Electrical impedance imaging method, storage medium, and device based on digital twin drive

Abstract

The present invention proposes an electrical impedance imaging method, storage medium, and device based on digital twin drive, which belongs to the field of medical imaging technology. The technical key points are: the WOVAE model is used to find the optimal potential features that match the read boundary voltage data set, and then the VAE model constructed by the virtual layer obtains the best conductivity image, and then the image output by the VAE model is interactively fed back through the communication layer. The whole method compensates for the invalid voltage data collected by the physical layer EIT device, and finally realizes highly robust lung EIT imaging. The use of the electrical impedance imaging method, storage medium, and device based on digital twin drive of the present application can effectively improve the robustness of imaging.

Description

[0001] This application is a divisional application with application number 202410330738.8, application date March 22, 2024, applicant Jilun Medical Engineering Intelligent Technology (Nanjing) Co., Ltd., and invention name "Method for seeking optimal electrical characteristics based on WOWAE, electrical impedance imaging method, storage medium, and device". Technical Field

[0002] The present invention belongs to the technical field of medical imaging and deep learning, and specifically relates to an electrical impedance imaging method, storage medium, and device based on digital twin drive. Background Art

[0003] EIT (full name: Electrical Impedance Tomography) is a visualization imaging technology that reconstructs the internal conductivity distribution based on the boundary voltage of biological tissues. It has unique application advantages, especially in bedside continuous real-time monitoring scenarios (such as: CN115245322A, CN117338278A, CN117179732A). However, EIT technology still faces some challenges, especially the problem of severely limited spatial resolution due to its soft field characteristics.

[0004] To overcome the above problems, researchers have adopted various advanced regularization methods to improve the EIT reconstruction quality.

[0005] The first technical route: Static EIT explores shape-driven reconstruction methods to restore the shape information of internal targets through spatial prior knowledge (such as: CN112401865A, CN110251130A, CN112401866A).

[0006] The second technical route: Dynamic EIT involves time series analysis of voltage measurement or image reconstruction, introduces the concept of time-space correlation, and uses time-space prior knowledge to obtain higher resolution reconstruction results (such as: CN111192337, CN110910466A).

[0007] The third technical route: Compared with the previous two traditional regularization technical routes, the EIT method based on deep learning can flexibly introduce prior information through training data, so that the model can better understand and reflect the potential characteristics of the object to be imaged. At the same time, once the model training is completed, the imaging speed is significantly improved. At the same time, the EIT method based on deep learning can also improve the resolution of the reconstructed image, making the imaging results clearer and more accurate. (For example: CN114663544A, CN112001977A, CN115670421A, CN114052702A).

[0008] Based on the above existing technology research, it is found that the EIT method based on deep learning still has the following problems:

[0009] (1) In long-term respiratory monitoring scenarios, changes in patient position or improper operation often lead to poor electrode contact or partial disconnection. When there are one or more faulty electrodes, the collected voltage data will be corrupted, resulting in severe image distortion. In this case, blindly reconstructing conductivity images from inaccurate measurements may lead to incorrect treatment decisions.

[0010] (2) Deep learning-based lung EIT imaging methods have made significant progress, but they still face some challenges, especially the adverse effects of model errors such as measurement noise, contact impedance, chest deformation, and electrode failure. In this case, deep learning-based electrical impedance tomography methods need to obtain reliable reconstruction results in the presence of faulty electrodes to ensure accurate diagnosis and treatment. Summary of the invention

[0011] The purpose of the present invention is to solve the problems existing in the above-mentioned prior art and to provide a method for seeking optimal electrical characteristics based on the WOVAE model.

[0012] Another object of the present invention is to provide an electrical impedance imaging method based on digital twin driving.

[0013] Another object of the present invention is to provide a storage medium.

[0014] Another object of the present invention is to provide a computer device.

