Sound velocity diagram reconstruction model training and reconstruction method and device, equipment and storage medium

Abstract

The embodiment of the invention relates to the technical field of image processing, and provides a sound velocity diagram reconstruction model training and reconstruction method and device, equipment and a storage medium, and the training method comprises the steps: obtaining a sound velocity diagram training data set, extracting an ultrasonic radio-frequency signal corresponding to each real sound velocity diagram from the sound velocity diagram training data set, obtaining an ultrasonic radio frequency signal training data set; training a pre-constructed reconstruction model by using the ultrasonic radio frequency signal training data set to obtain a trained reconstruction model; wherein the reconstruction model comprises an encoder, a wavefront attention mechanism module and a decoder, the wavefront attention mechanism module is used for receiving a low-layer feature map and a high-layer feature map from the encoder and the decoder, generating a wavefront attention weight according to the low-layer feature map and the high-layer feature map, mapping the wavefront attention weight to the low-layer feature map, and reconstructing the wavefront attention weight according to the low-layer feature map and the high-layer feature map. Therefore, the perception capability of the reconstruction model on the wavefront distortion area in the ultrasonic radio-frequency signal is enhanced.

Description

Technical Field

[0001] The embodiments in this specification relate to the field of image processing technology, and in particular to a method, apparatus, device and storage medium for training and reconstructing a sound velocity map reconstruction model. Background Technology

[0002] Due to the complex structural characteristics of the skull, including anisotropy, heterogeneity, and porosity, ultrasound waves undergo significant absorption and scattering when passing through the skull, leading to severe signal attenuation and distortion, which in turn affects image quality. In particular, changes in skull structure and sound velocity can cause phase distortion, severely impacting the accuracy of transcranial ultrasound imaging. Therefore, accurately reconstructing the sound velocity distribution of the skull is crucial for achieving high-quality transcranial ultrasound imaging. However, existing sound velocity map reconstruction methods only perform coarse processing, failing to fully exploit its structural and phase correlations, resulting in low information utilization. Furthermore, existing reconstruction models are susceptible to redundant signal interference during the learning process, leading to low accuracy in the reconstructed sound velocity map and limiting further improvements in transcranial imaging quality. Summary of the Invention

[0003] In view of the above-mentioned problems in the prior art, the purpose of the embodiments of this specification is to provide a method, apparatus, device and storage medium for training and reconstructing a basic sound velocity map reconstruction model, so as to improve the accuracy and efficiency of sound velocity map reconstruction.

[0004] To solve the above-mentioned technical problems, the specific technical solutions of the embodiments in this specification are as follows:

[0005] On the one hand, embodiments of this specification provide a method for training a sound velocity map reconstruction model, the method comprising:

[0006] Obtain a sound velocity map training dataset, which contains several real sound velocity maps;

[0007] The ultrasonic radio frequency signal corresponding to each real sound velocity map is extracted from the sound velocity map training dataset to obtain the ultrasonic radio frequency signal training dataset.

[0008] The pre-built reconstruction model is trained using the ultrasound radio frequency signal training dataset to obtain a trained reconstruction model. The reconstruction model includes an encoder, a wavefront attention mechanism module, and a decoder. The wavefront attention mechanism module receives low-level feature maps from the encoder and high-level feature maps from the decoder, generates wavefront attention weights based on the low-level and high-level feature maps, and maps the wavefront attention weights to the low-level feature maps to enhance the reconstruction model's ability to perceive wavefront distortion regions in ultrasound radio frequency signals.

[0009] Further, the step of extracting the ultrasonic radio frequency signal corresponding to each real sound velocity map from the sound velocity map training dataset includes:

[0010] Extract the target medium parameters from the sound velocity map;

[0011] A wave equation is constructed based on the target medium parameters;

[0012] The wave equation was solved using the spatial Fourier pseudospectral method and the second-order central difference method of time to obtain the total sound field pressure distribution sequence data;

[0013] Based on the spatial coordinates of the preset virtual ultrasonic receiving array elements, the ultrasonic radio frequency signal is obtained by extracting the acoustic pressure distribution sequence data at the corresponding spatial location from the total acoustic pressure distribution sequence data.

[0014] Furthermore, the target medium parameters include the maximum propagation velocity, the actual ultrasonic vibration frequency, and the mass attenuation parameter;

[0015] The step of constructing the wave equation based on the target medium parameters includes:

[0016] The first attenuation parameter is calculated based on the mass attenuation parameter, the maximum propagation velocity, and the actual ultrasonic vibration frequency.

[0017] The ultrasonic phase velocity is calculated based on the first attenuation parameter, the maximum propagation velocity, the actual ultrasonic vibration frequency, and the preset ultrasonic reference vibration frequency.

[0018] The second and third attenuation parameters are calculated based on the maximum propagation velocity, the ultrasonic reference vibration frequency, and the first attenuation parameter.

[0019] The wave equation is constructed based on the ultrasonic phase velocity, the first attenuation parameter, the second attenuation parameter, and the third attenuation parameter.

[0020] Furthermore, the step of constructing the wave equation based on the ultrasonic phase velocity, the first attenuation parameter, the second attenuation parameter, and the third attenuation parameter includes:

[0021] The wave equation is constructed using the following formula:

[0022] ;

[0023] in, ;

[0024] ;

[0025] ;

[0026] ;

[0027] in, Indicates ultrasonic phase velocity, This represents the sequence data of total sound field pressure distribution, where x represents spatial location and t represents time. Indicates the first attenuation parameter. This represents the second decay parameter, and ▽ represents the second derivative. This represents the third attenuation parameter. This represents the mass decay parameter. Indicates the actual vibration frequency of the ultrasound. Indicates the maximum propagation speed. This represents the ultrasonic reference vibration frequency.

[0028] Further, training the pre-built reconstruction model using the ultrasound radio frequency signal training dataset includes:

[0029] The ultrasonic radio frequency signal training dataset is input into the reconstruction model to obtain the reconstructed sound velocity map corresponding to each ultrasonic radio frequency signal.

[0030] The loss value between each reconstructed sound velocity map and the corresponding real sound velocity map is calculated based on a pre-constructed loss function;

[0031] The parameters of the reconstruction model are updated based on the loss value, and the process of inputting the ultrasonic radio frequency signal training dataset into the reconstruction model is repeated until a preset termination condition is met, resulting in a trained reconstruction model.

