A method for optimizing EIT imaging algorithm based on neurocritical care

Abstract

The present invention relates to the field of medical imaging optimization technology, and discloses an EIT imaging algorithm optimization method based on neurocritical illness, comprising the following steps: S101, collecting the patient's physiological characteristic data, the anatomical characteristic data of the anatomical region, and the basic characteristic data, and representing them through a first characteristic sequence; S102, preprocessing the first characteristic sequence to obtain a second characteristic sequence; S103, constructing an impedance baseline distribution prediction model according to the second characteristic sequence, and outputting the patient's impedance baseline distribution in a second preset time period; S104, using an improved Parrot optimization algorithm to optimize the preset sensitivity matrix, constructing an objective function according to the impedance baseline distribution prediction model, and outputting an optimal sensitivity matrix. The present invention optimizes the sensitivity matrix through personalized impedance baseline distribution prediction and an improved Parrot optimization algorithm, thereby achieving precise imaging, improving the adaptability to complex pathological scenes, and effectively reducing artifact sensitivity.

Description

Technical Field

[0001] The present invention belongs to the technical field of medical imaging optimization, and in particular relates to an EIT imaging algorithm optimization method based on neurocritical care. Background Art

[0002] Electrical impedance tomography (EIT) is a non-invasive imaging technique that is widely used in medical monitoring and engineering detection by measuring the changes in the electrical impedance distribution in the target area. However, since EIT is a typical ill-posed inverse problem, the reconstruction quality of its electrical impedance distribution is easily affected by measurement noise and the accuracy of the sensitivity matrix. Therefore, optimizing the sensitivity matrix to improve the reconstruction accuracy becomes the key to solving the EIT inverse problem.

[0003] Currently, the sensitivity matrix is ​​usually obtained through numerical simulation or experimental measurement, but these methods cannot dynamically adapt to individual differences or nonlinear changes in complex scenarios. Especially in the medical field, the diversity of individual anatomical characteristics and physiological states makes it difficult for EIT reconstruction based on a fixed sensitivity matrix to meet high-precision requirements. Summary of the invention

[0004] The present invention provides an EIT imaging algorithm optimization method based on neurocritical care, which solves the technical problems in the related technology that the sensitivity matrix is ​​fixed and lacks dynamic optimization capability, making it difficult to adapt to individual differences and nonlinear changes in complex scenes, resulting in insufficient accuracy of electrical impedance distribution reconstruction.

[0005] The present invention provides an EIT imaging algorithm optimization method based on neurological critical care, comprising the following steps:

[0006] S101, collecting physiological characteristic data of a patient, anatomical characteristic data of an anatomical region, and basic characteristic data within a first preset time period according to a first preset time interval, and representing them through a first characteristic sequence;

[0007] The nth sequence unit of the first feature sequence represents the physiological feature data, anatomical feature data and basic feature data of the patient at the nth time point, and n is the time point index;

[0008] S102, preprocessing the first feature sequence to obtain a second feature sequence;

[0009] S103, constructing an impedance baseline distribution prediction model according to the second feature sequence, and outputting the impedance baseline distribution of the patient in the second preset time period, where the impedance baseline distribution represents the impedance distribution of the patient in a healthy state;

[0010] S104, using the improved Parrot optimization algorithm to optimize the preset sensitivity matrix, constructing the objective function according to the impedance baseline distribution prediction model, and outputting the optimal sensitivity matrix.

[0011] Furthermore, the patient's anatomical region is divided into M sub-regions, the electrical impedance value of each sub-region of the anatomical region is collected, and the electrical impedance value of each sub-region is spliced ​​to obtain a comprehensive electrical impedance vector, which is represented by anatomical feature data. The physiological feature data include: respiratory rate, heart rate, oxygen saturation, intracranial pressure, and the basic feature data include: age, gender, weight and height, among which gender is represented by real number coding.

[0012] Furthermore, the impedance distribution is represented by anatomical feature data, namely, a comprehensive impedance vector, each vector element represents the impedance value of a sub-region, and the sensitivity matrix is ​​represented by matrix coding, each matrix element represents the sensitivity value of the electrode pair to the impedance change of a sub-region.

