Ultrasonic CT (Computed Tomography) rapid reconstruction method, device and equipment based on sparse visual angle

Abstract

The invention discloses an ultrasonic CT rapid reconstruction method, device and equipment based on a sparse view angle, and the method comprises the steps: obtaining an initial sparse view angle reconstruction image through a delay superposition beam forming method by using data collected by ultrasonic transducer array elements which are sparsely distributed in a space; constructing a deep neural network, wherein the deep neural network is composed of a coding module, a decoding module and a residual module; constructing a sound velocity model, obtaining full-view ultrasonic CT data by using a numerical simulation method, and generating paired sparse view reconstruction images and full-view reconstruction images by using different sparse view strategies; a deep neural network is trained, in the training process, an Adam optimizer is adopted, a simulated annealing algorithm is combined to dynamically adjust the learning rate, and a loss function is a mean square error; in an image reasoning stage, an initial sparse view angle reconstruction image is input into a trained deep neural network, rapid reconstruction of an ultrasonic CT image is realized through GPU calculation, and a full view angle ultrasonic image with high real-time performance and high quality is obtained.

Description

Technical Field

[0001] This specification relates to the field of computer technology, and in particular, to a fast reconstruction method, device, and equipment for ultrasonic CT based on a sparse perspective. Background Art

[0002] Ultrasonic CT (Computed Tomography with Ultrasound) is a technology that uses ultrasonic waves for tomographic imaging. Different from X-ray CT, ultrasonic CT uses ultrasonic waves to penetrate human tissues and reconstructs images by measuring the propagation time and direction of ultrasonic waves. Since the propagation speed of ultrasonic waves in human tissues is relatively slow and is affected by various factors, the image reconstruction process of ultrasonic CT is usually more complex and time-consuming than that of X-ray CT. In clinical applications, doctors usually need to observe the changes in the patient's condition in real time. Therefore, the fast reconstruction algorithm of ultrasonic CT is crucial for providing real-time feedback. Due to the slow propagation speed of ultrasonic waves, even a slight movement of the patient during the scanning process may cause image distortion. The fast reconstruction algorithm can reduce the artifacts caused by movement and optimize the image reconstruction process, improve the resolution and contrast of the image, and thus obtain more accurate diagnostic results.

[0003] Currently, the existing ultrasonic CT reconstruction algorithm technologies mainly include the following categories: iterative-based methods, transform-based methods, and deep learning-based methods. Iterative-based methods solve the linear equations in the image reconstruction process through iteration and gradually approximate the real image. Representative algorithms include algebraic reconstruction technique (ART) and simultaneous iterative reconstruction technique (SIRT), etc. Transform-based methods perform a certain transformation (such as Fourier transform, wavelet transform, etc.) on the projection data and then use the transformed data for image reconstruction. Representative algorithms include filtered back projection algorithm (FBP) and fast Fourier transform algorithm (FFT), etc. Although iterative-based methods and transform-based methods can improve the speed of image reconstruction to a certain extent, their computational complexity is still relatively high, especially when dealing with large-scale data. In addition, due to the interference of various factors in the propagation of ultrasonic waves in human tissues, the existing reconstruction algorithms still face challenges in dealing with artifacts and noise. In recent years, with the development of deep learning technology, more and more researchers have begun to attempt to use deep learning models for ultrasonic CT image reconstruction. Deep learning-based methods use deep learning models (such as convolutional neural networks) to learn the mapping relationship from raw ultrasonic data to images. This method can achieve fast and high-quality reconstruction, but requires a large amount of training data and computing resources. In addition, the direct mapping from raw data to images must ensure the consistency of the data acquisition system, and the application of the algorithm is restricted by actual medical devices of different specifications.

[0004] In summary, the fast reconstruction algorithm of ultrasonic CT is a continuously studied field, and new technologies and methods need to be continuously explored to improve the speed and quality of image reconstruction. Summary of the Invention

[0005] This application aims to overcome the above-mentioned disadvantages of the prior art and provides a fast reconstruction method, device and equipment for ultrasonic CT based on sparse views.

[0006] This application adopts the following technical solutions:

[0007] The first aspect of this application provides a fast reconstruction method for ultrasonic CT images based on sparse views, including:

[0008] S1. Using the data collected by ultrasonic transducer array elements that are sparsely distributed in space, an initial sparse view reconstruction image is obtained through the delay-and-sum beamforming method.

