Rapid generation method and system for transcranial focusing acoustical holographic lens based on deep learning

Abstract

The invention discloses a transcranial focusing acoustical holographic lens rapid generation method and system based on deep learning, and the method comprises the steps: employing a k-wave tool box in Matlab, employing a time reversal method to set a preset focusing point as a sound source, capturing a phase at a plane of an ultrasonic transducer, and designing an acoustical holographic lens; the position of the preset focus and the designed holographic lens serve as a data set to train a U-net network, the input of the U-net network is the position of the preset focus, and the output of the U-net network is the designed holographic lens; setting a target focus point, inputting the trained U-net network, and outputting and acquiring a holographic lens corresponding to the target focus point from the trained U-net network; and holographic lens design is output based on the U-net network, and the holographic lens is manufactured through 3D printing. Based on a deep learning framework, the acoustic holographic lens is quickly and accurately generated, and the problem that in the transcranial focusing process through the acoustic holographic lens, manufacturing and verification of the acoustic holographic lens consume too much computing resources is solved.

Description

Technical Field

[0001] The present invention belongs to the field of medical imaging technology, and in particular relates to a method and system for rapidly generating a transcranial focusing acoustic holographic lens based on deep learning. Background Art

[0002] Transcranial focused ultrasound has great development potential due to its advantages such as non-invasiveness, safety, and deep penetration. Ultrasound can maintain a high penetration depth when passing through the skull, allowing it to directly affect deep brain tissue, and the thermal effect generated by focused ultrasound helps to treat deep diseases. Therefore, compared with traditional transcranial surgical treatment methods, transcranial focused ultrasound technology can achieve surgical treatment without craniotomy, reducing the risk of infection and bleeding, and has broad application scenarios in the biomedical field. For example, at low intensity, transcranial focused ultrasound can reversibly open the blood-brain barrier without tissue ablation, and at high intensity, transcranial focused ultrasound can produce tissue ablation. However, when ultrasound passes through the skull, due to the nature of the skull itself and the complexity of its structure, it will cause strong ultrasonic scattering and attenuation, causing severe distortion and distortion of the transcranial sound field. At present, transcranial focusing is a robust and low-cost transcranial focusing method that modulates the phase of the transmitted plane wave through an ultrasonic holographic lens. The holographic lens used for transcranial focusing is currently designed based on the time reversal method. In the process of designing the hologram, three-dimensional simulation is required, which has a high computational cost, thus limiting the flexibility and practicality of the application of the acoustic holographic lens in transcranial focused ultrasound. Summary of the invention

[0003] To solve the above problems, the purpose of the present invention is to provide a method and system for quickly generating transcranial focusing acoustic holographic lenses based on deep learning, which realizes fast and accurate acoustic holographic lens generation based on the framework of deep learning, solves the problem of excessive computing resources consumed in the production and verification of acoustic holographic lenses in the process of transcranial focusing through acoustic holographic lenses, and can generate acoustic holographic lenses quickly and in real time.

[0004] The technical solution provided by the present invention is: a method for rapidly generating a transcranial focusing acoustic holographic lens based on deep learning, comprising the following steps:

[0005] Step S1, using the k-wave toolbox in Matlab, using the time inversion method to set the preset focal point as the sound source, capturing the phase at the ultrasonic transducer plane, and designing the acoustic holographic lens;

[0006] Step S2, using the preset focus position in step S1 and the designed holographic lens as a data set to train a U-net network, wherein the input of the U-net network is the preset focus position, and the output is the designed holographic lens;

[0007] Step S3, setting a target focal point, inputting the trained U-net network, and obtaining a holographic lens corresponding to the target focal point from the trained U-net network output;

[0008] Step S4, outputting the holographic lens design based on the U-net network, and manufacturing the holographic lens through 3D printing.

[0009] Preferably, in step S1, the acoustic holographic lens is designed by the time reversal method, the target focus position is preset, the focus is used as the sound source in the simulation, the signal is received transcranially at the ultrasonic transducer placement position, the signal is phase extracted, and the extracted phase information is converted into the height spectrum of each pixel of the acoustic holographic lens according to the following formula: h(i,j)=w(i,j).c water. c lens / 2πf.(c lens -c water ), where w(i,j) is the phase corresponding to pixel (i,j), c water and c lens is the speed of sound in water and in the holographic lens material, and f is the center frequency of ultrasound, which gives the height of each pixel of the holographic lens.

