Self-adaptive sound velocity tuning method and system for photoacoustic imaging

Abstract

The invention relates to the technical field of photoacoustic imaging, and provides a photoacoustic imaging sound velocity adaptive tuning method and system, and the method comprises the steps: S1, reconstructing a photoacoustic image based on a current sound velocity signal; s2, extracting a high-dimensional feature vector from the photoacoustic image based on an improved ViT model, predicting a sound velocity correction value according to the high-dimensional feature vector, and iteratively updating a sound velocity signal based on the sound velocity correction value; and S3, circulating S1 to S2 until an iteration termination condition is reached, and obtaining an optimized sound velocity signal. According to the method, the sound velocity error is regressed from the second level of the image by using the ViT model, the speed error can be reduced to be within 1m / s by performing closed-loop iteration for less than or equal to 5 times, and extra probes or manual test parameters and single-frame processing are not needed; and the image resolution and the contrast ratio are remarkably improved, so that the accurate sound velocity tuning is changed from expert operation to equipment standard configuration.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of photoacoustic imaging, and more particularly to a photoacoustic imaging sound velocity adaptive tuning method and system. BACKGROUND

[0002] Photoacoustic Imaging (PAI) is a technique that uses pulsed laser irradiation to excite ultrasonic waves in biological tissues. The ultrasonic waves are then received by an array of ultrasonic transducers and reconstructed into an image, achieving high contrast and high resolution imaging. The mainstream clinical equipment uses a back-projection algorithm for real-time imaging. The back-projection algorithm is an imaging algorithm that uses the position of the object and the distance of the transducer array. It analyzes the received ultrasonic signals in each transducer and calculates the sound pressure intensity at each position. This algorithm finds the ultrasonic signals emitted by the object at each transducer based on the position of the object and the distance of the transducer array, calculates the gradient of the signals, and then superimposes the gradients to determine the relative intensity of the sound pressure. By repeating this process, a photoacoustic image can be obtained. Through this image, the sound pressure intensity distribution at different positions can be clearly observed, and the imaging and analysis of the object structure can be realized. This algorithm is highly dependent on the sound velocity parameter in the medium. When the sound velocity error is large, the full width at half maximum of the reconstructed image expands, resulting in blurred images, structural misplacement, and decreased contrast of small structures.

[0003] Currently, the direction of optimizing the sound velocity parameter is generally artificial experience tuning. This method requires the operator to repeatedly adjust the sound velocity and visually assess the image quality, which is time-consuming and highly subjective. Alternatively, a sound velocity mapping based on ultrasound is used. This method requires the integration of an additional ultrasound probe for synchronous scanning, increasing the time cost. Another method is to match the tissue-temperature sound velocity database, but this method cannot adapt to individual differences and pathological state changes, resulting in high errors.

[0004] Therefore, there is an urgent need for a new solution that can automatically estimate and iteratively optimize the sound velocity within the reconstruction closed loop using the photoacoustic signal itself. SUMMARY

[0005] The present application provides a photoacoustic imaging sound velocity adaptive tuning method and system to solve the problem of blurred, misplaced, and low-efficiency photoacoustic images due to the extreme sensitivity of photoacoustic back-projection imaging to sound velocity and the difficulty of existing solutions to balance real-time, precision, and zero hardware increment.

[0006] According to a first aspect of the present application, a photoacoustic imaging sound velocity adaptive tuning method is provided, comprising: S1, reconstructing a photoacoustic image based on a current sound velocity signal; S2, extract high-dimensional feature vectors from the photoacoustic image based on the improved ViT model, predict the sound speed correction amount based on the high-dimensional feature vectors, and iteratively update the sound speed signal based on the sound speed correction amount; S3, repeat S1~S2 until the iteration termination condition is met, and the optimized sound speed signal is obtained.

[0007] Based on the above technical solution, the present invention can also be improved as follows.

[0008] Optionally, in step S1, the photoacoustic image is reconstructed using a delay-and-sum (DAS) algorithm.

