Non-contrast super-resolution ultrasonic imaging method and system based on diffusion model

Abstract

The invention discloses a non-contrast super-resolution ultrasonic imaging method and system based on a diffusion model, and the method comprises the steps: transmitting a multi-angle composite plane wave to a target object, collecting an echo signal, and carrying out the beam synthesis to obtain an erythrocyte B-type ultrasonic image sequence; a space-time filtering algorithm based on singular value decomposition is applied to the erythrocyte B-type ultrasonic image sequence, erythrocyte signals are separated from tissue signals and noise signals, and an ultrasonic image sequence with the high signal-to-noise ratio is obtained for erythrocyte PD image reconstruction; and carrying out super-resolution mapping on the red blood cell PD image based on the trained diffusion model so as to reconstruct a super-resolution ultrasonic image in a non-contrast mode. According to the method, the dependence on an ultrasonic contrast agent is avoided, super-resolution mapping is carried out on the low-resolution erythrocyte PD image based on the deep learning diffusion model completed through training, ultrasonic imaging with ultrahigh resolution is achieved, the application flexibility of the SR-US imaging technology is improved, and the application range of the SR-US imaging technology is expanded.

Description

Technical Field

[0001] The present invention belongs to the technical field of ultrasonic imaging, and in particular relates to a non-contrast super-resolution ultrasonic imaging method and system based on a diffusion model. Background Art

[0002] Ultrasound imaging technology plays an important role in clinical diagnosis. It is widely used in clinical practice because of its advantages such as non-invasiveness, safety, portability and low price. Power Doppler imaging (PDI) is a type of ultrasound imaging technology, which is generally used to measure the ultrasound backscattering energy of the acoustic scattering source in the imaging area and is sensitive to micro-blood flow. However, the spatial resolution of ultrasound imaging is constrained by the diffraction limit, and its upper limit is half the length of the ultrasound wave used.

[0003] In order to improve the spatial resolution of ultrasound imaging, super-resolution ultrasound (SR-US) imaging, also known as ultrasound localization microscopy (ULM) imaging technology, was proposed. This technology breaks through the acoustic diffraction limit and can achieve a spatial resolution of 10 μm without sacrificing penetration depth, so it can image the microvascular system of deep tissues. Specifically, SR-US imaging based on contrast power Doppler (PD) images calculates a PD image from multiple frames of ultrasound contrast images collected, and then maps the image to a super-resolution image through a deep learning neural network to achieve super-resolution reconstruction.

[0004] However, the existing SR-US imaging technology requires the injection of ultrasound contrast agents (microbubbles) to achieve ultrasound echo signal enhancement, which makes it face problems such as high computational complexity, limited application scenarios and low emission sound pressure, which greatly limits the practical application of SR-US technology. Specifically, first of all, ULM technology requires the positioning and tracking of microbubbles. This process has high computational complexity and is difficult to achieve real-time imaging. Secondly, although the SR-US technology based on contrast PD images avoids the positioning and tracking of microbubbles, it still requires the injection of ultrasound contrast agents. As an exogenous imaging marker, there may be certain risks of side effects, such as allergic reactions or microvascular embolism. Finally, in order to maintain the stability of microbubbles and prevent them from rupturing, the emission sound pressure is usually also limited, and its mechanical coefficient generally cannot exceed 0.2, which limits the signal-to-noise ratio of ultrasound images. Summary of the invention

[0005] To solve the above problems, the purpose of the present invention is to provide a non-contrast super-resolution ultrasound imaging method and system based on a diffusion model. Based on a deep learning diffusion model, super-resolution mapping of red blood cell PD images is performed to realize non-contrast super-resolution ultrasound imaging, so as to expand the application scope of SR-US imaging technology, reduce the time required for imaging, and improve the imaging time resolution.

[0006] The technical solution provided by the present invention is: a non-contrast super-resolution ultrasound imaging method based on a diffusion model, comprising the following steps:

[0007] Transmit multi-angle composite plane waves to the target object, collect echo signals and perform beam synthesis to obtain a red blood cell B-mode ultrasound image sequence;

[0008] A spatiotemporal filtering algorithm based on singular value decomposition is applied to the red blood cell B-mode ultrasound image sequence to separate the red blood cell signal from the tissue signal and the noise signal, and obtain an ultrasound image sequence with a high signal-to-noise ratio for red blood cell PD image reconstruction.