[0032] Further, the step of inputting the ultrasonic radio frequency signal training dataset into the reconstruction model to obtain the reconstructed sound velocity map corresponding to each ultrasonic radio frequency signal includes:

[0033] The ultrasonic radio frequency signal is input into the encoder to obtain a low-level feature map;

[0034] The low-level feature map is input into the decoder to obtain the high-level feature map;

[0035] The low-level feature map and the high-level feature map are input into the wavefront attention mechanism module to obtain a fused feature map.

[0036] The fused feature map is input into the decoder to obtain the reconstructed sound velocity map.

[0037] Furthermore, the loss function is expressed using the following formula:

[0038] ;

[0039] in, ;

[0040] in, Represents the loss function. This represents the mean squared error loss function. This represents the gradient-aware loss function, where N represents the total number of pixels. Indicates the weighting factor. and These represent the gradients of the predicted sound velocity field in the horizontal and vertical directions, respectively. and These represent the gradients of the real sound velocity field in the horizontal and vertical directions, respectively.

[0041] Further, the step of generating wavefront attention weights based on the low-level feature map and the high-level feature map includes:

[0042] The high-level feature map is upsampled to obtain an upsampled high-level feature map.

[0043] The low-level feature map and the upsampled high-level feature map are respectively input into the first weight extraction module to obtain the first feature weight map and the second feature weight map.

[0044] The first feature weight map and the second feature weight map are fused to obtain the initial fused weights:

[0045] The initial fusion weights are input into the second weight extraction module to generate wavefront attention weights.

[0046] On the other hand, embodiments of this specification provide a sound velocity map reconstruction method, the method comprising:

[0047] Acquire the target ultrasonic radio frequency signal;

[0048] The target ultrasonic radio frequency signal is input into any of the above-mentioned trained reconstruction models to obtain the reconstructed sound velocity map corresponding to the target ultrasonic radio frequency signal.

[0049] Furthermore, embodiments of this specification provide a sound velocity map reconstruction model training device, the device comprising:

[0050] The first acquisition module is used to acquire a sound velocity map training dataset, which contains several real sound velocity maps.

[0051] The extraction module is used to extract the ultrasonic radio frequency signal corresponding to each real sound velocity map from the sound velocity map training dataset to obtain the ultrasonic radio frequency signal training dataset.

[0052] The training module is used to train a pre-built reconstruction model using the ultrasound radio frequency signal training dataset to obtain a trained reconstruction model. The reconstruction model includes an encoder, a wavefront attention mechanism module, and a decoder. The wavefront attention mechanism module receives low-level feature maps from the encoder and high-level feature maps from the decoder, generates wavefront attention weights based on the low-level and high-level feature maps, and maps the wavefront attention weights to the low-level feature maps to enhance the reconstruction model's ability to perceive wavefront distortion regions in ultrasound radio frequency signals.

[0053] In another aspect, embodiments of this specification provide a sound velocity map reconstruction apparatus, the apparatus comprising:

[0054] The second acquisition module is used to acquire the target ultrasonic radio frequency signal;

[0055] The reconstruction module is used to input the target ultrasonic radio frequency signal into any of the above-mentioned trained reconstruction models to obtain the reconstructed sound velocity map corresponding to the target ultrasonic radio frequency signal.

[0056] In another aspect, embodiments of this specification also provide a computer device, including a memory, a processor, and a computer program stored in the memory, wherein the computer program, when executed by the processor, performs instructions of any of the methods described above.

[0057] In another aspect, embodiments of this specification also provide a computer-readable storage medium having a computer program stored thereon, the computer program being executed by a processor of a computer device to perform instructions for any of the methods described above.

[0058] In another aspect, embodiments of this specification also provide a computer program product, which, when run by the processor of a computer device, executes instructions for any of the methods described above.

[0059] By employing the above technical solutions, the sound velocity map reconstruction model training and reconstruction method provided in this specification can firstly construct a highly reliable training data foundation based on real sound velocity maps and their corresponding ultrasound radio frequency signals, avoiding physical law deviations caused by relying on synthetic data. This ensures that the mapping relationship between the radio frequency signal and sound velocity distribution learned by the model closely matches the actual ultrasound propagation scenario, significantly improving the model's generalization ability and reducing reconstruction errors caused by data distortion. Simultaneously, through the wavefront attention mechanism module, the reconstruction model can efficiently fuse low-level local detail features output by the encoder with high-level global structural features output by the decoder, accurately generating exclusive attention weights for wavefront distortion regions and empowering low-level feature learning. This improves the perception and feature capture capabilities of wavefront distortion regions in ultrasound radio frequency signals, enabling the reconstruction model to effectively focus on wavefront distortion regions and improve the recognition ability of skull boundaries and internal structures. This effectively solves the problem of low accuracy caused by traditional reconstruction models neglecting key information in distortion regions.

[0060] In addition, during the inference phase, ultrasound radio frequency signals are used as direct input to the model, and wavefront attention mechanism and structural perception loss function are combined to significantly improve reconstruction efficiency while enhancing the ability to identify skull boundaries and internal structures.

[0061] The above description is merely an overview of some embodiments of the technical solutions in this specification. In order to better understand the technical means of some embodiments of this specification and to implement them in accordance with the content of the specification, and to make the above and other objects, features and advantages of the embodiments of this specification more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0062] To more clearly illustrate the technical solutions in the embodiments or prior art of this specification, the drawings used in the description of the embodiments or prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this specification. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0063] Figure 1 This specification illustrates the steps of a sound velocity map reconstruction model training method in some embodiments.

[0064] Figure 2 This specification illustrates the steps for extracting ultrasonic radio frequency signals in some embodiments;

[0065] Figure 3 This specification illustrates schematic diagrams of the steps for constructing wave equations based on target medium parameters in some embodiments.

[0066] Figure 4The diagram illustrates the steps of training a pre-built reconstruction model using an ultrasonic radio frequency signal training dataset in some embodiments of this specification.

[0067] Figure 5 The diagram illustrates the training process of the reconstruction model in some embodiments of this specification;

[0068] Figure 6 This specification illustrates the steps of inputting an ultrasonic radio frequency signal training dataset into a reconstruction model to obtain a reconstructed sound velocity map corresponding to each ultrasonic radio frequency signal in some embodiments of this specification.