[0009] Optionally, the photoacoustic image can be reconstructed using the following formula:

[0010] in, The signal represents the synthesized photoacoustic image signal, where i is the transducer number and N is the number of transducers. Let i be the weight of the i-th transducer. Let r be the signal sequence of the i-th transducer, and r be the position vector of the target pixel in the photoacoustic image. Given the known coordinates of the i-th transducer element, Let the speed of sound be the speed of sound in the k-th iteration. This indicates that the sound wave arrives at the array element from the image point r. The required flight time for the original signal s i Extract the corresponding sampled value from (t).

[0011] Optionally, improvements to the ViT model may also be included, specifically: Modify the input channel of the Vision Transformer model to a single-channel grayscale input; The optimization objective of the Vision Transformer model is set as minimizing the mean square error between the predicted and actual sound speed values. The classification decision layer of the Vision Transformer model was modified to a linear regression layer to output continuous sound speed correction values. The improved Vision Transformer model was pre-trained using the ImageNet-21k dataset.

[0012] Optionally, the extraction of high-dimensional feature vectors from the photoacoustic image based on the improved ViT model includes: A single-channel grayscale photoacoustic image is input into the encoder module of the ViT model, and a non-overlapping sliding window is used to crop the input photoacoustic image into N sub-images of size P×P. Flatten each subgraph into a P shape by row.2 A 1-dimensional vector is then linearly projected onto a multi-dimensional feature space to obtain N multi-dimensional feature vectors. Each of the multidimensional feature vectors is given a unique positional code to clarify the spatial order of each multidimensional feature vector in the original image; The N multidimensional feature vectors with positional encoding are concatenated to obtain a learnable classification feature vector for global representation, namely the high-dimensional feature vector F.

[0013] Optionally, the step of predicting the sound velocity correction amount based on the high-dimensional feature vector and iteratively updating the sound velocity signal based on the sound velocity correction amount includes: The high-dimensional feature vector F is fed into the sound velocity correction regression branch of the improved ViT model. The high-dimensional feature vector F is linearly transformed using a first fully connected layer to obtain the dimension-reduced latent feature H; The hidden feature H is mapped to a scalar, namely the sound speed correction amount Δc, using a second fully connected layer; The sound speed correction value Δc is added to the current sound speed signal to obtain an updated sound speed signal.

[0014] Optionally, the iteration termination condition is: The sound velocity correction amount is less than the threshold, or the maximum number of iterations is reached.

[0015] According to a second aspect of the present invention, a photoacoustic imaging sound velocity adaptive tuning system is provided, comprising: The reconstruction module is used to reconstruct the photoacoustic image based on the current sound speed signal; The update module is used to extract high-dimensional feature vectors from the photoacoustic image based on the improved ViT model, predict sound velocity correction based on the high-dimensional feature vectors, and iteratively update the sound velocity signal based on the sound velocity correction. The iterative control module controls the reconstructing and updating modules to run in a loop until the iteration termination condition is met, thus obtaining the optimized sound velocity signal.

[0016] According to a third aspect of the present invention, an electronic device is provided, including a memory and a processor, wherein the processor is configured to execute a computer management program stored in the memory to implement the steps of the above-described photoacoustic imaging sound speed adaptive tuning method.

[0017] According to a fourth aspect of the present invention, a computer-readable storage medium is provided, on which a computer management program is stored, wherein the computer management program, when executed by a processor, implements the steps of the above-described photoacoustic imaging sound speed adaptive tuning method.

[0018] This invention provides a photoacoustic imaging sound velocity adaptive tuning method, system, electronic device, and storage medium. Based on the ViT model, it regresses the sound velocity error from the photoacoustic image itself at the second level. The sound velocity error can be reduced to a very small range (e.g., within 1 m / s) with ≤5 closed-loop iterations. It completely reuses existing photoacoustic channels and transducers, without the need for additional probes or manual parameter testing. It has high processing efficiency, with a single frame processing time of <1s. The image resolution and contrast are significantly improved, making precise sound velocity tuning a standard feature of equipment instead of an "expert operation". While improving imaging accuracy and efficiency, it reduces the complexity of operation and saves labor and time costs. Attached Figure Description

[0019] Figure 1 A flowchart of a photoacoustic imaging sound velocity adaptive tuning method provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of a classic ViT model; Figure 3 A block diagram of a photoacoustic imaging sound velocity adaptive tuning system provided in an embodiment of the present invention; Figure 4 A schematic diagram of the hardware structure of a possible electronic device provided by the present invention; Figure 5 This is a schematic diagram of the hardware structure of a possible computer-readable storage medium provided by the present invention. Detailed Implementation

[0020] The specific embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples. The following examples are for illustrative purposes only and are not intended to limit the scope of the invention.