[0009] The red blood cell PD image is super-resolution mapped based on the trained diffusion model to achieve reconstruction of a super-resolution ultrasound image in a non-contrast mode.

[0010] Preferably, the basic framework of the diffusion model is U-Net, which includes 2 residual blocks and introduces a self-attention mechanism. The high-resolution noise image and the low-resolution red blood cell PD image are bicubicly interpolated to the same resolution and then connected in the channel dimension as the network input of the diffusion model. The network output of the diffusion model is a high-resolution denoised image. The network is trained based on in vivo red blood cell ultrasound image data in a supervised learning manner, with low-resolution in vivo red blood cell PD images as network input and super-resolution vascular structure images obtained by ULM technology as network target output.

[0011] Preferably, the diffusion model:

[0012] In T optimization steps, the pure noise image y T ~N(0, I) gradually iteratively optimizes the output image, and finally obtains the target image y0~p(y|x) conditioned on the source image x. The distribution of the intermediate images in this iterative optimization process is defined as the forward diffusion process. Gaussian noise is gradually added to the output image through the Markov chain, which is recorded as q(y t |y t-1 ), the forward diffusion process is expressed as:

[0013]

[0014] In the formula, the scalar parameter α t Determines the amount of noise added in each iteration, and 0<αt <1;

[0015] In the middle process, y t The distribution of can be calculated as follows:

[0016]

[0017] In the formula,

[0018] Given (y0,y t ),y t-1 The posterior distribution of can be calculated as follows:

[0019] q(y t-1 |y0,y t )=N(y t-1 |μ,σ 2 I)

[0020]

[0021] Input source image x and noisy image y of the current optimization step t , the approximate target image y0 can be calculated by the following formula:

[0022]

[0023] p θ (y t-1 |y t , the average value of x) is:

[0024]

[0025] Each iterative optimization has the following form:

[0026]

[0027] In the formula, ε t ~N(0,I);

[0028] In this iterative optimization process, the denoising model f θ This provides an estimate of the gradient of the log density of the data, i.e. the noise level at that step.

[0029] Preferably, the loss function of the diffusion model is as follows:

[0030]

[0031] In the formula, εεN (0, I), p∈{1, 2},

[0032] The noise level at each iteration step can be predicted and denoising can be performed accordingly to optimize image quality.

[0033] Preferably, applying a spatiotemporal filtering algorithm based on singular value decomposition to the red blood cell B-mode ultrasound image sequence to separate the red blood cell signal from the tissue signal and the noise signal to obtain an ultrasound image sequence with a high signal-to-noise ratio for red blood cell PD image reconstruction further comprises:

[0034] The ultrasound image sequence I calculates the red blood cell PD image by the following formula:

[0035]

[0036] In the formula, I i is the i-th frame of the red blood cell B-mode ultrasound image sequence after singular value decomposition filtering, N is the total number of frames used to generate the PD image of this frame, which is set to 400.

[0037] Preferably, the training of the diffusion model further includes:

[0038] During the network training process, the Adam optimized gradient descent algorithm was used to train 2000 pairs of red blood cell PD images and corresponding ULM images for 75 cycles. The batch size was set to 1 and the initial learning rate was 5×10 -6 .

[0039] Preferably, the training of the diffusion model further includes:

[0040] Apply the learning rate decay strategy to gradually reduce the learning rate to 0 and gradually approach the optimal parameters.

[0041] Based on the same concept, the present invention also provides a non-contrast super-resolution ultrasound imaging system based on a diffusion model, comprising:

[0042] An image sequence acquisition module is used to transmit multi-angle composite plane waves to the target object, collect echo signals and perform beam synthesis to obtain a red blood cell B-type ultrasonic image sequence;

[0043] A PD image reconstruction module is used to apply a spatiotemporal filtering algorithm based on singular value decomposition to the red blood cell B-mode ultrasound image sequence, separate the red blood cell signal from the tissue signal and the noise signal, and obtain an ultrasound image sequence with a high signal-to-noise ratio for red blood cell PD image reconstruction;

[0044] The super-resolution mapping module is used to perform super-resolution mapping on the red blood cell PD image based on the trained diffusion model, so as to reconstruct a super-resolution ultrasound image in a non-contrast mode.

[0045] Based on the same concept, the present invention also provides an electronic device, comprising:

[0046] A memory, the memory being used to store a processing program;

[0047] A processor, wherein when executing the processing program, the processor implements any one of the above-mentioned non-contrast super-resolution ultrasound imaging methods based on the diffusion model.

