A data processing method for ultrasound-guided nerve block anesthesia

Abstract

The present invention relates to a data processing method for ultrasound-guided nerve block anesthesia. First, the ultrasound image data is preprocessed, and a noise reduction algorithm based on deep learning is used to remove image noise, and a local contrast enhancement algorithm is used to enhance the visibility of key structures, including nerves and blood vessels; secondly, deep learning is used to extract features, identify and segment nerves, blood vessels and muscle tissues in the image, and data from different ultrasound modes are integrated to extract tissue features; then the processing flow is optimized by a recursive learning algorithm and an adaptive algorithm, wherein the recursive learning algorithm optimizes the accuracy of feature extraction and image analysis based on previous injection effects and ultrasound images, and the adaptive algorithm adjusts processing parameters according to the body shape and tissue characteristics of different patients; finally, a volume reconstruction algorithm is used to construct a complete three-dimensional tissue model from a continuous two-dimensional ultrasound image sequence, and real-time position sensing technology is combined to control the positioning and real-time navigation of the needle.

Description

Technical Field

[0001] The present invention relates to an ultrasound-guided data processing method, in particular to a nerve block anesthesia ultrasound-guided data processing method. Background Art

[0002] Although the existing block anesthesia ultrasound-guided data processing methods, such as the method described in Chinese Patent Publication No.: CN2024105715803, have made some progress in image data enhancement, there are still some shortcomings and disadvantages that need to be further improved and optimized. These shortcomings and disadvantages are mainly reflected in the following aspects: First, the existing image enhancement technology mainly relies on histogram equalization and basic image processing algorithms. Although these methods can improve the contrast and clarity of the image to a certain extent, they often ignore the complex relationship between different tissue and organ structures. This method will lead to over-enhancement or under-enhancement when processing tissue structures, thereby affecting the authenticity and detail retention of the image. For example, when processing bones and soft tissues, due to their different reflection characteristics and motion states, simple enhancement methods will cause the bone area to be too bright and the soft tissue details to be lost. Secondly, the existing methods have limitations in processing the stability of tissue and organ structures. The stability of different tissue and organ structures varies. The position of bones is relatively stable, while soft tissues will undergo subtle changes due to movements such as breathing. The existing technology judges tissue stability by quantifying the brightness changes of pixels, but this method is not accurate enough in practical applications. Especially when dealing with anatomical structures, this stability judgment method based on brightness changes is easily affected by noise and artifacts, leading to misjudgment.

[0003] In addition, the existing technology also has shortcomings in multimodal data fusion. Although ultrasound images have advantages in real-time and convenience, their resolution and contrast are low, making it difficult to clearly display deep structures. In contrast, CT and MRI images have advantages in resolution and contrast, but lack real-time performance. Fusion of ultrasound images with CT or MRI data, and utilization of the complementarity of multimodal data, can significantly improve the clarity and contrast of images. However, the existing methods still face great challenges in the algorithms and implementation of data fusion, especially in terms of how to effectively align and fuse different modal data, lacking efficient and accurate solutions. In terms of real-time feedback and dynamic adjustment, the existing methods also have obvious shortcomings. During block anesthesia, doctors need to adjust operations based on real-time images and feedback. However, the existing technology lacks an effective real-time feedback mechanism, which makes it difficult for doctors to adjust operations in a timely and accurate manner. This not only increases the complexity of the operation, but also increases the risk of surgery. The lack of real-time feedback and dynamic adjustment mechanisms causes doctors to rely more on experience and intuition during the operation, making it difficult to ensure the accuracy and consistency of each operation. In terms of computational efficiency and algorithm optimization, the existing technology also needs to be improved. Although some algorithms based on deep learning and artificial intelligence have been applied to medical image processing, these algorithms usually require a lot of computing resources and time, making them difficult to apply in real-time operations. Especially when processing large-scale three-dimensional image data, the computational efficiency of existing algorithms often cannot meet clinical needs. This not only limits its application in actual operations, but also increases the complexity and cost of the system. Summary of the invention

[0004] The purpose of the present invention is to provide a data processing method for ultrasound-guided nerve block anesthesia, thereby solving some of the drawbacks and shortcomings pointed out in the background technology.

[0005] The present invention solves the above-mentioned technical problems by adopting the following technical solutions: firstly, the ultrasound image data is preprocessed, a noise reduction algorithm based on deep learning is used to remove image noise, and a local contrast enhancement algorithm is used to enhance the visibility of key structures, including nerves and blood vessels;

[0006] Secondly, deep learning is used for feature extraction to identify and segment nerves, blood vessels, and muscle tissues in the image, while integrating data from different ultrasound modes to extract tissue features;

[0007] The processing flow is then optimized using a recursive learning algorithm that optimizes the accuracy of feature extraction and image analysis based on previous injection results and ultrasound images, and an adaptive algorithm that adjusts processing parameters based on the body shape and tissue characteristics of different patients.

[0008] Finally, a volume reconstruction algorithm is used to construct a complete three-dimensional tissue model from a continuous two-dimensional ultrasound image sequence, and real-time position sensing technology is combined to control the positioning and real-time navigation of the needle.

[0009] Furthermore, the process of preprocessing the ultrasound image data includes:

[0010] S1, using a high-order partial differential equation with adaptive adjustment coefficients to remove image noise; the specific formula is:

[0011]

[0012] Among them, i and j are the orders of differentiation, λ and σ are parameters that are dynamically adjusted based on the image content;

[0013] S2. Then, for the visual enhancement of key structures including nerves and blood vessels, the stacking integral transform is used to combine information of different scales and directions:

[0014]

[0015] Where Ω is the image domain, w k is the weight of scale k, G k is the Gaussian blur function, ω(x,y) is the position-dependent weight function, adjusting the enhanced locality;

[0016] S3, then use the network structure containing local and global information feedback to control nerve and blood vessel segmentation:

[0017]

[0018] Among them, F i (x, y) are features extracted by a deep neural network, and K is a kernel function used to enhance the spatial context information in the image;

[0019] S4. Finally, according to the image contrast differences of different patients, a dynamic adjustment model that depends on image gradient and local brightness characteristics is designed:

[0020]

[0021] in, is the Laplace operator, which represents the second-order derivative of the image and is used to detect boundaries and details. γ(x,y) is a parameter adjusted according to the characteristics of the image.

[0022] Furthermore, the construction process of deep learning for feature extraction includes:

[0023] S1. First, multimodal data fusion is used through the function:

[0024]

[0025] Integration of different ultrasound modes i data, where K i (x,y) is for mode M i The designed convolution kernel, σ i Controlling spatial scale, α i are the weights obtained through training;

[0026] S2. Then perform hierarchical feature extraction through the deep learning model, applying the formula:

[0027]

[0028] Network L j Each layer output is weighted high-order differential, where β j is the level weight;

[0029] S3. Finally, an end-to-end deep learning model is used for segmentation, using the formula:

[0030]

[0031] Where V(u,v) is the trained transformation kernel used to extract the saliency of key tissue features such as nerves and blood vessels in image I.