[0013] Furthermore, the specific steps of S102 include:

[0014] S201, use 3 The principle is to identify outliers in the first feature sequence and replace the outliers with the average value of the values ​​other than the outliers;

[0015] S202, extracting a dynamic respiratory resistance feature according to the processed first feature sequence, and splicing it after the sequence unit of the first sequence feature. The calculation formula of the dynamic respiratory resistance feature is: ,in, represents the dynamic respiratory impedance feature, N represents the number of time points, n represents the time point index, M represents the number of sub-regions, m represents the sub-region index, e represents the natural constant, represents the electrical impedance value of the mth sub-region at the nth time point, represents the respiratory rate at the nth time point;

[0016] S203, extracting the physiological impedance multi-mode cross feature according to the processed first feature sequence, and splicing it after the sequence unit of the first sequence feature. The calculation formula of the physiological impedance multi-mode cross feature is: ,in, Indicates the multi-mode crossover characteristics of physiological resistance, represents the heart rate at the nth time point, represents the oxygen saturation at the nth time point, represents the heart rate weight coefficient, represents the intracranial pressure at the nth time point;

[0017] S204, normalizing the concatenated first feature sequence using a z-score normalization method to obtain a second feature sequence.

[0018] Furthermore, the electrical impedance baseline distribution prediction model is constructed based on the gated recurrent unit and improved by combining it with the frog effect. The structure of the electrical impedance baseline distribution prediction model includes: an input enhancement module, a dynamic gating module, a drift detection module and an output module;

[0019] The input enhancement module is used to input the second feature sequence and the sequence unit change rate feature, and the sequence unit change rate feature represents the difference between the current sequence unit and the previous sequence unit;

[0020] The dynamic gating module includes a reset gate, an update gate, and a hidden state update, and introduces the sequence unit change rate characteristics into the reset gate and the update gate;

[0021] The drift detection module is used to monitor the deviation between the current hidden state and the first hidden state, and when the deviation is greater than a first preset threshold, use a correction formula to perform correction, wherein the deviation between the current hidden state and the first hidden state is represented by the Euclidean distance between the two, and the first hidden state represents the mean value of the hidden state obtained by training in a sample of healthy people;

[0022] The output module is used to output the patient's electrical impedance baseline distribution in a second preset time period.

[0023] Furthermore, the calculation formula of the dynamic gating module includes:

[0024] ;

[0025] ;

[0026] ;

[0027] in, and represent the nth and n-1th sequence units of the second characteristic sequence respectively, , and Respectively represent the hidden states output at the nth, n-1th, and n-2th time points, and Represent the reset gate and update gate respectively, and Respectively represent the first weight coefficient and bias parameter of the reset gate, Represents the rate-of-change adjustment coefficient of the reset gate, Represents a splicing operation, and Respectively represent the first weight coefficient and bias parameter of the update gate, represents the rate-of-change adjustment coefficient of the update gate, and Respectively represent the first weight coefficient and bias parameter of the hidden state, Represents the adjustment coefficient of the residual connection, sigmoid represents the sigmoid activation function, and tanh represents the tanh activation function.

[0028] Furthermore, the correction formula is: ;

[0029] in, represents the hidden state at the nth time point after correction, represents the coefficient that controls the magnitude of the correction, represents the first hidden state.

[0030] Furthermore, the specific steps of S104 include:

[0031] S301, generating an initialization population satisfying constraint conditions based on normal distribution perturbation, wherein the initialization population includes P individuals;

[0032] S302, constructing an objective function according to the impedance baseline distribution prediction model, the calculation formula of the objective function is: ,in, represents the fitness value of an individual, represents the electrical impedance distribution of the individual after reconstruction, It indicates the electrical impedance distribution actually measured by the EIT device when the patient is in a diseased state. represents the pth individual, represents the baseline sensitivity matrix calculated based on the impedance baseline distribution output by the impedance baseline distribution prediction model, Represents the control baseline constraint The weight coefficient of express and The Euclidean distance of express and The Frobenius norm of ;

[0033] S303, calculating the fitness value of each individual in the population according to the objective function;

[0034] S304, updating the individual by simulating the behavior of the parrot;

[0035] S305, when the current number of iterations reaches the maximum number of iterations, the individual with the largest current fitness value is output as the optimal sensitivity matrix, otherwise return to S303, where the maximum number of iterations is a custom parameter.