[0048] Based on the same concept, the present invention also provides a readable storage medium, on which a processing program is stored. When the processing program is executed by a processor, any one of the above-mentioned non-contrast super-resolution ultrasound imaging methods based on a diffusion model is implemented.

[0049] Due to the adoption of the above technical solution, the present invention has the following advantages and positive effects compared with the prior art:

[0050] Compared with the existing SR-US imaging methods, the present invention avoids the dependence on ultrasound contrast agents, and realizes ultrasound imaging with ultra-high resolution by super-resolution mapping of low-resolution red blood cell PD images based on a trained deep learning diffusion model, thereby improving the application flexibility of SR-US imaging technology and expanding its application scope; in addition, since PD images are sensitive to microblood flow, the number of frames required for reconstruction is effectively reduced, thereby improving the temporal resolution of the method; in addition, the method avoids parameter dependence and does not require additional manual intervention. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] The specific embodiments of the present invention are further described in detail below with reference to the accompanying drawings, wherein:

[0052] Figure 1 It is a schematic diagram of the flow of the non-contrast super-resolution ultrasound imaging method based on the diffusion model of the present invention;

[0053] Figure 2 This is a network architecture diagram of the diffusion model of the present invention;

[0054] Figure 3 The present invention reconstructs a super-resolution ultrasound image in a non-contrast mode. DETAILED DESCRIPTION

[0055] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. The advantages and features of the present invention will become more apparent from the following description and claims. It should be noted that the drawings are all in a very simplified form and use non-precise ratios, which are only used to conveniently and clearly assist in explaining the purpose of the embodiments of the present invention.

[0056] It should be noted that all directional indications in the embodiments of the present invention (such as up, down, left, right, front, back, etc.) are only used to explain the relative position relationship, movement status, etc. between the components under a certain specific posture (as shown in the accompanying drawings). If the specific posture changes, the directional indication will also change accordingly.

[0057] The concept of the present invention is to construct a red blood cell energy Doppler super-resolution mapping model based on a deep learning network, including a diffusion model network framework, an iterative optimization process and a loss function; in a non-contrast mode, by emitting multi-angle ultrafast ultrasonic plane waves, collecting ultrasonic echo signals, and then compounding the multi-angle data and performing beam synthesis to obtain a red blood cell ultrasonic image sequence, and processing the sequence to obtain a low-resolution energy Doppler image and a super-resolution ultrasonic image reconstructed by a traditional method; based on the pairing relationship between the low-resolution energy Doppler image and the super-resolution ultrasonic image reconstructed by the traditional method, the network is trained; finally, the low-resolution energy Doppler image is processed by the trained deep learning model to reconstruct a super-resolution image, thereby realizing non-contrast super-resolution ultrasonic imaging.

[0058] Example

[0059] like Figure 1 As shown, a non-contrast super-resolution ultrasound imaging method based on a diffusion model comprises the following steps:

[0060] Step 1: emit multi-angle composite plane waves to the target object, collect echo signals and perform beam synthesis to obtain a red blood cell B-mode ultrasound image sequence;

[0061] Step 2: Apply a spatiotemporal filtering algorithm based on singular value decomposition to the red blood cell B-mode ultrasound image sequence to separate the red blood cell signal from the tissue signal and the noise signal, and obtain an ultrasound image sequence with a high signal-to-noise ratio for red blood cell PD image reconstruction;

[0062] Step 3: Perform super-resolution mapping on the red blood cell PD image based on the trained diffusion model to achieve reconstruction of a super-resolution ultrasound image in a non-contrast mode.

[0063] Compared with the existing SR-US imaging methods, the present invention avoids the dependence on ultrasound contrast agents, and realizes ultrasound imaging with ultra-high resolution by super-resolution mapping of low-resolution red blood cell PD images based on the trained deep learning diffusion model, thereby improving the application flexibility of SR-US imaging technology and expanding its application scope; in addition, since PD images are sensitive to microblood flow, the number of frames required for reconstruction is effectively reduced, thereby improving the temporal resolution of the method; in addition, the method avoids parameter dependence and does not require additional manual intervention.