[0069] Figure 7 This specification shows schematic diagrams illustrating the steps involved in generating wavefront attention weights in some embodiments.

[0070] Figure 8 A schematic diagram of the processing flow of the wavefront attention mechanism module in some embodiments of this specification is shown;

[0071] Figure 9 This specification illustrates the steps of a sound velocity map reconstruction method in some embodiments.

[0072] Figure 10 This specification shows a schematic diagram of the structure of a sound velocity map reconstruction model training device in some embodiments;

[0073] Figure 11 A schematic diagram of a sound velocity map reconstruction device is shown in some embodiments of this specification;

[0074] Figure 12 A schematic diagram of the structure of a computer device is shown in this specification.

[0075] Explanation of symbols in the attached drawings:

[0076] 1001, First Acquisition Module;

[0077] 1002. Extraction module;

[0078] 1003, Training Module;

[0079] 1101. Second Acquisition Module;

[0080] 1102. Rebuild module;

[0081] 1202. Computer equipment;

[0082] 1204, Processor;

[0083] 1206. Memory;

[0084] 1208. Drive mechanism;

[0085] 1210. Input / output module;

[0086] 1212. Input devices;

[0087] 1214. Output devices;

[0088] 1216. Presentation equipment;

[0089] 1218. Graphical User Interface;

[0090] 1220. Network interface;

[0091] 1222. Communication link;

[0092] 1224. Communication bus. Detailed Implementation

[0093] The technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this specification, and not all embodiments. Based on the embodiments in this specification, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this specification.

[0094] It should be noted that the terms "first," "second," etc., used in this specification, claims, and the foregoing drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, apparatus, product, or device that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0095] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties. Furthermore, the acquisition, storage, use, and processing of data in the technical solutions described in the embodiments of this application all comply with relevant regulations.

[0096] Due to the complex variations in skull tissue density, structure, and internal porosity, as well as its significant heterogeneity with surrounding soft tissues, the sound velocity in the skull (2700-3500 m / s) differs considerably from that in surrounding tissues (1540 m / s). Traditional ultrasound imaging, based on a uniform sound velocity distribution (1540 m / s), performs beamforming, neglecting these differences in skull sound velocity. This leads to structural location distortions in transcranial ultrasound images. Furthermore, incorrect signal superposition weakens the signal amplitude. Therefore, accurately characterizing the skull sound velocity before beamforming helps improve beamforming and thus enhances image quality. Full-waveform inversion (FWI) methods, based on ultrasound radio frequency signals, iteratively find the optimal sound velocity distribution using optimization methods, achieving accurate estimation of local sound velocities. However, this method typically requires approximately 1200 seconds to calculate a single frame, which is time-consuming, and its accuracy is low for skull structures with complex internal porosity, making it difficult to meet the needs of real-time or near-real-time clinical imaging. Therefore, rapid and accurate reconstruction of the sound velocity distribution of the skull is of great significance for achieving high-quality transcranial ultrasound imaging.

[0097] With the development of deep learning, sound velocity map reconstruction based on deep learning has become a trend. However, deep learning methods are highly sensitive to data quality and require large-scale datasets that closely resemble real physical processes. Therefore, constructing high-precision, high-fidelity simulation data has become one of the key bottlenecks in current deep learning-driven inversion methods. Existing numerical simulation methods mostly rely on the k-Wave platform, but its ability to model the complex spatial non-uniform attenuation characteristics in the skull is insufficient, leading to differences between training data and real imaging conditions, thus affecting the model's generalization ability and practical application performance. Secondly, in terms of network structure design, existing deep learning-based sound velocity map inversion methods mostly adopt general architectures, such as U-Net and ResNet. While these have some applicability in medical imaging tasks, they lack the ability to model the physical properties of ultrasound propagation (such as wavefront distortion). This structure ignores the unique velocity changes and directional shifts of sound waves propagating in the skull, limiting the model's sensitivity and discrimination ability to acoustically anomalous regions. Furthermore, although ultrasound radio frequency signals contain rich wavefront features and propagation path information, most existing methods only perform coarse processing on them, failing to fully explore their structural and phase correlations, resulting in low information utilization and the model being susceptible to interference from redundant signals during the learning process.

[0098] To address the aforementioned problems, this specification provides a method for training a sound velocity map reconstruction model. Figure 1This diagram illustrates the steps of a sound velocity map reconstruction model training method provided in the embodiments of this specification. This specification provides the operational steps of the method described in the embodiments or flowcharts, but based on conventional or non-inventive labor, more or fewer operational steps may be included. The order of steps listed in the embodiments is merely one possible execution order among many and does not represent the only possible execution order. In actual system or device products, the methods shown in the embodiments or drawings can be executed sequentially or in parallel. Specifically, as shown in the attached diagrams... Figure 1 As shown, the method may include the following steps:

[0099] S101: Obtain the sound velocity map training dataset, which contains several real sound velocity maps;

[0100] S102: Extract the ultrasonic radio frequency signal corresponding to each real sound velocity map from the sound velocity map training dataset to obtain the ultrasonic radio frequency signal training dataset.

[0101] S103: The pre-built reconstruction model is trained using the ultrasonic radio frequency signal training dataset to obtain a trained reconstruction model.

[0102] The reconstruction model includes an encoder, a wavefront attention mechanism module, and a decoder. The wavefront attention mechanism module receives a low-level feature map from the encoder and a high-level feature map from the decoder, generates wavefront attention weights based on the low-level and high-level feature maps, and maps the wavefront attention weights onto the low-level feature map to enhance the reconstruction model's ability to perceive wavefront distortion regions in ultrasound radio frequency signals.

[0103] By employing the above technical solutions, the sound velocity map reconstruction model training and reconstruction method provided in this specification can firstly construct a highly reliable training data foundation based on real sound velocity maps and their corresponding ultrasound radio frequency signals, avoiding physical law deviations caused by relying on synthetic data. This ensures that the mapping relationship between the radio frequency signal and sound velocity distribution learned by the model closely matches the actual ultrasound propagation scenario, significantly improving the model's generalization ability and reducing reconstruction errors caused by data distortion. Simultaneously, through the wavefront attention mechanism module, the reconstruction model can efficiently fuse low-level local detail features output by the encoder with high-level global structural features output by the decoder, accurately generating exclusive attention weights for wavefront distortion regions and empowering low-level feature learning. This improves the perception and feature capture capabilities of wavefront distortion regions in ultrasound radio frequency signals, enabling the reconstruction model to effectively focus on wavefront distortion regions and improve the recognition ability of skull boundaries and internal structures. This effectively solves the problem of low accuracy caused by traditional reconstruction models neglecting key information in distortion regions.