[0010] Preferably, the U-net network is composed of upsampling and downsampling modules. The downsampling module reduces the dimension of a fixed-size image to make it fit the size of the display area and obtains deeper image features. The upsampling module enlarges the image based on this and fuses the features of each layer by copying and cropping. The U-net network jump connection can effectively fuse deep and shallow layer information of each scale, thereby improving network training efficiency.

[0011] Preferably, the loss function of the U-net network is: in is the nth hologram calculated by the time reversal method, as the true value, is the hologram output by the corresponding U-net network, N represents the number of target images in the small batch training scale, and the optimizer is used to minimize the loss function during the training process. After each training, the network parameters will be automatically updated after iterative optimization.

[0012] Preferably, a U-net network is trained, the target focus image is used as the input of the U-net network, and the Ground Truth is a hologram generated by a time reversal method as the output of the U-net network. In the simulation, the image size is 64×64 pixels, and the pixel size of a single pixel is 750 μm, with a size of 48 mm×48 mm.

[0013] Preferably, the U-net network framework is implemented using Tensorflow 2.13 and Python 3.8. The U-net network framework is trained and tested on a Tesla V100-16G graphics processor, the encoder is optimized using Adam optimizer, the learning rate is set to 0.001, the mini batch is set to 16, and the training is performed 1500 times.

[0014] Preferably, after the U-net network training is completed, the U-net network is verified using the data in the test set, and the structural similarity SSIM between the acoustic hologram output by the U-net network and the acoustic hologram generated by the time reversal method is calculated. The calculation formula of the structural similarity SSIM is:

[0015] where μ x is the acoustic hologram generated by the proposed network, μ y is the acoustic hologram generated by the time reversal method, σ x and σ y Represent their respective variances, σ xy It represents the covariance between the two images. c1 and c2 are two regularization parameters. These two parameters are introduced to ensure that the structural similarity SSIM will not make mistakes when encountering a value of 0 during calculation, which improves the numerical stability of the calculation. c1 and c2 are defined as c1 = (α × L) 2 , c2=(β×L) 2 , where L=64, α=0.01, β=0.03.

[0016] Based on the same concept, the present invention also provides a deep learning-based transcranial focusing acoustic holographic lens rapid generation system, comprising:

[0017] A data acquisition module is used to use the k-wave toolbox in Matlab to set the preset focal point as the sound source using the time inversion method and capture the phase at the ultrasonic transducer plane to design the acoustic holographic lens;

[0018] A training module, used to train a U-net network using the preset focus position in the step data acquisition module and the designed holographic lens as a data set, wherein the input of the U-net network is the preset focus position, and the output is the designed holographic lens;

[0019] A lens generation module, used to set a target focal point, input the trained U-net network, and obtain a holographic lens corresponding to the target focal point from the trained U-net network output;

[0020] The lens printing module is used to output the holographic lens design based on the U-net network and manufacture the holographic lens through 3D printing.

[0021] Based on the same concept, the present invention also provides an electronic device, comprising:

[0022] A memory, the memory being used to store a processing program;

[0023] A processor, wherein when executing the processing program, the processor implements any one of the above-mentioned methods for rapidly generating a transcranial focusing acoustic holographic lens based on deep learning.

[0024] 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 methods for rapidly generating a transcranial focusing acoustic holographic lens based on deep learning is implemented.