[0021] Figure 1 A flowchart of a photoacoustic imaging sound velocity adaptive tuning method provided by the present invention is shown below. Figure 1 As shown, the method includes steps S1 to S3: S1, Reconstruct the photoacoustic image based on the current sound speed signal; S2, extract high-dimensional feature vectors from the photoacoustic image based on the improved ViT model, predict the sound speed correction amount based on the high-dimensional feature vectors, and iteratively update the sound speed signal based on the sound speed correction amount; S3, repeat S1~S2 until the iteration termination condition is met, and the optimized sound speed signal is obtained.

[0022] Understandably, given the shortcomings in the background technology, this invention proposes an adaptive sound velocity tuning method for photoacoustic imaging. This method proposes a closed-loop optimized adaptive sound velocity tuning framework, applicable to commonly used photoacoustic reconstruction algorithms and easily transferable; based on the ViT model, it regresses the sound velocity error from the photoacoustic image itself at the second level, and the sound velocity error can be compressed to a very small range (e.g., within 1 m / s) with ≤5 closed-loop iterations; it completely reuses existing photoacoustic channels and transducers, eliminating the need for additional probes or manual parameter testing; it has high processing efficiency, with a single frame processing time of <1 second; and it significantly improves image resolution and contrast, transforming precise sound velocity tuning from "expert operation" to "standard equipment feature."

[0023] This invention improves imaging accuracy and efficiency while reducing operational complexity and saving labor and time costs.

[0024] In one possible embodiment, in step S1, the photoacoustic image is reconstructed using a Delay-and-Sum (DAS) algorithm. For example, the photoacoustic image is reconstructed using the following formula:

[0025] in, The signal represents the synthesized photoacoustic image signal, where i is the transducer number and N is the number of transducers. Let i be the weight of the i-th transducer. Let r be the signal sequence of the i-th transducer, and r be the position vector of the target pixel in the photoacoustic image. Given the known coordinates of the i-th transducer element, Let be the speed of sound in the k-th iteration, when k=0. This represents the default initial sound speed signal. This indicates that the sound wave arrives at the array element from the image point r. The required flight time for the original signal s i Extract the corresponding sampled value from (t).

[0026] Understandably, the Delay-and-Sum (DAS) algorithm is the most intuitive photoacoustic reconstruction method. Its principle is to treat each pixel in the photoacoustic image as a "virtual sound source" and calculate its flight time to all transducers. Then, the sampled values ​​of each channel at that moment are weighted and summed, and the resulting sum is the initial sound pressure level of that pixel. By traversing the entire imaging area, a photoacoustic image is obtained.

[0027] In this embodiment, the reconstructed photoacoustic image may be based on a default initial sound velocity signal or on an iteratively obtained sound velocity signal. For example, the default sound velocity signal is used during the initial image reconstruction; during iterative updates, the new sound velocity signal obtained from the previous iteration is used for image reconstruction. Experiments have verified that the reconstruction time for a single frame of photoacoustic image in this embodiment is less than 100ms, demonstrating high imaging efficiency.

[0028] like Figure 2 The diagram shows the classic Vision Transformer Base / 16 (ViT_b_16) model structure. To implement the method of this invention and make the ViT model more suitable for photoacoustic image tuning, this embodiment of the invention also improves the ViT model. The improvements mainly include the following four aspects: 1. Modify the input channel of the Vision Transformer model to a single-channel grayscale input.

