A data compensation method for reconstructing conductivity distribution in cerebral ischemia

Abstract

The present invention discloses a data compensation method for reconstructing the conductivity distribution of cerebral ischemia. The method determines the degree of dehydration by feeding the measured voltage value of the full-field boundary of the brain into a trained fully connected neural network. The corresponding model boundary measured voltage value caused by individual scalp dehydration is found in a priori sequence based on the dehydration degree. Compensation is then performed and combined with the coordinate information of the pixels to achieve reconstruction of the intracranial cerebral ischemia image. Finally, the image reconstruction result is binarized using an adaptive threshold filtering algorithm. By directly processing the boundary voltage data, the method can effectively improve the image reconstruction quality.

Description

Technical Field

[0001] The invention belongs to the technical field of electrical tomography, and in particular relates to a data compensation method for reconstructing cerebral ischemia conductivity distribution. Background Art

[0002] Medical imaging is a key part of clinical procedures, providing a high accuracy rate for monitoring human diseases and providing a strong scientific basis for clinical diagnosis and biomedical research. Currently, many medical imaging technologies have been developed. Traditional imaging methods include computer tomography (CT) and magnetic resonance imaging (MRI). However, CT is radioactive and MRI is time-consuming and expensive. In recent years, electrical impedance tomography (EIT) has attracted widespread attention from many researchers due to its low price and high temporal resolution. Compared with CT and MRI, EIT is more suitable for bedside monitoring of patients. Therefore, EIT shows great potential in medical applications such as lung ventilation, brain imaging and cardiac function imaging.

[0003] Brain edema is a serious complication of acute cerebral ischemia. It is estimated that ischemia-induced cerebral edema accounts for 50% of patient mortality. Therefore, accurate detection of cerebral ischemia is crucial for patient recovery. To reduce patient mortality and improve prognosis, accurate imaging examinations are crucial in clinical treatment. However, conventional imaging methods cannot continuously monitor intracranial pathological changes. To demonstrate real-time therapeutic efficacy during treatment, bedside diagnostic methods for monitoring cerebral edema are urgently needed. As a visualization technique, EIT has gained popularity due to its ability to monitor intracranial pathological changes in real time. This technique reconstructs images representing changes in conductivity distribution and provides insights into physiological changes within brain regions, offering an alternative to brain imaging and making it a promising candidate for monitoring edema treatment. However, during dehydration procedures for cerebral edema treatment, it has been found that dehydration of the brain tissue also leads to scalp dehydration, significantly affecting boundary measurements. Consequently, the quality of cerebral edema reconstruction is severely affected. To address this technical issue, the present invention proposes a data compensation method for conductivity distribution reconstruction during cerebral ischemia to reduce the impact of scalp dehydration on intracranial monitoring.

[0004] In the existing technology, Bin Yang et al. published an article titled "Comparison of electrical impedance tomography and intracranial pressure during dehydration treatment of cerebral edema" in Volume 23, Page 101909 of the Elsevier NeuroImage Clinical in 2019. The article shows that EIT can monitor changes in brain water content to reveal the severity of cerebral edema. It can provide real-time and non-invasive imaging tools for early identification of cerebral edema and evaluation of mannitol dehydration. Although mannitol dehydration is effective in treating cerebral edema, during the mannitol dehydration treatment of cerebral edema, both the brain tissue layer and the scalp layer will be dehydrated. Scalp dehydration will lead to an increase in the impedance of the scalp layer, further hindering the injection of current into the brain layer, which will have a certain impact on the accurate monitoring of intracranial dehydration. However, the above methods mainly focus on studying the relationship between intracranial pressure and EIT images during dehydration. There are relatively few studies on suppressing the influence of scalp dehydration and using direct processing of voltage data to accurately reconstruct the distribution of cerebral ischemia conductivity.