[0032] Furthermore, the recursive learning algorithm optimizes feature extraction, based on the results of previous surgeries and ultrasound images, using the formula:

[0033]

[0034] where θ t is the model parameter at time t, η is the adaptive learning rate, y t It is the surgical effect data. is the effect predicted by the model, p(y t |x t θ t ) is the conditional probability density function;

[0035] Then an adaptive algorithm is applied to adjust the image processing parameters according to the patient's body size and tissue characteristics, using the formula:

[0036]

[0037] Where P original is the original image processing parameter, δ is the adjustment intensity, α k ,β k is the adjustment factor, D k (s) represents the kth feature extracted from the patient's ultrasound image, μ k is the expected value of the feature;

[0038] Finally, the real-time feedback and parameter optimization mechanism is realized, using the formula:

[0039]

[0040] where θ current and θ new is the model parameter before and after adjustment, γ is the real-time adjustment factor, y observed and predicted are the results observed during surgery and predicted by the model, φ(x data ,θ) is a parameterized feature extraction function.

[0041] Furthermore, the method for controlling the positioning and real-time navigation of the needle includes:

[0042] First, we implemented enhanced volume reconstruction technology based on machine learning to identify and construct three-dimensional models of nerves, blood vessels, and surrounding tissues by analyzing continuous two-dimensional ultrasound image sequences;

[0043] Then, through data fusion technology, the ultrasound image is combined with other imaging technology data including MRI or CT to enhance the detail richness of the obtained 3D model;

[0044] Finally, combined with real-time position sensing technology, an electromagnetic tracking system is used to monitor and transmit the needle's position information in real time, which is updated synchronously with the three-dimensional tissue model, and the real-time navigation system is controlled to accurately guide the needle to safely reach the target area along the predetermined path.

[0045] Furthermore, the enhanced volume reconstruction technique adopts the following method:

[0046] First, we implement the volume reconstruction technique based on deep learning, using the convolutional neural network CNN through the formula:

[0047]

[0048] Identify and predict the three-dimensional structure of nerves and blood vessels from continuous two-dimensional ultrasound images, where Θ represents the network parameters, α is the learning rate, N is the number of images, and x i is the input image, y i is the target structure label, f(x i ) represents the image x i Feature extraction function;

[0049] Then, the long short-term memory network LSTM is used to process the ultrasound image sequence, using the formula:

[0050] H t =σ s (W·[H t-1 ,Xt ]+b+∫ Ω κ(s,t)·H s ds)

[0051] Analyze the temporal relationship between images and adjust the algorithm to match patient-specific tissue variations, where H t is the hidden state at time t, X t is the input image, W and b are learning parameters, σ s is the activation function, κ(s,t) is the kernel function used to simulate H t With the previous state H s The dependency relationship of Ω is the set of past states;

[0052] Finally, ultrasound data is combined with other imaging techniques including CT or MRI through a multimodal data fusion network, formula:

[0053]

[0054] Create comprehensive 3D views with X US ,X CT ,X MRI Represent the image data of ultrasound, CT, and MRI respectively, Φ j represents the fusion parameter, φ j and ψ j is the function used for j-dimensional data processing, The entire data domain.

[0055] Furthermore, the method for constructing the enhanced detail richness of the obtained three-dimensional model is as follows:

[0056] First, we implement the multimodal data fusion network MMF-NN based on deep learning, through the algorithm:

[0057]

[0058] Identify and predict the three-dimensional structure of tissues from ultrasound US, MRI and CT images, where Θ represents the network parameters, η is the learning rate, N is the number of images, and x i,m is the input image from mode m, y i is the target structure label, f m represents the feature extraction function for mode m;

[0059] Then apply the feature extraction and matching algorithm, through the formula:

[0060]

[0061] Realize the alignment and feature matching of ultrasound US, MRI and CT images, represents the set of images from mode m, ω(x) is the weight function used to adjust the contribution of features from different imaging techniques, δ is the Dirac function that controls the overlap of images from each modality, and μ is the function that calculates the center of cross-modal features.

[0062] Furthermore, the real-time position sensing technology involves the following steps:

[0063] S1. First, we use the integrated electromagnetic tracking system EMT and real-time data processing algorithm. Through the algorithm:

[0064]

[0065] Update and adjust the three-dimensional tissue model, where Θ represents the parameters of the three-dimensional model, γ is the adjustment factor, is the loss function, which measures the deviation between the model and the electromagnetic tracking data, EMT t represents the electromagnetic tracking data at time t;

[0066] S2. Then apply dynamic 3D path planning and feedback mechanism to make path corrections based on real-time data, using the algorithm:

[0067]

[0068] Where Π represents the path planning parameter, η is the learning rate, is the predicted path output, y path is the actual needle path, is the gradient of the path output with respect to the planning parameters.

[0069] Beneficial effects of the present invention:

[0070] 1. By adopting an image processing algorithm based on deep learning, the present invention can adaptively enhance image data during ultrasound guidance, effectively improving the clarity and contrast of the image. This is especially important for identifying and locating tissue and organ structures, allowing anesthesiologists to see the details of the surgical area more clearly.

[0071] 2. The present invention can accurately segment different tissue and organ structures in ultrasound images, and reduce the loss of image details by independently processing and enhancing each organ and tissue area to be enhanced. This precise segmentation is of great significance to improving the success rate and safety of surgery, especially in complex tissue environments.

[0072] 3. By quantifying the brightness change index corresponding to the pixel points at the same position, the present invention can determine whether the tissue and organ structure to which the pixel points at the position belong is stable. For relatively stable tissue structures, the present invention can perform more accurate enhancement processing, thereby improving the overall image quality and enhancing the anesthesiologist's confidence in operation.

[0073] 4. The present invention can dynamically respond to changes in different tissue structures, such as position changes caused by breathing, and ensure that high accuracy and stability can be maintained during the image enhancement process through real-time analysis and processing. This dynamic response capability greatly reduces the risks during surgery and improves the anesthesia effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] Figure 1 The present invention is a flow chart of a data processing method for ultrasound-guided nerve block anesthesia.

[0075] Figure 2 This is a flow chart of the process of preprocessing ultrasound image data in the present invention.

[0076] Figure 3 A flowchart of the construction process of feature extraction for deep learning in the present invention.

[0077] Figure 4 This is a flow chart of the method for controlling the positioning and real-time navigation of the needle according to the present invention. DETAILED DESCRIPTION

[0078] The specific implementation modes of the present invention will be described in detail below in conjunction with the accompanying drawings.

