

[0011] The single-element super-resolution ultrasound imaging system based on deep learning provided by the present invention may also have the following features: wherein the calculation expression for reconstructing the microbubble image according to the Moore-Penrose pseudo-inverse method is: In the formula To reconstruct the true position of microbubbles in the microbubble image, H is the system measurement matrix, + is the Moore-Penrose pseudo-inverse matrix calculated by SVD decomposition, and u is the microbubble echo signal corresponding to the reconstructed microbubble image.

[0012] The single-element super-resolution ultrasound imaging system based on deep learning provided by the present invention may also have the following features: wherein, the training process of the microbubble localization network includes the following steps: step T1, randomly placing multiple points in the imaging area as simulated microbubbles, and changing the positions of the simulated microbubbles frame by frame; step T2, imaging the imaging area through a single-element ultrasound probe to obtain multiple simulated microbubble echo signals; step T3, performing image reconstruction through the Moore-Penrose pseudo-inverse method based on the simulated microbubble echo signals and the system measurement matrix to obtain corresponding reconstructed simulated microbubble images; step T4, taking all reconstructed simulated microbubble images and the corresponding simulated microbubble real positioning points as training data sets; step T5, constructing a deep convolutional neural network; step T6, training the deep convolutional neural network based on the training data set, combined with the Adam optimized gradient descent algorithm and the learning rate decay strategy, to obtain a trained deep convolutional neural network as the microbubble localization network.

[0013] The single-element super-resolution ultrasound imaging system based on deep learning provided by the present invention may also have the following features: wherein, in step T6, the deep convolutional neural network is updated by a loss function, and the expression of the loss function is: In the formula is the calculation result of the loss function, N is the number of batch images traversed by the deep convolutional neural network each time, y is the real positioning point of the simulated microbubble, This is the positioning result of the deep convolutional neural network for the true positioning point y of the simulated microbubble.

[0014] The single-element super-resolution ultrasound imaging system based on deep learning provided by the present invention may also have the following features: the deep convolutional neural network includes 6 residual modules, each residual module includes 2 convolutional layers and their corresponding batch normalization layers and activation functions for residual learning.

[0015] Functions and effects of the invention

[0016] According to the single-element super-resolution ultrasound imaging system based on deep learning involved in the present invention, first, the ultrasound acquisition module 10 acquires microbubble echo signals at a single imaging angle, and then generates corresponding super-resolution ultrasound microscopic images based on the microbubble echo signals, shortening the time for three-dimensional data acquisition and processing and significantly improving the ultrasound imaging speed; second, the ultrasound signal is acquired by the single-element ultrasound transducer and the sound field phase encoding plate, which significantly reduces the dependence of ultrasound imaging on the accuracy of imaging hardware and reduces the imaging cost; third, the microbubble localization network obtained by training with simulation data can generate microbubble localization images with high accuracy, thereby improving the imaging quality of super-resolution ultrasound microscopic images. Therefore, the single-element super-resolution ultrasound imaging system based on deep learning of the present invention can achieve fast imaging with high imaging quality based on simple imaging equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 is a block diagram of a single-element super-resolution ultrasound imaging system in an embodiment of the present invention;

[0018] Figure 2 Schematic diagram of the structure and working principle of a single-element ultrasonic probe in an embodiment of the present invention;

[0019] Figure 3 is a schematic diagram of a process for obtaining a system measurement matrix in an embodiment of the present invention;

[0020] Figure 4 1 is a schematic diagram of a process for training a microbubble localization network in an embodiment of the present invention;

[0021] Figure 5 Schematic diagram of the principle of training a deep convolutional neural network in an embodiment of the present invention;

[0022] Figure 6 is a schematic diagram of reconstructed microbubble images, true microbubble positioning, and microbubble positioning images in an embodiment of the present invention;

[0023] Figure 7 FIG2 is a flow chart of ultrasonic imaging performed using a single-element super-resolution ultrasonic imaging system in an embodiment of the present invention. DETAILED DESCRIPTION

[0024] In order to make the technical means, creative features, objectives and effects achieved by the present invention easy to understand, the following embodiments and the accompanying drawings specifically illustrate the single-element super-resolution ultrasound imaging system based on deep learning of the present invention.

