High-frame-rate electrical impedance imaging method and system based on sparse representation

Abstract

The invention relates to the technical field of electrical impedance imaging, and discloses a high-frame-rate electrical impedance imaging method and system based on sparse representation. The method comprises the steps of collecting original electrical impedance data, and performing preprocessing to obtain a training sample matrix; carrying out dictionary learning algorithm training by utilizing the training sample matrix to obtain an optimized dictionary matrix; setting a mask matrix for electrical impedance data acquisition, and performing sparse sampling on the original electrical impedance data according to the mask matrix to obtain incomplete data frames; and carrying out sparse representation on the incomplete data frame by using the optimized dictionary matrix, solving a corresponding sparse vector, and recovering the original electrical impedance data by using the dictionary matrix and the sparse vector. The method can effectively improve the system frame rate, is suitable for different electrical impedance imaging platforms, and does not depend on specific hardware.

Description

Technical Field

[0001] The present application relates to the technical field of electrical impedance imaging, and in particular to a high frame rate electrical impedance imaging method based on sparse representation. Background Art

[0002] Electrical impedance tomography (EIT) is a detection technology that can image the internal electrical conductivity of an object. This distribution of electrical conductivity accurately reflects the object's internal structure. Compared to existing medical imaging technologies such as CT, X-rays, and ultrasound, EIT offers advantages such as being non-invasive, radiation-free, and capable of instant imaging. It is a research hotspot in related fields such as smart wearables and biomedical applications, with applications including lung monitoring, breast cancer monitoring, gesture recognition, and epilepsy monitoring.

[0003] The basic principle of electrical impedance tomography is to attach a circle of electrodes to the surface of an object. A weak current is injected through selected electrode pairs as an excitation, and the voltage response of other electrode pairs is measured. This yields the impedance distribution at different locations within the object. Image reconstruction algorithms are then used to convert the measured impedance data into an image of the object's internal conductivity, revealing the tissue structure.

[0004] However, the entire electrical impedance imaging process requires numerous impedance measurements, which is extremely time-consuming, especially in scenarios with multiple electrodes and low-frequency excitation. For example, a 16-electrode electrical impedance system in adjacent excitation-measurement mode requires 16 x 13 = 208 independent impedance measurements per frame. This severely limits the system's imaging frame rate.

[0005] Current approaches to improving the frame rate of electrical impedance imaging primarily include using higher-speed hardware, such as high-frequency data acquisition systems and parallel acquisition and processing, and developing and optimizing more complex excitation and acquisition patterns. However, these approaches rely on specific hardware conditions and require redesign when the scenario changes, resulting in high implementation costs.

[0006] Therefore, a method that can be applied to different electrical impedance imaging platforms, does not rely on specific hardware, and can effectively improve the system frame rate has important research value and application needs. Summary of the Invention

[0007] The purpose of this application is to provide a high frame rate electrical impedance imaging method and system based on sparse representation to solve the problems raised in the above background technology.

[0008] The present application discloses a high frame rate electrical impedance tomography method based on sparse representation, comprising:

[0009] Collect raw electrical impedance data and preprocess them to obtain a training sample matrix;

[0010] Using the training sample matrix to perform dictionary learning algorithm training to obtain an optimized dictionary matrix;

[0011] Setting a mask matrix for electrical impedance data acquisition, and performing sparse sampling on the original electrical impedance data according to the mask matrix to obtain an incomplete data frame;

[0012] The optimized dictionary matrix is ​​used to perform sparse representation on the incomplete data frame, the corresponding sparse vector is solved, and the original electrical impedance data is restored using the dictionary matrix and the sparse vector.

[0013] In a preferred embodiment, the raw electrical impedance data is collected and preprocessed to obtain a training sample matrix, raw electrical impedance data frames are collected through an electrical impedance imaging platform, and each frame data is modulus normalized to form a training sample matrix.

[0014] In a preferred example, the dictionary learning algorithm is trained using the training sample matrix to obtain an optimized dictionary matrix, and the dictionary learning algorithm is iteratively trained using the training sample matrix and the set number of dictionary matrix atoms, wherein the process includes fixing the dictionary matrix to solve the sparse representation vector of the sample, and fixing the sparse vector to update the dictionary matrix until the convergence condition is met, and finally obtaining the optimized dictionary matrix.

[0015] In a preferred example, the mask matrix for electrical impedance data acquisition is set, and the original electrical impedance data is sparsely sampled according to the mask matrix to obtain an incomplete data frame. According to the number of stimulation modes E of the electrical impedance imaging platform and the number of measurement values ​​M of each stimulation mode, the mask matrix Φ for electrical impedance data acquisition is set, where the size of Φ is E x M, and each element indicates whether the corresponding measurement value is collected, where 1 indicates collection and 0 indicates non-collection; then, the original electrical impedance data is sparsely sampled row by row according to the set mask matrix Φ to obtain an incomplete data frame containing the collected measurement values.

[0016] In a preferred example, the optimized dictionary matrix is ​​used to perform sparse representation on the incomplete data frame, the corresponding sparse vector is solved, and the original electrical impedance data is restored using the dictionary matrix and the sparse vector. The optimized dictionary matrix is ​​used to perform sparse representation on the incomplete data frame, and an optimization model is constructed to solve the corresponding sparse vector α, wherein the optimization model is the L1 norm or L0 norm of the least squares model regularization; then, based on the obtained sparse vector α and the dictionary matrix, the product of the two matrices is calculated as the recovery value of the original electrical impedance data, and finally the recovery of the original electrical impedance data is completed.

[0017] In a preferred embodiment, the dictionary learning algorithm training using the training sample matrix to obtain an optimized dictionary matrix includes the following processing steps:

[0018] Perform singular value decomposition on the training sample matrix X and obtain

[0019] X=U,S,V T

[0020] Where U is a left unitary matrix of size N×N;

[0021] The first P columns of the left unitary matrix U are used as the initialized dictionary matrix D.

[0022] In a preferred embodiment, in the step of setting the mask matrix Φ for electrical impedance data acquisition, the mask matrix is:

[0023]

[0024] Among them, φ i，j (i=1,2,…,E;j=1,2,…,M) indicates whether the jth measurement value is collected in the i-th stimulation mode, φ i，j The value is 1 or 0, 1 means collection, 0 means no collection.

[0025] In a preferred embodiment, the optimized dictionary matrix is ​​used to sparsely represent the incomplete data frame, and the corresponding sparse vector is solved. The following processing is used to obtain the sparse vector α \Π :

[0026] Will solve for the sparse vector α \Π It is formulated as the following sparse optimization problem:

[0027]

[0028] Among them, λ is the error balance parameter, R(α) is the sparse constraint term, and takes the L0 norm or L1 norm.

