An electrical impedance imaging method, medium and apparatus for muscle dynamic imaging

Abstract

The application discloses a method, medium and device for electrical impedance imaging of muscle dynamics, and belongs to the technical field of image processing. The method comprises the following steps: collecting a muscle dynamic boundary voltage data sequence of a target in a walking process; inputting the muscle dynamic boundary voltage data sequence into a trained neural network model to reconstruct a corresponding muscle internal conductivity dynamic image sequence, and then obtaining an electrical impedance imaging result. The neural network model extracts spatial features of the muscle dynamic boundary voltage data sequence by using a convolution layer, and predicts a time evolution mode of a conductivity distribution caused by muscle activity by using a recurrent neural network layer. The application can effectively capture complex time sequence dynamics of muscle activity, and improves the speed and time resolution of electrical impedance imaging.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and more specifically, to an electrical impedance imaging method, medium, and device for dynamic muscle imaging. Background Technology

[0002] Electrical Impedance Tomography (EIT) is a novel imaging technique that infers the conductivity, dielectric constant, and impedance of a part of the body from surface electrode measurements, and uses this information to create a tomographic image of that area. It offers advantages such as non-invasiveness, repeatability, low cost, and functional imaging, and is widely used in biomedicine, agricultural geology, and industrial inspection. For example, in the medical field, EIT can be used to monitor the functional state of organs such as the lungs and heart, as well as to detect changes in the morphology and properties of muscle tissue. EIT's high temporal resolution allows for continuous monitoring imaging to capture the dynamic changes in muscle activity. This is invaluable for studying the responses and states of muscles under different exercise conditions. By monitoring changes in muscle impedance in real time, the temporal sequence and patterns of muscle activity can be analyzed in detail, revealing the dynamic characteristics of muscle contraction and relaxation.

[0003] Electrical impedance tomography (EIT) reconstructs an image of the body's conductivity or electrical impedance distribution by applying small alternating currents (typically in the frequency range of a few kHz to hundreds of kHz) to the body surface and then measuring the voltage differences between multiple locations. The basic principles of EIT include: 1) Current injection: Current is injected into the body through one or more selected pairs of electrodes; these currents are safe, small-amplitude alternating currents. 2) Voltage measurement: As the current passes through different tissues, different voltage drops are created due to the different conductivity of these tissues (e.g., muscle, fat, blood, and air have different conductivity levels); other electrodes are used to measure these voltage changes. 3) Data acquisition: In a complete scan, current is injected through different electrode combinations, and the corresponding voltage responses are recorded. This process generates a large number of voltage-current data points, called boundary measurement data. 4) Image reconstruction: The boundary measurement data is converted into an estimate of the internal conductivity distribution using a mathematical model. This step involves complex calculations because it requires solving the so-called "inverse problem."

[0004] The inverse problem refers to inferring the conductivity distribution (system characteristics) inside an object from a known voltage measurement result (output) and applied current pattern (input). This is a much more difficult problem, mainly because multiple different conductivity distributions may produce the same surface voltage measurement result, even small measurement noise can lead to huge fluctuations in the solution (i.e., solution instability), and the nonlinear relationship between conductivity and voltage increases the complexity of the solution. To solve the inverse problem, researchers have developed various reconstruction algorithms, such as iterative reconstruction algorithms. These algorithms are usually based on physical models (such as finite element models FEM or boundary element models BEM) to describe the conduction law of current inside the measured object (forward problem model). Then, through iterative optimization, the guessed value of the conductivity distribution inside the model is continuously adjusted to minimize the difference between the boundary voltage calculated based on the guessed value (forward problem solution) and the actual measured boundary voltage. Commonly used iterative algorithms include linearization methods and nonlinear iterative optimization methods. These iterative algorithms are theoretically mature and can obtain reconstructed images with high spatial resolution under certain conditions. However, they have significant limitations when applied to muscle dynamic imaging, mainly because:

[0005] 1) High computational complexity and long processing time. Each iteration requires solving one or more large systems of linear equations (derived from forward problems or Jacobian matrix calculations), resulting in a huge computational burden. For dynamic imaging requiring high temporal resolution (such as capturing millisecond-level muscle contractions), the reconstruction speed of these algorithms often cannot keep up (single-frame reconstruction may take seconds or even minutes), making it impossible to achieve real-time or near real-time monitoring.

[0006] 2) Sensitive to initial guesses and prior information. The convergence and final result of the iterative algorithm may depend on the guessed value of the initial conductivity distribution and the selected prior model or regularization parameter. Inappropriate selection may lead to slow convergence or getting trapped in local optima.