[0009] S2. Construct a deep neural network, which consists of an encoding module, a decoding module and a residual module.

[0010] S3. Construct a sound speed model, obtain full-view ultrasonic CT data using numerical simulation methods, and then generate paired sparse view reconstruction images and full-view reconstruction images using different sparse view strategies.

[0011] S4. Train the deep neural network. During the training process, the Adam optimizer is adopted, and the simulated annealing algorithm is combined to dynamically adjust the learning rate. The loss function is the mean square error.

[0012] S5. Input the initial sparse view reconstruction image into the deep neural network, and realize the fast reconstruction of ultrasonic CT images through the deep neural network that determines the real-time sparse image and performs high-quality reconstruction.

[0013] Optionally, the step S1 of using the data collected by ultrasonic transducer array elements that are sparsely distributed in space to obtain an initial sparse view reconstruction image through the delay-and-sum beamforming method specifically includes:

[0014] S11. According to the preset radius of the circular array and the number of circular arrays, determine the array elements that make up the circular array from each array element of the sensor.

[0015] S12. Determine the initial sparse view reconstruction image according to the data collected by the array elements that make up the circular array. Specifically, it includes:

[0016] S121. According to the preset number of ultrasonic transducer array elements, determine a plurality of excited array elements that are evenly or randomly distributed from the array;

[0017] S122. For each excitation element, when transmitting ultrasonic waves, the preset ultrasonic transducer elements receive the ultrasonic waves until the elements can no longer receive the ultrasonic waves, and then activate the next excitation element to transmit ultrasonic waves;

[0018] S13. When all excitation elements have transmitted ultrasonic waves, stop activating the excitation elements, and solve the problem that ultrasonic waves in different directions arrive at different times due to different propagation paths by determining the delay time and performing time-delay correction on the received ultrasonic signals. Among them, the delay time is determined by the following formula:

[0019] (1)

[0020] where, represents the coordinate of the nth excitation element, represents the coordinate of the mth receiving element, represents the coordinate of the imaging point,

[0021] S14. According to the delay time, superimpose the RF signals received by each single element and each circular array each time an excitation is made to determine the sparse image. The formula for the delay superposition result is as follows:

[0022] (2)

[0023] In the formula, represents the coordinate of the nth excitation element, represents the coordinate of the mth receiving element, represents the number of excitation times, represents Figure 2 the number of receiving elements, and the obtained sparse image is as shown in the input example of the deep neural network.

[0024] Optionally, the dynamic range of the initial sparse view reconstruction image is 60 db.

[0025] Optionally, the encoding module and the decoding module at least include: a convolutional layer, a batch normalization layer, and a rectified linear unit layer.

[0026] Optionally, the residual module at least includes: a convolutional layer and a rectified linear unit layer.

[0027] Optionally, for each decoding layer of the decoding module, the input data of the decoding layer is obtained by fusing the encoded features output by the corresponding encoding layer and the output result of the previous layer of the decoding layer, and the output result is determined through the decoding layer.

[0028] Optionally, the deep neural network adopts the Adam optimization algorithm during training, and the learning rate is artificially decayed using the simulated annealing method. After a certain number of iterations, the learning rate becomes one-tenth of the original.

[0029] The initial sparse-view reconstructed image is preprocessed before being input into the network. Each pixel value is divided by the maximum value of the entire training data set to rescale the values between 0 and 1. The weights of the convolutional layers are initialized with a random normal distribution with a mean of 0 and a standard deviation of 0.01 to prevent the gradients from vanishing or exploding during the initial training stage. The mean squared error (MSE) between the network output and the training ground truth is used as the training loss function. The formula for the mean squared error is:

[0030] (3)

[0031] In the formula, represents the pixel value at the position of the full-view image, represents the pixel value at the position of the sparse-view image, represents the total number of pixel points.