[0064] In the process of obtaining the technical solution of the present invention, firstly, the diffusion model network framework is constructed:

[0065] Aiming at the ultrasonic echo signal characteristics of red blood cells, the present invention constructs a red blood cell PD super-resolution mapping network based on the denoising diffusion probability model (DDPM). The basic framework of the network is U-Net. The high-resolution noise image and the low-resolution red blood cell PD image are bicubicly interpolated to the same resolution, and then connected in the channel dimension as the network input. The expected output is a high-resolution denoised image. The network is trained based on in vivo red blood cell ultrasound image data in a supervised learning manner, with low-resolution in vivo red blood cell PD images as network input and super-resolution vascular structure images obtained by ULM technology as network target output. The U-Net network used in the present invention contains 2 residual blocks, and introduces a self-attention mechanism to better capture the global features and long-distance dependencies in the image, thereby improving the quality of image generation. The overall structure of the proposed network is shown in the attached figure. Figure 2 shown.

[0066] Second, construct an iterative optimization process:

[0067] The cDDPM constructed in this project is a pure noise image y in T optimization steps. T ~N(0, I) gradually iteratively optimizes the output image, and finally obtains the target image y0~p(y|x) conditioned on the source image x. The distribution of the intermediate images in this iterative optimization process is defined as the forward diffusion process, that is, Gaussian noise is gradually added to the output image through the Markov chain, denoted as q(y t |y t-1 ). The forward diffusion process can be expressed by equation (1) and equation (2):

[0068]

[0069] In the formula, the scalar parameter α t Determines the amount of noise added in each iteration, and 0<α t <1.

[0070] Furthermore, in the middle process y t The distribution of can be calculated as follows:

[0071]

[0072] In the formula,

[0073] Therefore, given (y0,y t ),y t-1 The posterior distribution of can be calculated as follows:

[0074] q(y t-1 |y0,y t)=N(y t-1 |μ,σ 2 I) (4)

[0075]

[0076] For the denoising model f constructed by the present invention θ , the input is the source image x and the noisy image y of the current optimization step t , the approximate target image y0 can be calculated by the following formula:

[0077]

[0078] After substituting formula (5) into formula (4), p θ (y t-1 |y t ,x) is:

[0079]

[0080] Finally, each iterative optimization can be obtained in the following form:

[0081]

[0082] In the formula, ε t ~N(0,I).

[0083] In this iterative optimization process, f θ It provides an estimate of the gradient of the logarithmic density of the data, i.e., the noise level at that step, similar to the one-step update process in Langevin dynamics.

[0084] Next, construct the loss function:

[0085] The loss function constructed by the present invention is as follows:

[0086]

[0087] In the formula, ε~N(0, I), p∈{1, 2},

[0088] With the above settings, the model can predict the noise level at each iteration step and perform denoising accordingly to optimize image quality.

[0089] In step 1, the specific method of obtaining the red blood cell PD image and ULM image is as follows.

[0090] Acquisition of red blood cell PD images: plane wave technology is used to achieve ultrafast ultrasonic imaging, and the signal-to-noise ratio of the ultrasonic image is improved by compounding multi-angle ultrasonic signals. The collected RF data is then beamformed to obtain a red blood cell B-type ultrasonic image sequence, realizing the acquisition of high-density, weak-echo red blood cell ultrasonic echo signals.

[0091] For this ultrasound image sequence, a spatiotemporal filtering algorithm based on singular value decomposition (SVD) is applied to separate the red blood cell signal from the tissue signal and the noise signal to obtain an ultrasound image sequence I with a high signal-to-noise ratio. The PD image PD can be calculated by the following formula:

[0092]

[0093] In the formula, I i is the i-th frame of the red blood cell B-mode ultrasound image after SVD filtering, N is the total number of frames used to generate the PD image of this frame, which is set to 400 in the present invention.

[0094] Collecting ULM images: After completing the acquisition of the ultrasound image sequence of the non-contrast modality, the rats were injected with contrast agent and the ultrasound image sequence of the contrast modality was acquired in the same way. Subsequently, for the contrast ultrasound image sequence, the ULM technology was applied to obtain the super-resolution ultrasound image of the imaging surface as the target image for further deep learning training. Specifically, the ULM technology performs SVD filtering, microbubble detection, positioning and tracking on the contrast ultrasound image sequence, and finally accumulates all microbubble sub-pixel trajectories and reconstructs the super-resolution ultrasound image.

[0095] Considering that the ultrasound image sequences of non-contrast and contrast modes are acquired asynchronously, the non-contrast red blood cell PD images do not necessarily correspond accurately to the ULM images. The training set of the present invention is not entirely composed of non-contrast PD images. The ratio of non-contrast to contrast PD images is approximately 1:9, and the time series of the contrast PD images is included in the time series of the corresponding ULM label images.