[0007] 3) High accuracy requirements for the model. The algorithm's performance depends on the accuracy of the forward problem model. For complex, heterogeneous, and dynamically changing objects like the human body, establishing accurate geometric and physical models is a challenge in itself.

[0008] To overcome the shortcomings of traditional iterative algorithms, especially in dynamic imaging and real-time requirements, researchers have explored new methods. For example, patent application CN111938641A proposes an optimized adaptive extended Kalman filter (EKF) bioelectrical impedance imaging method. This approach aims to address the problems of low noise resistance, potential accuracy loss, and filter divergence that may exist in traditional EKF (extended Kalman filter) applications in EIT, and attempts to balance reconstruction accuracy and computational cost. However, although EKF is designed for dynamic systems, it still requires matrix operations (such as Jacobian matrix, covariance update, and Kalman gain calculation), which limits its maximum achievable temporal resolution and may be a bottleneck for capturing rapid muscle activity (e.g., millisecond-level). Furthermore, EKF relies on simplified state transition models (e.g., random walks) and local linearization assumptions, making it difficult to accurately capture the complex, nonlinear, and patterned dynamic changes of muscles during activities such as gait cycles. Another example is patent application CN119055216A, which proposes an upper arm muscle strength estimation method based on a two-branch self-attention mechanism network. This scheme estimates upper arm muscle strength more accurately by fusing EIT image information and raw voltage frequency domain information, and utilizing a self-attention mechanism. However, this scheme relies on pre-reconstructed EIT images using traditional algorithms (such as Gauss-Newton), which becomes a speed bottleneck in its overall process, and it is not a fully end-to-end dynamic imaging solution. Furthermore, although the attention mechanism can handle sequences, it is not effective in simulating the naturalness of temporal evolution and state memory for muscle activities with clear sequence and stage characteristics, such as gait cycles.

[0009] Analysis reveals the following main shortcomings in existing technologies:

[0010] 1) The contradiction between imaging speed and temporal resolution. During movement (e.g., walking, running, or rehabilitation training), muscles undergo rapid contraction and relaxation, causing significant changes in their internal conductivity distribution on a timescale of milliseconds to seconds. To accurately capture these transient changes, EIT imaging algorithms must have sufficiently high temporal resolution, enabling the reconstruction of a single frame within a short time. However, traditional EIT reconstruction algorithms, especially iterative optimization-based numerical methods (such as the Gauss-Newton method and the conjugate gradient method), while achieving relatively high image quality, typically involve repeatedly solving complex forward problems (calculating boundary voltages from a given conductivity distribution) and inverse problems (inverting the internal conductivity distribution from measured boundary voltages). This iterative process is computationally intensive, resulting in long reconstruction times for a single frame, often on the order of seconds or even minutes. Such long computational delays are far from meeting the needs of real-time or near-real-time monitoring of rapid muscle dynamics, causing the reconstructed image sequence to fail to accurately reflect the true temporal progression of muscle activity and lose crucial dynamic information.

[0011] 2) Training data lacks physiological relevance and dynamic realism. Most current neural network imaging algorithms used for EIT suffer from a common limitation during the training phase: they typically rely on simulation data generated from simplified, regular geometric shapes (such as circles, rectangles, or triangles). These general-purpose models often fail to accurately simulate the complex anatomical structures of human tissues, especially in irregularly shaped areas like the thigh, which contains multiple tissues (muscle groups, fat, and bone). More importantly, these models are usually static and cannot reflect the complex dynamic changes that muscles undergo during physiological activities (such as different phases of the gait cycle), such as muscle expansion and morphological changes (bulging) due to contraction, or thinning and lengthening due to stretching. These morphological changes in muscles directly affect their conductivity distribution and current pathways, and are a significant source of EIT signal variations. Using training data generated from such an oversimplified model that lacks dynamics and physiological relevance to train neural networks results in imaging algorithms that, while fast, suffer from significantly reduced accuracy in reconstructed images, inability to distinguish specific muscle structures (such as the quadriceps and hamstrings), and inability to reflect real dynamic muscle behavior (such as contraction / relaxation) when applied to real, dynamic human muscle imaging tasks. This makes it difficult to meet the needs of refined functional assessment.

[0012] 3) Insufficient capture of the temporal dependencies of muscle dynamic processes. Traditional iterative algorithms usually reconstruct independently frame by frame, or use simple regularization constraints on adjacent frames, which makes it difficult to effectively utilize the strong temporal continuity and state dependency information inherent in muscle activity (such as gait cycles). Summary of the Invention

[0013] The purpose of this invention is to overcome the shortcomings of the prior art and provide an electrical impedance imaging method, medium and device for muscle dynamic imaging that can meet the high temporal resolution requirements of muscle dynamic imaging and accurately reconstruct the complex spatiotemporal conductivity changes of muscles during real movement.