[0032] The second aspect of this application provides a fast reconstruction device for ultrasonic CT based on sparse views, including:

[0033] An initial sparse-view reconstructed image acquisition module, which is used to obtain an initial sparse-view reconstructed image by using the data collected by the ultrasonic transducer array elements sparsely distributed in space through the delay-and-sum beamforming method;

[0034] A deep neural network construction module, which is used to construct a deep neural network, and the deep neural network is composed of an encoding module, a decoding module, and a residual module;

[0035] An image generation module, which is used to construct a sound speed model to obtain full-view ultrasonic CT data by using the numerical simulation method, and then generate paired sparse-view reconstructed images and full-view reconstructed images by using different sparse-view strategies;

[0036] A deep neural network training module, which is used to train the deep neural network. During the training process, the Adam optimizer is adopted, and the simulated annealing algorithm is combined to dynamically adjust the learning rate, and the loss function is the mean squared error;

[0037] An ultrasonic CT image rapid reconstruction module, which is used in the image inference stage to input the initial sparse-view reconstruction image into a trained deep neural network, and through GPU calculation, rapidly reconstruct the ultrasonic CT image to obtain a full-view ultrasonic image with high real-time performance and high quality.

[0038] The third aspect of the present application provides a computer-readable storage medium storing a computer program, which when executed by a processor, implements the above-mentioned rapid imaging method for ultrasonic CT images based on sparse views.

[0039] The fourth aspect of the present application provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, it implements the above-mentioned rapid reconstruction method for ultrasonic CT images based on sparse views.

[0040] The present application uses the data collected by ultrasonic transducer array elements sparsely distributed in space, obtains the initial sparse-view reconstruction image through the delay-and-sum beamforming method, inputs it into a deep neural network, extracts different depth features by an encoding module, and fuses the output of the encoding layer and different depth features through a decoding module to obtain a reconstruction image; a residual module adds the reconstruction image output by the decoding layer to the initial sparse-view reconstruction image to finally obtain the required ultrasonic CT image. By determining the real-time sparse image and the deep neural network for high-quality reconstruction, the rapid reconstruction of ultrasonic CT images is achieved.

[0041] The present application achieves the rapid reconstruction of ultrasonic CT images by determining the real-time sparse image and the deep neural network for high-quality reconstruction.

[0042] The present invention has the following beneficial effects:

[0043] Based on the data collected by some excited array elements in the sensor, a sparse image is determined, and the sparse-view image is input into a deep neural network to obtain the output results of each encoding layer. Then, for each decoding layer of the decoding module, according to the encoding features of the encoding layer corresponding to this decoding layer and the output result of the previous layer of this decoding layer, the input data and the corresponding output result of this decoding layer are fused. A residual module is introduced to add the reconstruction image of the last decoding layer and the initial sparse-view image to generate the final ultrasonic CT image. Description of the Drawings

[0044] The drawings described herein are used to provide a further understanding of the present specification, form a part of the present specification, and the schematic embodiments and descriptions thereof are used to explain the present specification and do not constitute an improper limitation to the present specification. In the drawings:

[0045] Figure 1Schematic flow diagram of a fast reconstruction method for ultrasonic CT based on a sparse perspective provided by this application;

[0046] Figure 2 Input example of the deep neural network for a fast reconstruction method for ultrasonic CT based on a sparse perspective provided by this application;

[0047] Figure 3 Schematic diagram of the deep neural network structure for a fast reconstruction method for ultrasonic CT based on a sparse perspective provided by this application;

[0048] Figure 4 Output example of the deep neural network for a fast reconstruction method for ultrasonic CT based on a sparse perspective provided by this application;

[0049] Figure 5 Schematic diagram of a fast reconstruction device for ultrasonic CT based on a sparse perspective provided by this application;

[0050] Figure 6 For the application to provide a corresponding to Figure 1 Schematic diagram of the electronic device. Specific embodiments

[0051] To make the objectives, technical solutions, and advantages of this specification clearer, the technical solutions of this specification will be clearly and completely described below in conjunction with specific embodiments of this specification and the corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this specification, rather than all of them. Based on the embodiments in this specification, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of this application.

[0052] It should be noted here that the fast imaging method for ultrasonic CT images provided in this specification can be executed by a computer or a server, and for the specific entity executing the method, this specification does not make any restrictions here. And for the convenience of description in this specification, this specification takes the server executing this fast imaging method for ultrasonic CT images as an example for illustration.

[0053] The following will detail the technical solutions provided in each embodiment of this specification in conjunction with the drawings.

[0054] Embodiment 1

[0055] As Figure 1 shown, a schematic flow diagram of a fast imaging method for ultrasonic CT images based on a sparse perspective provided in this embodiment specifically includes the following steps:

[0056] S1: Using the data collected by the ultrasonic transducer array elements that are sparsely distributed in space, obtain an initial sparse perspective reconstruction image through the delay-and-sum beamforming method.