[0025] This embodiment provides a single-element super-resolution ultrasound imaging system based on deep learning, which is used to obtain super-resolution ultrasound microscopic images of imaging target tissues.

[0026] Figure 1 4 is a block diagram of a single-element super-resolution ultrasound imaging system in an embodiment of the present invention.

[0027] like Figure 1 As shown, the single-element super-resolution ultrasound imaging system 100 includes an ultrasound acquisition module 10 , a system measurement matrix storage module 20 , a microbubble reconstruction module 30 , a microbubble positioning module 40 and an image reconstruction module 50 .

[0028] The ultrasound acquisition module 10 includes a single-element ultrasound probe, which is used to acquire multiple microbubble echo signals of the imaging target tissue at a single imaging angle.

[0029] Figure 2 Schematic diagram of the structure and working principle of a single-element ultrasonic probe in an embodiment of the present invention.

[0030] like Figure 2 As shown, the single-element ultrasonic probe includes a single-element ultrasonic transducer and an acoustic field phase encoding plate made of acoustic conductive material, also known as the acoustic field encoding plate. The acoustic field phase encoding plate is the same size as the single-element ultrasonic transducer and is positioned in front of the single-element ultrasonic transducer. The surface of the acoustic field phase encoding plate consists of a highly random grid. In this embodiment, the acoustic field phase encoding plate is manufactured using 3D printing technology.

[0031] The working principle of a single-element ultrasound probe is as follows: a single-element ultrasound transducer is used to transmit ultrasound waves to the imaging area to construct a transmission field, and receive the echo field corresponding to the ultrasound contrast microbubbles generated by the ultrasound contrast agent. During this process, the grid causes the ultrasound sound waves to be locally delayed, thereby encoding the ultrasound sound field and obtaining the microbubble echo signal.

[0032] The system measurement matrix storage module 20 is used to store the system measurement matrix corresponding to the single-element ultrasound probe at a single imaging angle.

[0033] Figure 3 4 is a flow chart of obtaining a system measurement matrix in an embodiment of the present invention.

[0034] like Figure 3 As shown in Figure 2, the process of obtaining the system measurement matrix includes the following steps:

[0035] Step S1: using a three-dimensional translation stage and a hydrophone, a plane sound field distribution parallel to a sound field phase encoding plate in an imaging area is acquired.

[0036] In step S1, the expression of the plane sound field distribution is:

[0037]

[0038] Where p(x,y,z0) is the plane sound field distribution, is the sound field amplitude at the coordinate (x, y, z0), and φ(x, y, z0) is the sound field phase at the coordinate (x, y, z0).

[0039] Step S2: Perform a two-dimensional Fourier transform on the plane sound field distribution to obtain a corresponding plane angular spectrum.

[0040] Step S3: Calculate the sound field angular spectrum of each plane in the imaging area according to the angular spectrum method and the plane angular spectrum.

[0041] In step S3, the sound field angular spectrum of each plane in the imaging area is expressed as follows:

[0042]

[0043] k=(k x ,k y ,k z ),

[0044]

[0045] Where P(k x ,k y ,z) is the angular spectrum of the sound field at the coordinate (x, y, z), P(k x ,k y ,z0) is the angular spectrum of the sound field at the coordinate (x,y,z0), |k| is the wave number, c is the wave velocity, and ω is the angular frequency.

[0046] Step S4: performing a two-dimensional inverse Fourier transform on each sound field angular spectrum to obtain a corresponding sound field distribution.

[0047] Step S5: According to all the sound field distributions, a system measurement matrix is ​​calculated by the reciprocity principle.

[0048] The microbubble reconstruction module 30 is used to perform image reconstruction on each microbubble echo signal according to the system measurement matrix by using the Moore-Penrose pseudo-inverse method to obtain a corresponding reconstructed microbubble image.