[0014] According to a first aspect of the present invention, a method for electrical impedance tomography (EIT) for dynamic muscle imaging is provided. The method includes the following steps:

[0015] For the target, collect dynamic boundary voltage data sequences of muscles during walking;

[0016] The muscle dynamic boundary voltage data sequence is input into a trained neural network model to reconstruct the corresponding dynamic image sequence of muscle internal conductivity, thereby obtaining electrical impedance imaging results.

[0017] The neural network model uses convolutional layers to extract the spatial features of the muscle dynamic boundary voltage data sequence and recurrent neural network layers to predict the temporal evolution pattern of conductivity distribution caused by muscle activity.

[0018] In one embodiment, the training set for training the neural network model is obtained according to the following steps:

[0019] Based on the dynamic changes of the thigh muscles during human walking, the gait cycle is divided into resting state, early support phase, late support phase and swing phase. The resting state is characterized by the relaxed balance of the quadriceps and hamstring muscles. The early support phase refers to the heel strike to the middle stage. The late support phase refers to the toes leaving the ground. The swing phase refers to the foot leaving the ground and touching the ground again.

[0020] Using the resting state as the baseline model, multiple sets of models are obtained by constructing a finite element model. The finite element model consists of multiple finite element meshes, with the skeleton as the geometric center. The quadriceps femoris on the front and the hamstrings on the back evolve dynamically in a crescent shape. Each set of models is divided into four groups, representing the resting state, the early support phase, the late support phase, and the oscillating phase, respectively.

[0021] For the dynamic phases of the early support phase, the late support phase, and the swing phase, the area of ​​the quadriceps and hamstring muscles is adjusted parametrically to simulate the continuous process of muscle morphology evolving with gait, and the electrical conductivity values ​​of each tissue fluctuate randomly within a set normal physiological range.

[0022] The surface voltage distribution of the multiple models under the electrode excitation mode is calculated by finite element simulation, thereby obtaining a training set containing multiple sets of data. Each set of data corresponds to the conductivity and voltage pairing data of the four gait cycles of the resting state, the early stage of the support phase, the late stage of the support phase, and the oscillating phase.

[0023] In one embodiment, the neural network model includes multiple convolutional layers, recurrent neural network layers, and fully connected layers. Each convolutional layer is followed by a nonlinear activation function and a pooling layer. The convolutional layers generate convolutional features by performing convolution operations on the input data and convolutional kernels. The pooling layers generate a dimensionality reduction matrix by extracting set values ​​from the convolutional features. The recurrent neural network layer captures the dynamic correlation between the four gait cycles—the resting state, the early support phase, the late support phase, and the oscillating phase—through a memory unit mechanism.

[0024] In one embodiment, the recurrent neural network layer is a long short-term memory network layer.

[0025] According to a second aspect of the present invention, a computer-readable storage medium is provided having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the above-described electrical impedance imaging method for dynamic muscle imaging.

[0026] According to a third aspect of the present invention, a computer device is provided, including a memory and a processor, wherein a computer program capable of running on the processor is stored in the memory, wherein the processor executes the computer program to implement the steps of the above-described electrical impedance tomography method for muscle dynamic imaging.

[0027] Compared with existing technologies, the advantages of this invention are that, addressing the problems of slow imaging speed, insufficient capture of temporal dependencies, and lack of physiological correlation between models and training data in existing technologies, this invention provides a novel electrical impedance tomography (EIT) method for muscle dynamic imaging. This method achieves high-speed, near real-time muscle dynamic imaging, overcoming the computational speed bottlenecks of traditional iterative algorithms and filtering-based methods. Leveraging the advantages of end-to-end mapping in deep learning, it significantly shortens reconstruction time, meeting the high temporal resolution requirements for monitoring rapid muscle dynamic changes. It effectively captures the complex temporal dynamics of muscle activity by introducing specialized neural network structures (such as recurrent neural networks, especially long short-term memory networks, LSTM) for processing sequential data. These structures enable the invention to learn and utilize the continuous evolution and long-range dependencies of muscle conductivity distribution over time (e.g., the correlation and transition patterns of muscle states at different stages of the gait cycle). Improving the physiological relevance and accuracy of imaging results: A dynamic simulation model based on physiological principles is constructed to simulate the morphological and electrical conductivity changes of target muscle groups (such as the quadriceps and hamstrings) during specific dynamic activities (such as the gait cycle of walking). This model is then used to generate a large amount of training data with high physiological relevance and dynamic realism to train the neural network. This enables the trained algorithm to more accurately reconstruct the spatiotemporal distribution of muscles during real movement, improving its ability to distinguish different tissue structures and reflect muscle functional states (contraction and relaxation).