[0057] In the embodiment of this specification, the server obtains an initial sparse view reconstruction image through the delay-and-sum beamforming method based on the ultrasonic data excited by partially sparse array elements, and uses it as the input of the neural deep neural network.

[0058] Specifically, in the embodiment of this specification, according to the preset number of ultrasonic transducer array elements, a plurality of excited array elements with uniform distribution are determined from the array. For each excited array element, when the excited array element emits ultrasonic waves, each array element receives the ultrasonic waves until each array element can no longer receive the ultrasonic waves, and then the next excited array element is activated to emit ultrasonic waves. When each excited array element has emitted ultrasonic waves, the activation of the excited array element is stopped.

[0059] Furthermore, the server solves the problem that ultrasonic waves in different directions arrive at different times due to different propagation paths by determining the delay time and performing time delay correction on the received ultrasonic signals. Among them, the delay time is determined by the following formula:

[0060] (1)

[0061] where represents the coordinate of the nth excited array element, represents the coordinate of the mth receiving array element, represents the coordinate of the imaging point, represents the background sound speed.

[0062] Furthermore, the server superimposes the RF signals received by each single array element and each circular array each time an excitation occurs according to the delay time to obtain a sparse image. The calculation formula of the delay-and-sum result is as follows:

[0063] (2)

[0064] In the formula, represents the coordinate of the nth excited array element, represents the coordinate of the mth receiving array element, represents the coordinate of the imaging point, represents the number of excitation times, represents the number of receiving array elements, and the obtained sparse image is as shown in the Figure 2 input example of the deep neural network.

[0065] In this embodiment, the dynamic range of the initial sparse view reconstruction image is 60 db.

[0066] S2: Construct a deep neural network, which consists of an encoding module, a decoding module, and a residual module.

[0067] In the embodiments of this specification, the encoding module at least includes: an upsampling layer, a batch normalization layer, and a rectified linear unit layer.

[0068] Specifically, the sparse image passes through the convolutional layer in the sampling layer adjusted by the up rectified linear unit to obtain the features of the image, and then passes through the batch normalization layer combined with the convolutional layer to improve the speed and stability of obtaining features.

[0069] In the embodiments of this specification, the decoding module at least includes: a rectified linear unit layer, a batch normalization layer, and an upsampling layer.

[0070] Specifically, the structures of the decoding layers in the decoding module are symmetric to those of the encoding layers.

[0071] For example, in the trained deep neural network, 64 feature maps are generated through the convolutional layer of the encoding layer in the first encoding module. Then, in the convolutional steps of the convolutional layers in each subsequent encoding layer, this number doubles until the number of channels at the bottom layer reaches 512. The size of the convolutional kernels in all convolutional layers is 3×3.

[0072] S3: Construct a sound speed model, use numerical simulation methods to obtain full-view ultrasonic CT data, and then use different sparse view strategies to generate paired sparse view reconstruction images and full-view reconstruction images.

[0073] Specifically, the construction of the sound speed model includes randomly generating 100 sound speed models and screening 200 OA-breast slices from the public dataset, and using the k-wave simulation toolbox to obtain 300 sets of simulation array data. By extracting the array data of random sparse views, the initial reconstruction image is obtained using the delay-and-sum beamforming method, and then the gold standard image is obtained based on the full-view array data. A total of 50,000 pairs of training datasets for composing this deep neural network are obtained.

[0074] S4: Train the deep neural network. During the training process, the Adam optimizer is adopted, and the simulated annealing algorithm is combined to dynamically adjust the learning rate. The loss function is the mean square error. The mean square error formula is:

[0075] (3)

[0076] In the formula, represents the pixel value at the position of the full-view image, represents the pixel value at the position of the sparse view image, represents the total number of pixel points.

[0077] Specifically, the initial sparse-view reconstructed image is preprocessed before being input into the network. Each pixel value is divided by the maximum value of the entire training dataset to rescale the values between 0 and 1. The weights of the convolutional layers are initialized with a random normal distribution with a mean of 0 and a standard deviation of 0.01 to prevent the gradients from vanishing or exploding during the initial training phase.

[0078] Specifically, for all training processes, the Stochastic Gradient Descent method with a momentum value of 0.99 is used. The learning rate is artificially decayed using simulated annealing. At the 50th iteration, the learning rate is changed to one-tenth of the original value, reducing the learning rate from 0.01 to 0.00001, and then the learning rate remains unchanged.