[0029] It is understandable that, such as Figure 2 As shown, the conventional ViT model input channels are RGB three channels. The RGB three channels contain color information, while the grayscale values ​​of a photoacoustic image represent the distribution of light absorption energy. Their physical meanings are completely different, and forcing the use of three channels will introduce misleading noise. This embodiment changes the RGB three-channel input to a single-channel grayscale input, corresponding to... Figure 2 The Patch Embedding module is used in the Transformer Encoder. This is the first step in getting the data to be processed into the Transformer Encoder. The reason for this modification is that the photoacoustic imaging reconstruction result is a single-channel grayscale image, and changing the input channel to a single channel avoids unnecessary parameter redundancy and computational waste.

[0030] 2. Set the optimization objective of the Vision Transformer model to minimize the mean square error between the predicted and actual sound speed values.

[0031] 3. Modify the classification decision layer of the Vision Transformer model to a linear regression layer to output continuous sound speed correction values.

[0032] It is understood that this embodiment is based on... Figure 2 The classification head of the network structure shown has been replaced with a linear regression head, corresponding to... Figure 2The MLP Head module in the Transformer is the final decision layer after the encoder. The modifications in points 2 and 3 are made because the task of this invention is to regress a continuous sound velocity correction value (Δc, unit: m / s), rather than classification. The regression head (i.e., the linear regression layer) directly outputs numerical values, perfectly matching the task objective. Furthermore, transforming the classification problem into a regression problem changes the model optimization objective from "maximizing the classification probability" to "minimizing the mean squared error between the predicted and actual sound velocity values," making model training more direct and more accurate.

[0033] In a certain implementation scenario, the improved Vision Transformer model is also preset to have a cropping scale of 16×16 pixels when cropping the photoacoustic image into multiple sub-image patches, in order to match the resolution of photoacoustic imaging; there are 12 encoder layers to balance depth and real-time performance; and the hidden layer dimension is set to 768 to give it strong high-dimensional feature representation capabilities.

[0034] 4. The improved Vision Transformer model was pre-trained using the ImageNet-21k dataset.

[0035] Understandably, the improved Vision Transformer model, trained on the ImageNet-21k dataset, learns general texture features. Pre-training significantly reduces training time and improves subsequent model efficiency. Experiments have verified that the model converges quickly during training, requiring minimal time investment. Convergence occurs in no more than 30 epochs, with a training time of less than 10 hours.

[0036] In one possible embodiment, step S2, in which the high-dimensional feature vector is extracted from the photoacoustic image based on the improved ViT model, includes sub-steps S201 to S204.

[0037] S201, inputs a single-channel grayscale photoacoustic image into the encoder module of the ViT model, and uses a non-overlapping sliding window to crop the input photoacoustic image into N sub-image patches of size P×P.

[0038] For example, if the input photoacoustic image is uniformly sized as H×W (e.g., 224×224 pixels), and a non-overlapping sliding window is used to cut the input photoacoustic image into small squares of fixed size 16×16 pixels, a total of N=(H / 16)×(W / 16) sub-image patches can be obtained. Then, the 224×224 pixel input image can yield 14×14=196 sub-image patches, and each sub-image patch is a "visual word".

[0039] S202, flatten each subgraph Patch into P by row. 2The 1-dimensional vector is then linearly projected onto a multi-dimensional feature space to obtain N multi-dimensional feature vectors.

[0040] For example, each 16×16 pixel sub-image patch is stretched into a 256-dimensional vector by rows, and then mapped to a 768-dimensional feature space by multiplying and adding biases through a learnable projection matrix, forming 196 768-dimensional vectors, called patch tokens. That is, 196 multi-dimensional feature vectors are obtained through this step.

[0041] S203, add a unique position code to each of the multidimensional feature vectors to clarify the spatial order of each multidimensional feature vector in the original image.

[0042] For example, by adding a unique positional code (a 768-dimensional learnable vector) to each 768-dimensional vector (patch token), the ViT model can know the spatial order of each patch in the original graph.

[0043] S204, based on step S203, an additional learnable classification feature vector (cls token) for global representation is concatenated from the N multidimensional feature vectors with positional encoding.