[0049] In this embodiment, there is a linear relationship between the microbubble echo signal and the imaging target, which is expressed as follows:

[0050] u=Hv+n,

[0051] Where u is the microbubble echo signal, v is the true position of the ultrasound contrast-enhanced microbubble, n is the M-dimensional additive noise, which is equal to the number of time sampling points, H is the system measurement matrix with a size of M × N, and N represents the number of pixels in the imaging plane.

[0052] When the ultrasound contrast-enhanced microbubbles are sparsely distributed and the system measurement matrix H is known, the linear inversion algorithm can be used to recover the true position v of the ultrasound contrast-enhanced microbubbles from the microbubble echo signal u according to the above formula, and thus reconstruct the corresponding image.

[0053] Therefore, in this embodiment, the image reconstruction is achieved by using the Moore-Penrose pseudo-inverse method based on the received microbubble echo signal. Specifically, when the linear model is valid and n is additive white Gaussian noise, the expression for solving the problem is:

[0054]

[0055] The optimal solution to this problem, that is, the calculation expression for reconstructing the microbubble image according to the Moore-Penrose pseudo-inverse method, is:

[0056]

[0057] In the formula To reconstruct the true position of microbubbles in the microbubble image, H is the system measurement matrix, + is the Moore-Penrose pseudo-inverse matrix calculated by SVD decomposition, and u is the microbubble echo signal corresponding to the reconstructed microbubble image.

[0058] The microbubble localization module 40 includes a trained microbubble localization network, which is used to obtain a corresponding microbubble localization image based on the reconstructed microbubble image.

[0059] Figure 4 Schematic diagram of the process of training a microbubble localization network in an embodiment of the present invention.

[0060] like Figure 4 As shown in Figure 2, the training process of the microbubble localization network includes the following steps:

[0061] Step T1: randomly placing multiple points in the imaging area as simulated microbubbles, and changing the positions of the simulated microbubbles frame by frame.

[0062] Step T2: imaging the imaging area using a single-element ultrasonic probe to obtain multiple simulated microbubble echo signals.

[0063] Step T3: performing image reconstruction using the Moore-Penrose pseudo-inverse method based on the simulated microbubble echo signal and the system measurement matrix to obtain a corresponding reconstructed simulated microbubble image.

[0064] Step T4: All reconstructed simulated microbubble images and corresponding simulated microbubble real positioning points are used as a training data set. In this embodiment, the training data set includes 4000 pairs of reconstructed simulated microbubble images and corresponding simulated microbubble real positioning points.

[0065] Step T5: construct a deep convolutional neural network. In this embodiment, the deep convolutional neural network includes 6 residual modules connected in sequence.

[0066] Each residual module is composed of a first convolutional layer, a first normalization layer, a ReLU activation function, a second convolutional layer, a second normalization layer, and a feature fusion layer, connected in sequence for residual learning. The feature fusion layer is used to implement skip connections, fusing the features input to the residual module with the features output by the second normalization layer to obtain the features output by the residual module.

[0067] Step T6: Based on the training data set, the deep convolutional neural network is trained in combination with the Adam optimized gradient descent algorithm and the learning rate decay strategy to obtain a trained deep convolutional neural network as a microbubble localization network. In this embodiment, the deep convolutional neural network is trained for 100 cycles, i.e., the training data set is traversed 100 times, wherein the data batch size N for each cycle is 8, the initial learning rate is 0.01, and the learning rate is gradually reduced to 0, gradually approaching the optimal parameters.

[0068] Figure 5 Schematic diagram of the principle of training a deep convolutional neural network in an embodiment of the present invention.

[0069] like Figure 5 As shown, the reconstructed simulated microbubble image is input into the deep convolutional neural network to obtain the corresponding positioning result. The loss function constructed based on the MSE loss and the L1 regularization term is calculated according to the positioning result and the real positioning point of the simulated microbubble corresponding to the reconstructed simulated microbubble image. The deep convolutional neural network is updated by combining the loss function with stochastic gradient descent to train the microbubble positioning network.

[0070] Among them, the expression of the loss function is:

[0071]

[0072] In the formula is the calculation result of the loss function, N is the number of batch images traversed by the deep convolutional neural network each time, y is the real positioning point of the simulated microbubble, This is the positioning result of the deep convolutional neural network for the true positioning point y of the simulated microbubble.