[0025] Due to the adoption of the above technical solution, the present invention has the following advantages and positive effects compared with the prior art:

[0026] The network parameters are trained by deep learning methods to replace the three-dimensional simulation performed in the computer to improve the calculation speed. It has been verified that the effect of reconstructing the focus point is equivalent to that of the traditional time reversal method, and the acoustic holographic lens for transcranial focusing can be quickly produced. In this embodiment, the data of the acoustic holographic lens used for transcranial focusing can be produced at a calculation speed of more than ten milliseconds through the preset focus position through the deep learning model, and after comparison, the reconstruction result of the acoustic holographic lens generated by this method is equivalent to the reconstruction result of the acoustic holographic lens generated by the traditional time reversal method, and the generation speed is greatly improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] The specific embodiments of the present invention are further described in detail below with reference to the accompanying drawings, wherein:

[0028] Figure 1 This is a flow chart of a method for rapidly generating a transcranial focusing acoustic holographic lens based on deep learning according to an embodiment of the present invention;

[0029] Figure 2 It is a schematic diagram of the U-net network structure of the present invention;

[0030] Figure 3 The phase information of the holographic surface used to make the acoustic holographic lens;

[0031] Figure 4 This is the focusing effect of the holographic lens generated in real time by the U-net network in the simulation and experiment. DETAILED DESCRIPTION

[0032] 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.

[0033] 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.

[0034] First embodiment

[0035] like Figure 1 As shown, a method for quickly generating a transcranial focusing acoustic holographic lens based on deep learning in this embodiment includes the following steps:

[0036] Step S1, using the k-wave toolbox in Matlab, using the time inversion method to set the preset focal point as the sound source, capturing the phase at the ultrasonic transducer plane, and designing the acoustic holographic lens;

[0037] Step S2, using the preset focus position in step S1 and the designed holographic lens as a data set to train a U-net network, wherein the input of the U-net network is the preset focus position, and the output is the designed holographic lens;

[0038] Step S3, setting a target focal point, inputting the trained U-net network, and obtaining a holographic lens corresponding to the target focal point from the trained U-net network output;

[0039] Step S4, outputting the holographic lens design based on the U-net network, and manufacturing the holographic lens through 3D printing.

[0040] The present invention uses a deep learning method to train network parameters to replace the three-dimensional simulation performed in the computer to improve the calculation speed. It has been verified that the effect of reconstructing the focus point is equivalent to that of the traditional time reversal method, and the acoustic holographic lens for transcranial focusing can be quickly produced. In this embodiment, the data of the acoustic holographic lens used for transcranial focusing can be produced at a calculation speed of more than ten milliseconds through a preset focus position through a deep learning model, and after comparison, the reconstruction result of the acoustic holographic lens generated by this method is equivalent to the reconstruction result of the acoustic holographic lens generated by the traditional time reversal method, and the generation speed is greatly improved.

[0041] Preferably, in step S1, the acoustic holographic lens is designed by the time reversal method, the target focus position is preset, the focus is used as the sound source in the simulation, the signal is received transcranially at the ultrasonic transducer placement position, the signal is phase extracted, and the extracted phase information is converted into the height spectrum of each pixel of the acoustic holographic lens according to the following formula: h(i,j)=w(i,j).c water .c lens / 2πf.(c lens -c water ), where w(i,j) is the phase corresponding to pixel (i,j), c water and c lens is the speed of sound in water and in the holographic lens material, and f is the center frequency of ultrasound, which gives the height of each pixel of the holographic lens.

[0042] Step S1, using the k-wave toolbox in Matlab, using the time inversion method, setting the preset focal point as the sound source, capturing the phase at the plane where the ultrasonic transducer is placed, to design the acoustic holographic lens. In the simulation process of this embodiment, the grid pixel size is set to 250um, the simulation area size is 192×192×220 pixels, the ultrasonic center frequency is 2MHz, and the Courant number is 0.2. Based on this, the holographic lens is designed by the time inversion method. First, the target focus position is preset, and the focus is used as the sound source in the simulation. After the transcranial, the signal is received at the position where the ultrasonic transducer is placed, and the phase of the signal is extracted. The phase of each pixel point of the holographic surface is set to the arithmetic mean of the surrounding three pixel points. The position of the preset focus in step S1 and the designed holographic lens are used as data sets, and the U-net network is trained. The input is the preset focus position, and the output is the phase map. The height of each pixel point of the acoustic holographic lens can be calculated by the phase map.

[0043] Preferably, the U-net network is composed of upsampling and downsampling modules. The downsampling module reduces the dimension of a fixed-size image to make it fit the size of the display area and obtains deeper image features. The upsampling module enlarges the image based on this and fuses the features of each layer by copying and cropping. The U-net network jump connection can effectively fuse deep and shallow layer information of each scale, thereby improving network training efficiency.