[0015] The technical solution of this application is:

[0016] A method for seeking optimal electrical characteristics based on a WOVAE model comprises the following steps:

[0017] S1, input boundary voltage data set u;

[0018] S2, initialization: initialize a group of whale individuals in the latent space, each whale represents a set of electrical features;

[0019] S3, calculate the fitness of each whale's position and find the optimal position of the whale;

[0020] S3-1, converting each set of electrical characteristics of each whale into a corresponding simulated boundary voltage data set; wherein a method of converting a set of electrical characteristics into a corresponding simulated boundary voltage data set is:

[0021] First, the electrical features are input into a trained VAE model, and the trained VAE model outputs a high-precision reconstructed EIT image;

[0022] Then, the reconstructed EIT image is forward modeled and solved to obtain a simulated boundary voltage data set v;

[0023] S3-2, calculate the fitness value f(σ) of each whale, and select the optimal whale position and retain it;

[0024]

[0025] Among them, cov represents covariance, var represents variance, v and u are the simulated boundary voltage data set and the input boundary voltage data set respectively;

[0026] S4, based on the optimal whale position determined in step S3, surround the prey, attack with a bubble net and search for the prey, and update the whale's position;

[0027] S5: Determine whether the iteration is completed;

[0028] Determine whether the termination condition is met: if so, output the optimal set of electrical characteristics; otherwise, return to step S3.

[0029] An electrical impedance tomography method based on digital twin driving comprises the following steps:

[0030] S100, reading a boundary voltage data set;

[0031] S200, using the aforementioned method to seek the best electrical characteristics;

[0032] S300, inputting the optimal electrical features obtained in step S200 into the trained VAE model, and the VAE model outputs the reconstructed EIT imaging.

[0033] Furthermore, step S100 also includes the following sub-steps:

[0034] S101, read the electrical impedance monitoring device to obtain the boundary voltage data set u 实测 ;

[0035] S102, obtain the input boundary voltage data set u:

[0036]

[0037] Among them, λ is a regularization parameter, is an identity matrix, is a submatrix consisting of all the columns corresponding to the triangular elements that lie near the boundary.

[0038] A storage medium stores a computer program, wherein the computer program is used to execute the steps of the aforementioned method for seeking optimal electrical characteristics.

[0039] A storage medium stores a computer program, wherein the computer program is used to execute the steps of the aforementioned electrical impedance imaging method.

[0040] A computer device includes a memory and a processor, wherein the memory stores computer-readable instructions, and when the processor executes the computer-readable instructions, the steps of the method for seeking optimal electrical characteristics based on the WOVAE model as described above are implemented.

[0041] A computer device includes a memory and a processor, wherein the memory stores computer-readable instructions, and when the processor executes the computer-readable instructions, the steps of the electrical impedance imaging method based on digital twin drive as described above are implemented.

[0042] The advantages of the technical solution of the present invention are mainly reflected in:

[0043] (1) This application develops a new twin EIT framework, which aims to compensate for the invalid data of faulty electrodes and achieve highly robust lung imaging. The twin EIT framework proposed in this application is driven by a continuous circulation flow of information and data from physical space to virtual space and then back to physical space. At the same time, it deeply integrates EIT perception technology and deep convolutional evolutionary algorithm to establish a digital mapping between the virtual WOWAE model and the physical EIT device. This realizes data sharing, interactive feedback and visual analysis in the entire respiratory monitoring process.

[0044] (2) For the WOVAE algorithm, the fundamental problem in its application is how to determine the fitness function (the fitness function is used to evaluate the quality of the whale's position).

[0045] The fitness function determined in this application is:

[0046]

[0047] Among them, cov represents covariance, var represents variance, v and u are the simulated boundary voltage dataset and the input boundary voltage dataset respectively. In the process of latent space exploration, the larger the objective function f(σ), the higher the similarity between the boundary voltage dataset calculated by the latent feature and the input boundary voltage dataset, and the more likely the whale position is to be the optimal solution.

[0048] (3) After the fitness function is determined, another problem arises. The WOA algorithm is initialized to obtain simulated electrical characteristics, not simulated boundary voltage data sets. Therefore, the simulated electrical characteristics obtained by initialization need to be converted into boundary voltage data sets before the fitness function can be calculated.