[0079] Combined with Figure 1 , a data processing method for ultrasound-guided nerve block anesthesia involves the following steps: first, the ultrasound image is preprocessed, including using a deep learning model to remove random noise and structural noise in the image, and a denoising algorithm based on a generative adversarial network (GAN) is used, which generates a clear denoised image by training an adversarial model. In addition, local contrast enhancement techniques, including local adaptive histogram equalization (CLAHE), are used to specifically enhance the contrast of key structures in the ultrasound image, including nerves and blood vessels, making these structures more prominent and easy to identify in the image.

[0080] Secondly, feature extraction is performed through deep learning technology, which involves the use of convolutional neural networks (CNNs) to accurately identify and segment nerves, blood vessels, and muscle tissues in images. This process not only relies on a single ultrasound image mode, but also integrates data from multiple ultrasound modes, including blood flow information provided by Doppler mode, to comprehensively capture and analyze tissue characteristics. Multimodal data integration is achieved through specifically designed network layers, where each layer is responsible for extracting and fusing information from different modes, ensuring that the final output feature map has a high degree of information richness and relevance.

[0081] Next, recursive learning and adaptive algorithms are used to further optimize the processing flow. The recursive learning algorithm uses data collected from previous surgeries to continuously optimize the parameters of the model, making the system's feature extraction and image analysis more accurate after each operation. At the same time, the adaptive algorithm dynamically adjusts the image processing parameters according to the patient's specific body shape and tissue characteristics, and can adjust the intensity of image enhancement according to the patient's muscle density, which ensures the personalization and optimization of the processing process.

[0082] Finally, the continuous two-dimensional ultrasound image sequence is converted into a complete three-dimensional tissue model through a volume reconstruction algorithm. This step uses an algorithm to estimate the spatial relationship between each two-dimensional image layer and reconstruct a detailed three-dimensional structure. At the same time, the electromagnetic tracking system with real-time position sensing technology is used to monitor and transmit the position information of the needle in real time. The system synchronizes this information with the three-dimensional model and guides the needle to accurately reach the target area along the optimal path through the control algorithm, thereby greatly improving the safety and success rate of the operation. This method of combining advanced image processing technology with a real-time navigation system provides unprecedented accuracy and ease of operation for ultrasound guidance of block anesthesia.

[0083] Embodiment 1:

[0084] The embodiment adopts a deep learning-based block anesthesia ultrasound-guided data processing method to perform lumbar nerve block anesthesia on a patient with chronic low back pain. First, the ultrasound image data is preprocessed. A high-order partial differential equation with adaptive adjustment coefficients is used to remove image noise. The specific formula is:

[0085]

[0086] Among them, i and j are the orders of differentiation, and λ and σ are parameters that are dynamically adjusted based on the image content. In actual operation, the value range of i and j is selected to be 1 to 2, the value range of λ is 0.1 to 1, and the value range of σ is 1 to 5.

[0087] By this method, the noise in the ultrasound image is effectively removed, and the boundaries of nerves and blood vessels become clearer. Taking a certain ultrasound image pixel as an example, set λ = 0.5, σ = 2, and select i = 1, j = 1 for calculation. The original image gradient value is 2, and the value of the pixel after denoising N(x, y) is calculated by the formula:

[0088] 0.5*(2 2 )*exp(-((1 2 +1 2 ) / 8))=1.21

[0089] Significantly reduces noise impact.

[0090] Combined with Figure 2Next, a deep learning model is used for feature extraction. The convolutional neural network (CNN) is used to identify and segment nerves, blood vessels, and muscle tissues in ultrasound images, while integrating data from different ultrasound modes. The preprocessed ultrasound image is input, and the model identifies nerves and blood vessels and generates a segmented image showing the specific location of each tissue. The image processing process is further optimized through recursive learning algorithms and adaptive algorithms. The recursive learning algorithm continuously adjusts the model parameters based on the previous injection effect and ultrasound image. The formula is:

[0091]

[0092] Among them, θ t is the current model parameter, η is the learning rate, ranging from 0.01 to 0.1, is the loss function, and f is the model function. If the initial parameter θ0 is set to 1, the learning rate η = 0.05, and the loss function value is 0.5, then the new parameter θ1 = 1 + 0.05 * 0.5 = 1.025, indicating that the parameters are gradually optimized to improve the accuracy of image processing.

[0093] The adaptive algorithm adjusts the processing parameters according to body shape and tissue characteristics to ensure personalization and precision. The formula is:

[0094]

[0095] Among them, δ is the adjustment strength, ranging from 0.1 to 0.5, α k and β k is the adjustment coefficient, ranging from 0.1 to 1, μ k is the characteristic expected value. Set the initial parameter P original = 2, δ = 0.3, α1 = 0.5, β1 = 0.2, characteristic difference D1(s) = 0.8, expected value μ1 = 1, then P adjusted =2*(1+0.3*0.5*exp(-0.2*(0.8-1) 2 ))=2.15, showing the dynamic adjustment process of parameters.

[0096] Finally, a volume reconstruction algorithm is used to construct a complete three-dimensional tissue model from a continuous two-dimensional ultrasound image sequence. Combined with the electromagnetic tracking system EMT, the position information of the needle is monitored and transmitted in real time to ensure accurate positioning of the needle. Through the formula:

[0097]

[0098] Dynamically adjust the path planning, η is the learning rate, the value range is 0.01 to 0.1, Π is the path planning parameter. Set the initial path parameter Π0 to 3, η = 0.05, and the predicted path error to 0.4, then the new path parameter Π1 = 3 + 0.05 * 0.4 = 3.02, indicating that the path is gradually optimized.

[0099] Through these steps, doctors can accurately guide the needle to the target nerve area in real time during surgery, ensuring the success of anesthesia and the safety of patients. The whole process demonstrates the significant advantages of the new method in improving image processing accuracy, real-time and personalization.

[0100] This embodiment then uses a visual enhancement technique based on superposition integral transform to further improve the visibility of nerves and blood vessels in ultrasound images, and uses information of different scales and directions to enhance the image, making the key structures more prominent in the image. The specific formula is as follows:

[0101]

[0102] Ω is the image domain, w k is the weight of scale k, G k is a Gaussian blur function, and ω(x,y) is a position-dependent weight function used to adjust the enhanced locality.

[0103] Further improve the clarity of nerves and blood vessels in the image to more accurately position the needle. The specific operation is as follows:

[0104] The ultrasound image of the patient's lumbar spine area is selected as the processing object. The image domain Ω contains the entire lumbar spine area, including key nerve and vascular structures.