[0036] Furthermore, the specific steps of S301 include:

[0037] S501, determining the range of sensitivity matrix elements;

[0038] S502, generating an initial sensitivity matrix using uniform distribution according to the range of sensitivity matrix elements;

[0039] S503, generating P individuals by adding normal distribution disturbance to the initial sensitivity matrix element by element to form an initialization population. Specifically, the calculation formula for generating individuals is: ,in, , represents the pth individual generated, represents the initial sensitivity matrix, represents a random noise matrix of the same dimension as the initial sensitivity matrix, Each element of follows a normal distribution with a mean of 0 and a variance of .

[0040] The beneficial effects of the present invention are as follows: the present invention constructs an electrical impedance baseline distribution prediction model based on health status by combining the patient's physiological characteristics, anatomical characteristics and basic characteristics, making the sensitivity matrix optimization more personalized and accurate, and significantly improving the adaptability of electrical impedance imaging in complex pathological scenarios;

[0041] The present invention optimizes the sensitivity matrix by using the improved Parrot optimization algorithm and constructs the objective function in combination with the impedance baseline distribution prediction model, so that the reconstruction result can reflect the pathological state without deviating from the healthy state benchmark, thereby reducing the sensitivity to artifacts. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 It is a flow chart of an EIT imaging algorithm optimization method based on neurocritical illness of the present invention. DETAILED DESCRIPTION

[0043] The subject matter described herein will now be discussed with reference to example embodiments. It should be understood that the discussion of these embodiments is only to enable those skilled in the art to better understand and implement the subject matter described herein, and the functions and arrangements of the elements discussed may be changed without departing from the scope of protection of the contents of this specification. Each example may omit, replace or add various processes or components as needed. In addition, the features described relative to some examples may also be combined in other examples.

[0044] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in one or more embodiments of the present invention should be understood by people with ordinary skills in the field to which the present invention belongs. The words "first", "second" and similar words used in one or more embodiments of the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. "Include" or "comprise" and similar words mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0045] like Figure 1 As shown, a method for optimizing an EIT imaging algorithm based on neurocritical illness includes the following steps:

[0046] S101, collecting physiological characteristic data of a patient, anatomical characteristic data of an anatomical region, and basic characteristic data within a first preset time period according to a first preset time interval, and representing them through a first characteristic sequence;

[0047] The nth sequence unit of the first feature sequence represents the physiological feature data, anatomical feature data and basic feature data of the patient at the nth time point, and n is the time point index;

[0048] S102, preprocessing the first feature sequence to obtain a second feature sequence;

[0049] S103, constructing an impedance baseline distribution prediction model according to the second feature sequence, and outputting the impedance baseline distribution of the patient in the second preset time period, where the impedance baseline distribution represents the impedance distribution of the patient in a healthy state;

[0050] S104, using the improved Parrot optimization algorithm to optimize the preset sensitivity matrix, constructing the objective function according to the impedance baseline distribution prediction model, and outputting the optimal sensitivity matrix.

[0051] In the present invention, directly using the impedance distribution in the diseased state to calculate the sensitivity matrix will cause the sensitivity matrix to be too close to the patient's current pathological characteristics, causing the reconstruction to be too sensitive to artifacts. By predicting the impedance distribution in the patient's healthy state through the impedance baseline distribution prediction model, the sensitivity can be constrained so that the reconstruction can reflect the diseased state without deviating from the reference value of the healthy state.

[0052] In one embodiment of the present invention, the patient's anatomical region is divided into M sub-regions, the electrical impedance value of each sub-region of the anatomical region is collected, and the electrical impedance value of each sub-region is spliced ​​to obtain a comprehensive electrical impedance vector, which is represented by anatomical feature data. The physiological feature data include: respiratory rate, heart rate, oxygen saturation, intracranial pressure, and basic feature data include: age, gender, weight and height, among which gender is represented by real number coding.