[0029] In a preferred embodiment, the process of restoring the original electrical impedance data using the dictionary matrix and the sparse vector includes the following processing:

[0030] Original signal recovery process: For each data point i, if i is not in the missing data index set π, then x i Remain unchanged; if i is in π, take the dictionary matrix D and the sparse vector α \Π The product of x i The recovery value of

[0031] Acceleration ratio calculation process: Acceleration ratio Among them, N is the total amount of original data, φ i，j It is a mask matrix element, indicating whether the corresponding data is collected.

[0032] The present application also discloses a high frame rate electrical impedance imaging system based on sparse representation, comprising:

[0033] The training sample acquisition module is used to collect the original electrical impedance data and perform preprocessing to obtain the training sample matrix;

[0034] A dictionary training module is used to train a dictionary learning algorithm using the training sample matrix to obtain an optimized dictionary matrix;

[0035] The mask matrix setting and data acquisition module is used to set the mask matrix for electrical impedance data acquisition and perform sparse sampling on the original electrical impedance data according to the mask matrix to obtain incomplete data frames;

[0036] The data recovery module is used to perform sparse representation on the incomplete data frame using the optimized dictionary matrix, solve the corresponding sparse vector, and use the dictionary matrix and the sparse vector to restore the original electrical impedance data.

[0037] In the embodiment of the present application, a set of high frame rate electrical impedance imaging methods based entirely on numerical methods is proposed, which has the advantages of not requiring any upgrade or modification of existing hardware, being unrelated to the specific electrical impedance imaging stimulation-measurement protocol, being low in cost, easy to use, adaptable to various electrical impedance imaging hardware systems, breaking through the limitations of the hardware capabilities themselves, and greatly improving the electrical impedance imaging frame rate. Among them, the mask matrix of signal acquisition reduces hardware requirements and increases processing speed by reducing the amount of data collected, thereby achieving high efficiency and low-cost signal acquisition. At the same time, the signal recovery part, based on these small amounts of collected data, applies advanced algorithms to recover complete signal information. This method not only reduces the necessary amount of data collection, but also ensures the accuracy of data recovery, making the entire system more efficient and economical, and being suitable for various occasions requiring fast and efficient signal processing.

[0038] The specification of this application records a large number of technical features, which are distributed in various technical solutions. If all possible combinations of technical features of this application (i.e., technical solutions) are to be listed, the specification will be too lengthy. In order to avoid this problem, the various technical features disclosed in the above-mentioned invention content of this application, the various technical features disclosed in the various embodiments and examples below, and the various technical features disclosed in the accompanying drawings can be freely combined with each other to form various new technical solutions (these technical solutions are all deemed to have been recorded in this specification), unless such a combination of technical features is technically infeasible. For example, in one example, feature A+B+C is disclosed, and in another example, feature A+B+D+E is disclosed. Features C and D are equivalent technical means that play the same role. Technically, only one of them can be used, and it is impossible to use them at the same time. Feature E can be technically combined with feature C. Then, the solution of A+B+C+D should not be considered as having been recorded because it is technically infeasible, while the solution of A+B+C+E should be considered as having been recorded. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 1 is a flow chart of a high frame rate electrical impedance tomography method based on sparse representation according to the first embodiment of the present application;

[0040] Figure 2 This is a process diagram of an example of a high frame rate electrical impedance imaging method based on sparse representation according to the first embodiment of the present application.

[0041] Figure 3 It is a structural schematic diagram of a high frame rate electrical impedance imaging system based on sparse representation according to the first embodiment of the present application. DETAILED DESCRIPTION

[0042] In the following description, many technical details are provided to help readers better understand this application. However, those skilled in the art will understand that even without these technical details and various changes and modifications based on the following embodiments, the technical solutions claimed in this application can be implemented.

[0043] The following is a summary of some of the innovative features of this application:

[0044] The present invention discloses a high-frame-rate electrical impedance imaging method based on sparse representation, comprising: a user specifies an electrical impedance imaging platform and provides a data sample for dictionary learning; a dictionary learning algorithm is selected and an initial dictionary value is set; the dictionary is trained and the dictionary value is adjusted using the dictionary learning algorithm and the training data sample; the user specifies an electrical impedance data acquisition mask matrix to reduce data acquisition for electrical impedance imaging; data acquisition is performed based on the mask matrix, and the collected data is recorded as a defective signal; a sparse representation is performed on the defective signal using the trained dictionary to restore the original signal; and imaging is performed using the restored original signal and the electrical impedance imaging algorithm. The present invention is based entirely on numerical methods to achieve an improvement in the electrical impedance imaging frame rate, does not require any upgrade or modification of existing hardware, and is unrelated to the stimulation-measurement protocol of electrical impedance imaging.

[0045] In order to make the objectives, technical solutions and advantages of this application clearer, the implementation methods of this application will be further described in detail below with reference to the accompanying drawings.

[0046] The first embodiment of the present application relates to a high frame rate electrical impedance imaging method based on sparse representation, the process of which is as follows: Figure 1 As shown, the method includes the following steps:

[0047] Step 100: Collect raw electrical impedance data and perform preprocessing to obtain a training sample matrix.

[0048] In this step, the original electrical impedance data frames are collected through the electrical impedance imaging platform, and the modulus normalization processing is performed on each frame data to form a training sample matrix.

[0049] Specifically, collecting raw impedance data frames through an impedance imaging platform refers to using a specific impedance imaging device or system to obtain raw impedance measurement data of a target area according to a certain pattern, and the collected data is a data set containing multiple frames.

[0050] Normalizing the modulus of each frame of data means normalizing each frame of collected electrical impedance data so that the modulus of each sample is normalized to a fixed range. This is to eliminate the influence of the sample scale and facilitate subsequent processing.

[0051] Constructing a training sample matrix involves organizing the preprocessed electrical impedance data frame into a matrix format, where each column represents a training sample. This training sample matrix will serve as the input to the dictionary learning algorithm.

[0052] In summary, the purpose of this step is to prepare electrical impedance data samples and perform necessary preprocessing to construct the training sample matrix required for dictionary learning.

[0053] Step 200: Using the training sample matrix to perform dictionary learning algorithm training to obtain an optimized dictionary matrix.

[0054] Dictionary learning algorithms are a class of unsupervised machine learning algorithms used to learn a sparse representation dictionary from training samples. The main idea of ​​this algorithm is that, assuming a signal can be linearly represented by several atoms in the dictionary, a dictionary containing these atoms can be learned to achieve a sparse representation of the signal.

[0055] This step uses the training sample matrix and the set number of dictionary matrix atoms to perform iterative training of the dictionary learning algorithm. The process includes fixing the dictionary matrix to solve the sparse representation vector of the sample, and fixing the sparse vector to update the dictionary matrix until the convergence condition is met, and finally obtaining the optimized dictionary matrix.