[0044] U-net network architecture Figure 2As shown in the figure, it mainly consists of upsampling and downsampling. The main idea is to reduce the dimension of a fixed-size image to make it fit the size of the display area, and generate a thumbnail of the corresponding image to extract deeper image features. Then, upsampling is used to enlarge the image, and the features of each layer of downsampling and upsampling are fused by copying and cropping. Finally, the convolution layer can learn to combine more accurate outputs from this information. An important advantage of the U-net network is that the upsampling part has a large number of feature channels, which enables the network to propagate contextual information to higher resolution layers. We prepared 10,000 sets of data sets, of which 9,000 sets were used as training sets and 1,000 sets were used as test sets. The input and output sizes were both 64×64 pixels. The network was trained, and the loss function used was the MSE loss function.

[0045] Preferably, the loss function of the U-net network is: in is the nth hologram calculated by the time reversal method, as the true value, is the hologram output by the corresponding U-net network, N represents the number of target images in the small batch training scale, and the optimizer is used to minimize the loss function during the training process. After each training, the network parameters will be automatically updated after iterative optimization.

[0046] The mean squared error (MSE) is used as the loss function in the network. The mean squared error loss function guides the network training process by minimizing the squared difference between the predicted value and the true value, so that the similarity between the generated hologram and the target image is maximized. In this way, high-quality holograms suitable for transcranial focusing can be effectively generated. An optimizer is used to minimize the loss function during training. After each training, the network parameters are automatically updated after iterative optimization.

[0047] Preferably, a U-net network is trained, the target focus image is used as the input of the U-net network, and the Ground Truth is a hologram generated by a time reversal method as the output of the U-net network. In the simulation, the image size is 64×64 pixels, and the pixel size of a single pixel is 750 μm, with a size of 48 mm×48 mm.

[0048] Preferably, the U-net network framework is implemented using Tensorflow 2.13 and Python 3.8. The U-net network framework is trained and tested on a Tesla V100-16G graphics processor, the encoder is optimized using Adam optimizer, the learning rate is set to 0.001, the mini batch is set to 16, and the training is performed 1500 times.

[0049] In order to train the network, 20,000 pairs of data were prepared. Each set of data includes a target focus image and an acoustic hologram generated by time reversal at a frequency of 2 MHz (as ground truth). The network takes the target focus image as input and outputs the corresponding acoustic hologram, which is suitable for an ultrasonic frequency of 2 MHz. For the network, the target focus image (64×64 binary image) is the input of the network, and the ground truth is a hologram generated by time reversal. Among them, 18,000 sets of data are used as training sets to train the network, and 2,000 sets of data are used as test sets. In the simulation, the image size is 64×64 pixels, and the pixel size of a single pixel is 750 μm, so the actual size is 48 mm×48 mm. The network framework is implemented using Tensorflow 2.13

[78] and Python 3.8. The framework was trained and tested on the graphics processing unit (GPU) of Tesla V100-16G. The encoder is optimized using the Adam optimizer

[79] with the learning rate set to 0.001 and the mini-batch size set to 16. Training is performed for 1500 epochs.

[0050] Preferably, after the U-net network training is completed, the U-net network is verified using the data in the test set, and the structural similarity SSIM between the acoustic hologram output by the U-net network and the acoustic hologram generated by the time reversal method is calculated. The calculation formula of the structural similarity SSIM is:

[0051] where μ x is the acoustic hologram generated by the proposed network, μ y is the acoustic hologram generated by the time reversal method, σ x and σ y Represent their respective variances, σ xy It represents the covariance between the two images. c1 and c2 are two regularization parameters. These two parameters are introduced to ensure that the structural similarity SSIM will not make mistakes when encountering a value of 0 during calculation, which improves the numerical stability of the calculation. c1 and c2 are defined as c1 = (α × L) 2 , c2=(β×L) 2 , where L=64, α=0.01, β=0.03.

[0052] In this embodiment, as the number of loop iterations increases, the loss function gradually converges. After 1500 loop iterations, the model has completely converged. The SSIM value ranges from -1 to 1. The closer it is to 1, the higher the similarity between the two images. In the performance of the 2000 test sets, the average SSIM value is 0.972.