[0053] Specifically, for anatomical feature data, a three-dimensional anatomical structure map of the patient's anatomical region is extracted through medical imaging data such as CT and MRI, and the three-dimensional anatomical structure map is evenly divided into different sub-regions according to a grid. The electrical impedance signal is collected through the EIT device to obtain the electrical impedance value of each sub-region, and then the comprehensive electrical impedance vector is obtained.

[0054] In one embodiment of the present invention, the impedance distribution is represented by anatomical feature data, i.e., a comprehensive impedance vector, each vector element represents the impedance value of a sub-region, and the sensitivity matrix is ​​represented by matrix coding, each matrix element represents the sensitivity value of the electrode pair to the impedance change of a sub-region.

[0055] In one embodiment of the present invention, the specific steps of S102 include:

[0056] S201, use 3 The principle is to identify outliers in the first feature sequence and replace them with the average value of the values ​​other than the outliers. Specifically, the mean and standard deviation of the features in the first feature sequence are calculated, and the normal data range is obtained as follows: , identify values ​​outside the normal data range as outliers, and replace the identified outliers with the average value of the values ​​outside the range to ensure the stability and accuracy of the data, where is the mean, is the standard deviation;

[0057] S202, extracting a dynamic respiratory resistance feature according to the processed first feature sequence, and splicing it after the sequence unit of the first sequence feature, wherein the dynamic respiratory resistance feature can capture the relationship between physiological changes and electrical impedance distribution by weighting the dynamic characteristics of the time series. Specifically, the calculation formula of the dynamic respiratory resistance feature is: ,in, represents the dynamic respiratory impedance feature, N represents the number of time points, n represents the time point index, M represents the number of sub-regions, m represents the sub-region index, e represents the natural constant, represents the electrical impedance value of the mth sub-region at the nth time point, represents the respiratory rate at the nth time point;

[0058] S203, extracting a physiological impedance multimode cross feature according to the processed first feature sequence, and splicing it after the sequence unit of the first sequence feature, wherein the physiological impedance multimode cross feature can quantify the comprehensive influence of the physiological signal on the electrical impedance distribution by combining the relationship between heart rate, oxygen saturation and intracranial pressure and electrical impedance distribution. Specifically, the calculation formula of the physiological impedance multimode cross feature is: ,in, Indicates the multi-mode crossover characteristics of physiological resistance, represents the heart rate at the nth time point, represents the oxygen saturation at the nth time point, represents the heart rate weight coefficient, represents the intracranial pressure at the nth time point;

[0059] S204, normalizing the concatenated first feature sequence using a z-score normalization method to obtain a second feature sequence.

[0060] In one embodiment of the present invention, an impedance baseline distribution prediction model is constructed based on a gated cyclic unit and is improved by combining it with the frog effect. The frog effect emphasizes that the system is insensitive to slowly changing signals, which may result in key changes not being captured. In the prediction of impedance baseline distribution, certain slow changes such as gradual changes in pathological development or gradually drifting baselines may be ignored by the gated cyclic unit, resulting in inaccurate predictions. Therefore, the gated cyclic unit is improved. The structure of the impedance baseline distribution prediction model includes: an input enhancement module, a dynamic gating module, a drift detection module and an output module. The input enhancement module increases the sequence unit change rate feature, the dynamic gating module enhances the selective memory of slowly changing features, and the drift detection module dynamically monitors the drift of hidden states to prevent cumulative errors.