[0079] Optionally, this specification also provides a method for training a neural network, including: the server determines a preset deep neural network to be trained and a preset pair of ultrasound CT images. According to the full-view ultrasound CT image, the corresponding sparse-view images are determined. The sparse-view images are sequentially input into the deep neural network to be trained, and the mean square error between each simulated image output by the deep neural network to be trained and the full-view ultrasound CT image is calculated. Taking the minimum mean square error as the optimization goal, the deep neural network is trained.

[0080] S5: In the image inference stage, the initial sparse-view reconstructed image is input into the trained deep neural network, and the fast reconstruction of the ultrasound CT image is realized through GPU calculation, obtaining a full-view ultrasound image with high real-time performance and high quality.

[0081] Based on Figure 1 The provided fast imaging method of the ultrasound CT image, the server obtains the initial sparse-view reconstructed image according to the data collected by some array elements in the sensor through the delay-and-sum beamforming method. And the reconstructed image is input into the trained deep neural network to generate the final ultrasound CT image. By determining the sparse image with real-time performance and the deep neural network for high-quality reconstruction, the high-quality and fast reconstruction of the ultrasound CT image is realized, meeting the existing medical needs.

[0082] In the embodiments of this specification, based on a type of breast numerical phantom, numerical tests are carried out based on the matlab platform and the k-wave toolbox.

[0083] Furthermore, the server receives a deep neural network model based on UNET and trains it under the conditions of a Tesla V100 GPU, 32GB of memory, and a 6-core CPU cloud computing resource.

[0084] It should be noted here that when multiple trained deep neural networks are obtained according to the above training method, the server can use the mean squared error (MSE),

[0085] peak signal-to-noise ratio (PSNR), structural similarity

[0086] (Structural Similarity Index Measure, SSIM), and running time to compare with the traditional delay superposition algorithm to evaluate the performance of the ultrasound CT fast reconstruction method based on the deep neural network.

[0087] Among them, PSNR is defined based on the mean squared error. The larger its value, the better the quality of the image. The formula is as follows:

[0088] (4)

[0089] In the formula, is the maximum value of the image, is the mean squared error.

[0090] Among them, SSIM is a number between 0 and 1. The larger it is, the smaller the gap between the output image and the distortion-free image, that is, the better the image quality. When the two images are identical, SSIM = 1. The calculation formula is as follows:

[0091] (5)

[0092] In the formula, l is the luminance comparison function, c is the contrast comparison function, and s is the structure comparison function. 、 and respectively represent the proportions of different features in the SSIM measurement, and the default is 1 for all. This specification does not limit which method to specifically use to evaluate the deep neural network.

[0093] For the traditional delay superposition beamforming algorithm with 128 excitations, the running time is 89.1 s, MSE = 0.024, PSNR = 16.19, SSIM = 0.77. For the ultrasound CT fast reconstruction method based on sparse views, the running time is 13.4 s, MSE = 0.007, PSNR = 21.55, SSIM = 0.82. From the numerical indicators of MSE, PSNR, SSIM, and running time, it can be seen that the fast reconstruction algorithm improves the image quality while reducing the running time from 89.1 s to 13.4 s, and the computational cost is 1 / 6 of the original.

[0094] Example 2

[0095] This embodiment provides a fast reconstruction device for ultrasonic CT images corresponding to a fast reconstruction method of ultrasonic CT based on a sparse perspective according to Embodiment 1, as Figure 5 shown below:

[0096] A reconstruction module 201 determines a sparse image based on data collected by some elements in the sensor;

[0097] An encoding module 203 inputs the sparse image into a trained deep neural network, and through each encoding layer in the encoding module of the deep neural network, obtains the output results output by each encoding layer, where the output results correspond to each other in the encoding layer and the decoding layer with the same number of channels;

[0098] A decoding module 205, for each decoding layer of each decoding module, fuses the encoding features output by the corresponding encoding layer of this decoding layer and the output result of the previous layer of this decoding layer to obtain the input data of this decoding layer, and determines the output result through this decoding layer;

[0099] A residual module 207 splices the reconstructed image output by the last decoding layer in the decoding module and the sparse image, and inputs them into the residual learning module of the decoding module, and generates an ultrasonic CT image through the residual learning module.