[0105] According to the different scales and directions of the image, different weights w are set k In order to improve the processing accuracy, three different scales are selected, and the weights are set as w1 = 0.4, w2 = 0.3, and w3 = 0.3. For each scale, a different Gaussian blur function G is used. k (x, y), these functions are used to smooth the image and extract the image features with different blur levels. The selected Gaussian blur function parameters are: the standard deviation of G1 σ1 = 1, the standard deviation of G2 σ2 = 2, and the standard deviation of G3 σ3 = 3.

[0106] The position-dependent weight function ω(x,y) adjusts the enhancement effect by analyzing the position of each pixel in the image. A weight function based on the image gradient is selected, which identifies the edges and important structures in the image, making the enhancement effect in these areas more significant. The specific definition is:

[0107]

[0108] Represents the gradient of the image, and σ=1.5 is the adjustment parameter.

[0109] When processing ultrasound images, an area in the lumbar spine image is selected for detailed calculation. The initial image intensity of a certain pixel in this area is set to I(x,y) = 150. According to the weights of different scales and the Gaussian blur function parameters, the calculation process of the processed pixel intensity C(I) is as follows:

[0110] 1. Gaussian blur processing:

[0111] Scale 1: I1(x,y)*G1(x,y)=120;

[0112] Scale 2: I2(x,y)*G2(x,y)=130;

[0113] Scale 3: I3(x,y)*G3(x,y)=140.

[0114] 2. Overlay weight:

[0115]

[0116] 3. Position-dependent weight calculation:

[0117] Image Gradient

[0118]

[0119] 4. Final enhancement results:

[0120]

[0121] After processing in this way, the nerve and blood vessel structures in the image are significantly enhanced, and doctors can more clearly identify and locate these key structures, thereby accurately guiding the needle to the target area, ensuring the success of the operation and the safety of the patient. This process demonstrates the feasibility and effectiveness of visual enhancement technology based on superposition integral transform in practical applications.

[0122] The embodiment further utilizes a network structure including local and global information feedback to control the segmentation of nerves and blood vessels. The network structure used can be expressed by the following formula:

[0123]

[0124] Among them, F i (x,y) is the feature extracted by the deep neural network, β iis the weight coefficient of the feature, ranging from 0.1 to 1. K is the kernel function used to enhance the spatial context information in the image, σ s It is the activation function, Sigmoid function.

[0125] The deep neural network DNN is used to extract local and global features from the enhanced ultrasound images. Each layer of the network extracts features of different scales and complexities, including edges, textures, and high-level semantic information. The number of features m is set to 5, and each feature F i The weight β of (x,y) i It is randomly initialized between 0.1 and 1 and adjusted through training.

[0126] The role of the kernel function K is to combine local and global information to increase the accuracy of segmentation. Select the Gaussian kernel function:

[0127]

[0128] Where σ=2.

[0129] When processing the patient's lumbar ultrasound image, an image block is selected for detailed calculation. The initial image intensity of a pixel in the image block is set to I(x,y) = 150. The features F1(x,y) to F5(x,y) extracted by the network are 0.6, 0.8, 0.7, 0.9, 0.5, respectively, and the corresponding weights are β1 = 0.4, β2 = 0.5, β3 = 0.3, β4 = 0.6, β5 = 0.2.

[0130] Calculate local feature contribution:

[0131]

[0132] Contribution of kernel function, setting the kernel range R = 3 and selecting Gaussian kernel parameter σ = 2:

[0133]

[0134] The average eigenvalue is set to 0.7, and the integral result is approximately:

[0135]

[0136] Calculation of the final segmentation result:

[0137] S(x,y)=σ(1.25+3.402)=σ(4.652)≈0.99

[0138] The segmentation result of each pixel in the ultrasound image is obtained. The result of σ is close to 1, indicating a high probability of being a nerve or a blood vessel.

[0139] The final embodiment uses a dynamic adjustment model that depends on image gradient and local brightness characteristics to optimize image contrast. The dynamic contrast adjustment model used can be expressed by the following formula:

[0140]

[0141] in, is the Laplacian operator used to detect boundaries and details in an image, γ(x,y) is a parameter adjusted according to the image characteristics, μ(I) is the global average intensity of the image, and θ is a parameter that controls the intensity of contrast adjustment, with a value range of 1 to 5.

[0142] Image gradient and local brightness characteristics are extracted from the enhanced and segmented image to determine the contrast adjustment parameter γ(x,y) for each pixel. The value of γ(x,K) is set to range from 0.5 to 2, depending on the local brightness and gradient information.

[0143] When processing the patient's lumbar ultrasound image, a key area is selected for detailed calculation. In this area, the initial image intensity of a certain pixel point is set to I(x,y) = 180, the global average intensity μ(I) = 150, and the parameter θ that controls the contrast adjustment intensity is set to 2.

[0144] Calculate the Laplacian operator for this pixel:

[0145]

[0146] According to the image characteristics, the contrast adjustment parameter γ(x, y) of the pixel is determined to be 1.5.

[0147] Substitute into the formula to calculate:

[0148]

[0149] Since exp(-15) is very close to 0, the simplified result is:

[0150] A(I,x,y)≈180+30×1=210

[0151] The contrast of this pixel is increased, making the edges of nerves and blood vessels more clearly visible.

[0152] Embodiment 2:

[0153] The embodiment uses multimodal data fusion technology to combine data from different ultrasound modes and extract features through deep learning. This process helps to fully understand and identify the anatomical structure of the target area. The multimodal data fusion formula used is as follows:

[0154]

[0155] M i (x, y) are data from different ultrasound modes, K i (x,y) is for mode M i The designed convolution kernel, σ i Controlling spatial scale, α i are the weights obtained through training.

[0156] Three different ultrasound modes are selected for data fusion, namely B-mode, Doppler mode and Elastography. For each mode, a different convolution kernel K is designed. i , where the convolution kernel parameter of B-mode is σ1=1.5, the convolution kernel parameter of Dopplermode is σ2=2.0, and the convolution kernel parameter of Elastography is σ3=1.0.

[0157] Through deep learning model training, the weight α of each mode is obtained i During the training process, the weights obtained are set to α1 = 0.5, α2 = 0.3, and α3 = 0.2.

[0158] When processing the patient's lumbar ultrasound image, a key area is selected for detailed calculation. The initial image intensity of a certain pixel in this area is set to I(x,y)=180.

[0159] For the B-mode data M1(x,K), after being processed by the convolution kernel L1(x,y) and Gaussian weighted:

[0160]

[0161] Weight α1 = 0.5;

[0162] For Doppler mode data M2(x,y), after being processed by convolution kernel K2(x,y) and Gaussian weighted:

[0163]

[0164] Weight α2 = 0.3;

[0165] For the Elastography data M3(x,y), after being processed by the convolution kernel K3(x,y) and Gaussian weighting:

[0166]

[0167] Weight α3=0.2.