[0096] In some embodiments, during the training process of the network, the Adam optimized gradient descent algorithm is used, and 75 cycles of training are performed (traversing the training set 75 times) based on 2000 pairs of red blood cell PD images and corresponding ULM images. The batch size is set to 1 and the initial learning rate is 5×10-6.

[0097] In some embodiments, in order to speed up the convergence of the loss function, a learning rate decay strategy is applied to gradually reduce the learning rate to 0 and gradually approach the optimal parameters. The network training process takes about 50 hours in total.

[0098] Based on the trained deep learning model, the PD image obtained from the red blood cell ultrasound image sequence obtained in the above experiment is super-resolution mapped, and the super-resolution ultrasound image can be reconstructed in the non-contrast mode as shown in the attached figure. Figure 3 shown.

[0099] From the non-contrast super-resolution reconstruction results proposed in the present invention, it can be found that, compared with the existing SR-US imaging method, the present invention avoids the dependence on ultrasound contrast agents, and realizes ultrasound imaging with ultra-high resolution by super-resolution mapping of low-resolution red blood cell PD images, thereby improving the application flexibility of SR-US imaging technology and expanding its application range; in addition, since PD images are sensitive to microblood flow, the number of frames required for reconstruction is effectively reduced, thereby improving the temporal resolution of the method; finally, the method avoids parameter dependence and does not require additional manual intervention.

[0100] Based on the same concept, the present invention also provides a non-contrast super-resolution ultrasound imaging system based on a diffusion model, comprising:

[0101] An image sequence acquisition module is used to transmit multi-angle composite plane waves to the target object, collect echo signals and perform beam synthesis to obtain a red blood cell B-type ultrasonic image sequence;

[0102] A PD image reconstruction module is used to apply a spatiotemporal filtering algorithm based on singular value decomposition to the red blood cell B-mode ultrasound image sequence, separate the red blood cell signal from the tissue signal and the noise signal, and obtain an ultrasound image sequence with a high signal-to-noise ratio for red blood cell PD image reconstruction;

[0103] The super-resolution mapping module is used to perform super-resolution mapping on the red blood cell PD image based on the trained diffusion model, so as to reconstruct a super-resolution ultrasound image in a non-contrast mode.

[0104] Based on the same concept, the present invention also provides an electronic device, comprising: a memory, the memory being used to store a processing program; and a processor, the processor implementing any one of the above-mentioned non-contrast super-resolution ultrasound imaging methods based on a diffusion model when executing the processing program.

[0105] Based on the same concept, the present invention also provides a readable storage medium, on which a processing program is stored. When the processing program is executed by a processor, any one of the above-mentioned non-contrast super-resolution ultrasound imaging methods based on a diffusion model is implemented.

[0106] If the non-contrast super-resolution ultrasound imaging method based on the diffusion model is implemented in the form of program instructions and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this embodiment is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of software, and the computer software is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present disclosure. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program codes.

[0107] Those skilled in the art can clearly understand that, for the sake of convenience and brevity in description, the identification content specifically executed by the above-described system and device can refer to the corresponding process in the aforementioned method embodiment.

[0108] The embodiments of the present invention are described in detail above in conjunction with the accompanying drawings, but the present invention is not limited to the above embodiments. Even if various changes are made to the present invention, if these changes fall within the scope of the claims of the present invention and their equivalents, they still fall within the protection scope of the present invention.

Claims

1. A non-contrast super-resolution ultrasound imaging method based on a diffusion model, characterized in that: The following steps are involved: Transmit multi-angle composite plane waves to the target object, collect echo signals and perform beam synthesis to obtain a red blood cell B-mode ultrasound image sequence; A spatiotemporal filtering algorithm based on singular value decomposition is applied to the red blood cell B-mode ultrasound image sequence to separate the red blood cell signal from the tissue signal and the noise signal, and obtain an ultrasound image sequence with a high signal-to-noise ratio for red blood cell PD image reconstruction. The red blood cell PD image is super-resolution mapped based on the trained diffusion model to achieve reconstruction of a super-resolution ultrasound image in a non-contrast mode.