[0056] Specifically, using the training sample matrix means using the constructed impedance training sample matrix as the input of the dictionary learning algorithm. The training sample matrix contains sufficient impedance signal feature information and is the basic data for dictionary learning. Setting the number of dictionary matrix atoms means that the number of columns of the dictionary matrix represents the number of atoms contained in the dictionary. The appropriate number of atoms is set according to the complexity of the signal. The more atoms, the more complex the signal can be represented. Iterative training of the dictionary learning algorithm means that the typical dictionary learning uses iterative optimization to update the dictionary matrix. There are two main steps: (1) Fix the dictionary matrix D and learn the sparse representation coefficient α of the sample through the sparse coding algorithm. (2) Fix the sparse coefficient α and update the dictionary matrix D so that it can better represent the sample. The above two processes are iterated until the convergence condition is met. Obtaining the optimized dictionary matrix means that through iterative training, a dictionary matrix that is optimal for representing the given training sample and can capture the signal characteristics can be learned. Convergence condition and iteration stop means that the maximum number of iterations can be set, or the error function is less than the convergence condition such as SettingAccuracy, and the iteration process is stopped when it is met.

[0057] In summary, this step iteratively optimizes the initial dictionary matrix through the dictionary learning algorithm to obtain a dictionary matrix that best represents the electrical impedance signal, providing support for subsequent signal reconstruction.

[0058] Step 300: Setting a mask matrix for electrical impedance data acquisition, and performing sparse sampling on the original electrical impedance data according to the mask matrix to obtain an incomplete data frame.

[0059] Specifically, according to the number of stimulation modes E of the electrical impedance imaging platform and the number of measurement values ​​M for each stimulation mode, a mask matrix Φ for electrical impedance data acquisition is set, where the size of Φ is E x M, and each element indicates whether the corresponding measurement value is acquired, where 1 indicates acquisition and 0 indicates non-acquisition; then, according to the set mask matrix Φ, the original electrical impedance data is sparsely sampled row by row to obtain incomplete data frames containing the acquired measurement values.

[0060] Step 300 involves setting a mask matrix for electrical impedance data acquisition and using the mask matrix to sparsely sample the original electrical impedance data to obtain an incomplete data frame. Specifically, this step takes into account the number of stimulation modes (E) of the electrical impedance imaging platform and the number of measurements (M) under each stimulation mode. Based on these two parameters, a mask matrix Φ of size E x M is set. In this matrix, the value of each element determines whether the corresponding measurement value is collected, where 1 indicates that the corresponding measurement value will be collected, and 0 indicates that the corresponding measurement value is not collected. Then, according to the set mask matrix Φ, sparse sampling is performed on the original electrical impedance data row by row to obtain an incomplete data frame containing the collected measurement values. This process speeds up data processing and improves imaging rate by reducing the amount of data that must be collected, while retaining sufficient information to allow subsequent data recovery and imaging processing.

[0061] For example, in a 16-electrode system, if adjacent patterns are used, there are typically 16 stimulation patterns, each with 13 measurements, for a total of 16 × 13 = 208 measurements. When these measurements are sparsely sampled using the mask matrix Φ, a sampling rate of 50% means that only approximately half of the 13 measurements per stimulation pattern are actually acquired, meaning only approximately 6 or 7 measurements per pattern are acquired. Therefore, at this sampling rate, the entire system will only acquire approximately 104 measurements in total, which is half the original 208 measurements.

[0062] Therefore, if only half of the raw data frames are acquired in this scenario, the electrical impedance tomography system's frame rate can theoretically be doubled, achieving a 200% acceleration. This is because the amount of data acquired is halved, and the time required to process this data is also reduced accordingly, allowing more data frames to be acquired and processed in the same amount of time, thereby accelerating the imaging rate.

[0063] Step 400: Use the optimized dictionary matrix to perform sparse representation on the obtained incomplete data frame, solve the corresponding sparse vector, and use the dictionary matrix and sparse vector to restore the original electrical impedance data.

[0064] Specifically, the optimized dictionary matrix is ​​used to perform sparse representation on the incomplete data frame, and an optimization model is constructed to solve the corresponding sparse vector α, where the optimization model is the L1 norm or L0 norm of the least squares model regularization; then, based on the obtained sparse vector α and the dictionary matrix, the product of the two matrices is calculated as the recovery value of the original electrical impedance data, and finally the recovery of the original electrical impedance data is completed.

[0065] Step 400 involves using the optimized dictionary matrix to perform a sparse representation on the obtained incomplete data frame, then solving for the corresponding sparse vector, and finally using the dictionary matrix and sparse vector to restore the original electrical impedance data. Specifically, this process includes the following key steps:

[0066] 1) Sparse representation: First, the incomplete data frame is sparsely represented using an optimized dictionary matrix. This means that each incomplete data frame is represented as a linear combination of some columns (dictionary atoms) in the dictionary matrix, and the coefficients of these columns constitute the sparse vector α.

[0067] 2) Optimization Model Construction: In this step, an optimization model is constructed to solve the sparse vector α. This optimization model is typically based on a least squares problem and constrains the sparsity of the sparse vector by regularizing the L1 norm (to promote sparsity) or the L0 norm (to directly count non-zero elements). Such an optimization model aims to find the sparse vector α that best represents the incomplete data frame while maintaining the vector's sparsity.

[0068] 3) Data recovery: Once the sparse vector α is obtained, the original electrical impedance data is recovered by calculating the product of this vector and the optimized dictionary matrix. This product represents an estimate of the original electrical impedance data recovered from the incomplete data frame.

[0069] Through these steps, step 400 realizes the restoration process from the incomplete data frame to the complete original electrical impedance data frame, so that even when the acquired data is incomplete, a complete electrical impedance image can be effectively reconstructed. This is of great significance for improving the speed and efficiency of electrical impedance imaging.

[0070] The high frame rate electrical impedance imaging method based on sparse representation of the above-described embodiment is fully realized based on numerical methods, does not require any upgrading and modification of existing hardware, and is unrelated to the stimulation protocol of electrical impedance imaging, has broken through the limitation of hardware capabilities, and greatly improved the electrical impedance imaging frame rate. Among them, the mask matrix of signal acquisition reduces hardware requirements and increases processing speed by reducing the amount of data collected, thereby achieving high-efficiency and low-cost signal acquisition. At the same time, the signal recovery part is based on these small amounts of collected data, and uses advanced algorithms to recover complete signal information. This method not only reduces the necessary amount of data collection, but also ensures the accuracy of data recovery, making the entire system more efficient and economical, and is applicable to various occasions that need fast and efficient signal processing.

[0071] In order to better understand the technical solution of the present application, three specific examples are provided below for illustration. The details listed in the examples are mainly for ease of understanding and are not intended to limit the scope of protection of the present application.

[0072] According to the first example, the high frame rate electrical impedance tomography method based on sparse representation includes the following steps:

[0073] Step 110: The user specifies an electrical impedance imaging platform. Assume that the platform has E stimulation modes, each stimulation mode has M measurement values, obtains the original electrical impedance data frame, and forms a training sample for dictionary learning after normalization of the modulus value:

[0074] X=[x1 x2 ... x L ]∈R N×L

[0075] Each training sample has the same dimension N=E×M, and there are L training samples in total.