[0104] In the embodiments of this specification, the several real sound velocity maps in the sound velocity map training dataset in step S101 refer to the distribution maps of spatial position-sound velocity values ​​obtained through precise measurement or multimodal calibration based on the physical properties of actual media, such as human skulls, soft tissues, or ex vivo experimental samples, rather than computer-generated simulation data. In some embodiments, real sound velocity maps can be obtained from skull CT images.

[0105] In the embodiments of this specification, refer to Figure 2 Step S102 extracts the ultrasonic radio frequency signal corresponding to each real sound velocity map from the sound velocity map training dataset, including the following steps:

[0106] S201: Extract the target medium parameters from the sound velocity map;

[0107] S202: Construct a wave equation based on the target medium parameters;

[0108] S203: Solve the wave equation using the spatial Fourier pseudospectral method and the second-order central difference method of time to obtain the total sound field pressure distribution sequence data;

[0109] S204: Extract the sound field pressure distribution sequence data at the corresponding spatial location from the total sound field pressure distribution sequence data according to the spatial coordinates of the preset virtual ultrasonic receiving array element to obtain the ultrasonic radio frequency signal.

[0110] It is understandable that ultrasound radio frequency (RF) signals are time-domain pressure change signals received by the probe after sound waves propagate in the medium corresponding to the real sound velocity map. The spatial heterogeneity of the sound velocity map directly leads to wavefront distortion, amplitude attenuation, and phase shift of the ultrasound RF signal. Therefore, it is necessary to extract the corresponding ultrasound RF signal from the real sound velocity map through physical mapping. In order to accurately simulate the transcranial propagation behavior of plane wave ultrasound in heterogeneous skull structures, this specification proposes a high-precision numerical simulation method to extract ultrasound RF signals from real sound velocity maps. Specifically, a time-domain wave propagation model based on a constant Q attenuation model is used to extract ultrasound RF signals. This model can effectively capture complex wave propagation phenomena such as diffraction, scattering, and mode conversion, and is particularly suitable for multi-scale wavefront distortion caused by multi-layered skull structures and non-uniform media.

[0111] In the embodiments of this specification, step S201 extracts target medium parameters from the sound velocity map. The purpose is to transform the spatial-sound velocity static distribution data of the sound velocity map into physical parameters of the medium that can be calculated using the subsequent wave equation, providing a data basis for subsequent simulation of ultrasonic radio frequency signals. In some embodiments, the target medium parameters include the maximum propagation velocity, the actual ultrasonic vibration frequency, and the mass attenuation parameter. The maximum propagation velocity is the maximum value among all voxel sound velocity values ​​in the sound velocity map; the actual ultrasonic vibration frequency refers to the actual angular frequency of the transmitted signal, determined by the probe and the excitation signal; and the mass attenuation parameter corresponds to the attenuation characteristics of the medium, determined by the medium type.

[0112] In the embodiments of this specification, refer to Figure 3 Step S202 involves constructing a wave equation based on the target medium parameters, including the following steps:

[0113] S301: The first attenuation parameter is calculated based on the mass attenuation parameter, the maximum propagation velocity, and the actual ultrasonic vibration frequency;

[0114] S302: The ultrasonic phase velocity is calculated based on the first attenuation parameter, the maximum propagation velocity, the actual ultrasonic vibration frequency, and the preset ultrasonic reference vibration frequency;

[0115] S303: The second attenuation parameter and the third attenuation parameter are calculated based on the maximum propagation velocity, the ultrasonic reference vibration frequency and the first attenuation parameter;

[0116] S304: The wave equation is constructed based on the ultrasonic phase velocity, the first attenuation parameter, the second attenuation parameter, and the third attenuation parameter.

[0117] Specifically, in the embodiments of this specification, step S301 calculates the first attenuation parameter using the following formula based on the mass attenuation parameter, the maximum propagation velocity, and the actual ultrasonic vibration frequency:

[0118]

[0119] in, Indicates the first attenuation parameter. This represents the mass decay parameter. Indicates the actual vibration frequency of the ultrasound. This indicates the maximum propagation speed.

[0120] Step S302 calculates the ultrasonic phase velocity using the following formula based on the first attenuation parameter, maximum propagation velocity, ultrasonic reference vibration frequency, and actual ultrasonic vibration frequency:

[0121]

[0122] in, Indicates ultrasonic phase velocity, Indicates the first attenuation parameter. Indicates the actual vibration frequency of the ultrasound. This represents the ultrasonic reference vibration frequency.

[0123] Step S303 uses the following formula to calculate the second and third attenuation parameters based on the maximum propagation velocity, the ultrasonic reference vibration frequency, and the first attenuation parameter:

[0124]

[0125]

[0126] in, This represents the second attenuation parameter. This represents the third attenuation parameter. Indicates the first attenuation parameter. Indicates the actual vibration frequency of the ultrasound. Indicates the maximum propagation speed. This represents the ultrasonic reference vibration frequency.

[0127] Step S304 constructs the wave equation based on the ultrasonic phase velocity, the first attenuation parameter, the second attenuation parameter, and the third attenuation parameter, as shown below:

[0128]

[0129] in, Indicates ultrasonic phase velocity, This represents the sequence data of total sound field pressure distribution, where x represents spatial location and t represents time. Indicates the first attenuation parameter. This represents the second decay parameter, and ▽ represents the second derivative. This represents the third attenuation parameter. This represents the mass decay parameter. Indicates the actual vibration frequency of the ultrasound. Indicates the maximum propagation speed. This represents the ultrasonic reference vibration frequency.