[0044] After this step, the total sequence length of the multidimensional feature vectors becomes N+1, meaning the total number of multidimensional feature vectors changes from 196 to 197. The sequence of these 197 768-dimensional vectors is then sequentially fed into a 12-layer Transformer Encoder. Each layer uses multi-head self-attention and a feedforward network to allow global context information to interact fully between different blocks, ultimately outputting 197 768-dimensional vectors of the same length.

[0045] The 768-dimensional vector of the learnable classification feature vector cls token corresponding to the last layer is taken as the compressed representation of the entire photoacoustic image, that is, the required high-dimensional feature vector F, which can be directly used for subsequent sound speed migration regression tasks.

[0046] Understandably, the image features extracted in step S2 mainly include the structural and texture features of the tested object. Structural features primarily include macroscopic structural information such as vascular networks, tissue boundaries, and lesion morphology; distortion of structural information is directly related to sound velocity error. Texture features are mainly used to quantify image uniformity, contrast, and entropy; abnormal textures are often caused by artifacts resulting from sound velocity mismatch. Therefore, optimizing the sound velocity signal based on the extracted high-dimensional feature vector F is of significant importance.

[0047] In one possible embodiment, step S2, which involves predicting the sound velocity correction amount based on the high-dimensional feature vector and iteratively updating the sound velocity signal based on the sound velocity correction amount, includes sub-steps S205 to S209.

[0048] S205, Feature Preparation: The high-dimensional feature vector F is fed into the sound velocity correction regression branch of the improved ViT model without any additional convolution or pooling.

[0049] Flattening and Dimension Check: If the high-dimensional feature vector F still maintains a sequence form (e.g., 197×768), then only the 768-dimensional vector corresponding to the position of the learnable classification feature vector cls token used for global representation is extracted; if it has already been extracted, then the vector is used directly.

[0050] S206, First Fully Connected Dimensionality Reduction: The extracted high-dimensional feature vector F is linearly transformed using the first fully connected layer to obtain the dimension-reduced hidden feature H.

[0051] For example, in a certain implementation scenario, a linear transformation is performed between the weight matrix W1 (768→256) and the bias b1: H = ReLU(W1·F + b1), The 256-dimensional latent feature H is obtained, and ReLU is selected as the activation function to introduce nonlinearity.

[0052] S207, First fully connected mapping: The hidden feature H is mapped to a scalar, namely the sound speed correction amount Δc, using a second fully connected layer.

[0053] For example, in a certain implementation scenario, based on S206, the weight matrix W2 (256→1) is further mapped to the bias b2: Δc = W²·H + b² Output a scalar Δc, which is the sound velocity correction amount corresponding to the current image (unit: m / s, range: ±50m / s, continuous and can be positive or negative).

[0054] S208, Range Limitation (Optional): During the training phase, use tanh activation and then multiply by 50 to limit the output range to [-50, +50] to prevent excessively large outlier predictions; during the inference phase, the result of step S207 can be used directly.

[0055] S209, Result Delivery: Add the obtained sound velocity correction Δc to the current sound velocity signal to obtain the updated sound velocity signal: c (k+1) =c k +Δc, Update the speed of sound signal c (k+1) Step S1 of the main process is sent back to complete one closed-loop correction prediction.

[0056] In one possible embodiment, the iteration termination condition for the sound speed signal is: The sound velocity correction amount is less than the threshold ε: |Δc|<ε, and the threshold ε is preferably set to 1m / s; or, The number of iterations k reaches the maximum number of iterations k0, for example, the maximum number of iterations k0 is set to 5.

[0057] In a certain experimental verification scenario, the following steps are performed sequentially: initial sound velocity c o → Back projection reconstruction of photoacoustic image → High-dimensional feature extraction using ViT model → Prediction of sound velocity offset Δc → Update sound velocity c k+1 =c k +Δc→ loop until |Δc|<ε, and the experimental data are shown in Table 1.

[0058] Table 1 Experimental Data

[0059] The above experiments verified that an optimal sound velocity can be obtained after two rounds of iteration (three image reconstructions). In the experiment, the offset in the first round was less than 10 m / s, and the offset in the second round was less than 1 m / s. In extreme cases, the optimized sound velocity signal can be obtained in no more than five rounds of iteration.