[0005] In order to improve the spatial resolution of reconstructed images and the quality of brain imaging in patients with cerebral ischemia, the present invention proposes a data compensation method for reconstructing the conductivity distribution of cerebral ischemia. This method uses the compensated boundary voltage for image reconstruction. Compared with the previous linear image reconstruction algorithm, this method can effectively improve the image reconstruction quality by directly processing the boundary voltage data. Currently, there are no related reports in this regard. Summary of the Invention

[0006] The technical problem solved by the present invention is to provide a data compensation method for reconstructing the conductivity distribution of cerebral ischemia. The method determines the degree of dehydration by sending the measured voltage value of the full-field boundary of the brain into a trained fully connected neural network. The corresponding model boundary measured voltage value caused by individual scalp dehydration is found in the prior sequence based on the dehydration degree. Compensation is then performed and combined with the coordinate information of the pixel to achieve reconstruction of the intracranial cerebral ischemia image. Finally, the image reconstruction result is binarized using an adaptive threshold filtering algorithm. By directly processing the boundary voltage data, the method can effectively improve the image reconstruction quality.

[0007] The present invention adopts the following technical solution to solve the above technical problems: a data compensation method for reconstructing the conductivity distribution of cerebral ischemia, characterized by the following specific steps:

[0008] Step S1: Determine the shape and structure of the brain, construct a standard 2D elliptical brain model on a computer, and integrate the conductivity information of different brain tissues into the tissue structure of different layers. The model has a three-layer structure of scalp, skull, and brain tissue. The conductivity of the scalp layer, skull layer, and brain tissue layer is set to 0.44S / m, 0.012S / m, and 0.163S / m, respectively, to simulate brain edema.

[0009] Step S2: Using a 16-electrode electrical impedance tomography system, electrode No. 1 is placed at the most anterior point of the cranial model, and then 16 electrodes are equally spaced and attached to the scalp layer in a counterclockwise direction. In the mode of relative current excitation and adjacent voltage measurement, a safe current excitation is first applied to electrode No. 1, while electrode No. 9 is grounded. Voltage measurements are taken for a total of 12 electrode pairs, 2-3, 3-4...6-7, 7-8; 10-11, 11-12...14-15, and 15-16, obtaining a total of 12 voltage measurement values ​​as the first set of measurement data. Similarly, according to the same method, electrodes 2 to 16 are stimulated in sequence, the electrode opposite to the stimulation electrode is grounded, and voltage values ​​are measured on the remaining electrode pairs, obtaining a total of 16 sets of measurement data, each set of measurement data containing 12 voltage measurement values. After traversing and stimulating each electrode, a total of 192 voltage measurement values ​​are obtained.

[0010] Step S3, using the electrical impedance tomography method to obtain the brain full-field boundary voltage measurement value U when the brain tissue and scalp layer are dehydrated at the same time bs , where the degree of dehydration d={d1,d2,…d i}, the subscript i is used to mark the corresponding dehydration degree, the value of i is: i={1,2,···9,10}, the interval step is set to 1%, that is, d1 represents the dehydration degree of 1%, d2 represents the dehydration degree of 2%, and so on, the brain full-field boundary voltage measurement value U bs and its corresponding dehydration degree d i Construct a sample S, measure multiple groups of samples to form a training data set D, and use the training data set D to train the network. Secondly, measure the model boundary voltage value U caused by the single scalp dehydration d∈[1%,10%] s ;Measure the voltage value U at the model boundary s and its corresponding dehydration degree d i Construct a prior sequence P s ;

[0011] Step S4, constructing a fully connected neural network, which mainly includes a straightening layer, an input layer, a hidden layer and an output layer;

[0012] Step S401: In the forward propagation process, in order to match the input layer, the full-field boundary voltage measurement value of the brain is converted to By straightening the layer into The brain full field boundary voltage measurement value As input to a fully connected neural network;

[0013] In step S402, the hidden layer has 60 neurons, which are activated by the Rectified Linear Unit (ReLU) function. The ReLU function is used to generate nonlinear mapping, and its mathematical expression is:

[0014]

[0015] Where m represents the input. If m is negative or equal to 0, the neuron is not activated. Otherwise, the output is m.