[0100] Optionally, the reconstruction module 201 is further configured to determine a sparse image based on data collected by some elements in the sensor. According to the preset radius of the circular array and the number of circular arrays, the elements constituting the multi-circular array are determined from each element of the sensor. A sparse image is determined based on the data collected by the elements constituting the multi-circular array.

[0101] Optionally, the reconstruction module 201 is further configured to determine a sparse image based on data collected by some elements in the sensor. According to the preset number of ultrasonic transducer elements, a plurality of uniformly distributed excitation elements are determined from the array. For each excitation element, when the excitation element emits ultrasonic waves, each element receives the ultrasonic waves until each element can no longer receive the ultrasonic waves, and the next excitation element is activated to emit ultrasonic waves. When each excitation element has emitted ultrasonic waves, the activation of the excitation element is stopped, and a sparse image is determined.

[0102] Optionally, the encoding layers in the encoding module 203 at least include: an upsampling layer, a batch normalization layer, and a rectified linear unit layer.

[0103] Optionally, the decoding layers in the decoding module 205 at least include: a rectified linear unit layer, a batch normalization layer, and an upsampling layer.

[0104] Optionally, the decoding module 205 is further configured to, for each decoding layer of the decoding module, fuse the encoded features output by the encoding layer corresponding to the decoding layer and the output result of the previous layer of the decoding layer to obtain the input data of the decoding layer, and determine the output result through the decoding layer. Determine the encoding layer corresponding to the first decoding layer in the decoding module, fuse the encoded features output by the encoding layer corresponding to the first decoding layer and the encoded features output by the previous encoding layer of the encoding layer corresponding to the first decoding layer to obtain the input decoded features of the first decoding layer, and determine the decoded features output by the encoding layer corresponding to the first decoding layer through the first decoding layer. For each non-first decoding layer in the decoding module, fuse the encoded features output by the encoding layer corresponding to the layer and the decoded features output by the previous decoding layer of the layer to obtain the input decoded features of the layer, and determine the decoded features of the layer output by the encoding layer corresponding to the layer through the decoding layer.

[0105] Optionally, the decoding module 205 is further configured to, for each decoding layer of the decoding module, fuse the encoded features output by the encoding layer corresponding to the decoding layer and the output result of the previous layer of the decoding layer to obtain the input data of the decoding layer, and determine the output result through the decoding layer. Determine the encoding layer corresponding to the first decoding layer in the decoding module, fuse the encoded features output by the encoding layer corresponding to the first decoding layer and the encoded features output by the previous encoding layer of the encoding layer corresponding to the first decoding layer to obtain the input decoded features with the size of the encoded features output by the previous encoding layer of the first decoding layer, and determine the decoded features with the size of the encoded features output by the encoding layer corresponding to the first decoding layer through the decoding layer. For each non-first decoding layer in the decoding module, fuse the encoded features output by the encoding layer corresponding to the input layer and the decoded features output by the previous decoding layer of the layer to obtain the input decoded features with the size of the decoded features output by the previous decoding layer of the layer, and determine the decoded features with the size of the encoded features output by the encoding layer corresponding to the layer through the decoding layer.

[0106] Embodiment 3

[0107] This embodiment provides a computer-readable storage medium storing a computer program that can be used to execute the fast reconstruction method for ultrasonic CT images based on the sparse perspective in Embodiment 1.

[0108] Embodiment 4

[0109] This embodiment provides Figure 6 the schematic structural diagram of an electronic device corresponding to Figure 1 As Figure 6 described, at the hardware level, the electronic device includes a processor, an internal bus, a network interface, a memory, and a non-volatile memory. Of course, it may also include other hardware required for other services. The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it to implement the above Figure 1The rapid imaging method for ultrasonic CT images. Of course, in addition to the software implementation, this specification does not exclude other implementation methods, such as logic devices or a combination of software and hardware, etc. That is to say, the execution subject of the following processing flow is not limited to each logic unit, and can also be hardware or a logic device.