[0076] Wherein, X=[x1, x2, ... xL] indicates that the matrix X is formed by stacking L column vectors.

[0077] Each column vector xi represents a training sample, and the vector length is N.

[0078] N=E x M, where E represents the number of stimulation modes and M represents the number of measurement channels in each stimulation mode.

[0079] Therefore, the length of each sample xi is Ex M, which contains all the measurement values ​​of a frame of electrical impedance data.

[0080] A total of L frames of electrical impedance data are collected, so there are L training samples.

[0081] The matrix X dimension is N x L, where N represents the sample dimension and L represents the number of samples.

[0082] In summary, the matrix X combines L electrical impedance data samples of length N=ExM by columns to form a training data set for the dictionary learning algorithm.

[0083] Each sample includes all measurement values ​​of one frame of electrical impedance data that can be obtained according to the electrical impedance platform parameters (number of stimulation modes E, number of measurement channels M).

[0084] Optionally, this step includes the following processes:

[0085] EIT platform setup: The user first needs to specify the EIT platform, which includes determining the number of stimulation modes on the platform (E) and the number of measurements for each stimulation mode (M). This is because different EIT systems may have different configurations and measurement modes.

[0086] Raw data collection: Users collect raw electrical impedance data frames through the electrical impedance imaging platform. These data frames represent the electrical impedance measurement results under different stimulation modes.

[0087] Data preprocessing: The collected raw impedance data frames undergo modulus normalization to eliminate scale differences between measurements and ensure that data is compared and processed on the same scale. Normalization is often a crucial step in data preprocessing, improving the performance and stability of subsequent learning algorithms.

[0088] Formation of the training sample matrix: The preprocessed data is integrated into a training sample matrix X, which consists of L training samples, each of dimension N = E × M. This means that each training sample is obtained from the corresponding electrical impedance data frame, and all samples have the same dimension, ensuring data consistency and comparability.

[0089] Through the above steps, data can be effectively collected and preprocessed from the EIT platform, providing the necessary training samples for subsequent dictionary learning and sparse representation of EIT data. These training samples are the basis for efficient sparse representation and data recovery in subsequent steps.

[0090] Step 120: Select a dictionary learning algorithm, set the number of dictionary atoms P (P ≥ N), perform singular value decomposition on the training sample matrix, and initialize the dictionary matrix D∈R N×P are the first P columns of the left unitary matrix.

[0091] Optionally, this step includes the following:

[0092] 1) Select a dictionary learning algorithm: First, select a dictionary learning algorithm that is suitable for the training sample matrix. The choice of dictionary learning algorithm depends on the characteristics of the data and the required sparse representation performance.

[0093] 2) Set the number of dictionary atoms, P: During dictionary learning, you need to set the number of atoms, P, in the dictionary matrix. This is a key parameter when constructing the dictionary matrix. The number of atoms, P, must be at least equal to the dimension, N, of the training sample, i.e., P ≥ N, to ensure that the dictionary has sufficient capacity to represent different data samples.

[0094] 3) Singular Value Decomposition: Perform singular value decomposition (SVD) on the training sample matrix. This is a matrix decomposition method that can extract the main features of the matrix. The decomposition results include the left unitary matrix U, the singular value matrix S, and the right unitary matrix V^T.

[0095] 4) Initialize the dictionary matrix D: Based on the results of the singular value decomposition, initialize the dictionary matrix D to the first P columns of the left unitary matrix U, that is, D = U[:,:P]. This initialized dictionary matrix contains the most important features of the training sample matrix and provides a good starting point for the subsequent dictionary learning process.

[0096] Through this step, a preliminary dictionary matrix based on the main features of the training data can be obtained, which will be further optimized and trained in subsequent steps to better suit specific data representation requirements.

[0097] Step 130: Use the training samples and the selected dictionary learning algorithm to train the initialized dictionary matrix D. The processing process is as follows:

[0098] Step 131: Fix the dictionary D and solve for each sample x i (i=1, 2, ..., N) sparse vector α i (i=1, 2, ..., N);

[0099] Step 132: Fix the sparse vector α i , update the dictionary matrix D;

[0100] Step 133: Select the mean square error estimator as the performance index in the dictionary learning process:

[0101]

[0102] in,

[0103] e, refers to the performance index of the dictionary learning process, namely the mean square error (MSE). It is an indicator used to evaluate the effect of dictionary learning, which represents the difference between the original training sample and the training sample obtained through the dictionary matrix D and the sparse vector α. i The error between the reconstructed samples.

[0104] L refers to the total number of training samples, that is, the number of samples used for training in the dictionary learning process.

[0105] x i , refers to the i-th training sample, which is the original data actually obtained from the dataset.

[0106] D refers to the dictionary matrix, which is optimized by the dictionary learning algorithm and contains all dictionary atoms (or bases).

[0107] α i , refers to the i-th training sample x i A sparse vector when using the dictionary D for sparse representation. It contains the coefficients of each dictionary atom when representing sample x_i.

[0108] (x i -Dα i ) 2 , represents the square of the error between the i-th training sample and its sparse representation. This error is calculated to measure the relationship between the dictionary D and the sparse vector α i How well the original sample x is reconstructed i .

[0109] Step 134: When e meets the convergence condition or reaches the maximum number of training times, go to step 140, otherwise repeat steps 131 to 133.

[0110] In other words, in step 130, the initialized dictionary matrix D is trained using the training samples and the selected dictionary learning algorithm. The process includes the following specific steps:

[0111] Step 131: First, fix the dictionary matrix D, then for each sample: x i (i=1,2,…N solve its corresponding sparse vector α i This means that when the dictionary matrix D remains unchanged, the optimization algorithm is used to find the best representation of each training sample x. i The sparse coefficient vector α i .

[0112] Step 132: Next, fix the sparse vector α i , update the dictionary matrix D. In this step, the sparse vector α has been obtained i is used to update the dictionary matrix D to better fit the training data. This may involve adjusting the atoms (columns) in the dictionary matrix so that they more effectively represent the training examples.

[0113] Step 133: Use the mean square error e as the performance index in the dictionary learning process. The mean square error here is calculated by calculating the difference between each sample and its dictionary matrix D and sparse vector α i The performance index e is the average of these errors and is used to evaluate the effectiveness of dictionary learning.

[0114] Step 134: Determine whether the performance index e meets the convergence condition or whether the maximum number of training times has been reached. If either condition is met, proceed to the next step; otherwise, repeat steps 131 to 133.

[0115] Through this series of steps, the dictionary matrix D is updated and optimized through selection, and finally an optimized dictionary matrix that can effectively represent the training samples is obtained, providing a basis for subsequent sparse representation and signal recovery.

[0116] Step 140: The user specifies a mask matrix for electrical impedance data acquisition.