[0049] To solve this problem, the VAE model is called into the WOA algorithm, the initialized simulated electrical characteristics are input into the trained VAE model, and the VAE model outputs a reconstructed EIT image. The reconstructed EIT image is forward modeled and solved to obtain a simulated boundary voltage dataset (that is, a simulated boundary voltage dataset corresponding to each whale is obtained).

[0050] That is, in the process of seeking the optimal electrical features in the WOA algorithm, VAE features are embedded (WOA embeds VAE features to form a WOAE model) in order to solve the fitness. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] The present application is further described in detail below in conjunction with the embodiments in the accompanying drawings, but does not constitute any limitation to the present application.

[0052] Figure 1 This is an electrical impedance imaging method based on digital twin driving in the present invention.

[0053] Figure 2 These are the EIT reconstruction results from 5 patients in the training sample in the present invention.

[0054] Figure 3 These are the EIT reconstruction results from 5 patients in the test samples in the present invention.

[0055] Figure 4 Statistical diagram of EIT reconstruction results from test samples in the present invention.

[0056] Figure 5 This is a diagram showing the robustness analysis of EIT reconstruction to measurement noise in the present invention.

[0057] Figure 6 This is a diagram showing the robustness analysis of EIT reconstruction to chest deformation in the present invention.

[0058] Figure 7 This is a diagram showing the robustness analysis of EIT reconstruction for a faulty electrode in the present invention. DETAILED DESCRIPTION

[0059] The purpose, advantages and features of the present invention will be explained by the non-limiting description of the following preferred embodiments. These embodiments are only typical examples of the application of the technical solution of the present invention, and any technical solution formed by equivalent replacement or equivalent transformation falls within the scope of protection claimed by the present invention.

[0060] <1: A method of electrical impedance imaging based on digital twin drive>

[0061] <1.1、Technical issues solved>

[0062] During EIT imaging, the boundary voltage data matrix collected by several electrodes is used to achieve EIT imaging. If one or more electrodes have unpredictable abnormalities, the incorrect voltage collection will seriously affect the imaging results, resulting in the loss of EIT imaging shape.

[0063] <1.2、Technical ideas>

[0064] The technical idea of ​​EIT imaging based on digital twin drive proposed in this application is that the WOVAE model is used to find the optimal potential features that match the read boundary voltage data set, and then the VAE model constructed by the virtual layer obtains the best conductivity image, and then the image output by the VAE model is interactively fed back through the communication layer. The whole method compensates for the invalid voltage data collected by the physical layer EIT device, and finally realizes highly robust lung EIT imaging.

[0065] <1.3、Overall technical solution>

[0066] The analytical framework of the method of this application is as follows Figure 1 As shown, it is divided into three layers:

[0067] (1) Physical layer, which implements the following two functions:

[0068] Data collection: The boundary voltage data set is obtained through the electrode assembly attached to the surface of the organism.

[0069] Display EIT imaging: EIT imaging is performed using a display screen or other device.

[0070] (2) Communication layer, which is used for communication from the virtual layer to the physical layer. It is mainly divided into two aspects:

[0071] The collected data of the physical layer is transferred to the virtual layer, that is, the boundary voltage data set is transferred to the virtual layer.

[0072] The result obtained by the virtual layer is passed to the display screen.

[0073] (3) Virtual layer:

[0074] WOVAE: Aims to find the optimal potential electrical signature with the smallest difference between the acquired voltage dataset and the simulated voltage dataset.

[0075] VAE model: The conductivity image is solved by the results obtained by WOVAE.

[0076] An electrical impedance tomography method based on digital twin driving comprises the following steps:

[0077] S100, reading a boundary voltage data set;

[0078] S200, sending the read boundary voltage data set to the WOWAE model, and WOWAE obtains the best electrical features (i.e., the feature information required by the VAE model in S300, the electrical features represent the pixels or information of the VAE input image, that is, a set of codes used to represent the EIT image in the potential low-dimensional space);

[0079] S300, the electrical features obtained by the WOVAE model are input into the trained VAE model, and the VAE model outputs imaging.