[0073] Figure 6Schematic diagram of reconstructed microbubble image, real microbubble positioning and microbubble positioning image in an embodiment of the present invention.

[0074] like Figure 6 As shown in the figure, the four images in the first row are different reconstructed microbubble images, the four images in the second row are the actual microbubble locations corresponding to each reconstructed microbubble image, namely the white dots in the images, and the four images in the third row are the corresponding microbubble localization images obtained by the microbubble localization network based on each reconstructed microbubble image. The white dots in the microbubble localization images are the predicted microbubble locations. This shows that the microbubble locations identified by the microbubble localization network based on the reconstructed microbubble images are consistent with the actual microbubble locations, indicating that the microbubble localization network has a high localization accuracy.

[0075] The image reconstruction module 50 is used to perform superposition processing on all microbubble positioning images to obtain a super-resolution ultrasound microscopic image.

[0076] Figure 7 FIG2 is a flow chart of ultrasonic imaging performed using a single-element super-resolution ultrasonic imaging system in an embodiment of the present invention.

[0077] like Figure 7 As shown, performing ultrasound imaging using the single-element super-resolution ultrasound imaging system 100 includes the following steps:

[0078] In step R1 , the ultrasound acquisition module 10 is used to acquire multiple microbubble echo signals of the imaging target tissue at a single imaging angle.

[0079] In step R2, the microbubble reconstruction module 30 reconstructs each microbubble echo signal using the Moore-Penrose pseudo-inverse method according to the system measurement matrix stored in the system measurement matrix storage module 20 to obtain a corresponding reconstructed microbubble image.

[0080] In step R3, the microbubble positioning module 40 is used to obtain a corresponding microbubble positioning image based on the reconstructed microbubble image.

[0081] In step R4, the image reconstruction module 50 is used to perform superposition processing on all microbubble positioning images to obtain a super-resolution ultrasound microscopic image.

[0082] Functions and Effects of the Embodiments

[0083] According to the single-element super-resolution ultrasound imaging system based on deep learning involved in this embodiment, first, the ultrasound acquisition module 10 collects microbubble echo signals at a single imaging angle, and then generates corresponding super-resolution ultrasound microscopic images based on the microbubble echo signals, shortening the time for three-dimensional data acquisition and processing and significantly improving the speed of ultrasound imaging. Second, the ultrasound signal is collected by a single-element ultrasound transducer and an acoustic field phase encoding plate, which greatly reduces the dependence of ultrasound imaging on the accuracy of imaging hardware and reduces imaging costs. Third, the microbubble localization network obtained through simulation data training can generate microbubble localization images with high accuracy, thereby improving the imaging quality of super-resolution ultrasound microscopic images. In short, this method can achieve fast imaging with high imaging quality based on simple imaging equipment.

[0084] Those skilled in the art will appreciate that the present invention is not limited to the foregoing embodiments. The foregoing embodiments and descriptions are merely illustrative of the principles of the present invention. Various changes and modifications may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and modifications are intended to fall within the scope of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A single-element super-resolution ultrasound imaging system based on deep learning, used to obtain super-resolution ultrasound microscopic images of imaging target tissues, characterized in that: include: An ultrasound acquisition module, comprising a single-element ultrasound probe, for acquiring multiple microbubble echo signals of the imaging target tissue at a single imaging angle; A system measurement matrix storage module, configured to store the system measurement matrix corresponding to the single-element ultrasound probe at the single imaging angle; a microbubble reconstruction module, configured to reconstruct the image of each microbubble echo signal using the Moore-Penrose pseudo-inverse method according to the system measurement matrix to obtain a corresponding reconstructed microbubble image; a microbubble localization module, comprising a trained microbubble localization network, for obtaining a corresponding microbubble localization image based on the reconstructed microbubble image; An image reconstruction module is used to superimpose all the microbubble positioning images to obtain the super-resolution ultrasound microscopy image. The single-element ultrasonic probe includes a single-element ultrasonic transducer and a sound field phase encoding plate made of acoustic conductive material. The sound field phase encoding plate has the same size as the single-element ultrasonic transducer and is arranged in front of the single-element ultrasonic transducer. The surface of the acoustic field phase encoding plate is composed of highly random grids, which locally delay ultrasonic sound waves and encode the ultrasonic sound field.