[0168] Substitute these data into the multimodal fusion formula for calculation:

[0169]

[0170] Calculate the contribution of B-mode:

[0171]

[0172] Calculate the contribution of Dopplermode:

[0173]

[0174] Calculate the contribution of Elastography:

[0175]

[0176] The final fusion result:

[0177] F(x,y)=72.25+45.6+22.5=140.35

[0178] Through multimodal data fusion, doctors can combine information from different ultrasound modes to generate more accurate and comprehensive feature maps.

[0179] Combined with Figure 3 , the embodiment then performs hierarchical feature extraction through a deep learning model to further improve the visibility and segmentation accuracy of nerves and blood vessels in the image. The hierarchical feature extraction formula used is as follows:

[0180]

[0181] in, is a high-order Laplacian operator of the image, which is used to detect multi-level boundaries and details in the image. j (x,y) is the network L j Each layer output, β j is the level weight, ranging from 0.1 to 1.

[0182] Each layer of the deep learning model extracts features of different scales and complexities. The shallower layers extract edge and texture information, and the deeper layers extract high-level semantic information. Set the number of feature extraction layers m = 4, and the weight of each layer β j It is randomly initialized between 0.1 and 1 and adjusted through training.

[0183] When processing lumbar ultrasound images, a key area is selected for detailed calculation, and the initial image intensity of a certain pixel point in the area is set to I(x,y)=180.

[0184] The feature L1(x,y) extracted in the first layer is processed by a high-order Laplacian operator:

[0185]

[0186] The feature L2(x,y) extracted in the second layer is processed by a high-order Laplacian operator:

[0187]

[0188] The feature L3(x,y) extracted in the third layer is processed by a high-order Laplacian operator:

[0189]

[0190] The feature L4(x,y) extracted from the fourth layer is processed by a high-order Laplace operator:

[0191]

[0192] Substitute these data into the hierarchical feature extraction formula for calculation:

[0193]

[0194] Calculate the contribution of the first layer:

[0195]

[0196] Calculate the contribution of the second layer:

[0197]

[0198] Calculate the contribution of the third layer:

[0199]

[0200] Calculate the contribution of the fourth layer:

[0201]

[0202] Final feature extraction results:

[0203] T(x,y)=9+8+5+1.5=23.5

[0204] Through hierarchical feature extraction, doctors can extract multi-level features in ultrasound images, making the boundaries and details of nerves and blood vessels clearer.

[0205] The embodiment further uses an end-to-end deep learning model to accurately segment nerves and blood vessels, and the formula is as follows:

[0206]

[0207] Among them, V(u,v) is the trained transformation kernel, which is used to extract key tissue features in image I, including the saliency of nerves and blood vessels.

[0208] The transformation kernel V(u,v) is obtained by training with ultrasound image data. The transformation kernel size is set to 3×3, and each value is fixed to the following matrix after training:

[0209]

[0210] When processing lumbar ultrasound images, a key area is selected for detailed calculation. The initial image intensity matrix I of a 3×3 pixel point in this area is set to be:

[0211]

[0212] Substitute these data into the segmentation formula to calculate:

[0213]

[0214] Calculate the specific value:

[0215] =36+95+40+68+144+76+32+85+36=612

[0216] Apply the activation function:

[0217]

[0218] Each pixel is calculated through the convolution of the transformation kernel V(u,v) and the initial image intensity I(xu,yv), and finally generates a probability value close to 1, indicating that the area is highly likely to be a nerve or blood vessel.

[0219] Embodiment 3:

[0220] The embodiment uses a recursive learning algorithm to adjust and optimize the model parameters of feature extraction, with the goal of dynamically adjusting the model based on previous surgical results and ultrasound image data to more accurately match actual surgical needs. The core formula of the algorithm is as follows:

[0221]

[0222] θ t is the model parameter at time t, η is the adaptive learning rate, which is set in the range of 0.01 to 0.1 to ensure a smooth learning process, and y t It is the surgical effect data. is the effect predicted by the model, p(y t ∣x t θ t ) is the conditional probability density function.

[0223] During the patient's surgery, the doctor recorded the effect data of the previous surgery. t, and through real-time ultrasound images x t The effect prediction was carried out Set at a specific moment, the actual surgical effect is y t = 0.8 (complete success is 1), and the surgical effect predicted by the model is

[0224] Set the learning rate η = 0.05;

[0225] Set the initial value of the model parameter θ t =1.0;

[0226] Set the gradient of the conditional probability density function Estimated to be 0.4.

[0227] Substitute these data into the recursive learning algorithm formula for calculation:

[0228]

[0229] The calculation process is simplified (the integral result is set to 1, which represents the entire image area):

[0230] θ t+1 =1.0+0.05×(0.2×0.4)=1.0+0.05×0.08=1.004

[0231] Through the above recursive learning algorithm, doctors can dynamically adjust the model parameters θ based on the actual surgical results and ultrasound image data. t , making the model closer to the actual surgical situation.

[0232] The embodiment further adaptively adjusts the image processing parameters to better visualize key structures, including nerves and blood vessels, thereby improving the accuracy and safety of surgery. The adaptive algorithm formula used is as follows:

[0233]

[0234] Among them, P original is the original image processing parameter, δ is the adjustment strength, and the value range is 0.05 to 0.2 to ensure that the adjustment is not excessive; α k and β k is the adjustment coefficient, which controls the weight and adjustment rate of the feature respectively, and the value range is generally 0.1 to 1; D k (s) represents the kth feature extracted from the patient's ultrasound image, μ k is the expected value of the feature.

[0235] During the patient's preoperative ultrasound examination, the doctor determined several key features, including muscle tissue density, fat layer thickness, and blood vessel location. The number of features was set to n = 3, and the adjustment factor for each feature was calculated based on the patient's specific data.

[0236] Set the original image processing parameters P original =1.0;

[0237] The adjustment strength δ=0.1 was selected.

[0238] Feature weight α k and the adjustment rate β k Set to:

[0239] α1=0.8,β1=0.5 (muscle density)

[0240] α2=0.7,β2=0.6(fat layer thickness)

[0241] α3=0.6,β3=0.7 (blood vessel position)

[0242] Set actual feature D k (s) and expected value μ k for:

[0243] D1(s)=2.0,μ1=1.5(muscle density)

[0244] D2(s)=1.0,μ2=1.0(fat layer thickness)

[0245] D3(s)=0.5,μ3=0.8(blood vessel position)

[0246] Calculate the adjustment factor:

[0247]

[0248] Calculate the adjusted parameters:

[0249] P adjusted =1.0×(1+0.1×1.969)=1.0×1.197=1.197

[0250] Adaptive algorithms enable physicians to precisely adjust ultrasound image processing parameters based on the patient's individual body shape and tissue characteristics.