[0016] There are ten neurons in the output layer, and the activation function is SoftMax, which converts multiple inputs into output data with a sum of 1. For the i-th neuron, SoftMax is described as:

[0017]

[0018] Where S i is the output of the output layer, e is the output of the neuron, j is the number of neurons, and after activation by the SoftMax function, the output of the output layer shows the probabilities of ten different classifications;

[0019] Step S403: The brain full field boundary voltage measurement value U bs The corresponding dehydration degree d i As the output of the fully connected neural network, it is set to 10 layers;

[0020] Step S5: Train the fully connected neural network. The loss function H used in the training is:

[0021]

[0022] In the formula, p(n) represents the label value, and q(n) represents the network output value;

[0023] Step S6: Use the Adam (Adaptive Momentum Estimation) optimizer for training, with the learning rate and regularization parameters set to 0.0001 and 0.000001 respectively;

[0024] Step S7, obtaining the actual dehydration boundary voltage measurement value U of different cerebral ischemia patients or the same cerebral ischemia patient at different time periods according to the above EIT measurement method. real , the actual dehydration boundary voltage measurement value U real Input into the trained fully connected neural network, and obtain the dehydration degree d at this time through forward propagation i;

[0025] Step S8, in the prior sequence P s Find the corresponding dehydration degree d i The measured voltage value U at the model boundary s , get the compensated voltage u', that is, u'=U real -U s ;

[0026] Step S9: The electrical tomography problem is considered as an inverse problem u'≈A·g, where A is the sensitivity matrix and g is the conductivity change. Based on the L1 regularization method, the conductivity distribution is estimated as Where λ is the regularization parameter used to balance the fidelity term The weight between ||g||1 and the penalty term;

[0027] Step S10, using the alternating direction multiplier method to solve the optimal conductivity distribution of step S9, fusing the obtained optimal conductivity distribution with the position information of the image pixels to reconstruct a conductivity distribution image of the compensated cerebral ischemia;

[0028] In step S11 , the compensated image reconstruction result is binarized using an adaptive threshold filtering algorithm, where values ​​above the threshold are set to 1 and values ​​below the threshold are set to 0.

[0029] Compared with existing technologies, the present invention offers the following advantages and benefits: It proposes, for the first time, a data compensation method for reconstructing the conductivity distribution of cerebral ischemia. This method determines the degree of dehydration by feeding the measured voltage values ​​at the full-field boundary of the brain into a trained fully connected neural network. The dehydration level is then used to find the corresponding model boundary measured voltage value caused by scalp dehydration alone in a priori sequences. Compensation is then performed, and the pixel coordinate information is combined to reconstruct the intracranial image. Finally, the reconstructed image is binarized using an adaptive threshold filtering algorithm. This method can effectively improve the spatial resolution of reconstructed images of intracranial hemorrhage, thereby improving brain imaging quality. It has great potential for application in functional brain medical imaging. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 A flowchart of a data compensation method for cerebral ischemia conductivity distribution reconstruction provided by the present invention;

[0031] Figure 2 The pattern of the measured field, electrode distribution, excitation current and measurement voltage of a single cross-section of the 2D elliptical rabbit head model of the present invention;

[0032] Figure 3 It is the structural diagram of the fully connected neural network;

[0033] Figure 4Schematic diagram of image reconstruction results of three models under noise-free conditions and without compensation and the method proposed by the present invention;

[0034] Figure 5 Correlation coefficient (CC) and mean square error (RMSE) of the reconstruction results of the three models are respectively without compensation and the reconstruction results of the present method. DETAILED DESCRIPTION

[0035] The present invention provides a data compensation method for cerebral ischemia conductivity distribution reconstruction, which is described in detail with reference to the accompanying drawings and embodiments.

[0036] The data compensation method for reconstructing the conductivity distribution of cerebral ischemia described in the present invention aims to improve the quality of cerebral ischemia image reconstruction when the scalp and brain tissue are dehydrated at the same time. It addresses the problems of reconstructed image artifacts and the inability to present the target object. By using the voltage after data compensation for image reconstruction, the spatial resolution of the reconstruction of intracranial cerebral ischemia images is improved, effectively improving the quality of the reconstructed images.

[0037] like Figure 1 FIG. 1 is a flowchart of a data compensation method for cerebral ischemia conductivity distribution reconstruction provided by the present invention, and the specific steps are as follows:

[0038] Step S1: Determine the shape and structure of the brain, construct a standard 2D elliptical brain model on a computer, and integrate the conductivity information of different brain tissues into the different layers of the tissue structure. The model consists of three layers: scalp, skull, and brain tissue. The conductivity of the scalp layer, skull layer, and brain tissue layer is set to 0.44 S / m, 0.012 S / m, and 0.163 S / m, respectively, to simulate brain edema.