[0061] In one embodiment of the present invention, the electrical impedance baseline distribution prediction model is constructed based on the gated cycle unit and improved by combining it with the frog effect. The structure of the electrical impedance baseline distribution prediction model includes: an input enhancement module, a dynamic gating module, a drift detection module and an output module;

[0062] The input enhancement module is used to input the second feature sequence and the sequence unit change rate feature, and the sequence unit change rate feature represents the difference between the current sequence unit and the previous sequence unit;

[0063] The dynamic gating module includes a reset gate, an update gate, and a hidden state update, and introduces the sequence unit change rate characteristics into the reset gate and the update gate;

[0064] The drift detection module is used to monitor the deviation between the current hidden state and the first hidden state, and when the deviation is greater than a first preset threshold, use a correction formula to perform correction, wherein the deviation between the current hidden state and the first hidden state is represented by the Euclidean distance between the two, and the first hidden state represents the mean value of the hidden state obtained by training in a sample of healthy people;

[0065] The output module is used to output the patient's electrical impedance baseline distribution in a second preset time period.

[0066] In one embodiment of the present invention, the calculation formula of the dynamic gating module includes:

[0067] ;

[0068] ;

[0069] ;

[0070] in, and represent the nth and n-1th sequence units of the second characteristic sequence respectively, , and Respectively represent the hidden states output at the nth, n-1th, and n-2th time points, and Represent the reset gate and update gate respectively, and Respectively represent the first weight coefficient and bias parameter of the reset gate, Represents the rate-of-change adjustment coefficient of the reset gate, Represents a splicing operation, and Respectively represent the first weight coefficient and bias parameter of the update gate, represents the rate-of-change adjustment coefficient of the update gate, and Respectively represent the first weight coefficient and bias parameter of the hidden state, Represents the adjustment coefficient of the residual connection, sigmoid represents the sigmoid activation function, and tanh represents the tanh activation function.

[0071] In one embodiment of the present invention, the correction formula is: ;

[0072] in, represents the hidden state at the nth time point after correction, represents the coefficient that controls the magnitude of the correction, represents the first hidden state.

[0073] In one embodiment of the present invention, the specific steps of S104 include:

[0074] S301, generating an initialization population satisfying constraint conditions based on normal distribution perturbation, wherein the initialization population includes P individuals;

[0075] The individual is represented by a sensitivity matrix, the dimension of which is B×M, where B represents the number of electrode pairs, M represents the number of sub-regions, and the elements of the sensitivity matrix represent the sensitivity of the electrode pair to the change in the electrical impedance of a sub-region;

[0076] The constraints include:

[0077] The values ​​of the elements of the sensitivity matrix are within a preset range, for example, no sensitivity is 0, the highest sensitivity is 1, and the value of the element ranges from 0 to 1;

[0078] The sum of each row of the sensitivity matrix is ​​1, that is, the total sensitivity value of each electrode pair is 1;

[0079] S302, constructing an objective function according to the impedance baseline distribution prediction model, the calculation formula of the objective function is: ,in, represents the fitness value of an individual, Represents the electrical impedance distribution of an individual after reconstruction. Specifically, reconstruction refers to the process of inferring the electrical impedance distribution of a sub-region by measuring the electrode potential difference V and the sensitivity matrix. , represents the pseudo-inverse operation of the pth individual, V represents the electrode potential difference, It indicates the electrical impedance distribution actually measured by the EIT device when the patient is in a diseased state. represents the pth individual, represents the baseline sensitivity matrix calculated based on the impedance baseline distribution output by the impedance baseline distribution prediction model, Represents the control baseline constraint The weight coefficient of express and The Euclidean distance of express and The Frobenius norm of ;

[0080] S303, calculating the fitness value of each individual in the population according to the objective function;

[0081] S304, updating the individual by simulating the behavior of the parrot;

[0082] Specifically, the steps for updating an individual include:

[0083] S401, simulating the foraging behavior of the parrot, and using the first update formula to update the individual, the first update formula is: ,in, and They represent the sensitivity matrix of the kth individual when the number of iterations is t+1 and t, respectively. represents the Levy flight distribution, represents the individual with the largest fitness value in the population. Represents a random number between 0 and 1. represents the exponential decay coefficient, t represents the current number of iterations, and k represents the individual index of the population;

[0084] S402, simulating the parrot's staying behavior, using the second updating formula to update the individual, the second updating formula is: , where c represents the iteration index;