[0053] Step S3, select several groups of holographic lenses generated by the network with preset focus in the test set for verification in simulation, and the phase diagram of the generated acoustic holographic lens is as follows: Figure 3 As shown, the phase map is then converted into the height of each pixel of the acoustic holographic lens and manufactured via 3D printing.

[0054] In step S4, the settings in the simulation are consistent with those in step S1. In the experiment, the size of the ultrasonic transducer is 5 cm × 5 cm, and it emits a sine wave at 2 MHz. The sine wave is phase modulated after passing through the holographic lens, and then focuses behind the skull after passing through the skull. The results after simulation and experimental reconstruction are shown in Figure 2. Figure 4 As shown, it can be seen that the positions of the two in the numerical simulation and the real experiment are basically the same. The present invention selects the SSIM value and PSNR value of the target point position and the normalized sound field intensity reconstructed in the simulation and experiment as quantitative indicators to evaluate the quality of the reconstructed sound field. In the simulation, the SSIM value of the reconstruction result of the acoustic holographic lens made by the time reversal method based on the traditional method has an average value of 0.91 in all examples, and the PSNR value is 32.69, while the SSIM value of the reconstruction result of the acoustic holographic lens made based on the U-net method is 0.82, and the PSNR value is 32.94. In the experiment, the SSIM value of the reconstruction effect of the acoustic holographic lens made based on the traditional time reversal method is 0.81 in all examples, and the PSNR value is 27.47. The reconstruction effect of the acoustic holographic lens made by the deep learning method based on U-net is 0.74, and the PSNR value is 26.96. It is not difficult to see that the two perform equally well in reconstructing the transcranial focused sound field. However, it takes about 120 seconds to generate an acoustic holographic lens for transcranial focusing based on the traditional time reversal method in the computing environment of NVIDIA Tesla V100-16G, but it only takes 15ms to generate an acoustic holographic lens for transcranial focusing based on the U-net method, which increases the generation speed by several orders of magnitude.

[0055] Second embodiment

[0056] Based on the same concept, the present invention also provides a deep learning-based transcranial focusing acoustic holographic lens rapid generation system, comprising:

[0057] A data acquisition module is used to use the k-wave toolbox in Matlab to set the preset focal point as the sound source using the time inversion method and capture the phase at the ultrasonic transducer plane to design the acoustic holographic lens;

[0058] A training module, used to train a U-net network using the preset focus position in the step data acquisition module and the designed holographic lens as a data set, wherein the input of the U-net network is the preset focus position, and the output is the designed holographic lens;

[0059] A lens generation module, used to set a target focal point, input the trained U-net network, and obtain a holographic lens corresponding to the target focal point from the trained U-net network output;

[0060] The lens printing module is used to output the holographic lens design based on the U-net network and manufacture the holographic lens through 3D printing.

[0061] 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 methods for rapidly generating a transcranial focusing acoustic holographic lens based on deep learning when executing the processing program.

[0062] 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 methods for rapidly generating a transcranial focusing acoustic holographic lens based on deep learning is implemented.

[0063] If the rapid generation method of transcranial focusing acoustic holographic lens based on deep learning 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 part of the contribution to the prior art or all or part of the technical solution can be embodied in the form of software. The computer software is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, server, or 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, etc., and other media that can store program codes.

[0064] 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.

[0065] 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 method for rapidly generating transcranial focusing acoustic holographic lens based on deep learning, characterized in that: The following steps are involved: Step S1, using the k-wave toolbox in Matlab, using the time inversion method to set the preset focal point as the sound source, capturing the phase at the ultrasonic transducer plane, and designing the acoustic holographic lens; Step S2, using the preset focus position in step S1 and the designed holographic lens as a data set to train a U-net network, wherein the input of the U-net network is the preset focus position, and the output is the designed holographic lens; Step S3, setting a target focal point, inputting the trained U-net network, and obtaining a holographic lens corresponding to the target focal point from the trained U-net network output; Step S4, outputting the holographic lens design based on the U-net network, and manufacturing the holographic lens through 3D printing.