[0080] <1.4、WOVAE model>

[0081] A method for seeking optimal electrical characteristics based on a WOVAE model comprises the following steps:

[0082] S1, input boundary voltage dataset;

[0083] S2, initialization: Initialize a group of whale individuals in the latent space, each whale represents a set of electrical features (i.e., the feature information required by VAE, the electrical features are the pixels or information of the input image, which is a set of codes used to represent the EIT image in the latent low-dimensional space);

[0084] S3, calculate the fitness of each whale's position and find the optimal position of the whale;

[0085] S3-1, convert each set of electrical characteristics of each whale into a corresponding simulated boundary voltage dataset;

[0086] Among them, the method of converting a set of electrical characteristics into a corresponding simulated boundary voltage data set is:

[0087] First, the electrical features are input into the trained VAE model, and the trained VAE model outputs a high-precision reconstructed EIT image (the variational autoencoder restores and reconstructs the high-precision EIT image through the trained decoder);

[0088] Then, the reconstructed EIT image is forward modeled and solved to obtain a simulated boundary voltage data set (it should be noted that the method of obtaining a boundary voltage data set from an EIT image belongs to the prior art and will not be described here, such as: V-ShapedDense Denoising Convolutional Neural Network for Electrical ImpedanceTomography. IEEE Transactions on Instrumentation and Measurement. Issue: 71, Onpage(s): 1-14, Apr. 28.2022);

[0089] S3-2, determine the fitness function and calculate the fitness value f(σ) of each whale through the fitness function, and select the optimal whale position and retain it;

[0090] The fitness function is expressed as:

[0091] Among them, cov represents covariance, var represents variance, v and u represent the simulated boundary voltage dataset and the input boundary voltage dataset respectively. In the process of latent space exploration, the larger the objective function f(σ), the higher the similarity between the boundary voltage dataset calculated by the latent feature and the input boundary voltage dataset, and the more likely the whale position is to be the optimal solution;

[0092] S4, based on the optimal whale position determined in step S3, surround the prey, attack with a bubble net and search for the prey, and update the whale's position;

[0093] Among them, in the encirclement stage: when the whale actively swims towards the prey position, its position update formula is as follows:

[0094]

[0095] in:

[0096] D b represents the encircling step length;

[0097] A and C both represent coefficient vectors;

[0098] Indicates the position of the whale with the best current fitness when iterating to the kth generation;

[0099] Indicates the jth whale position when iterating to the kth generation;

[0100] Indicates the jth whale position when iterating to the k+1th generation;

[0101] r1 and r2 are random vectors in [0,1];

[0102] The value of a decreases linearly from 2 to 0, achieving a shrinking encirclement;

[0103] The formula of the bubble network attack is as follows:

[0104]

[0105] Among them, D p represents the distance between the whale and its prey, b is a constant that controls the shape of the logarithmic spiral, and l is a random number between [0,1];

[0106] It is worth noting that when hunting prey, whales swim around their prey in a spiral trajectory in a shrinking encirclement and launch bubble net attacks. Therefore, the behavior probability p is set to choose to shrink the encirclement or spiral swimming to update the whale's position, that is:

[0107]

[0108] The search for prey: In the search phase, the whale no longer selects the current best individual as the target for position update, but randomly selects a whale from the current population to approach (the purpose of this is to force the whale to deviate from the prey, expand the search range, and enhance the global search ability by increasing the randomness of the exploration). The position update formula is as follows:

[0109]

[0110] in: represents the randomly selected whale position, D r is the search step length.

[0111] S5: Determine whether the iteration is completed:

[0112] Determine whether the termination condition is met (eg, determine whether the current calculation has reached the maximum number of iterations): if so, output the optimal set of electrical characteristics; otherwise, return to step S3.