[0085] S403, simulating the communication behavior of the parrot, and using the third update formula to update the individual, the third update formula is: ,in, represents the disturbance coefficient, represents a random matrix that follows a normal distribution;

[0086] In one embodiment of the present invention, the foraging behavior is used to simulate the parrot's extensive exploration of the solution space, so as to ensure the global optimization of the sensitivity matrix and the electrical impedance distribution during the optimization process, and effectively avoid the imaging quality being affected by premature convergence to the local solution; the stay behavior is used to help the optimization algorithm to further conduct local search after finding the potential optimal solution, thereby improving the accuracy of the sensitivity matrix and ensuring that the reconstruction result is consistent with the baseline distribution of the healthy state; the communication behavior is used to enhance the diversity of the population and the global search capability, accelerate the optimization convergence, and improve the quality of the final imaging result.

[0087] S305, when the current number of iterations reaches the maximum number of iterations, the individual with the largest current fitness value is output as the optimal sensitivity matrix, otherwise return to S303, where the maximum number of iterations is a custom parameter, preferably, the maximum number of iterations is set to 200.

[0088] In one embodiment of the present invention, the specific steps of S301 include:

[0089] S501, determining the range of sensitivity matrix elements;

[0090] S502, generating an initial sensitivity matrix using uniform distribution according to the range of sensitivity matrix elements;

[0091] S503, generating P individuals by adding normal distribution disturbance to the initial sensitivity matrix element by element to form an initialization population. Specifically, the calculation formula for generating individuals is: ,in, , represents the pth individual generated, represents the initial sensitivity matrix, represents a random noise matrix of the same dimension as the initial sensitivity matrix, Each element of follows a normal distribution with a mean of 0 and a variance of .

[0092] In one embodiment of the present invention, by adding a normal distribution disturbance based on the initial sensitivity matrix, it is possible to effectively avoid individuals in the population from being overly concentrated on a certain solution, thereby improving the breadth of the algorithm in the global search process and improving the optimization efficiency.

[0093] The above describes an embodiment of the present embodiment, but the present embodiment is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present embodiment, ordinary technicians in this field can also make many forms, all of which are within the protection of the present embodiment.

Claims

1. A method for optimizing an EIT imaging algorithm based on neurological critical care, characterized in that: The following steps are involved: S101, collecting physiological characteristic data of a patient, anatomical characteristic data of an anatomical region, and basic characteristic data according to a first preset time interval within a first preset time period, and representing them through a first characteristic sequence; The nth sequence unit of the first feature sequence represents the physiological feature data, anatomical feature data and basic feature data of the patient at the nth time point, and n is the time point index; S102, preprocessing the first feature sequence to obtain a second feature sequence, the specific steps include: S201, use 3 The principle is to identify outliers in the first feature sequence and replace the outliers with the average value of the values ​​other than the outliers; S202, extracting a respiratory dynamic resistance feature according to the processed first feature sequence, and splicing it after the sequence unit of the first sequence feature; S203, extracting physiological impedance multi-mode cross features according to the processed first feature sequence, and splicing the features after the sequence unit of the first sequence features; S204, normalizing the concatenated first feature sequences using a z-score normalization method to obtain a second feature sequence; S103, constructing an impedance baseline distribution prediction model according to the second feature sequence, and outputting the impedance baseline distribution of the patient in the second preset time period, where the impedance baseline distribution represents the impedance distribution of the patient in a healthy state; The impedance baseline distribution prediction model is constructed based on the gated recurrent unit and improved by combining it with the frog effect. The structure of the impedance baseline distribution prediction model includes: input enhancement module, dynamic gating module, drift detection module and output module; The input enhancement module is used to input the second feature sequence and the sequence unit change rate feature, and the sequence unit change rate feature represents the difference between the current sequence unit and the previous sequence unit; The dynamic gating module includes a reset gate, an update gate and a hidden state update, and introduces the sequence unit change rate feature in the reset gate and the update gate; the drift detection module is used to monitor the deviation between the current hidden state and the first hidden state, and when the deviation is greater than the first preset threshold, a correction formula is used for correction, wherein the deviation between the current hidden state and the first hidden state is represented by the Euclidean distance between the two, and the first hidden state represents the hidden state mean obtained by training in the healthy population sample, wherein the correction formula is: ;in, represents the hidden state at the nth time point after correction, represents the coefficient that controls the magnitude of the correction, represents a first hidden state; the output module is used to output the patient's electrical impedance baseline distribution in the second preset time period; S104, using the improved Parrot optimization algorithm to optimize the preset sensitivity matrix, constructing the objective function according to the impedance baseline distribution prediction model, and outputting the optimal sensitivity matrix.