2. The non-contrast super-resolution ultrasound imaging method based on the diffusion model according to claim 1, characterized in that: The basic framework of the diffusion model is U-Net. The U-Net network contains 2 residual blocks and introduces a self-attention mechanism. The high-resolution noise image and the low-resolution red blood cell PD image are bicubicly interpolated to the same resolution and then connected in the channel dimension as the network input of the diffusion model. The network output of the diffusion model is a high-resolution denoised image. The network is trained based on in vivo red blood cell ultrasound image data in a supervised learning manner, with low-resolution in vivo red blood cell PD images as network input and super-resolution vascular structure images obtained by ULM technology as network target output.

3. The non-contrast super-resolution ultrasound imaging method based on diffusion model according to claim 2, characterized in that: The diffusion model: In T optimization steps, the pure noise image y T ~N(0, I) gradually iteratively optimizes the output image, and finally obtains the target image y0~p(y|x) conditioned on the source image x. The distribution of the intermediate images in this iterative optimization process is defined as the forward diffusion process. Gaussian noise is gradually added to the output image through the Markov chain, which is recorded as q(y t |y t-1 ), the forward diffusion process is expressed as: In the formula, the scalar parameter α t Determines the amount of noise added in each iteration, and 0<α t <1; In the middle process, y t The distribution of can be calculated as follows: In the formula, Given (y0, y t ),y t-1 The posterior distribution of can be calculated as follows: q(y t-1 |y0,y t )=N(y t-1 |μ,σ 2 I) Input source image x and noisy image y of the current optimization step t , the approximate target image y0 can be calculated by the following formula: p θ (y t-1 |y t , the average value of x) is: Each iterative optimization has the following form: In the formula, ε t ~N(0,I); In this iterative optimization process, the denoising model f θ This provides an estimate of the gradient of the log density of the data, i.e. the noise level at that step.

4. The non-contrast super-resolution ultrasound imaging method based on diffusion model according to claim 3, characterized in that: The loss function of the diffusion model is as follows: where ε~N(0,I), p∈{1,2}, The noise level at each iteration step can be predicted and denoising can be performed accordingly to optimize image quality.

5. The non-contrast super-resolution ultrasound imaging method based on diffusion model according to claim 1, characterized in that: Applying a spatiotemporal filtering algorithm based on singular value decomposition to the red blood cell B-mode ultrasound image sequence to separate the red blood cell signal from the tissue signal and the noise signal, and obtaining an ultrasound image sequence with a high signal-to-noise ratio for red blood cell PD image reconstruction further includes: The ultrasound image sequence I calculates the red blood cell PD image by the following formula: In the formula, I i is the i-th frame of the red blood cell B-mode ultrasound image sequence after singular value decomposition filtering, N is the total number of frames used to generate the PD image of this frame, which is set to 400.

6. The non-contrast super-resolution ultrasound imaging method based on diffusion model according to claim 1, characterized in that: The training of the diffusion model further includes: During the network training process, the Adam optimized gradient descent algorithm was used to train 2000 pairs of red blood cell PD images and corresponding ULM images for 75 cycles. The batch size was set to 1 and the initial learning rate was 5×10 -6 .

7. The non-contrast super-resolution ultrasound imaging method based on diffusion model according to claim 1, characterized in that: The training of the diffusion model further includes: Apply the learning rate decay strategy to gradually reduce the learning rate to 0 and gradually approach the optimal parameters.

8. A non-contrast super-resolution ultrasound imaging system based on a diffusion model, characterized in that: include: An image sequence acquisition module is used to transmit multi-angle composite plane waves to the target object, collect echo signals and perform beam synthesis to obtain a red blood cell B-type ultrasonic image sequence; A PD image reconstruction module is used to apply a spatiotemporal filtering algorithm based on singular value decomposition to the red blood cell B-mode ultrasound image sequence, separate the red blood cell signal from the tissue signal and the noise signal, and obtain an ultrasound image sequence with a high signal-to-noise ratio for red blood cell PD image reconstruction; The super-resolution mapping module is used to perform super-resolution mapping on the red blood cell PD image based on the trained diffusion model, so as to reconstruct a super-resolution ultrasound image in a non-contrast mode.

9. An electronic device, characterized in that: include: A memory, the memory being used to store a processing program; A processor, wherein when executing the processing program, the processor implements the non-contrast super-resolution ultrasound imaging method based on the diffusion model as described in any one of claims 1 to 7.

10. A readable storage medium, characterized in that: The readable storage medium stores a processing program, and when the processing program is executed by a processor, the non-contrast super-resolution ultrasound imaging method based on a diffusion model as described in any one of claims 1 to 7 is implemented.

Images