[0113] <1.5、VAE model>

[0114] VAE (full name: Variational Autoencoder) is a deep convolutional reconstruction network used to construct latent spatial features of complex data under given conditions, which is used to reconstruct high-precision EIT images.

[0115] VAE mainly consists of an encoder module and a decoder module, each of which is built from multiple convolutional neural networks.

[0116] VAE learns high-level features of EIT images through the encoding-decoding process and maps them into a low-dimensional feature space with an approximate Gaussian distribution, introducing stronger spatial prior constraints for EIT reconstruction.

[0117] Both the encoder and decoder are deep convolutional reconstruction networks used to construct latent space features of complex data under given conditions. The encoder network takes the conditional voltage data as additional input and reconstructs the original conductivity image based on the sampled latent codes.

[0118] The training of VAE is an existing technology and will not be described here (https: / / blog.csdn.net / baidu_35692628 / article / details / 125360283).

[0119] <1.6. Pre-processing of measured boundary voltage data set>

[0120] In actual EIT imaging, it is extremely challenging and time-consuming to accurately estimate the geometric information such as the contours of biological organs and electrode positions between different patients. In addition, the boundary voltage dataset u obtained by real measurement at the physical layer 实测 It is also easily disturbed by artifacts such as respiratory motion. It is necessary to filter out these uncertain geometric information as much as possible and only extract the voltage signal related to lung ventilation, represented by u. Therefore, for the boundary voltage dataset u 实测 Preprocessing is performed using a boundary sensitive matrix Extract boundary error, using u err express.

[0121]

[0122] Among them, λ is a regularization parameter, is an identity matrix, is a submatrix (which can be solved by the EIT forward problem, assuming that the chest area is uniformly distributed and calculated by the finite element method) consisting of all columns corresponding to triangular elements located near the boundary. Then, the preprocessed voltage data u is used l Instead of the original data u 实测 To reconstruct the image in order to mitigate the impact of boundary errors and motion artifacts.

[0123] <2: Comparison of EIT imaging results of multiple models>

[0124] 2.1 Training the VAE network.

[0125] There are nearly 90,000 EIT images used to train the VAE network, and about 1,000 test images. In the VAE network model training, the balance coefficient α is set to 10 -2 , the initial learning rate β is set to 10 -4 , and is reduced when the validation loss stops decreasing to prevent the network from overfitting. The number of training epochs is set to 1000, and the network parameters with the lowest validation loss are saved for feature-based inversion.

[0126] 2.2 Experiment 1.

[0127] In the first experiment, CT scan data of 5 patients were randomly selected from a training cohort of 50 patients to generate an EIT model, and traditional regularization (TR), convolutional neural network (CNN), VAE and the WOWAE proposed in this invention were tested.

[0128] like Figure 2 As shown in the figure, the TR method can only roughly reconstruct the shape of the lungs. The CNN method can basically reconstruct the contour of the lungs, but there are some artifacts. The VAE method can reconstruct the shape of the lungs more clearly, and there are almost no artifacts. In contrast, the WOWAE method based on feature inversion is highly similar to the actual situation and can clearly reconstruct the sharp edges and concave parts of the lungs.

[0129] 2.3 Experiment 2.

[0130] like Figure 3 In the second experiment, CT scans were selected for each patient from the test cohort and EIT models were generated for testing. The results showed that the feature-based inversion results were slightly different from the actual situation, because the VAE did not see the conductivity images of the organs for these models.

[0131] However, the similarity between these WOVAE reconstruction results and the actual situation is satisfactory, in which the damaged part is obviously reversed. For the second experiment, the reconstruction performance of the WOVAE network can be further improved by increasing the diversity of training samples.

[0132] 2.4 Comparison of results.

[0133] Four reconstruction methods (TR, CNN, VAE, and WOVAE) are compared in terms of EIT reconstruction accuracy. The accuracy of these methods is measured using several different metrics, including root mean square error (RMSE), image correlation sparsity coefficient (ICC), structural similarity index (SSIM), and peak signal-to-noise ratio (PSNR).