[0039] In step S2, a 16-electrode electrical impedance tomography system is used to place electrode 1 at the most anterior point of the brain model. Then, 16 electrodes are evenly spaced and placed counterclockwise on the scalp. In the mode of relative current excitation and adjacent voltage measurement, a safe current excitation is first applied to electrode 1, while electrode 9 is grounded. Voltage measurements are taken for a total of 12 electrode pairs: 2-3, 3...6-7, 7-8; 10-11, 11-12...14-15, 15-16, obtaining a total of 12 voltage measurement values ​​as the first set of measurement data. Similarly, according to the same method, electrodes 2 to 16 are stimulated in sequence, the electrode opposite to the excitation electrode is grounded, and the voltage values ​​are measured on the remaining electrode pairs, obtaining a total of 16 sets of measurement data. Each set of measurement data contains 12 voltage measurement values. After traversing and stimulating each electrode, a total of 192 voltage measurement values ​​are obtained.

[0040] Step S3, using the electrical impedance tomography method to obtain the brain full-field boundary voltage measurement value U when the brain tissue and scalp layer are dehydrated at the same timebs , where the degree of dehydration d={d1,d2,…d i}, the subscript i is used to mark the corresponding degree of dehydration. The value of i is: i={1,2,···9,10}, and the interval step is set to 1%, that is, d1 represents the degree of dehydration of 1%, d2 represents the degree of dehydration of 2%, and so on. The measured value of the full-field boundary voltage of the brain U bs and its corresponding dehydration degree d i Construct a sample S, measure multiple groups of samples to form a training data set D. Use the training data set D to train the network. Secondly, measure the model boundary voltage value U caused by the single scalp dehydration d∈[1%,10%] s ;Measure the voltage value U at the model boundary s and its corresponding dehydration degree d i Construct a prior sequence P s .

[0041] Step S4, constructing a fully connected neural network, which mainly includes a straightening layer, an input layer, a hidden layer and an output layer.

[0042] Step S401: In the forward propagation process, in order to match the input layer, the full-field boundary voltage measurement value of the brain is converted to By straightening the layer into The brain full field boundary voltage measurement value As the input of the fully connected neural network.

[0043] In step S402, the hidden layer has 60 neurons, which are activated by the ReLU function. The ReLU function is used to generate nonlinear mapping, and its mathematical expression is:

[0044]

[0045] Where m represents the input. If m is negative or equal to 0, the neuron is not activated. Otherwise, the output is m;

[0046] There are ten neurons in the output layer, and the activation function is SoftMax, which converts multiple inputs into output data with a sum of 1. For the i-th neuron, SoftMax is described as:

[0047]

[0048] Where S i is the output of the output layer, e is the output of the neuron, and j is the number of neurons. After activation by the SoftMax function, the output of the output layer shows the probabilities of ten different classifications;

[0049] Step S403: The brain full field boundary voltage measurement value Ubs The corresponding dehydration degree d i As the output of the fully connected neural network, it is set to 10 layers;

[0050] Step S5: Train the fully connected neural network. The loss function H used in the training is:

[0051]

[0052] In the formula, p(n) represents the label value, and q(n) represents the network output value;

[0053] Step S6: Use the Adam (Adaptive Momentum Estimation) optimizer for training, with the learning rate and regularization parameters set to 0.0001 and 0.000001 respectively;

[0054] Step S7, obtaining the actual dehydration boundary voltage measurement value U of different cerebral ischemia patients or the same cerebral ischemia patient at different time periods according to the above EIT measurement method. real , the actual dehydration boundary voltage measurement value U real Input into the trained fully connected neural network, and after forward propagation, the dehydration degree d at this time can be obtained i ;

[0055] Step S8, in the prior sequence P s Find the corresponding dehydration degree d i The measured voltage value U at the model boundary s , get the compensated voltage u', that is, u'=U real -U s ;