2. The EIT imaging algorithm optimization method based on neurological critical care according to claim 1, characterized in that: The patient's anatomical region is divided into M sub-regions, the electrical impedance value of each sub-region of the anatomical region is collected, and the electrical impedance values ​​of each sub-region are spliced ​​to obtain a comprehensive electrical impedance vector, which is represented by anatomical feature data. The physiological feature data include: respiratory rate, heart rate, oxygen saturation, intracranial pressure, and basic feature data include: age, gender, weight and height, among which gender is represented by real number coding.

3. The EIT imaging algorithm optimization method based on neurological critical care according to claim 2, characterized in that: The impedance distribution is represented by anatomical feature data, namely, a comprehensive impedance vector, each vector element represents the impedance value of a sub-region, and the sensitivity matrix is ​​represented by matrix coding, each matrix element represents the sensitivity value of the electrode pair to the impedance change of a sub-region.

4. The EIT imaging algorithm optimization method based on neurological critical care according to claim 1, characterized in that: The calculation formula of the dynamic gating module includes: ； ； ;in, and represent the nth and n-1th sequence units of the second characteristic sequence respectively, , and Respectively represent the hidden states output at the nth, n-1th, and n-2th time points, and Represent the reset gate and update gate respectively, and Respectively represent the first weight coefficient and bias parameter of the reset gate, Represents the rate-of-change adjustment coefficient of the reset gate, Represents a splicing operation, and Respectively represent the first weight coefficient and bias parameter of the update gate, represents the rate-of-change adjustment coefficient of the update gate, and Respectively represent the first weight coefficient and bias parameter of the hidden state, Represents the adjustment coefficient of the residual connection, sigmoid represents the sigmoid activation function, and tanh represents the tanh activation function.

5. The EIT imaging algorithm optimization method based on neurological critical care according to claim 1, characterized in that: The specific steps of S104 include: S301, generating an initialization population that meets the constraint conditions based on normal distribution disturbance, the initialization population includes P individuals; S302, constructing an objective function according to the impedance baseline distribution prediction model, the calculation formula of the objective function is: , where F represents the fitness value of the individual, represents the electrical impedance distribution of the individual after reconstruction, It indicates the electrical impedance distribution actually measured by the EIT device when the patient is in a diseased state. represents the pth individual, represents the baseline sensitivity matrix calculated based on the impedance baseline distribution output by the impedance baseline distribution prediction model, Represents the control baseline constraint The weight coefficient of express and The Euclidean distance of express and Frobenius norm; S303, calculate the fitness value of each individual in the population according to the objective function; S304, updating the individual by simulating the behavior of the parrot; S305, when the current number of iterations reaches the maximum number of iterations, the individual with the largest current fitness value is output as the optimal sensitivity matrix, otherwise return to S303, where the maximum number of iterations is a custom parameter.

6. The EIT imaging algorithm optimization method based on neurological critical care according to claim 5, characterized in that: The specific steps of S301 include: S501, determining the range of sensitivity matrix elements; S502, generating an initial sensitivity matrix using uniform distribution according to the range of sensitivity matrix elements; S503, generating P individuals by adding normal distribution disturbance to the initial sensitivity matrix element by element to form an initialization population. Specifically, the calculation formula for generating individuals is: ,in, , represents the pth individual generated, represents the initial sensitivity matrix, represents a random noise matrix of the same dimension as the initial sensitivity matrix, Each element of follows a normal distribution with a mean of 0 and a variance of .

Images