[0134] RMSE (Root Mean Square Error) measures the error between the reconstructed image and the real image. The smaller the RMSE, the higher the reconstruction accuracy. In this experiment, it is 0.08±0.02, followed by 0.11±0.03 for VAE, then 0.15±0.02 for CNN, and finally 0.23±0.05 for TR. This is the lowest RMSE of WOWAE among the four methods, indicating that WOWAE performs best in reconstruction accuracy.

[0135] ICC (Intraclass Correlation Coefficient) measures the similarity between images. The higher the ICC value, the better the consistency between the reconstructed image and the real image. In this experiment, WOWAE's ICC is 0.97±0.03, the highest among the four methods, indicating that WOWAE performs best in maintaining image consistency. It is followed by VAE's 0.95±0.05, CNN's 0.93±0.04, and TR's 0.81±0.05.

[0136] SSIM (Structural Similarity Index) is also an indicator of image quality, which takes into account the similarity of brightness, contrast, and structure. The closer the SSIM value is to 1, the better the image quality. In this experiment, WOWAE's SSIM is 0.93±0.03, the highest among the four methods, indicating that WOWAE performs best in terms of structural similarity. It is followed by VAE's 0.90±0.04, CNN's 0.79±0.05, and TR's 0.66±0.04.

[0137] PSNR (Peak Signal-to-Noise Ratio) measures the signal-to-noise ratio of an image. The higher the PSNR value, the lower the noise level of the image and the better the quality. In this experiment, WOWAE's PSNR is 22.63±1.73, the highest among the four methods, indicating that WOWAE performs best in terms of signal-to-noise ratio. It is followed by VAE's 19.46±2.12, CNN's 16.78±1.30, and TR's 13.04±1.86.

[0138] Combining these indicators, it can be concluded that WOVAE outperforms TR, CNN, and VAE in terms of EIT reconstruction accuracy. This shows that WOWAE is a more effective EIT reconstruction method.

[0139] <3: EIT imaging robustness analysis of this application>

[0140] like Figure 5 As shown in the figure, in the third experiment, before each reconstruction, Gaussian white noise ranging from 60dB to 20dB was randomly added to the original voltage data, and the SSIM (structural similarity index) and PSNR (peak signal-to-noise ratio) performance of the four reconstruction methods (TR, CNN, VAE and WOVAE) at different noise levels were analyzed to evaluate their robustness to measurement noise.

[0141] In the left figure, it can be seen that as the signal-to-noise ratio (SNR) decreases, that is, the noise level increases, the SSIM values ​​of all methods decrease. At higher SNR levels (e.g., 60dB to 40dB), the SSIM values ​​of WOWAE and VAE are relatively high, indicating that they are more robust to noise under these conditions. However, as the SNR decreases further, the SSIM values ​​of all methods decrease significantly, indicating that the performance of all methods will be affected under high noise conditions. In the right figure, the analysis of PSNR shows a similar trend. At higher SNR levels, the PSNR values ​​of WOWAE and VAE are relatively high, indicating that they are more resistant to noise under these conditions. As the SNR decreases, the PSNR values ​​generally decrease, which reflects the negative impact of noise on image quality. Overall, WOWAE shows better robustness at higher SNR levels, but its performance is basically not affected as the noise level increases. This shows that WOWAE has some advantages in dealing with measurement noise. In addition, TR and CNN perform relatively poorly in these tests, especially under low SNR conditions.

[0142] like Figure 6 As shown in the fourth experiment, before each reconstruction, the patient's chest contour was randomly deformed to a certain extent to simulate the breathing changes in real situations. The reconstruction results of four reconstruction methods (TR, CNN, VAE and WOWAE) were analyzed to evaluate their robustness to chest deformation.

[0143] like Figure 7 As shown in the fifth experiment, before each reconstruction, multiple electrode detachment situations are randomly simulated to simulate the possible faulty electrode situations in real situations, and the reconstruction results of four reconstruction methods (TR, CNN, VAE and WOWAE) are analyzed to evaluate their robustness to faulty electrodes.