[0130] This specification introduces a frequency-dependent viscoelastic attenuation mechanism in its embodiments. By coupling a constant-Q power-law absorption term, it can accurately express the energy dissipation and frequency dispersion characteristics caused by the microstructure of the skull, thus balancing numerical accuracy and computational stability. Step S203 obtains the evolution of the entire sound field in time and space through numerical calculation, i.e., the data on the pressure distribution changing over time. Specifically, the spatial derivative is calculated using the Fourier pseudo-spectral method in the spatial dimension, and the pressure field is updated using the second-order central difference method in the temporal dimension to obtain the pressure distribution in the entire space at each moment, thereby obtaining the total sound field pressure distribution sequence data. Step S204 simulates the physical position and receiving characteristics of the real probe using virtual receiving array elements, extracting the ultrasound radio frequency signal from the probe receiving end from the global sound field to convert the spatiotemporal sound field data into ultrasound radio frequency signals that the model can train. In some embodiments, virtual array elements can be preset in the receiving area of ​​the simulation model, such as the number of array elements, the spacing between array elements, and the probe position, simulating the placement position of the real probe, based on the actual parameters of the clinical ultrasound probe, such as the number of array elements, the spacing between array elements, and the probe position. For each virtual array element, the pressure values ​​at all time points corresponding to its spatial coordinates are extracted. This involves filtering the pressure time series corresponding to the element's location from the total sound field pressure distribution sequence. This time series data is essentially the ultrasound radio frequency signal corresponding to the actual sound velocity map; its amplitude changes reflect sound wave energy attenuation, and its phase changes reflect wavefront distortion, perfectly matching the original ultrasound radio frequency signal received by the clinical probe. In some embodiments, to further approximate real clinical data, the extracted ultrasound radio frequency signal may undergo noise removal, amplitude normalization, and phase calibration.

[0131] In the embodiments of this specification, refer to Figure 4 Step S103, which involves training the pre-built reconstruction model using the ultrasonic radio frequency signal training dataset, includes the following steps:

[0132] S401: Input the ultrasonic radio frequency signal training dataset into the reconstruction model to obtain the reconstructed sound velocity map corresponding to each ultrasonic radio frequency signal;

[0133] S402: Calculate the loss value between each reconstructed sound velocity map and the corresponding real sound velocity map based on the pre-constructed loss function;

[0134] S403: Update the parameters of the reconstruction model according to the loss value, and repeat the step of inputting the ultrasonic radio frequency signal training dataset into the reconstruction model until the preset termination condition is reached to obtain the trained reconstruction model.

[0135] Specifically, Figure 5 A schematic diagram of the training process for reconstructing the model, as shown below. Figure 5As shown, a true sound velocity map can be obtained from a skull CT image. Ultrasonic radio frequency (RF) signals are extracted from the true sound velocity map using the high-precision numerical simulation method described above. These RF signals are then input into a pre-built reconstruction model to obtain a reconstructed sound velocity map. The true and reconstructed sound velocity maps are then input into a pre-built loss function to calculate the difference between the reconstruction result and the true baseline. This provides a directional basis for subsequent model parameter optimization. A smaller loss value indicates a smaller deviation between the reconstructed and true sound velocity maps, and higher model prediction accuracy. For each sample in the training dataset, its reconstructed and corresponding true sound velocity maps are input into the loss function. First, the loss value for a single sample is calculated. Then, the average loss values ​​of all single samples within the current training batch are taken to obtain the batch loss value, which serves as the core basis for subsequent parameter updates. After calculating the loss value, the model parameters are adjusted using the backpropagation algorithm to minimize the loss value. Iterative training improves the reconstruction model's ability to learn the mapping relationship between RF signals and sound velocity maps, ultimately resulting in a stable reconstruction model. In some embodiments, the training preset termination condition can be a preset training round, a preset validation set loss threshold, and a preset training set loss threshold, etc. When the training reaches the preset termination condition, all parameters of the current model, such as weights and biases, are saved. This model is the trained reconstruction model, which can stably receive new target ultrasonic radio frequency signals and output a reconstructed sound velocity map with the required accuracy.

[0136] In this embodiment of the specification, to effectively integrate the physical structural features of the skull interior with the spatial constraints in the reconstruction task, a structure-aware hybrid loss function is proposed. This function combines traditional mean squared error (MSE) with gradient loss to supervise the neural network in generating a more structurally accurate and clearly defined sound velocity distribution map during the training phase. In this embodiment of the specification, the loss function is expressed using the following formula:

[0137] ;

[0138] in, ;

[0139] in, Represents the loss function. This represents the mean squared error loss function. This represents the gradient-aware loss function, where N represents the total number of pixels. Indicates the weighting factor. and These represent the gradients of the predicted sound velocity field in the horizontal and vertical directions, respectively. and These represent the gradients of the true sound velocity field in the horizontal and vertical directions, respectively. In other embodiments, in addition to MSE and gradient loss, boundary constraint loss and structure preservation loss, such as SSIMLoss, can be introduced to further enhance the model's ability to preserve anatomical boundaries and local structures.

[0140] In the embodiments of this specification, refer to Figure 6 Step S401 involves inputting the ultrasonic radio frequency signal training dataset into the reconstruction model to obtain a reconstructed sound velocity map corresponding to each ultrasonic radio frequency signal, including the following steps:

[0141] S601: Input the ultrasonic radio frequency signal into the encoder to obtain a low-level feature map;

[0142] S602: Input the low-level feature map into the decoder to obtain the high-level feature map;

[0143] S603: Input the low-level feature map and the high-level feature map into the wavefront attention mechanism module to obtain a fused feature map;

[0144] S604: Input the fused feature map into the decoder to obtain the reconstructed sound velocity map.