[0117] In step 140, the user is asked to provide a mask matrix Φ for electrical impedance data acquisition. This mask matrix is ​​used to determine which electrical impedance data should be collected and which should be ignored. Specifically, the size of the mask matrix Φ is E x M, where E is the number of stimulation modes and M is the number of measurements under each stimulation mode. In the mask matrix, each element φ_ij (where i = 1, 2, ..., E; j = 1, 2, ..., M) indicates whether the jth measurement value under the i-th stimulation mode in each frame of electrical impedance data is collected. If the value of φ_ij is 1, it means that the data at the corresponding position in the electrical impedance data frame should be collected; if the value of φ_ij is 0, it means that the data at the corresponding position should be ignored.

[0118] This process allows users to selectively acquire electrical impedance data based on actual needs and data processing capabilities, thereby reducing data processing time and increasing imaging speed without sacrificing too much data quality. For example, in some applications, not all measurements may be required for a valid reconstruction. In this case, a mask matrix can be set to only acquire the necessary measurements.

[0119] Step 150: Collect the incomplete data frame x according to the mask matrix given by the user \Π , where π represents x \Π The set of index values ​​corresponding to the missing data relative to the original signal x.

[0120] In step 150, a user-specified mask matrix is ​​used to acquire incomplete data frames. The mask matrix Φ is used to determine which data should be acquired (marked with 1) and which should be ignored (marked with 0) during the electrical impedance tomography process. The mask matrix has dimensions E x M, where E represents the number of stimulation modes and M represents the number of measurements for each stimulation mode.

[0121] In this process, the incomplete data frame x \ΠIt is obtained by applying the mask matrix Φ to the original electrical impedance data frame x, where π represents the set of indices of missing data. Specifically, if the data at a certain position in the original data frame is not collected due to the mask matrix, then the data at that position will be missing in the missing data frame, and the corresponding index will be included in the set π.

[0122] Therefore, the core of step 150 is to filter the raw electrical impedance data using the mask matrix Φ so that only the data most useful for subsequent data processing and imaging is collected, thereby achieving sparse data acquisition. This not only reduces the data processing burden but also speeds up electrical impedance imaging, especially when hardware performance is limited.

[0123] Step 160: Using the incomplete data frame x \Π The corresponding dictionary D \Π x \Π Perform sparse representation to obtain a sparse vector α \Π .

[0124] Step 160 describes the use of a dictionary matrix D that is optimized and adjusted for the specific incomplete data frame. \Π To the incomplete data frame x \Π Perform sparse representation and solve to obtain the corresponding sparse vector α \Π process.

[0125] in,

[0126] x \Π It refers to the incomplete data frame obtained after the mask matrix processing, where π represents the set of index values ​​of the missing data in the original signal x.

[0127] D \Π Based on the incomplete data frame x \Π is the adjusted dictionary matrix, which means that the dictionary has been adapted to the missing pattern of the data, possibly constructed by selecting the columns corresponding to non-missing data from the original dictionary matrix D.

[0128] The process of sparse representation refers to the use of D \Π To express x \Π , that is, find a sparse vector α \Π , so that D \Π α \Π As close to x as possible \Π This is usually achieved by solving an optimization problem that aims to minimize the reconstruction error while maintaining α \Π The sparsity of .

[0129] Through this step, the data can be reconstructed based on the incomplete data frame and the corresponding adjusted dictionary matrix, which provides an effective method for restoring the original signal when the data is incomplete.

[0130] Step 170: Using dictionary D and sparse vector α \Π The original signal is restored, and the acceleration ratio is calculated and imaging is performed using the electrical impedance tomography method.

[0131] Step 170 uses the dictionary D and the sparse vector α \Π To restore the original signal, calculate the acceleration ratio and the imaging process using the electrical impedance tomography method. The specific steps include:

[0132] 1) Original signal recovery: Use the trained dictionary D and the corresponding sparse vector α \Π , restore the original electrical impedance signal. This process involves using the dictionary D to restore the data points in the index set Π of each missing data, that is, for each index i in the index set Π, use D and α \Π The product of and recovers the corresponding value in the original signal.

[0133] 2) Calculate the acceleration ratio: The acceleration ratio (AR) is calculated by comparing the total number of measurements in the original data frame with the number of measurements actually collected. It reflects the degree of improvement in the speed of data collection. The calculation formula is: where N is the total number of measurements in the raw electrical impedance data frame, φij is the element of the electrical impedance data acquisition mask matrix, E is the total number of stimulation patterns, and M is the number of measurements in each stimulation pattern.

[0134] 3) Imaging: The recovered raw signal is fed into the electrical impedance tomography reconstruction algorithm for imaging. This step uses the recovered electrical impedance data to generate an electrical impedance image, thereby visualizing the internal structure of the object under test.

[0135] In summary, step 170 restores the complete original signal by using the trained dictionary and sparse representation technology, evaluates the efficiency of data acquisition by calculating the acceleration ratio, and performs electrical impedance imaging using the restored signal, thereby achieving high frame rate electrical impedance imaging.

[0136] Optionally, training sample acquisition is performed, and as the state of the imaging target changes, electrical impedance data samples containing as many target states as possible are acquired. In the electrical impedance imaging method, in order to improve the accuracy and adaptability of imaging, electrical impedance data samples can be selectively acquired according to changes in the imaging target (such as different parts of the human body or different medical conditions). The purpose of this is to ensure that the training sample matrix covers various possible states of the imaging target, so that the electrical impedance characteristics of the imaging target can be better reflected and reconstructed in the subsequent dictionary learning and imaging process. By including as many electrical impedance data samples of target state changes as possible, the adaptability of the imaging system to different conditions and the accuracy of imaging can be improved.

[0137] Optionally, step 10020 includes the following processing:

[0138] Singular value decomposition of sample matrix X

[0139] X=U,S,V T

[0140] Where U∈R N×N is a left unitary matrix;

[0141] Initialize the dictionary matrix D to the first P columns of U

[0142] D=U[:,:P]

[0143] The details are as follows:

[0144] 1) Singular value decomposition of the sample matrix X: This process involves decomposing the training sample matrix X into the product of three matrices, namely: X = U, S, V T

[0145] in:

[0146] U is a left unitary matrix of size N×N, that is, the column vectors of U are orthogonal and have modulus 1.

[0147] S is a diagonal matrix containing singular values, representing the different eigenvalues ​​of the data.

[0148] V T is another unitary matrix, often used to represent different eigenvectors.

[0149] 2) Initialize the dictionary matrix D: In this step, the dictionary matrix D is initialized to the first P columns of U, i.e., D = U[:,:P]. Here, P is the pre-set number of dictionary atoms, which determines the size of the dictionary matrix D. Initializing the dictionary matrix in this way ensures that D contains the most important features of the training sample matrix X, providing a good starting point for subsequent dictionary learning.