[0028] Other features and advantages of the invention will become clear from the following detailed description of exemplary embodiments of the invention with reference to the accompanying drawings. Attached Figure Description

[0029] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments of the invention and, together with their description, serve to explain the principles of the invention.

[0030] Figure 1 This is a flowchart of an electrical impedance imaging method for dynamic muscle imaging according to an embodiment of the present invention;

[0031] Figure 2 This is a diagram of an RNN-LSTM (Recurrent Neural Network-Long Short-Term Memory) neural network model architecture according to an embodiment of the present invention;

[0032] Figure 3 This is a schematic diagram illustrating the process of constructing a training set according to an embodiment of the present invention;

[0033] Figure 4 This is a schematic diagram illustrating the process of training an RNN-LSTM neural network model according to an embodiment of the present invention. Detailed Implementation

[0034] Various exemplary embodiments of the present invention will now be described in detail with reference to the accompanying drawings. It should be noted that, unless otherwise specifically stated, the relative arrangement, numerical expressions, and values ​​of the components and steps set forth in these embodiments do not limit the scope of the invention.

[0035] The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit the invention or its application or use.

[0036] Techniques, methods, and equipment known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and equipment should be considered part of the specification.

[0037] In all the examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limitations. Therefore, other examples of exemplary embodiments may have different values.

[0038] It should be noted that similar labels and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be discussed further in subsequent figures.

[0039] To address the issues of low temporal resolution and insufficient physiological relevance in existing electrical impedance tomography (EIT) techniques for muscle dynamic imaging, this invention proposes a deep learning-based EIT muscle dynamic imaging method. An RNN-LSTM neural network model is designed, comprising convolutional layers and LSTM (Long Short-Term Memory) layers. The convolutional layers (and pooling layers) process the input boundary voltage data (e.g., data within each time step or short time window), extracting spatial features of the voltage distribution, effectively filtering information, and capturing patterns related to the spatial structure of the conductivity distribution across muscle, fat, and bone tissue sections at the current moment. The LSTM layers process the temporally ordered feature sequences extracted by the convolutional parts. Utilizing the unique gating mechanisms of LSTM (such as forget gates, input gates, and output gates) and memory cells, this model can effectively learn and memorize the dynamic evolution and long-term dependencies of muscle conductivity distribution over time. For example, it can capture the continuous changes in muscle contraction and relaxation states during different phases (such as the stance and swing phases) of the human walking gait cycle, overcoming the gradient problem that may exist in standard RNNs.

[0040] Specifically, see Figure 1 As shown, the provided electrical impedance tomography method for muscle dynamic imaging includes the following steps:

[0041] Step S110: Based on the temporal characteristics of the continuous change of muscle state over time during walking, design an RNN-LSTM neural network model.

[0042] Considering that when applying EIT to muscle dynamic imaging, since continuous surface voltage signals have the temporal characteristics of reflecting the continuous changes in muscle state over time, it is necessary to rely on a network model that can process serialized continuous data. Therefore, a neural network model RNN-LSTM based on Recurrent Neural Network (RNN) and Long Short-Term Memory (LSTM) units was designed. This model has advantages in time series modeling, and is especially suitable for processing dynamic test data generated during human walking.

[0043] Recurrent Neural Networks (RNNs) are artificial neural networks that maintain a "memory" of historical inputs through recurrent connections. Their core advantage lies in processing sequential data. Unlike feedforward neural networks, RNNs process inputs step-by-step over time, ensuring that the output at each time step depends not only on the current input but also on the historical hidden states. This characteristic makes them particularly suitable for time-dependent tasks, such as measuring continuous voltage signals generated by muscle activity during walking.

[0044] Sherstinsky derived the mathematical foundation of RNNs from the perspective of differential equations, modeling them as nonlinear first-order nonhomogeneous ordinary differential equations:

[0045] (1)

[0046] in, For state vectors, For the input vector, It is a nonlinear vector function. Let be the bias vector, and t be a continuous-time variable. After discretization using the backward Euler method, the time-domain expression of the RNN is:

[0047] (2)

[0048] (3)

[0049] in, , , These are the weight matrices for the state loop, output feedback, and input mapping, respectively. G represents the state bias, and G(·) represents the activation function. This represents the output / activation vector of the previous time step. It is the state vector of the previous time step. This represents the state vector at the current time step. This represents the output / activation vector of the previous time step. Let represent the current input vector at time step , and n represent the time step. This recursive structure enables RNNs to capture temporal patterns, making them suitable for modeling the continuous evolution of muscle EIT data.