[0060] Figure 3 A structural diagram of a photoacoustic imaging sound velocity adaptive tuning system provided in an embodiment of the present invention is shown below. Figure 3 As shown, a photoacoustic imaging sound velocity adaptive tuning system includes a reconstruction module, an update module, and an iterative control module, wherein: The reconstruction module is used to reconstruct the photoacoustic image based on the current sound speed signal; The update module is used to extract high-dimensional feature vectors from the photoacoustic image based on the improved ViT model, predict sound velocity correction based on the high-dimensional feature vectors, and iteratively update the sound velocity signal based on the sound velocity correction. The iterative control module controls the reconstructing and updating modules to run in a loop until the iteration termination condition is met, thus obtaining the optimized sound velocity signal.

[0061] It is understood that the photoacoustic imaging sound speed adaptive tuning system provided by the present invention corresponds to the photoacoustic imaging sound speed adaptive tuning method provided in the foregoing embodiments. The relevant technical features of the photoacoustic imaging sound speed adaptive tuning system can be referred to the relevant technical features of the photoacoustic imaging sound speed adaptive tuning method, and will not be repeated here.

[0062] Please see Figure 4 , Figure 4 A schematic diagram illustrating an embodiment of the electronic device provided in this invention. For example... Figure 4As shown, this embodiment of the invention provides an electronic device 400, including a memory 410, a processor 420, and a computer program 411 stored in the memory 410 and executable on the processor 420. When the processor 420 executes the computer program 411, it performs the following steps: S1, Reconstruct the photoacoustic image based on the current sound speed signal; S2, extract high-dimensional feature vectors from the photoacoustic image based on the improved ViT model, predict the sound speed correction amount based on the high-dimensional feature vectors, and iteratively update the sound speed signal based on the sound speed correction amount; S3, repeat S1~S2 until the iteration termination condition is met, and the optimized sound speed signal is obtained.

[0063] Please see Figure 5 , Figure 5 This is a schematic diagram illustrating an embodiment of a computer-readable storage medium provided by the present invention. (See diagram below.) Figure 5 As shown, this embodiment provides a computer-readable storage medium 500 on which a computer program 511 is stored. When the computer program 511 is executed by a processor, it performs the following steps: S1, Reconstruct the photoacoustic image based on the current sound speed signal; S2, extract high-dimensional feature vectors from the photoacoustic image based on the improved ViT model, predict the sound speed correction amount based on the high-dimensional feature vectors, and iteratively update the sound speed signal based on the sound speed correction amount; S3, repeat S1~S2 until the iteration termination condition is met, and the optimized sound speed signal is obtained.

[0064] This invention provides a photoacoustic imaging sound velocity adaptive tuning method, system, and storage medium. It proposes a closed-loop optimization framework for sound velocity adaptive tuning, applicable to commonly used photoacoustic reconstruction algorithms and easily transferable. Based on the ViT model, it regresses the sound velocity error from the photoacoustic image itself at the second level; closed-loop iterations of ≤5 times can reduce the sound velocity error to a very small range (e.g., within 1 m / s). It completely reuses existing photoacoustic channels and transducers, eliminating the need for additional probes or manual parameter testing. It boasts high processing efficiency, with a single frame processing time of <1 second. Image resolution and contrast are significantly improved, transforming precise sound velocity tuning from "expert operation" to "standard equipment feature." This invention improves imaging accuracy and efficiency while reducing operational complexity and saving labor and time costs.

[0065] It should be noted that the descriptions of each embodiment in the above embodiments have different focuses. For parts that are not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0066] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied 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.

[0067] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0068] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0069] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0070] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0071] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A photoacoustic imaging sound velocity adaptive tuning method, characterized in that, include: S1, Reconstruct the photoacoustic image based on the current sound speed signal; S2, extract high-dimensional feature vectors from the photoacoustic image based on the improved ViT model, predict the sound speed correction amount based on the high-dimensional feature vectors, and iteratively update the sound speed signal based on the sound speed correction amount; S3, repeat S1~S2 until the iteration termination condition is met, and the optimized sound speed signal is obtained.