[0150] Optionally, the mask matrix in step 140 is:

[0151]

[0152] Among them, φ i，j (i=1, 2, ..., E; j=1, 2, ..., M) indicates whether the jth measurement value in the i-th stimulation mode in each frame of electrical impedance data is collected or not, φ i，j The value of is 1 or 0, if φ i，j If φ is 1, the data collection of the corresponding position in the electrical impedance data frame is retained. i，j If it is 0, the data collection of the corresponding position in the electrical impedance data frame will be skipped;

[0153] The mask matrix Φ in step 140 defines which data should be collected during the electrical impedance imaging process. This mask matrix has a structure of E rows and M columns, where E represents the number of stimulation modes and M represents the number of measurements under each stimulation mode. Each element φ(i, j) in the matrix (i ranges from 1 to E, j ranges from 1 to M) represents whether the jth measurement value under the i-th stimulation mode should be collected. If the value of φ(i, j) is 1, it indicates that the corresponding measurement value in the electrical impedance data frame should be collected; if the value of φ(i, j) is 0, it indicates that the corresponding measurement value will not be collected.

[0154] This mask matrix Φ allows users to customize which data should be acquired, reducing the amount of data required for processing and analysis, thereby speeding up electrical impedance imaging. For example, if the user is only interested in measurements taken under certain stimulation patterns, they can set the corresponding positions in the mask matrix Φ to 1 and all other positions to 0. This approach allows for sparse data acquisition, helping to increase imaging rates and reduce data processing requirements.

[0155] Optionally, the processing process of step 160 is as follows:

[0156] Sparse vector α \Π It is obtained by solving the following sparse optimization problem:

[0157]

[0158] Among them, λ is the error balance parameter, R(α) is the sparse constraint term, generally L0-Norm or L1-Norm;

[0159] More specifically, in step 160, the sparse vector α \Π It is obtained by solving the following sparse optimization problem:

[0160]

[0161] in:

[0162] α \Π is the sparse vector obtained after the solution, which is used to best sparsely represent the incomplete data frame x \Π .

[0163] x \Π is the incomplete data frame, i.e., the data obtained from the original electrical impedance signal by applying the mask matrix Φ.

[0164] D \Π It is the dictionary matrix corresponding to the incomplete data frame, which may be obtained by filtering out the columns corresponding to non-missing data from the full dictionary matrix D.

[0165] is the L2 norm of the reconstruction error, which measures the error between the sparse vector α and the dictionary matrix D \Π The reconstructed data and the original incomplete data frame x \Π The difference between.

[0166] λ is the error balance parameter, which is used to balance the weight between the reconstruction error and the sparsity constraint term.

[0167] R(α) is a sparse constraint term, typically using the L0 norm or L1 norm. The L0 norm represents the number of nonzero elements, while the L1 norm is the sum of the absolute values ​​of nonzero elements, and is used to promote the sparsity of the solution.

[0168] By solving this optimization problem, the sparse representation vector α of the incomplete data frame can be obtained \Π , this vector can be used in subsequent signal reconstruction and imaging processes.

[0169] Optionally, the processing process of step 170 is as follows:

[0170] Original signal restoration:

[0171]

[0172] Acceleration Rate (AR) calculation:

[0173]

[0174] Specifically, in the processing of step S7 above, there are two main parts: restoration of the original signal and calculation of the acceleration ratio (AR), which are as follows:

[0175] 1) Original signal recovery: This step describes how to use the sparse vector α \Π And dictionary matrix D to restore the original signal. For each data point x i , if index i does not belong to the index set Π of missing data (i.e. ), then the original signal x i Remain unchanged; if index i belongs to the index set Π of missing data (i.e. i∈Π), use the dictionary matrix D and sparse vector α \Π This means that for those missing data points, their values ​​will be estimated through sparse representation technology and the trained dictionary matrix.

[0176] 2) Acceleration ratio (AR) calculation: The acceleration ratio is an indicator to measure the sampling efficiency. It is calculated by dividing the total number of measurement values ​​N in the original data frame by the number of measurement values ​​actually collected (determined by the mask matrix Φ). Specifically, the acceleration ratio AR is calculated as Where φ_ij represents an element in the mask matrix, indicating whether the i-th stimulation pattern and the j-th measurement value are acquired (1 for acquisition, 0 for non-acquisition). A higher speedup ratio indicates a greater acceleration of the sampling process, which means that the number of data acquisitions is reduced while maintaining the same imaging quality.

[0177] According to the second example, the high frame rate electrical impedance tomography method based on sparse representation includes the following steps:

[0178] Step 210: obtaining the original electrical impedance dataset of the area to be measured, and forming a dictionary training dataset after normalization processing;

[0179] Step 220: Set the number of dictionary matrix atoms to P and create a dictionary matrix D;

[0180] Step 230: Perform singular value decomposition on the matrix composed of the dictionary training data set obtained in step 100, and assign the first P columns of the left unitary matrix of the singular value decomposition result to the dictionary matrix D to initialize the dictionary matrix;

[0181] Step 240: Based on the dictionary training data set in S1 and the initialized dictionary matrix in S3, a dictionary learning algorithm is used to train the dictionary matrix until the dictionary training algorithm converges (the performance index reaches a minimum value) or the maximum number of training times is reached.

[0182] Preferably, the performance index e of dictionary learning is:

[0183] Among them, α i is the i-th sample x i The sparse vector is sparsely represented using the dictionary D, where L is the number of training samples.

[0184] Step 250: creating an electrical impedance data acquisition mask matrix Φ, and acquiring electrical impedance data according to the mask matrix to obtain an incomplete data frame;

[0185] Step 260: Perform sparse representation on the incomplete data frame collected by S5 according to the dictionary matrix trained by S4, and restore the original electrical impedance data frame and calculate the acceleration ratio.

[0186] Preferably, the optimization function of sparse representation is:

[0187]

[0188] Where π represents the incomplete data frame x \Π The set of index values ​​corresponding to the missing data relative to the original data frame x, λ is the error balance parameter, and R(α) is the sparse constraint term, generally L0-Norm or L1-Norm.

[0189] Preferably, the original electrical impedance original data frame recovery method is:

[0190]

[0191] Preferably, the acceleration rate (AR) is calculated as follows:

[0192]

[0193] Among them, φ i，j (i=1, 2, ..., E; j=1, 2, ..., M) is the element value of the i-th row and j-th column of the electrical impedance data acquisition mask matrix Φ, indicating whether the j-th (each stimulation mode has M measurement values) measurement value under the i-th (there are E stimulation modes) stimulation mode in each frame of electrical impedance data is collected or not. i，j The value is 1 or 0.

[0194] Taking a 16-electrode impedance imaging system as an example, under the adjacent stimulation-measurement protocol, there are 16 stimulation modes, each with 13 impedance measurement values. A frame of unaccelerated impedance data contains 16 × 13 = 208 impedance measurement values. At this time, if the measurement values ​​under each stimulation mode are randomly lost at a ratio of 50%, that is, only half of the elements in each row of the Φ matrix are 1 and the other half are 0, then 200% acceleration can be achieved, and the frame rate of the impedance imaging system is twice that of the unaccelerated state.