[0110] In the 1990s, it was obvious to distinguish whether an improvement in a technology was an improvement in hardware (e.g., improvement in circuit structures such as diodes, transistors, switches, etc.) or an improvement in software (improvement in method processes). However, with the development of technology, many improvements in method processes today can be regarded as direct improvements in hardware circuit structures. Almost all designers obtain the corresponding hardware circuit structures by programming the improved method processes into the hardware circuits. Therefore, it cannot be said that an improvement in a method process cannot be implemented with a hardware entity module. For example, a programmable logic device (PLD) (e.g., a field programmable gate array (FPGA)) is such an integrated circuit whose logic function is determined by the user programming the device. The designer can program by himself to "integrate" a digital system on a PLD, without having to ask a chip manufacturer to design and fabricate a dedicated integrated circuit chip. Moreover, nowadays, instead of manually fabricating integrated circuit chips, this programming is mostly implemented using "logic compiler" software, which is similar to the software compiler used in program development and writing. The original code before compilation also has to be written in a specific programming language, which is called a hardware description language (HDL), and there is not only one kind of HDL, but many kinds, such as ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), Confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, RHDL (Ruby Hardware Description Language), etc. Currently, the most commonly used are VHDL (Very-High-Speed Integrated Circuit Hardware Description Language) and Verilog. Those skilled in the art should also be aware that only by slightly logically programming the method process with the above-mentioned several hardware description languages and programming it into the integrated circuit can the hardware circuit implementing the logic method process be easily obtained.

[0111] The controller can be implemented in any suitable manner. For example, the controller can take the form of, for example, a microprocessor or a processor and a computer-readable medium storing computer-readable program code (such as software or firmware) executable by the (micro)processor, logic gates, switches, an application specific integrated circuit (ASIC), a programmable logic controller, and an embedded microcontroller. Examples of the controller include, but are not limited to, the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicone Labs C8051F320. The memory controller can also be implemented as part of the control logic of the memory. Those skilled in the art also know that, in addition to implementing the controller in the form of pure computer-readable program code, it is entirely possible to implement the same function by logically programming the method steps so that the controller is in the form of logic gates, switches, application specific integrated circuits, programmable logic controllers, and embedded microcontrollers. Therefore, such a controller can be considered a hardware component, and the devices included therein for implementing various functions can also be regarded as structures within the hardware component. Or even, the devices for implementing various functions can be regarded as either software modules for implementing the method or structures within the hardware component.

[0112] The systems, devices, modules, or units illustrated in the above embodiments can be specifically implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, the computer can be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or any combination of these devices.

[0113] For the convenience of description, when describing the above devices, they are described separately as various units according to their functions. Of course, when implementing this specification, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0114] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0115] This application is described with reference to the flowcharts and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the application. It should be understood that each flow and / or block in the flowchart and / or block diagram, and combinations of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processors of general purpose computers, special purpose computers, embedded processors, or other programmable data processing devices to produce a machine, such that the instructions executed by the processors of the computer or other programmable data processing devices produce means for implementing the functions specified in one or more of the flows Figure 1 one or more of the flows and / or blocks Figure 1 or means for implementing the functions specified in one or more of the blocks.

[0116] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to operate in a particular manner, such that the instructions stored in the computer-readable memory produce a manufacture including instruction means that implement the functions specified in one or more of the flows Figure 1 one or more of the flows and / or blocks Figure 1 or means for implementing the functions specified in one or more of the blocks.

[0117] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operational steps are performed on the computer or other programmable device to produce a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one or more of the flows Figure 1 one or more of the flows and / or blocks Figure 1 or means for implementing the functions specified in one or more of the blocks.

[0118] In a typical configuration, a computing device includes one or more processors (CPUs), an input / output interface, a network interface, and memory.

[0119] The memory may include non-permanent memory in the form of computer-readable media, random access memory (RAM), and / or non-volatile memory such as read-only memory (ROM) or flash RAM. The memory is an example of computer-readable media.

[0120] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined in this article, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0121] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0122] It should be understood by those skilled in the art that the embodiments of this specification may be provided as methods, systems or computer program products. Therefore, this specification may take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware. Moreover, this specification may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0123] This specification may be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. This specification may also be practiced in distributed computing environments where tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules may be located in local and remote computer storage media, including storage devices.

[0124] Each embodiment in this specification is described in a progressive manner. For the parts that are the same or similar among the embodiments, reference can be made to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and for the relevant parts, reference can be made to the partial description of the method embodiment.

[0125] The above are only the embodiments of this specification and are not intended to limit this specification. For those skilled in the art, various changes and modifications can be made to this specification. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of this specification shall be included within the scope of the claims of this specification.