[0251] The next step of the embodiment is to implement a real-time feedback and parameter optimization mechanism to adjust the model parameters of the ultrasound guidance system in real time, and dynamically optimize according to the difference between the actual situation observed during the operation and the results predicted by the pre-model. The real-time feedback and parameter optimization formula used is as follows:

[0252] θ new =θ current+γ(∫(y observed -y predicted )·φ(x data ,θ)dx data )

[0253] Among them, θ current and θ new is the model parameter before and after adjustment, γ is the real-time adjustment factor, and the value range is usually set to 0.01 to 0.05 to ensure stable adjustment; observed and predicted are the results observed during surgery and predicted by the model, respectively; φ(x data ,θ) is a parameterized feature extraction function.

[0254] Actual nerve responses monitored by doctors during patient surgery observed and the neural response y predicted by the model predicted There was some deviation in the early stage of the operation. Setting:

[0255] y observed =0.9 (nerve response intensity); y predicted =0.7; initial value of parameter θ current =2.0; real-time adjustment factor γ = 0.03; feature extraction function φ(x data ,θ) is calculated to be 0.8.

[0256] Substitute these data into the real-time feedback and parameter optimization formula for calculation:

[0257] θ new =2.0+0.03(∫(0.9-0.7)·0.8dx data )

[0258] The calculation process is simplified (the integral result is set to 1, which reflects the entire affected area):

[0259] θ new =2.0+0.03×(0.2×0.8)=2.0+0.03×0.16=2.0048

[0260] Through real-time feedback and parameter optimization mechanisms, doctors can ensure that the ultrasound guidance system adapts to the actual surgical conditions during the operation and adjust the model parameters in time to match the observed surgical effects.

[0261] Embodiment 4:

[0262] The embodiment adopts enhanced volume reconstruction technology based on machine learning, which identifies and constructs three-dimensional models of nerves, blood vessels and surrounding tissues by analyzing continuous two-dimensional ultrasound image sequences. The core is to use deep learning, especially convolutional neural network CNN, to ensure the accuracy and real-time performance of the reconstructed model. The formula of enhanced volume reconstruction technology is as follows:

[0263]

[0264] Θ represents the network parameters, α is the learning rate, which ranges from 0.001 to 0.01 to ensure a stable learning process, N is the number of images, and x i is the input image, y i is the target structure label, f(x i ) represents the image x i The feature extraction function.

[0265] During the preparation for surgery, the doctor used ultrasound equipment to collect multiple continuous two-dimensional images of the patient's waist area. These images covered different depths from the skin surface to the deep nerves and blood vessels. The trained CNN model was used to extract key features in each two-dimensional image, including the location of nerve bundles and blood vessels. The number of images was set to N = 30, and the feature extraction function f(x i ) can identify key structures.

[0266] Set the initial values ​​of the model parameters Set the learning rate α = 0.005;

[0267] Suppose that in a certain training process, for a certain image x i , the target structure label predicted by the model The actual label y i =0.8, then the loss function gradient is 0.4.

[0268] Substitute into the formula to calculate:

[0269]

[0270] The calculation process is simplified (assuming that the gradient contributed by each image is the same):

[0271]

[0272] Through deep learning-based enhanced volume reconstruction technology, doctors can not only build accurate 3D models, but also adjust and optimize the position of the needle in real time during the operation to ensure the safety and success rate of each step of the operation.

[0273] The embodiment further applies a long short-term memory network LSTM to process ultrasound image sequences and analyze the temporal relationship between images to match patient-specific tissue changes. The formula is as follows:

[0274]

[0275] H t is the hidden state at time t, X t is the input image, W and b are learning parameters with values ​​ranging from 0.01 to 0.1, σ s is the activation function, usually Sigmoid function, κ(s,t) is the kernel function used to simulate H t With the previous state H s The dependency of Ω is the set of past states.

[0276] The ultrasound device is used to collect a continuous ultrasound image sequence of the patient's waist area, with a total of 30 frames, and the size of each frame is 512x512 pixels.

[0277] Set the initial value of the hidden state H0 = 0, and the learning parameters W = 0.05, b = 0.1;

[0278] Select Gaussian kernel function Where σ=2.

[0279] For the first time step t=1:

[0280] H1=σ s (W·[H0,X1]+b+∫ Ω κ(s,1)·H s ds)

[0281] Set the feature extraction result of X1 to 200, and the integral part mainly depends on the initial state H0 being 0, and the calculation is simplified to:

[0282] H1=σ s (0.05·(0+200)+0.1)=σ(10.1)≈1

[0283] For the second time step t=2:

[0284] H2=σ s (W·[H1,X2]+b+∫ Ω κ(s,2)·H s ds)

[0285] Set the feature extraction result of X2 to 190, and H1≈1:

[0286]

[0287] because The main contribution at s = 1 is approximately the value of H1:

[0288] H2≈σ(0.05·191+0.1+1×0.6065)≈σ(9.65+0.1+0.6065)=σ(10.3565)≈1

[0289] This LSTM technology for processing ultrasound image sequences significantly improves the doctor's reaction speed and decision-making accuracy during surgery, ensuring that the needle reaches the target area along the optimal path.

[0290] The final example combines ultrasound data with imaging technology to create a comprehensive three-dimensional view through a multimodal data fusion network. The formula used is as follows:

[0291]

[0292] Among them, X US ,X CT ,X MRI Represent the image data of ultrasound, CT, and MRI respectively, Φ j represents the fusion parameter, φ j and ψ j is the function used for j-dimensional data processing, The entire data domain.

[0293] Ultrasound images, CT scan images, and MRI images of the patient's lumbar region were collected. The ultrasound images had 30 frames, and the CT and MRI images had 10 frames each, and the image size of each frame was 512x512 pixels.

[0294] In order to ensure the fusion of different modality images, image registration is first performed to ensure that ultrasound, CT and MRI images are aligned in the same spatial coordinate system.

[0295] Set the initial fusion parameter Φ j The value range is 0.1 to 1, and the specific setting is Φ j =0.5.

[0296] φ j and ψ j The settings are as follows:

[0297] φ j (x) = x 1.5 ,ψ j (C US ,X CT ,X MRI ; Φ j )=Φ j ·(X US +X CT +X MRI )

[0298] Set at a certain pixel position, the ultrasound image data is X US =150, CT image data is X CT =200, MRI image data is X MRI =180.

[0299] ψ j (X US ,X CT ,X MRI ; Φ j )=0.5·(150+200+180)=0.5·530=265

[0300]

[0301] Set J = 3, and each φ j The functions are the same:

[0302]

[0303] This three-dimensional view provides more detailed anatomical information than single-modality images, allowing doctors to see the patient's nerves, blood vessels and other important tissues more clearly.