[0056] Step S9: The electrical tomography problem is considered as an inverse problem u'≈A·g, where A is the sensitivity matrix and g is the conductivity change. Based on the L1 regularization method, the conductivity distribution is estimated as Where λ is the regularization parameter used to balance the fidelity term The weight between ||g||1 and the penalty term;

[0057] Step S10, using the alternating direction multiplication method to solve the optimal conductivity distribution of step S9. The obtained optimal conductivity distribution is fused with the position information of the image pixels to reconstruct the conductivity distribution image of the compensated cerebral ischemia;

[0058] In step S11 , the compensated image reconstruction result is binarized using an adaptive threshold filtering algorithm, where values ​​above the threshold are set to 1 and values ​​below the threshold are set to 0.

[0059] like Figure 2The figure shows a single cross-section of the EIT rabbit head model with 2D elliptical 16 electrodes, using a relative excitation and adjacent measurement mode, with the 16 electrodes evenly distributed outside the field.

[0060] like Figure 3 As shown in the figure, the constructed fully connected neural network diagram includes a straightening layer, an input layer, a hidden layer, and an output layer.

[0061] Take the case where the target is located at three different positions in the brain tissue layer as an example, the actual distribution of the target in the field is as follows Figure 4 As shown in the first row, this example uses COMSOL Multiphysics 5.4 and MATLAB R2016a for simulation modeling. The simulation parameters are set based on the parameters of real rabbit biological tissue. The brain background conductivity is set to the conductivity of rabbit cerebrospinal fluid, 0.149 S / m, and the conductivity of the ischemic target inclusions is set to 0.06 S / m. The first row shows the reconstructed image obtained without compensation. It can be seen that the background is unclear, a large number of electrode artifacts appear at the boundary, and the size and position of the target object can hardly be reconstructed. In contrast, the second row shows the reconstructed image obtained using the method proposed in this invention. For the three models A, B, and C, the target object can be well reconstructed, and there are almost no artifacts in the background. The results show that this method can effectively improve the quality of intracranial cerebral ischemia image reconstruction and can improve the resolution of image reconstruction, which provides important guidance for timely clinical diagnosis of the disease.

[0062] like Figure 5 Figure 2 shows the correlation coefficient (CC) and root mean square error (RMSE) of the reconstruction results of the four uncompensated models and the proposed method. (a) shows the CC values ​​for the four models, and (b) shows the RMSE values ​​for the four models. CC is expressed as follows: a larger correlation coefficient indicates better reconstructed image quality.

[0063]

[0064] Where g c is the calculated conductivity, g a represents the actual conductivity, and Represent the conductivity value of the e-th element, and Represent g c With g a The average value of .

[0065] The RMSE expression is shown in the following formula. The smaller the root mean square error value of the reconstructed image, the better the quality of the reconstructed image.

[0066]

[0067] Where N=7271 is the effective pixel points for reconstruction, and Δg i are the i-th pixel where the predicted and true conductivity distributions change in the target region, respectively.

[0068] It can be seen that the correlation coefficient of the image reconstructed by the proposed method is much greater than that of the image reconstructed before compensation, which further confirms the superiority of the proposed method and its ability to effectively improve the quality of the reconstructed image. Even at low levels of dehydration, it can still show good performance.

[0069] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A data compensation method for reconstructing the conductivity distribution of cerebral ischemia, characterized in that: The specific steps are: Step S1: Determine the shape and structure of the brain, construct a standard 2D elliptical brain model on a computer, and integrate the conductivity information of different brain tissues into the tissue structure of different layers. The model has a three-layer structure of scalp, skull, and brain tissue. The conductivity of the scalp layer, skull layer, and brain tissue layer is set to 0.44S / m, 0.012S / m, and 0.163S / m, respectively, to simulate brain edema. Step S2: Using a 16-electrode electrical impedance tomography system, place electrode No. 1 at the most anterior point of the brain model, and then place the 16 electrodes counterclockwise around the scalp at equal intervals. Under the mode of relative current excitation and adjacent voltage measurement, first apply a safe current excitation to electrode No. 1, while grounding electrode No.