[0195] Technical effect:

[0196] The above example addresses the problem that the frame rate of the electrical impedance imaging system is limited by hardware performance. A method for numerically improving the frame rate is proposed. This method does not require any upgrades or modifications to the existing hardware system and is independent of the stimulation-measurement protocol of electrical impedance imaging. It can be adapted to various electrical impedance application platforms and systems, breaking through the limitations of the hardware capabilities themselves and greatly improving the frame rate of electrical impedance imaging. Among them, the mask matrix of signal acquisition reduces hardware requirements and increases processing speed by reducing the amount of data collected, thereby achieving high-efficiency and low-cost signal acquisition. At the same time, the signal recovery part uses advanced algorithms based on these small amounts of collected data to recover complete signal information. This method not only reduces the necessary amount of data collection, but also ensures the accuracy of data recovery, making the entire system more efficient and economical, and suitable for various occasions requiring fast and efficient signal processing.

[0197] According to the third example, see Figure 2 , provides a high frame rate electrical impedance tomography method based on sparse representation, comprising the following steps:

[0198] The first step is dictionary learning and training:

[0199] Data samples are collected through the electrical impedance imaging platform, and dictionary learning algorithm training is performed to obtain a dictionary matrix. The steps 310 to 340 are included.

[0200] Step 310: collecting L frames of electrical impedance data through a 16-electrode electrical impedance imaging platform, and normalizing the modulus values ​​of each electrical impedance frame data to form a training data set for dictionary learning;

[0201] Step 320: performing singular value decomposition on the matrix formed by the sample set;

[0202] Step 330: Create a dictionary matrix D for P atoms, and assign the first P columns of the left unitary matrix obtained in step 320 to the dictionary matrix for initialization;

[0203] Step 340: Based on the electrical impedance data sample set collected in S1, the dictionary matrix D in step 330 is trained using a dictionary learning algorithm such as K-SVD until the performance index of the dictionary learning reaches a minimum value or the maximum number of training times is reached.

[0204] Among them, the performance index e of dictionary learning is:

[0205]

[0206] Among them, α i is the i-th sample x i A sparse vector sparsely represented using the dictionary D.

[0207] The second step is sparse sampling:

[0208] Sparse sampling is performed according to the acquisition mask matrix to obtain incomplete data, corresponding to step 350 .

[0209] Step 350: Create an electrical impedance data acquisition mask matrix Φ, and acquire electrical impedance data according to the mask matrix to obtain an incomplete data frame x \Π .

[0210] Among them, the mask matrix is:

[0211]

[0212] Among them, φ i，j (i=1, 2, ..., E; j=1, 2, ..., M) indicates whether the jth measurement value in the i-th stimulation mode in each frame of electrical impedance data is collected or not, φ i，j The value of is 1 or 0, if φ i，j If φ is 1, the data collection of the corresponding position in the electrical impedance data frame is retained. i，j If it is 0, the data collection of the corresponding position in the electrical impedance data frame will be skipped.

[0213] Step 3: Signal recovery:

[0214] The incomplete data is sparsely represented using a dictionary matrix, the sparse vector is solved, and the original complete signal is restored, including steps 360-370.

[0215] Step 360: Use the dictionary matrix D trained in step 340 to calculate x \Π Perform sparse representation solution.

[0216] The sparse optimization problem is:

[0217]

[0218] Where π represents the incomplete data frame x \Π The set of index values ​​corresponding to the missing data relative to the original data frame x, λ is the error balance parameter, and R(α) is the sparse constraint term, generally L0-Norm or L1-Norm.

[0219] Step 370: Use the dictionary matrix D trained in step 340 and the sparse vector α solved in step 360 \Π Restore the original signal.

[0220] The original signal recovery function is:

[0221]

[0222] Step 4: Imaging:

[0223] The restored original signal is input into the electrical impedance imaging reconstruction algorithm for imaging, corresponding to step 380 .

[0224] Step 380: Use the restored original signal to input into an electrical impedance imaging algorithm for imaging.

[0225] Through steps 310 to 340 , data samples of the corresponding electrical impedance imaging platform are collected, and prior knowledge is obtained through a dictionary learning algorithm.

[0226] Through steps 350 to 370, sparse sampling of the corresponding electrical impedance platform is achieved and the original signal is restored. Sparse sampling reduces the requirement on the number of electrical impedance data collected and greatly improves the frame rate of electrical impedance data collection.

[0227] Compared with methods such as using high-speed and high-performance hardware to improve the frame rate of electrical impedance imaging, the present invention proposes a high-frame-rate electrical impedance imaging method that is completely based on numerical methods. It has the advantages of not requiring any upgrades or modifications to existing hardware and being independent of the specific electrical impedance imaging stimulation-measurement protocol. It is low-cost, easy to use, and can be adapted to various electrical impedance imaging hardware systems, breaking through the limitations of the hardware capabilities themselves and greatly improving the frame rate of electrical impedance imaging.

[0228] The second embodiment of the present application relates to a high frame rate electrical impedance imaging system based on sparse representation, whose structure is as follows: Figure 3 As shown, the high frame rate electrical impedance imaging system based on sparse representation includes:

[0229] The training sample acquisition module is used to collect the original electrical impedance data and perform preprocessing to obtain the training sample matrix;

[0230] A dictionary training module is used to train a dictionary learning algorithm using the training sample matrix to obtain an optimized dictionary matrix;

[0231] The mask matrix setting and data acquisition module is used to set the mask matrix for electrical impedance data acquisition and perform sparse sampling on the original electrical impedance data according to the mask matrix to obtain incomplete data frames;

[0232] The data recovery module is used to perform sparse representation on the incomplete data frame using the optimized dictionary matrix, solve the corresponding sparse vector, and use the dictionary matrix and the sparse vector to restore the original electrical impedance data.

[0233] The first embodiment is a method embodiment corresponding to the present embodiment. The technical details in the first embodiment can be applied to the present embodiment, and the technical details in the present embodiment can also be applied to the first embodiment.

[0234] It should be noted that, those skilled in the art should understand that the implementation functions of the modules shown in the embodiment of the high frame rate electrical impedance imaging system based on sparse representation can be understood with reference to the relevant description of the high frame rate electrical impedance imaging method based on sparse representation. The functions of the modules shown in the embodiment of the high frame rate electrical impedance imaging system based on sparse representation can be implemented by a program (executable instruction) running on a processor, or by a specific logic circuit. If the high frame rate electrical impedance imaging system based on sparse representation in the embodiment of the present application is implemented in the form of a software function module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the embodiment of the present application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the methods described in each embodiment of the present application. The aforementioned storage medium includes various media that can store program code, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a disk, or an optical disk. Thus, embodiments of the present application are not limited to any specific combination of hardware and software.