[0304] Embodiment 5:

[0305] The embodiment uses a multimodal data fusion network MMF-NN based on deep learning to enhance the detail richness of the obtained three-dimensional model, and the formula used is:

[0306]

[0307] Θ represents the network parameters, η is the learning rate (usually ranging from 0.001 to 0.01), N is the number of images, and x i,m is the input image from modality m (ultrasound US, MRI, CT), y i is the target structure label, f m Represents the feature extraction function for pattern m.

[0308] Multimodal image data including ultrasound, MRI and CT were collected from the patient's waist area. Each imaging technique was assumed to provide 30 images. The images of each imaging technique were extracted by its corresponding feature extraction function f m Processed to identify and label nerves, blood vessels, and other critical structures.

[0309] set up The initial value is random; the learning rate η is set to 0.005.

[0310] The above formula is used to update the network parameters through an iterative process to optimize the prediction ability of the model and enhance the detail richness of the 3D model. It is set that in a certain iteration, for an ultrasound image x i,US , MRI image x i,MRI and CT image x i,CT The feature extraction result is f US (x i,US )=0.8,f MRI (x i,MRI )=0.85,f CT (x i,CT )=0.9.

[0311]

[0312] Assuming the average contribution of the simplified gradient calculation is 0.3, then:

[0313]

[0314] This deep learning multimodal data fusion network enables doctors to rely on more detailed and comprehensive 3D models to locate and navigate the needle during surgery, significantly improving the success rate and safety of the surgery. In real-time operation, doctors can observe detailed changes in real time, adjust surgical strategies, ensure precise operation of the needle during surgery, and reduce risks and discomfort to patients.

[0315] The embodiment further uses a feature extraction and matching algorithm to ensure accurate alignment and feature matching between ultrasound US, magnetic resonance imaging MRI and computed tomography CT images. The algorithm formula used is as follows:

[0316]

[0317] in, represents the set of images from mode m, ω(x) is the weight function used to adjust the contribution of features from different imaging techniques, δ is the Dirac function that controls the overlap of images from each modality, and μ is the function that calculates the center of cross-modal features.

[0318] Ultrasound, MRI, and CT images of the patient's lumbar region were collected. Each imaging technique provided detailed images of internal structures. Image registration was first performed using image processing techniques to ensure that the images of different modalities were aligned in spatial coordinates. Feature extraction algorithms were then applied to the images of each modality to identify key anatomical structures, including nerves and blood vessels. A feature matching algorithm was then used to calculate the correspondence and center position of each feature in different modalities to achieve high-precision feature alignment.

[0319] It is assumed that all three modality images show a key blood vessel in a specific area. USThe position in the MRI image x is 100. MRI The position in the CT image is 102. CT The position in is 101.

[0320] Set the weight function ω(x) to be uniformly distributed, that is, all features contribute equally. Calculate the cross-modal feature center:

[0321] Feature matching calculation:

[0322] Ω aligned =∫ {100,102,101} ω(x)δ(x-101)dx

[0323] Using the Dirac function δ ensures that only fully overlapping features are represented, which in fact shows that all modes are highly consistent at position 101.

[0324] This feature extraction and matching algorithm ensures that ultrasound, MRI, and CT images are accurately aligned during surgery, providing a comprehensive and unified view, making surgical navigation more accurate.

[0325] Embodiment 6:

[0326] Combined with Figure 4 The embodiment introduces real-time position sensing technology and combines it with the electromagnetic tracking system EMT to monitor and transmit the position information of the needle in real time. Using the integrated electromagnetic tracking system and real-time data processing algorithm, the three-dimensional tissue model is updated and adjusted through the following algorithm:

[0327]

[0328] Among them, Θ represents the parameters of the three-dimensional model, and γ is the adjustment factor, which is usually set in the range of 0.001 to 0.01 to ensure stability; is a loss function that measures the deviation between the model and the electromagnetic tracking data; EMT t represents the electromagnetic tracking data at time t.

[0329] Electromagnetic sensors are installed on the needle and the surgical area to ensure that all movements and position information are captured and transmitted in real time. Every movement and position change of the needle is recorded and synchronized to the 3D tissue model in real time, so that the model is continuously updated to reflect the actual surgical scene.

[0330] Set the initial parameter Θ old =1.0.

[0331] Loss Function Represents the deviation between the actual position of the needle and the position predicted by the model. The loss function calculates this deviation every time the needle position is updated.

[0332] Assume that at a certain time point t, the electromagnetic tracking data EMT t The needle deviates from the predetermined path, and the calculated loss function gradient is 0.2, then:

[0333]

[0334] Through real-time position sensing technology, doctors can monitor the position of the needle in real time during surgery and precisely control its movement to ensure that it follows the intended path.

[0335] The embodiment further applies dynamic three-dimensional path planning and feedback mechanism to perform path correction based on real-time data to ensure that the needle reaches the target area safely along the predetermined path. The path correction algorithm used is as follows:

[0336]

[0337] Π represents the path planning parameter, η is the learning rate, which is set in the range of 0.001 to 0.01 to ensure a stable learning process; Predicted path output, y path is the actual needle path, is the gradient of the path output with respect to the planning parameters.

[0338] The electromagnetic tracking system is used to monitor the position of the needle in real time and obtain the actual path of the needle at every moment during the operation. path Based on the current 3D model and electromagnetic tracking data, the algorithm predicts the needle's path And adjust the path planning parameter Π according to real-time feedback; set the initial parameter Π old =2.0; Set the predicted path output 1.8, actual path y path = 2.0, the gradient of the path output to the planning parameters Set the learning rate η = 0.005, then:

[0339] Π new =2.0+0.005∫(1.8-2.0) 2 0.5dx path

[0340] Since (1.8-2.0) 2 =0.04, calculate the integral part:

[0341] Π new =2.0+0.005·0.04·0.5=2.0+0.0001=2.0001

[0342] Through this dynamic three-dimensional path planning and feedback mechanism, doctors can continuously correct the needle's path based on real-time data during the operation to ensure that the needle safely reaches the target area along the predetermined path.