2. The photoacoustic imaging sound velocity adaptive tuning method according to claim 1, characterized in that, In step S1, the photoacoustic image is reconstructed using the Delay-and-Sum (DAS) algorithm.

3. The photoacoustic imaging sound velocity adaptive tuning method according to claim 2, characterized in that, The photoacoustic image is reconstructed using the following formula: in, The signal represents the synthesized photoacoustic image signal, where i is the transducer number and N is the number of transducers. Let i be the weight of the i-th transducer. Let r be the signal sequence of the i-th transducer, and r be the position vector of the target pixel in the photoacoustic image. Given the known coordinates of the i-th transducer element, Let the speed of sound be the speed of sound in the k-th iteration. This indicates that the sound wave arrives at the array element from the image point r. The required flight time for the original signal s i Extract the corresponding sampled value from (t).

4. The photoacoustic imaging sound velocity adaptive tuning method according to claim 1, characterized in that, This also includes improvements to the ViT model, specifically including: Modify the input channel of the Vision Transformer model to a single-channel grayscale input; The optimization objective of the Vision Transformer model is set as minimizing the mean square error between the predicted and actual sound speed values. The classification decision layer of the Vision Transformer model was modified to a linear regression layer to output continuous sound speed correction values. The improved Vision Transformer model was pre-trained using the ImageNet-21k dataset.

5. The photoacoustic imaging sound velocity adaptive tuning method according to claim 1, characterized in that, The improved ViT model extracts high-dimensional feature vectors from the photoacoustic image, including: A single-channel grayscale photoacoustic image is input into the encoder module of the ViT model, and a non-overlapping sliding window is used to crop the input photoacoustic image into N sub-images of size P×P. Flatten each subgraph into a P shape by row. 2 A 1-dimensional vector is then linearly projected onto a multi-dimensional feature space to obtain N multi-dimensional feature vectors. Each of the multidimensional feature vectors is given a unique positional code to clarify the spatial order of each multidimensional feature vector in the original image; The N multidimensional feature vectors with positional encoding are concatenated to obtain a learnable classification feature vector for global representation, namely the high-dimensional feature vector F.

6. The photoacoustic imaging sound velocity adaptive tuning method according to claim 1, characterized in that, The step of predicting the sound velocity correction based on the high-dimensional feature vector and iteratively updating the sound velocity signal based on the sound velocity correction includes: The high-dimensional feature vector F is fed into the sound velocity correction regression branch of the improved ViT model. The high-dimensional feature vector F is linearly transformed using a first fully connected layer to obtain the dimension-reduced latent feature H; The hidden feature H is mapped to a scalar, namely the sound speed correction amount Δc, using a second fully connected layer; The sound speed correction value Δc is added to the current sound speed signal to obtain an updated sound speed signal.

7. The photoacoustic imaging sound velocity adaptive tuning method according to claim 1, characterized in that, The iteration termination condition is: The sound velocity correction amount is less than the threshold, or the maximum number of iterations is reached.

8. A photoacoustic imaging sound velocity adaptive tuning system, characterized in that, include: The reconstruction module is used to reconstruct the photoacoustic image based on the current sound speed signal; The update module is used to extract high-dimensional feature vectors from the photoacoustic image based on the improved ViT model, predict sound speed correction based on the high-dimensional feature vectors, and iteratively update the sound speed signal based on the sound speed correction. The iterative control module controls the reconstructing and updating modules to run in a loop until the iteration termination condition is met, thus obtaining the optimized sound velocity signal.

9. An electronic device, characterized in that, It includes a memory and a processor, wherein the processor is used to implement the steps of the photoacoustic imaging sound velocity adaptive tuning method as described in any one of claims 1-7 when executing a computer management program stored in the memory.

10. A computer-readable storage medium, characterized in that, It stores a computer management program, which, when executed by a processor, implements the steps of the photoacoustic imaging sound speed adaptive tuning method as described in any one of claims 1-7.

Images