9. Then, place electrodes No. 2-3, No. 3-4, No. 4-5, No. 5-6, No. 6-7, No. 7-8, No. 10-11, No. 11-12, and No. 12-13. The voltage values ​​are measured at a total of 12 electrode pairs, namely, No. 1, No. 13-14, No. 14-15, and No. 15-16, and a total of 12 voltage measurement values ​​are obtained as the first set of measurement data. Similarly, according to the same method, electrodes 2 to 16 are stimulated in sequence, the electrode opposite to the stimulated electrode is grounded, and the voltage values ​​are measured on the remaining electrode pairs, and a total of 16 sets of measurement data are obtained. Each set of measurement data contains 12 voltage measurement values. After traversing and stimulating each electrode, a total of 192 voltage measurement values ​​are obtained; Step S3: First, the measurement value U of the full-field boundary voltage of the brain when the brain tissue layer and the scalp layer are simultaneously dehydrated is obtained by the measurement method of the electrical impedance tomography system. bs , where the degree of dehydration d={d1,d2,…d i }, the subscript i is used to mark the corresponding dehydration degree, the value of i is: i={1,2,···9,10}, the interval step is set to 1%, that is, d1 represents the dehydration degree of 1%, d2 represents the dehydration degree of 2%, and so on, the brain full-field boundary voltage measurement value U bs and its corresponding dehydration degree d i Construct a sample S, measure multiple groups of samples to form a training data set D, and use the training data set D to train the network. Secondly, measure the model boundary voltage value U caused by the single scalp dehydration d'∈[1%,10%] s ;Measure the voltage value U at the model boundary s and its corresponding dehydration degree d i 'Constitute a priori sequence P s ; Step S4, constructing a fully connected neural network, which includes a straightening layer, an input layer, a hidden layer and an output layer; Step S401: In the forward propagation process, in order to match the input layer, the full-field boundary voltage measurement value of the brain is converted to By straightening the layer into The brain full field boundary voltage measurement value As input to a fully connected neural network; In step S402, the hidden layer has 60 neurons, which are activated by the ReLU function. The ReLU function is used to generate nonlinear mapping, and its mathematical expression is: Where m represents the input. If m is negative or equal to 0, the neuron is not activated. Otherwise, the output is m. There are ten neurons in the output layer, and the activation function is SoftMax, which converts multiple inputs into output data with a sum of 1. For the t-th neuron, SoftMax is described as: Where S t is the output of the output layer, e t is the output of the neuron, j is the number of neurons, and after activation by the SoftMax function, the output of the output layer shows the probabilities of ten different classifications; Step S403: The brain full field boundary voltage measurement value U bs The corresponding dehydration degree d i As the output of the fully connected neural network, it is set to 10 layers; Step S5: Train the fully connected neural network. The loss function H used in the training is: In the formula, p(n) represents the label value, and q(n) represents the network output value; Step S6: Use the Adam (Adaptive Momentum Estimation) optimizer for training, with the learning rate and regularization parameters set to 0.0001 and 0.000001 respectively; Step S7, obtaining the actual dehydration boundary voltage measurement value U of different cerebral ischemia patients or the same cerebral ischemia patient at different time periods according to the measurement method of the electrical impedance tomography system. real , the actual dehydration boundary voltage measurement value U real Input into the trained fully connected neural network, and obtain the dehydration degree d at this time through forward propagation i ; Step S8, in the prior sequence P s Find the corresponding dehydration degree d i 'Model boundary measured voltage value U s , get the compensated voltage u', that is, u'=U real -U s ; Step S9: The electrical tomography problem is considered as an inverse problem u'≈A·g, where A is the sensitivity matrix and g is the conductivity change. Based on the L1 regularization method, the conductivity distribution is estimated as Where λ is the regularization parameter used to balance the fidelity term The weight between ||g||1 and the penalty term; Step S10, using the alternating direction multiplier method to solve the optimal conductivity distribution of step S9, fusing the obtained optimal conductivity distribution with the position information of the image pixels to reconstruct a conductivity distribution image of the compensated cerebral ischemia; In step S11 , the compensated image reconstruction result is binarized using an adaptive threshold filtering algorithm, where values ​​above the threshold are set to 1 and values ​​below the threshold are set to 0.

Images