[0343] The above shows and describes the basic principles, main features and advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments, and the above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A method for processing data under ultrasound guidance for nerve block anesthesia, characterized in that The following steps are involved: First, the ultrasound image data is preprocessed by using a deep learning-based denoising algorithm to remove image noise, and a local contrast enhancement algorithm is used to enhance the visibility of key structures, including nerves and blood vessels. Secondly, deep learning is used for feature extraction to identify and segment nerves, blood vessels, and muscle tissues in the image, while integrating data from different ultrasound modes to extract tissue features; The processing flow is then optimized using a recursive learning algorithm that optimizes the accuracy of feature extraction and image analysis based on previous injection results and ultrasound images, and an adaptive algorithm that adjusts processing parameters based on the body shape and tissue characteristics of different patients. Finally, a volume reconstruction algorithm is used to construct a complete three-dimensional tissue model from a continuous two-dimensional ultrasound image sequence, and real-time position sensing technology is combined to control the positioning and real-time navigation of the needle; The process of preprocessing the ultrasonic image data comprises: A high-order partial differential equation with adaptive adjustment coefficients is used to remove image noise; the specific formula is: Where N(x,y) is the value of the pixel after denoising, i and j are the orders of differentiation, and λ and σ are parameters that are dynamically adjusted based on the image content; Then, for the visual enhancement of key structures including nerves and blood vessels, the stacking integral transform is used to combine information of different scales and directions: Where Ω is the image domain, w k is the weight of scale k, G k is the Gaussian blur function, ω(x,y) is the position-dependent weight function, adjusting the enhanced locality; Then, a network structure with local and global information feedback is used to control nerve and blood vessel segmentation: Among them, F i (x, y) are features extracted by a deep neural network, and K is a kernel function used to enhance the spatial context information in the image; Finally, according to the image contrast differences of different patients, a dynamic adjustment model that depends on image gradient and local brightness characteristics is designed: in, is the Laplace operator, which represents the second-order derivative of the image and is used to detect boundaries and details. γ(x,y) is a parameter adjusted according to the image characteristics. μ(I) is the global average intensity of the image. θ is a parameter that controls the intensity of contrast adjustment. The method for controlling the positioning and real-time navigation of the needle comprises: First, we implemented enhanced volume reconstruction technology based on machine learning to identify and construct three-dimensional models of nerves, blood vessels, and surrounding tissues by analyzing continuous two-dimensional ultrasound image sequences; Then, through data fusion technology, the ultrasound image is combined with other imaging technology data including MRI or CT to enhance the detail richness of the obtained 3D model; Finally, combined with real-time position sensing technology, an electromagnetic tracking system is used to monitor and transmit the needle's position information in real time, which is updated synchronously with the three-dimensional tissue model, and the real-time navigation system is controlled to accurately guide the needle to safely reach the target area along the predetermined path.

2. The method for processing data under ultrasound guidance for nerve block anesthesia according to claim 1, characterized in that The construction process of deep learning for feature extraction includes: First, multimodal data fusion is used through the function: Integration of different ultrasound modes i data, where K i (x,y) is for mode M i The designed convolution kernel, σ i Controlling spatial scale, α i are the weights obtained through training; Then, hierarchical feature extraction is performed through the deep learning model, applying the formula: Network L j Each layer output is weighted high-order differential, where β j is the level weight, It is a high-order Laplacian operator of an image, used to detect multi-level boundaries and details in an image; Finally, an end-to-end deep learning model is used for segmentation, using the formula: Where V(u,v) is the trained transformation kernel, which is used to extract key tissue features in image I.

3. The method for processing data under ultrasound guidance for nerve block anesthesia according to claim 1, characterized in that The recursive learning algorithm optimizes feature extraction based on the results of previous surgeries and ultrasound images using the formula: where θ t is the model parameter at time t, η is the adaptive learning rate, y t It is the surgical effect data. is the effect predicted by the model, p(y t |x t θ t ) is the conditional probability density function, x t It is a real-time ultrasound image; Then an adaptive algorithm is applied to adjust the image processing parameters according to the patient's body size and tissue characteristics, using the formula: Where P original is the original image processing parameter, δ is the adjustment intensity, α k ,β k is the adjustment factor, D k (s) represents the kth feature extracted from the patient's ultrasound image, μ k is the expected value of the feature; Finally, the real-time feedback and parameter optimization mechanism is realized, using the formula: where θ current and θ new is the model parameter before and after adjustment, γ is the real-time adjustment factor, y observed and predicted are the results observed during surgery and predicted by the model, φ(x data ,θ) is a parameterized feature extraction function.

4. The method for processing data under ultrasound guidance for nerve block anesthesia according to claim 1, characterized in that The method adopted by the enhanced volume reconstruction technique is: First, we implement the volume reconstruction technique based on deep learning, using the convolutional neural network CNN through the formula: Identify and predict the three-dimensional structure of nerves and blood vessels from continuous two-dimensional ultrasound images, where Θ represents the network parameters, α is the learning rate, N is the number of images, and x i is the input image, y i is the target structure label, f(x i ) represents the image x i Feature extraction function; Then, the long short-term memory network LSTM is used to process the ultrasound image sequence, using the formula: Analyze the temporal relationship between images and adjust the algorithm to match patient-specific tissue variations, where H t is the hidden state at time t, X t is the input image, W and b are learning parameters, σ s is the activation function, κ(s,t) is the kernel function used to simulate H t With the previous state H s The dependency relationship of Ω is the set of past states; Finally, ultrasound data is combined with other imaging techniques including CT or MRI through a multimodal data fusion network, formula: Create comprehensive 3D views with X US ,X CT ,X MRI Represent the image data of ultrasound, CT, and MRI respectively, Φ j represents the fusion parameter, φ j and ψ j is the function used for j-dimensional data processing, The entire data domain.

5. The method for processing data under ultrasound guidance for nerve block anesthesia according to claim 1, characterized in that The method for constructing the enhanced detail richness of the obtained three-dimensional model: First, we implement the multimodal data fusion network MMF-NN based on deep learning, through the algorithm: Identify and predict the three-dimensional structure of tissues from ultrasound US, MRI and CT images, where Θ represents the network parameters, η is the learning rate, N is the number of images, and x i,m is the input image from mode m, y i is the target structure label, f m represents the feature extraction function for mode m; Then apply the feature extraction and matching algorithm, through the formula: Realize the alignment and feature matching of ultrasound US, MRI and CT images, represents the set of images from mode m, ω(x) is the weight function used to adjust the contribution of features from different imaging techniques, δ is the Dirac function that controls the overlap of images from each modality, and μ is the function that calculates the center of cross-modal features.

6. The method for processing data under ultrasound guidance for nerve block anesthesia according to claim 1, characterized in that The real-time position sensing technique involves the following steps: First, we use an integrated electromagnetic tracking system EMT and real-time data processing algorithm. Through the algorithm: Update and adjust the three-dimensional tissue model, where Θ represents the parameters of the three-dimensional model, γ is the adjustment factor, is the loss function, which measures the deviation between the model and the electromagnetic tracking data, EMT t represents the electromagnetic tracking data at time t; Then, dynamic 3D path planning and feedback mechanism are applied to make path corrections based on real-time data, using the following algorithm: Where Π represents the path planning parameter, η is the learning rate, is the predicted path output, y path is the actual needle path, is the gradient of the path output with respect to the planning parameters.

Images