2. The single-element super-resolution ultrasound imaging system based on deep learning according to claim 1, Its characteristics are: The process of obtaining the system measurement matrix includes the following steps: Step S1, using a three-dimensional translation stage and a hydrophone, collecting and obtaining a plane sound field distribution in an imaging area that is parallel to the sound field phase encoding plate; Step S2, performing a two-dimensional Fourier transform on the plane sound field distribution to obtain a corresponding plane angular spectrum; Step S3, calculating the sound field angular spectrum of each plane in the imaging area according to the angular spectrum method and the plane angular spectrum; Step S4, performing a two-dimensional inverse Fourier transform on each of the sound field angular spectra to obtain a corresponding sound field distribution; Step S5: Calculate the system measurement matrix based on the reciprocity principle according to all the sound field distributions.

3. The single-element super-resolution ultrasound imaging system based on deep learning according to claim 2, characterized in that: in, In step S1, the expression of the plane sound field distribution is: Where p(x,y,z0) is the plane sound field distribution, is the sound field amplitude at the coordinate (x, y, z0), and φ(x, y, z0) is the sound field phase at the coordinate (x, y, z0).

4. The single-element super-resolution ultrasound imaging system based on deep learning according to claim 2, characterized in that: in, In step S3, the expression of the sound field angular spectrum is: k=(k x ,k y ,k z ), Where P(k x , k y , z) is the angular spectrum of the sound field at the coordinate (x, y, z), P(k x , k y , z0) is the angular spectrum of the sound field at the coordinate (x, y, z0), |k| is the wave number, c is the wave velocity, and ω is the angular frequency.

5. The single-element super-resolution ultrasound imaging system based on deep learning according to claim 1, characterized in that: in, The calculation expression of the reconstructed microbubble image obtained according to the Moore-Penrose pseudo-inverse method is: In the formula is the true position of the microbubble in the reconstructed microbubble image, H is the system measurement matrix, + is the Moore-Penrose pseudo-inverse matrix calculated by SVD decomposition, and u is the microbubble echo signal corresponding to the reconstructed microbubble image.

6. The single-element super-resolution ultrasound imaging system based on deep learning according to claim 1, Its characteristics are: The training process of the microbubble localization network includes the following steps: Step T1, randomly placing multiple points in the imaging area as simulated microbubbles, and changing the positions of the simulated microbubbles frame by frame; Step T2, imaging the imaging area using the single-element ultrasonic probe to obtain a plurality of simulated microbubble echo signals; Step T3, performing image reconstruction using the Moore-Penrose pseudo-inverse method based on the simulated microbubble echo signal and the system measurement matrix to obtain a corresponding reconstructed simulated microbubble image; Step T4, taking all the reconstructed simulated microbubble images and the corresponding simulated microbubble real positioning points as a training data set; Step T5, constructing a deep convolutional neural network; Step T6: Based on the training data set, the deep convolutional neural network is trained in combination with the Adam optimized gradient descent algorithm and the learning rate decay strategy to obtain the trained deep convolutional neural network as the microbubble localization network.

7. The single-element super-resolution ultrasound imaging system based on deep learning according to claim 6, characterized in that: in, In step T6, the deep convolutional neural network is updated using a loss function, where the loss function is expressed as: In the formula is the calculation result of the loss function, N is the number of batch images traversed by the deep convolutional neural network each time, y is the real positioning point of the simulated microbubble, It is the positioning result of the deep convolutional neural network for the real positioning point y of the simulated microbubble.

8. The single-element super-resolution ultrasound imaging system based on deep learning according to claim 6, characterized in that: in, The deep convolutional neural network includes 6 residual modules. Each of the residual modules includes two convolutional layers and their corresponding batch normalization layers and activation functions for residual learning.

Images