[0144] Regardless of Figure 6 or Figure 7 It can be seen that when there is a certain model error, the first three traditional methods (TR, CNN and VAE) cannot accurately reconstruct the lung shape, and there is a large reconstruction deviation from the actual CT situation. This is because model errors such as chest deformation or electrode detachment will change the distribution of current to a certain extent, thereby affecting the accuracy of the measured voltage data. The first three traditional methods can only perform reliable EIT reconstruction based on valid voltage data. Therefore, when reliable voltage data cannot be accurately collected, erroneous voltage collection will seriously affect the imaging results and lead to missing lung shapes.

[0145] In contrast, the WOAE imaging method driven by digital twins can accurately reconstruct lung images, can well overcome model errors such as chest deformation or electrode detachment, and has high reconstruction robustness. This is because WOA uses the objective function based on the voltage correlation coefficient to find the optimal potential features, and then finds the best conductivity image through the WOAE model constructed in the virtual layer, so that the generated simulated voltage and the actual measured voltage are most closely matched, and interactive feedback is performed in the communication layer, thereby compensating for the invalid voltage data collected by the physical layer EIT device, and finally achieving highly robust lung EIT imaging.

[0146] The above embodiments are preferred implementation modes of the present invention and are only used to facilitate the description of the present invention. They are not intended to limit the present invention in any form. Any person with ordinary knowledge in the relevant technical field, if they do not depart from the scope of the technical features of the present invention, can make equivalent embodiments by partial changes or modifications to the technical contents disclosed in the present invention, and they still fall within the scope of the technical features of the present invention without departing from the technical features of the present invention.

Claims

1. An electrical impedance tomography method based on digital twin drive, characterized in that: The following steps are involved: S100, reading a boundary voltage data set; S200, seeking the best electrical characteristics; S300, inputting the best electrical features obtained in step S200 into the trained VAE model, and the VAE model outputs the reconstructed EIT imaging; Among them, the method of seeking the best electrical characteristics of S200 adopts the method of seeking the best electrical characteristics based on the WOVAE model, which includes the following steps: S1, input boundary voltage data set u; S2, initialization: Initialize a group of whale individuals in the latent space, each whale represents a set of electrical features; S3, calculate the fitness of each whale's location and find the optimal location of the whale; S3-1, converting each set of electrical characteristics of each whale into a corresponding simulated boundary voltage data set; wherein a method of converting a set of electrical characteristics into a corresponding simulated boundary voltage data set is: First, the electrical features are input into a trained VAE model, and the trained VAE model outputs a high-precision reconstructed EIT image; Then, the reconstructed EIT image is forward modeled and solved to obtain a simulated boundary voltage data set v; S3-2, calculate the fitness value f(σ) of each whale, and select the optimal whale position and retain it; Among them, cov represents covariance, var represents variance, v and u are the simulated boundary voltage data set and the input boundary voltage data set respectively; S4, based on the optimal whale position determined in step S3, surround the prey, attack with a bubble net and search for the prey, and update the whale's position; S5: Determine whether the iteration is completed; Determine whether the termination condition is met: if so, output the optimal set of electrical characteristics; otherwise, return to step S3.

2. The electrical impedance tomography method based on digital twin driving according to claim 1, characterized in that: Step S100 also includes the following sub-steps: S101, read the electrical impedance monitoring device to obtain the boundary voltage data set u 实测 ; S102, obtain the input boundary voltage data set u: Among them, λ is a regularization parameter, is an identity matrix, is a submatrix consisting of all the columns corresponding to the triangular elements that lie near the boundary.

3. A storage medium, characterized in that: It stores a computer program, which is used to execute the steps of the electrical impedance imaging method based on digital twin drive as described in claim 1 or 2.

4. A computer device, characterized in that: It includes a memory and a processor, wherein the memory stores computer-readable instructions, and when the processor executes the computer-readable instructions, the steps of the electrical impedance imaging method based on digital twin drive as described in claim 1 or 2 are implemented.

Images