[0145] Understandably, to fully exploit the spatial wavefront information in ultrasound radio frequency signals, this specification proposes a deep neural network architecture embedding a wavefront structure perception mechanism. In some embodiments, the neural network architecture uses U-Net as the main structure; in others, it can be replaced with a Transformer variant or a network with temporal modeling capabilities, such as LSTM, to adapt to different types of input features or improve model expressive power. Furthermore, a lightweight network version can be used to deploy this method on portable ultrasound devices. Addressing the problem that conventional feature fusion methods fail to effectively focus on wavefront distortion regions, a wavefront attention module is designed to jointly guide and enhance the structure of low-level encoder features and high-level semantic information of the decoder. Step S601 extracts features from the input ultrasound radio frequency signal using the encoder to obtain local detail features, which correspond to the local manifestations of wavefront distortion in the ultrasound radio frequency signal. Step S602 involves extracting features from the low-level feature map output by the encoder using the decoder. While preserving local features, a global understanding of the sound velocity distribution is generated, resulting in a high-level feature map. This high-level feature map describes the overall contour of the sound velocity distribution, reflecting the reconstruction model's understanding of the global structure of the sound velocity distribution, and can be used to guide the attention mechanism to focus on important regions. Step S603 involves the wavefront attention mechanism module receiving the low-level feature map F from the encoder. E With high-level feature map F from the decoder DTo enhance the perception of local wavefront variations, the attention mechanism module first receives low-level feature maps from the encoder and high-level feature maps from the decoder. By calculating the correlation between the low-level and high-level feature maps, wavefront attention weights are generated. Regions with high weights correspond to areas with significant wavefront distortion and a large contribution to sound velocity reconstruction. These wavefront attention weights are then mapped onto the low-level feature maps to enhance the reconstruction model's ability to perceive wavefront distortion regions in the ultrasound radio frequency signal.

[0146] In the embodiments of this specification, refer to Figure 7 The wavefront attention mechanism module generates wavefront attention weights based on low-level and high-level feature maps, specifically including the following steps:

[0147] S701: Perform an upsampling operation on the high-level feature map to obtain an upsampled high-level feature map;

[0148] S702: Input the low-level feature map and the upsampled high-level feature map into the first weight extraction module respectively to obtain the first feature weight map and the second feature weight map;

[0149] S703: Merge the first feature weight map and the second feature weight map to obtain the initial fused weights:

[0150] S704: Input the initial fusion weights into the second weight extraction module to generate wavefront attention weights.

[0151] It is understandable that, such as Figure 8 As shown, the specific processing flow of the wavefront attention mechanism module is as follows:

[0152] 1) Size alignment: First, align the high-level feature map F of the decoder. D Perform an upsampling operation to make its spatial resolution match that of the encoder's low-level feature map F. E Matching ensures spatial consistency for subsequent feature fusion.

[0153] 2) Channel-weighted extraction: For F respectively E Compared with the upsampled F D A first weight extraction module, combining a 1×1 convolutional kernel and batch normalization (BN), extracts the corresponding feature weight map. The two are then superimposed and activated by the ReLU function to complete the initial fusion process and obtain the initial fusion weights.

[0154] 3) Wavefront attention weight calculation: The initial fusion weights are input into the second weight extraction module to generate wavefront attention weights, and then normalized to the [0, 1] interval by the Sigmoid function.

[0155] 4) Feature Modulation: The wavefront attention weight map is used as a mask on the low-level feature map F of the encoder. E ,get This allows for explicit enhancement of the wavefront structure and suppression of the noise region.

[0156] The wavefront attention mechanism module can focus on locations with significant wavefront changes in the encoder features, guiding the network to adaptively learn local structural changes caused by boundary diffraction, refraction, and sound velocity gradient variations during training. This improves the model's ability to resolve non-uniform regions in the sound velocity map. The above process can be represented by the following formula:

[0157]

[0158] in, Represents the fused feature map. Represents the low-level feature map; This represents the high-level feature map, Upsample() represents the upsampling operation, and Weight E and Weight D Represents the first weight extraction module, ReLU represents the ReLU activation function, and Conv 1×1 This represents a 1x1 convolution and BN operation, i.e., the second weight extraction module, where σ represents the Sigmoid activation function. This indicates element-wise multiplication. In other embodiments, in addition to linear attention modules, multi-channel attention mechanisms, self-attention mechanisms, and other attention modules can also be used in the design of the wavefront attention mechanism module.

[0159] In the embodiments of this specification, to improve the generalization ability and cross-individual adaptability of the model, a training mechanism that takes into account multi-source data structures is proposed for the reconstruction model. In terms of data construction, a skull model based on monkey microCT images is used, and structural models from real human skull CT images are also collected and constructed, thus forming a representative and diverse training dataset. Using a simplified and efficient numerical simulator developed in the embodiments of this specification, simulated ultrasound propagation experiments were conducted on these structural models to generate corresponding radio frequency signals for model training. Experimental results show that the model has high stability and reconstruction accuracy under different individual structures, different imaging angles, and different beam conditions. In other embodiments of this specification, in addition to ultrasound radio frequency signals, the input of the reconstruction model can also be combined with various other supervisory signals, such as sound velocity, bone thickness, and boundary position, for joint training. This makes the trained reconstruction model easily extendable to three-dimensional reconstruction or multimodal input scenarios, improving the generalization ability of the reconstruction model.

[0160] Based on the aforementioned method for training a sound velocity map reconstruction model, this specification also provides a corresponding sound velocity map reconstruction model training device. The device may include a system (including a distributed system), software (application), module, component, server, client, etc., using the method described in this specification, combined with necessary hardware implementation. Based on the same innovative concept, the devices in one or more embodiments provided in this specification are as described in the following embodiments. Since the implementation schemes and methods for solving the problem are similar, the implementation of specific devices in this specification can refer to the implementation of the aforementioned method, and repeated details will not be repeated. As used below, the terms "unit" or "module" can refer to a combination of software and / or hardware that implements a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0161] Specifically, Figure 10 This is a schematic diagram of the module structure of one embodiment of the sound velocity map reconstruction model training device provided in this specification, with reference to... Figure 10 As shown in the embodiments of this specification, a sound velocity map reconstruction model training device includes:

[0162] The first acquisition module 1001 is used to acquire a sound velocity map training dataset, which contains several real sound velocity maps.

[0163] Extraction module 1002 is used to extract the ultrasonic radio frequency signal corresponding to each real sound velocity map from the sound velocity map training dataset to obtain ultrasonic radio frequency signal training dataset.

[0164] The training module 1003 is used to train a pre-built reconstruction model using the ultrasound radio frequency signal training dataset to obtain a trained reconstruction model. The reconstruction model includes an encoder, a wavefront attention mechanism module, and a decoder. The wavefront attention mechanism module receives low-level feature maps from the encoder and high-level feature maps from the decoder, generates wavefront attention weights based on the low-level and high-level feature maps, and maps the wavefront attention weights to the low-level feature maps to enhance the reconstruction model's ability to perceive wavefront distortion regions in ultrasound radio frequency signals.