[0235] Accordingly, an embodiment of the present application further provides a computer storage medium, in which computer-executable instructions are stored. When the computer-executable instructions are executed by a processor, the various method embodiments of the present application are implemented.

[0236] In addition, the present application embodiment also provides a high frame rate electrical impedance imaging system based on sparse representation, including a memory for storing computer executable instructions, and a processor; the processor is used to implement the steps in the above-mentioned various method implementations when executing the computer executable instructions in the memory. Wherein, the processor can be a central processing unit (Central Processing Unit, referred to as "CPU"), or other general-purpose processors, digital signal processors (Digital Signal Processor, referred to as "DSP"), application specific integrated circuits (Application Specific Integrated Circuit, referred to as "ASIC"), etc. The aforementioned memory can be a read-only memory (read-only memory, referred to as "ROM"), a random access memory (random access memory, referred to as "RAM"), a flash memory (Flash), a hard disk or a solid-state drive, etc. The steps of the method disclosed in each embodiment of the present invention can be directly embodied as a hardware processor for execution, or can be completed by a combination of hardware and software modules in the processor.

[0237] It should be noted that in this patent application, relational terms such as first and second, etc., are used solely to distinguish one entity or operation from another, and do not necessarily require or imply any actual relationship or order between these entities or operations. Furthermore, the terms "comprise," "include," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, article, or apparatus comprising a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus. Without further limitation, an element specified by the phrase "comprising a" does not preclude the presence of additional identical elements in the process, method, article, or apparatus comprising the element. In this patent application, reference to performing an action in accordance with an element means performing the action in accordance with at least that element, including two situations: performing the action in accordance with that element alone, and performing the action in accordance with that element and other elements. Expressions such as "plurality," "multiple times," and "many" include "two," "twice," "two kinds," and "more than two," "more than two times," and "more than two kinds."

[0238] All documents mentioned in this application are considered to be included in their entirety in the disclosure of this application so that they can be used as a basis for modification when necessary. In addition, it should be understood that after reading the above disclosure of this application, those skilled in the art may make various changes or modifications to this application, and these equivalent forms also fall within the scope of protection claimed in this application.

Claims

1. A high frame rate electrical impedance tomography method based on sparse representation, characterized in that: include: Collect raw electrical impedance data and preprocess them to obtain a training sample matrix; Using the training sample matrix to perform dictionary learning algorithm training to obtain an optimized dictionary matrix; Setting a mask matrix for electrical impedance data acquisition, and performing sparse sampling on the original electrical impedance data according to the mask matrix to obtain an incomplete data frame; The optimized dictionary matrix is ​​used to perform sparse representation on the incomplete data frame, the corresponding sparse vector is solved, and the original electrical impedance data is restored using the dictionary matrix and the sparse vector.

2. The method according to claim 1, wherein The raw electrical impedance data is collected and pre-processed to obtain a training sample matrix. The raw electrical impedance data frames are collected through the electrical impedance imaging platform, and the module value normalization processing is performed on each frame data to form a training sample matrix.

3. The method according to claim 1, wherein The dictionary learning algorithm is trained using the training sample matrix to obtain an optimized dictionary matrix, and the training sample matrix and the set number of dictionary matrix atoms are used to perform iterative training of the dictionary learning algorithm, wherein the process includes fixing the dictionary matrix to solve the sparse representation vector of the sample, and fixing the sparse vector to update the dictionary matrix until the convergence condition is met, and finally obtaining the optimized dictionary matrix.

4. The method according to claim 1, wherein The mask matrix for electrical impedance data acquisition is set, and the original electrical impedance data is sparsely sampled according to the mask matrix to obtain an incomplete data frame. According to the number of stimulation modes E of the electrical impedance imaging platform and the number of measurement values ​​M of each stimulation mode, the mask matrix Φ for electrical impedance data acquisition is set, where the size of Φ is E x M, and each element indicates whether the corresponding measurement value is collected, where 1 indicates collection and 0 indicates non-collection; then, the original electrical impedance data is sparsely sampled row by row according to the set mask matrix Φ to obtain an incomplete data frame containing the collected measurement values.

5. The method according to claim 1, wherein The optimized dictionary matrix is ​​used to perform sparse representation on the incomplete data frame, the corresponding sparse vector is solved, and the original electrical impedance data is restored using the dictionary matrix and the sparse vector. The incomplete data frame is sparsely represented using the optimized dictionary matrix, and an optimization model is constructed to solve the corresponding sparse vector α, wherein the optimization model is the L1 norm or L0 norm of the least squares model regularization; then, based on the obtained sparse vector α and the dictionary matrix, the product of the two matrices is calculated as the recovery value of the original electrical impedance data, and finally the recovery of the original electrical impedance data is completed.

6. The method according to claim 1 or 2, wherein: The dictionary learning algorithm training using the training sample matrix to obtain an optimized dictionary matrix includes the following processing steps: Perform singular value decomposition on the training sample matrix X and obtain X=U,S,V T Where U is a left unitary matrix of size N×N; The first P columns of the left unitary matrix U are used as the initialized dictionary matrix D.

7. The method according to claim 4, wherein In the step of setting the mask matrix Φ for electrical impedance data acquisition, the mask matrix is: Among them, φ i，j (i=1,2,…,E;j=1,2,…,M) indicates whether the jth measurement value is collected in the i-th stimulation mode, φ i，j The value is 1 or 0, 1 means collection, 0 means no collection.

8. The method according to claim 5, wherein The optimized dictionary matrix is ​​used to sparsely represent the incomplete data frame, and the corresponding sparse vector is solved. The sparse vector α is obtained by the following process. \Π : Will solve for the sparse vector α \∏ It is formulated as the following sparse optimization problem: Among them, λ is the error balance parameter, R(α) is the sparse constraint term, and takes the L0 norm or L1 norm.

9. The method according to claim 1 or 5, wherein: The process of restoring the original electrical impedance data by using the dictionary matrix and the sparse vector includes the following processing: Original signal recovery process: For each data point i, if i is not in the missing data index set π, then x i remain unchanged; If i is in π, then take the dictionary matrix D and the sparse vector α \П The product of x i The recovery value of Acceleration ratio calculation process: Acceleration ratio Among them, N is the total amount of original data, φ i，j It is a mask matrix element, indicating whether the corresponding data is collected.

10. A high frame rate electrical impedance tomography system based on sparse representation, characterized in that: include: The training sample acquisition module is used to collect the original electrical impedance data and perform preprocessing to obtain the training sample matrix; A dictionary training module is used to train a dictionary learning algorithm using the training sample matrix to obtain an optimized dictionary matrix; The mask matrix setting and data acquisition module is used to set the mask matrix for electrical impedance data acquisition and perform sparse sampling on the original electrical impedance data according to the mask matrix to obtain incomplete data frames; The data recovery module is used to perform sparse representation on the incomplete data frame using the optimized dictionary matrix, solve the corresponding sparse vector, and use the dictionary matrix and the sparse vector to restore the original electrical impedance data.

Images