Claims

1. A fast reconstruction method for ultrasonic CT based on sparse view, characterized in that: include: S1: Using the data collected by the ultrasonic transducer array elements sparsely distributed in space, the initial sparse perspective reconstructed image is obtained through the delay superposition beamforming method; S2: construct a deep neural network, wherein the deep neural network consists of an encoding module, a decoding module and a residual module; S3: Construct a sound velocity model and use numerical simulation methods to obtain full-view ultrasonic CT data, and then use different sparse view strategies to generate pairs of sparse view reconstructed images and full view reconstructed images; S4: Training deep neural networks. During the training process, the Adam optimizer was used, combined with the simulated annealing algorithm to dynamically adjust the learning rate. The loss function was the mean square error. S5: In the image inference stage, the initial sparse view reconstructed image is input into the trained deep neural network, and the ultrasonic CT image is quickly reconstructed through GPU computing to obtain a high-real-time and high-quality full-view ultrasonic image.

2. The method according to claim 1, characterized in that In step S1, data collected by ultrasonic transducer array elements sparsely distributed in space specifically includes: S11. According to a preset number of ultrasonic transducer array elements, a plurality of sparsely distributed excitation array elements are determined from the array by probability sampling; S12. For each excitation array element, when emitting an ultrasonic wave, each preset ultrasonic transducer array element receives the ultrasonic wave, and the sampling time is twice the propagation time of the direct wave between the two farthest array elements, and the next excitation array element is activated to emit an ultrasonic wave; S13. When each excitation array element has emitted ultrasonic waves, stop activating the excitation array element and determine a sparse viewing angle image.

3. The method according to claim 1, characterized in that In step S1, an initial sparse perspective reconstructed image is obtained by a delay-addition beamforming method, which specifically includes: S14. Based on the data collected by the array elements of the annular array under the sparse perspective, the delay time is determined and the received ultrasonic signal is corrected to solve the problem that ultrasonic waves in different directions have different arrival times after passing through different propagation paths; wherein the delay time The formula for determining is as follows: （1） in, express Serial number excitation array element coordinates, express Serial number receiving array element coordinates, represents the coordinates of the imaging point, represents the background sound speed; S15. According to the delay time, each single array element is excited each time, and the radio frequency signals received by each annular array are superimposed to determine the sparse image, and the delay superposition result The calculation formula is as follows: （2） In the formula, express Serial number excitation array element coordinates, express Serial number receiving array element coordinates, represents the coordinates of the imaging point, Indicates the number of excitations, Indicates the number of receiving array elements.

4. The method according to claim 1, characterized in that In step S2, the encoding module includes the following encoding layers: a convolutional layer, a batch normalization layer, and a rectified linear unit layer.

5. The method according to claim 1, characterized in that In step S2, the decoding module includes the following decoding layers: a convolutional layer, a rectified linear unit layer, a batch normalization layer, and an upsampling layer.

6. The method according to claim 5, characterized in that Each decoding layer of the decoding module obtains the input data of the decoding layer by fusing the coding features output by the coding layer corresponding to the decoding layer and the output result of the previous layer of the decoding layer, and determines the output result through the decoding layer.

7. The method according to claim 1, characterized in that In step S4, during the training process of the deep neural network, the Adam optimizer is used in combination with the simulated annealing algorithm to dynamically adjust the learning rate, and the learning rate is reduced to one tenth of the original rate every 50 iterations.

8. An ultrasonic CT fast reconstruction device based on sparse viewing angle, characterized in that: include: An initial sparse perspective reconstruction image acquisition module is used to obtain an initial sparse perspective reconstruction image by using the data collected by the ultrasonic transducer array elements through a time-delayed superposition beamforming method; A deep neural network building module, used to build a deep neural network, wherein the deep neural network consists of an encoding module, a decoding module and a residual module; An image generation module is used to construct a sound velocity model and obtain full-view ultrasonic CT data using a numerical simulation method, and then use different sparse view strategies to generate pairs of sparse view reconstruction images and full-view reconstruction images; The deep neural network training module is used to train the deep neural network. During the training process, the Adam optimizer is used, and the simulated annealing algorithm is combined to dynamically adjust the learning rate. The loss function is the mean square error. The ultrasonic CT image fast reconstruction module is used to input the initial sparse view reconstructed image into the trained deep neural network during the image inference stage, and realize the fast reconstruction of the ultrasonic CT image through GPU computing to obtain high real-time and high-quality full-view ultrasonic images.

9. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by a processor, the method described in any one of claims 1 to 7 is implemented.

10. An electronic device, characterized in that: The method comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the method according to any one of claims 1 to 7 when executing the program.

Images