[0165] The beneficial effects obtained by the apparatus provided in the embodiments of this specification are consistent with the beneficial effects obtained by the methods described above, and will not be repeated here.

[0166] Based on the sound velocity map reconstruction method described above, this specification also provides a corresponding sound velocity map reconstruction method, as detailed below. Figure 9 As shown, the reconstruction method may include the following steps:

[0167] S901: Acquire the target ultrasonic radio frequency signal;

[0168] S902: Input the target ultrasonic radio frequency signal into any of the above-mentioned trained reconstruction models to obtain the reconstructed sound velocity map corresponding to the target ultrasonic radio frequency signal.

[0169] Understandably, during the inference phase, inputting ultrasound radiofrequency signals into the trained reconstruction model directly outputs the corresponding two-dimensional skull sound velocity map, significantly improving reconstruction efficiency. This result can be further applied to downstream tasks such as ultrasound phase correction, transcranial imaging distortion compensation, and adaptive beam modulation, and has the potential to be integrated into clinical ultrasound equipment or imaging systems.

[0170] The acoustic velocity map reconstruction model training and reconstruction method proposed in the embodiments of this specification has good versatility and application expansion potential. First, it can be extended to soft tissue ultrasound imaging tasks, such as the construction of high-resolution acoustic velocity maps of tissues like the liver and breast, to assist in disease detection and tissue characterization. Second, it can be integrated into intraoperative ultrasound navigation systems for real-time reconstruction of bone tissue acoustic velocity maps, identification of tissue boundaries and intraoperative changes, and improvement of navigation accuracy. Third, the method can also be extended to reconstruction tasks of other acoustic properties, including seismic acoustic velocity inversion. Fourth, this method can be combined with magnetic resonance imaging (MRI), computed tomography (CT), and other modalities to build a cross-modal learning platform, realizing the collaborative reconstruction or registration of multimodal medical images and expanding its application prospects in comprehensive imaging diagnostic systems.

[0171] Based on the sound velocity map reconstruction method described above, this specification also provides a corresponding sound velocity map reconstruction device. (Refer to...) Figure 11 As shown in the embodiments of this specification, a sound velocity map reconstruction device includes:

[0172] The second acquisition module 1101 is used to acquire the target ultrasonic radio frequency signal;

[0173] The reconstruction module 1102 is used to input the target ultrasonic radio frequency signal into the trained reconstruction model described above to obtain the reconstructed sound velocity map corresponding to the target ultrasonic radio frequency signal.

[0174] Reference Figure 12As shown, based on the aforementioned method for training and reconstructing a sound velocity map reconstruction model, an embodiment of this specification also provides a computer device 1202, wherein the above-described method operates on the computer device 1202. The computer device 1202 may include one or more processors 1204, such as one or more central processing units (CPUs), each of which may implement one or more hardware threads. The computer device 1202 may also include any memory 1206 for storing any kind of information, such as code, settings, data, etc. Non-limitingly, for example, the memory 1206 may include any type of RAM, any type of ROM, flash memory, hard disk, optical disk, etc. More generally, any memory can use any technology to store information. Further, any memory may provide volatile or non-volatile retention of information. Further, any memory may represent a fixed or removable component of the computer device 1202. In one case, when the processor 1204 executes associated instructions stored in any memory or combination of memories, the computer device 1202 may perform any operation of the associated instructions. The computer device 1202 also includes one or more drive mechanisms 1208 for interacting with any memory, such as a hard disk drive mechanism, an optical disk drive mechanism, etc.

[0175] Computer device 1202 may also include an input / output module 1210 (I / O) for receiving various inputs (via input device 1212) and providing various outputs (via output device 1214). A specific output mechanism may include a presentation device 1216 and an associated graphical user interface (GUI) 1218. In other embodiments, the input / output module 1210 (I / O), input device 1212, and output device 1214 may be omitted, and the device may function solely as a computer device within a network. Computer device 1202 may also include one or more network interfaces 1220 for exchanging data with other devices via one or more communication links 1222. One or more communication buses 1224 couple the components described above together.

[0176] Communication link 1222 can be implemented in any way, such as via a local area network, a wide area network (e.g., the Internet), a point-to-point connection, or any combination thereof. Communication link 1222 may include any combination of hardwired links, wireless links, routers, gateway functions, name servers, etc., governed by any protocol or combination of protocols.

[0177] Corresponding to the methods shown in any of the above embodiments, this specification also provides a computer-readable storage medium storing a computer program that, when executed by a processor, performs the steps of the above methods.

[0178] This specification also provides computer-readable instructions, wherein when a processor executes the instructions, the program therein causes the processor to perform the method shown in any of the above embodiments.

[0179] This specification also provides a computer program product, including at least one instruction or at least one program segment, which is loaded and executed by a processor to implement the method shown in any of the above embodiments.

[0180] It should be understood that in the various embodiments of this specification, the sequence number of each process does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this specification.

[0181] It should also be understood that, in the embodiments of this specification, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this specification generally indicates that the preceding and following related objects have an "or" relationship.

[0182] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed in this specification can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of each example have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this specification.

[0183] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0184] In the several embodiments provided in this specification, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the couplings or direct couplings or communication connections shown or discussed may be indirect couplings or communication connections through some interfaces, devices, or units, or they may be electrical, mechanical, or other forms of connection.

[0185] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of the embodiments described in this specification, depending on actual needs.

[0186] Furthermore, the functional units in the various embodiments of this specification can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0187] If the integrated unit is implemented as a software functional unit 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 specification, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this specification. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0188] This specification uses specific embodiments to illustrate the principles and implementation methods of this specification. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this specification. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this specification. Therefore, the content of this specification should not be construed as a limitation of this specification.

Claims

1. A method for training a sound velocity map reconstruction model, characterized in that, The method includes: Obtain a sound velocity map training dataset, which contains several real sound velocity maps; The ultrasonic radio frequency signal corresponding to each real sound velocity map is extracted from the sound velocity map training dataset to obtain the ultrasonic radio frequency signal training dataset. The pre-built reconstruction model is trained using the ultrasound radio frequency signal training dataset to obtain a trained reconstruction model. The reconstruction model includes an encoder, a wavefront attention mechanism module, and a decoder. The wavefront attention mechanism module receives low-level feature maps from the encoder and high-level feature maps from the decoder, generates wavefront attention weights based on the low-level and high-level feature maps, and maps the wavefront attention weights to the low-level feature maps to enhance the reconstruction model's ability to perceive wavefront distortion regions in ultrasound radio frequency signals.