2. The method for rapidly generating a transcranial focusing acoustic holographic lens based on deep learning according to claim 1 is characterized in that: In step S1, the acoustic holographic lens is designed by the time reversal method. The target focus position is preset first. The focus is used as the sound source in the simulation. The signal is received transcranially at the position where the ultrasonic transducer is placed. The phase of the signal is extracted. The extracted phase information is converted into the height spectrum of each pixel of the acoustic holographic lens according to the following formula: h(i,j)=w(i,j).c water .c lens / 2πf.(c lens -c water ), where w(i,j) is the phase corresponding to pixel (i,j), c water and c lens is the speed of sound in water and in the holographic lens material, and f is the center frequency of ultrasound, which gives the height of each pixel of the holographic lens.

3. The method for rapidly generating a transcranial focusing acoustic holographic lens based on deep learning according to claim 1 is characterized in that: The U-net network consists of upsampling and downsampling modules. The downsampling module reduces the dimension of a fixed-size image to make it fit the size of the display area and obtains deeper image features. The upsampling module enlarges the image based on this and fuses the features of each layer by copying and cropping. The U-net network jump connection can effectively fuse deep and shallow layer information of each scale, thereby improving network training efficiency.

4. The method for rapidly generating a transcranial focusing acoustic holographic lens based on deep learning according to claim 3 is characterized in that: The loss function of the U-net network is: in is the nth hologram calculated by the time reversal method, as the true value, is the hologram output by the corresponding U-net network, N represents the number of target images in the small batch training scale, and the optimizer is used to minimize the loss function during the training process. After each training, the network parameters will be automatically updated after iterative optimization.

5. The method for rapidly generating a transcranial focusing acoustic holographic lens based on deep learning according to claim 4 is characterized in that: The U-net network is trained, the target focus image is used as the input of the U-net network, and the Ground Truth is a hologram generated by the time reversal method as the output of the U-net network. In the simulation, the image size is 64×64 pixels, and the pixel size of a single pixel is 750μm, with a size of 48mm×48mm.

6. The method for rapidly generating a transcranial focusing acoustic holographic lens based on deep learning according to claim 5 is characterized in that: The U-net network framework was implemented using Tensorflow 2.13 and Python 3.

8. The U-net network framework was trained and tested on a Tesla V100-16G graphics processor. The encoder was optimized using the Adam optimizer, the learning rate was set to 0.001, the mini batch was set to 16, and the training was performed 1500 times.

7. The method for rapidly generating a transcranial focusing acoustic holographic lens based on deep learning according to claim 1, characterized in that: After the U-net network training is completed, the U-net network is verified using the data in the test set, and the structural similarity SSIM between the acoustic hologram output by the U-net network and the acoustic hologram generated by the time reversal method is calculated. The calculation formula of the structural similarity SSIM is: where μ x is the acoustic hologram generated by the proposed network, μ y is the acoustic hologram generated by the time reversal method, σ x and σ y Represent their respective variances, σ xy It represents the covariance between the two images. c1 and c2 are two regularization parameters. These two parameters are introduced to ensure that the structural similarity SSIM will not make mistakes when encountering a value of 0 during calculation, which improves the numerical stability of the calculation. c1 and c2 are defined as c1 = (α × L) 2 , c2=(β×L) 2 , where L=64, α=0.01, β=0.

03.

8. A deep learning-based transcranial focusing acoustic holographic lens rapid generation system, characterized in that: include: A data acquisition module is used to use the k-wave toolbox in Matlab to set the preset focal point as the sound source using the time inversion method and capture the phase at the ultrasonic transducer plane to design the acoustic holographic lens; A training module, used to train a U-net network using the preset focus position in the step data acquisition module and the designed holographic lens as a data set, wherein the input of the U-net network is the preset focus position, and the output is the designed holographic lens; A lens generation module, used to set a target focal point, input the trained U-net network, and obtain a holographic lens corresponding to the target focal point from the trained U-net network output; The lens printing module is used to output the holographic lens design based on the U-net network and manufacture the holographic lens through 3D printing.

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 deep learning-based transcranial focusing acoustic holographic lens rapid generation method 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 the processor, the method for rapidly generating a transcranial focusing acoustic holographic lens based on deep learning as described in any one of claims 1 to 7 is implemented.

Images