2. The method according to claim 1, characterized in that, The step of extracting the ultrasonic radio frequency signal corresponding to each real sound velocity map from the sound velocity map training dataset includes: Extract the target medium parameters from the sound velocity map; A wave equation is constructed based on the target medium parameters; The wave equation was solved using the spatial Fourier pseudospectral method and the second-order central difference method of time to obtain the total sound field pressure distribution sequence data; Based on the spatial coordinates of the preset virtual ultrasonic receiving array elements, the ultrasonic radio frequency signal is obtained by extracting the acoustic pressure distribution sequence data at the corresponding spatial location from the total acoustic pressure distribution sequence data.

3. The method according to claim 2, characterized in that, The target medium parameters include maximum propagation velocity, actual ultrasonic vibration frequency, and mass attenuation parameters. The step of constructing the wave equation based on the target medium parameters includes: The first attenuation parameter is calculated based on the mass attenuation parameter, the maximum propagation velocity, and the actual ultrasonic vibration frequency. The ultrasonic phase velocity is calculated based on the first attenuation parameter, the maximum propagation velocity, the actual ultrasonic vibration frequency, and the preset ultrasonic reference vibration frequency. The second and third attenuation parameters are calculated based on the maximum propagation velocity, the ultrasonic reference vibration frequency, and the first attenuation parameter. The wave equation is constructed based on the ultrasonic phase velocity, the first attenuation parameter, the second attenuation parameter, and the third attenuation parameter.

4. The method according to claim 3, characterized in that, The wave equation constructed based on the ultrasonic phase velocity, the first attenuation parameter, the second attenuation parameter, and the third attenuation parameter includes: The wave equation is constructed using the following formula: ； in, ; ； ； ； in, Indicates ultrasonic phase velocity, This represents the sequence data of total sound field pressure distribution, where x represents spatial location and t represents time. Indicates the first attenuation parameter. This represents the second decay parameter, and ▽ represents the second derivative. This represents the third attenuation parameter. This represents the mass decay parameter. Indicates the actual vibration frequency of the ultrasound. Indicates the maximum propagation speed. This represents the ultrasonic reference vibration frequency.

5. The method according to claim 1, characterized in that, The step of training the pre-built reconstruction model using the ultrasonic radio frequency signal training dataset includes: The ultrasonic radio frequency signal training dataset is input into the reconstruction model to obtain the reconstructed sound velocity map corresponding to each ultrasonic radio frequency signal. The loss value between each reconstructed sound velocity map and the corresponding real sound velocity map is calculated based on a pre-built loss function; The parameters of the reconstruction model are updated based on the loss value, and the process of inputting the ultrasonic radio frequency signal training dataset into the reconstruction model is repeated until a preset termination condition is met, resulting in a trained reconstruction model.

6. The method according to claim 5, characterized in that, The step of inputting the ultrasonic radio frequency signal training dataset into the reconstruction model to obtain the reconstructed sound velocity map corresponding to each ultrasonic radio frequency signal includes: The ultrasonic radio frequency signal is input into the encoder to obtain a low-level feature map; The low-level feature map is input into the decoder to obtain the high-level feature map; The low-level feature map and the high-level feature map are input into the wavefront attention mechanism module to obtain a fused feature map. The fused feature map is input into the decoder to obtain the reconstructed sound velocity map.

7. The method according to claim 5, characterized in that, The loss function is expressed by the following formula: ； in, ; in, Represents the loss function. This represents the mean squared error loss function. This represents the gradient-aware loss function, where N represents the total number of pixels. Indicates the weighting factor. and These represent the gradients of the predicted sound velocity field in the horizontal and vertical directions, respectively. and These represent the gradients of the real sound velocity field in the horizontal and vertical directions, respectively.

8. The method according to claim 1, characterized in that, The step of generating wavefront attention weights based on the low-level feature maps and high-level feature maps includes: The high-level feature map is upsampled to obtain an upsampled high-level feature map. The low-level feature map and the upsampled high-level feature map are respectively input into the first weight extraction module to obtain the first feature weight map and the second feature weight map. The first feature weight map and the second feature weight map are fused to obtain the initial fused weights: The initial fusion weights are input into the second weight extraction module to generate wavefront attention weights.

9. A method for reconstructing a sound velocity map, characterized in that, The method includes: Acquire the target ultrasonic radio frequency signal; The target ultrasonic radio frequency signal is input into the trained reconstruction model according to any one of claims 1-8 to obtain the reconstructed sound velocity map corresponding to the target ultrasonic radio frequency signal.

10. A sound velocity map reconstruction model training device, characterized in that, The device includes: The first acquisition module is used to acquire a sound velocity map training dataset, which contains several real sound velocity maps. The extraction module is used to extract the ultrasonic radio frequency signal corresponding to each real sound velocity map from the sound velocity map training dataset to obtain the ultrasonic radio frequency signal training dataset. The training module is used to train a pre-built reconstruction model using the ultrasound radio frequency signal training dataset to obtain a trained reconstruction model. The reconstruction model includes an encoder, a wavefront attention mechanism module, and a decoder. The wavefront attention mechanism module receives low-level feature maps from the encoder and high-level feature maps from the decoder, generates wavefront attention weights based on the low-level and high-level feature maps, and maps the wavefront attention weights to the low-level feature maps to enhance the reconstruction model's ability to perceive wavefront distortion regions in ultrasound radio frequency signals.

11. A sound velocity map reconstruction device, characterized in that, The device includes: The second acquisition module is used to acquire the target ultrasonic radio frequency signal; The reconstruction module is used to input the target ultrasonic radio frequency signal into the trained reconstruction model according to any one of claims 1-8 to obtain the reconstructed sound velocity map corresponding to the target ultrasonic radio frequency signal.

12. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method as described in any one of claims 1 to 9.

13. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the method as described in any one of claims 1 to 9.

14. A computer program product, characterized in that, It includes at least one instruction or at least one program segment, said at least one instruction or said at least one program segment being loaded and executed by a processor to implement the method as claimed in any one of claims 1 to 9.

Images