[0050] However, standard RNNs face the vanishing / exploding gradient problem during training. During backpropagation, gradients over long sequences can decay or diverge exponentially, making it difficult to learn long-term dependencies. This is particularly critical for muscle dynamics imaging, as impedance signals may exhibit temporal correlations across multiple gait cycles, thus requiring a more stable model architecture.

[0051] To overcome the limitations of standard RNNs, researchers have proposed the LSTM unit, which achieves long-term dependency modeling by introducing memory cells and three types of gating mechanisms (forget gate, input gate, and output gate). Sherstinsky further theoretically proved that the constant error carousel (CEC) of LSTM can effectively alleviate the gradient vanishing problem. The core equation of LSTM is as follows:

[0052] The forget gate controls the percentage of historical state retention:

[0053] (4)

[0054] Where n is the time step. It is the forgetting gate vector (controls the proportion of the previous cell state that is forgotten, 0 = all forgotten, 1 = all retained). It is the weight matrix input to the forget gate. It is the current input vector. It is the weight matrix from the previous cell state to the forget gate. It is the weight matrix from the previous hidden state to the forget gate. It is the vector of the previous hidden state. It is the forget gate bias vector.

[0055] The input gate regulates the update of the state by new information:

[0056]

[0057] in, It is the input gate vector (which controls the update ratio of candidate values). : The weight matrix input to the input gate. It is the weight matrix from the previous cell state to the input gate. It is the weight matrix from the previous hidden state to the input gate. It is the input gate bias vector.

[0058] State update refers to merging historical state with new input:

[0059] (7)

[0060] in, It is the current cell state vector (core memory unit). It involves forgetting old information (regulated by the forget gate). This indicates the addition of new information (adjusted via the input gate).

[0061] Output gate control observable output:

[0062] (8)

[0063] (9)

[0064] in, This represents the output gate vector (which controls the output ratio of cell state). This represents the weight matrix from the input to the output gate. This represents the weight matrix from the current cell state to the output gate. This represents the weight matrix from the previous hidden state to the output gate. This represents the output gate bias vector. is the sigmoid activation function, and ⊙ represents element-wise multiplication. The gating mechanism of LSTM enables it to simultaneously capture short-term fluctuations and long-term trends in muscle impedance data, providing crucial support for dynamic imaging.

[0065] See Figure 2 As shown, the designed RNN-LSTM neural network model includes multiple convolutional layers, recurrent neural network layers (e.g., LSTM layers), and fully connected layers. Each convolutional layer is followed by a non-linear activation function and a pooling layer. The number of convolutional layers, kernel size, etc., can be set according to actual needs.

[0066] Besides the LSTM layer, convolutional and pooling layers are essentially specially designed data preprocessing layers. Their core function is to filter the input data and extract effective information. Convolutional layers generate new feature values ​​by performing convolution operations on the original input data and the convolution kernel. Since this technique was initially designed to extract features from image datasets, the input data needs to be presented in a structured matrix form. The convolution kernel (i.e., the filter) can be viewed as a miniature window relative to the input matrix, with its coefficients organized in a matrix. This window slides across the input matrix, performing convolution operations on each sub-region it passes through. All the results form a convolution matrix, whose feature values ​​are determined by the filter coefficients and their dimensionality. By applying different convolution kernels to the input data, multiple convolutional features with higher representational power can be generated, which are superior to the original input features, thus effectively improving model performance. The convolutional layers are followed by non-linear activation functions and pooling layers. Pooling layers, as a downsampling method, generate a dimensionality-reduced matrix by extracting specific values ​​from the convolutional features. Its operation is similar to that of a convolutional layer. It uses a small sliding window to take the value of each convolutional feature block as input, and generates a single new value as output through defined operations (such as max pooling or average pooling to calculate the maximum / average value of each feature block). Ultimately, the pooling layer produces new matrices, which can be viewed as compressed versions of the output features of the convolutional layer. The advantage of pooling is its enhanced stability; because the output pooled values ​​are insensitive to small changes in the input, this characteristic effectively improves the model's generalization ability.

[0067] Step S120: Obtain a training set by constructing a finite element simulation model, and train an RNN-LSTM neural network model based on a set loss function to learn the mapping relationship between dynamic voltage and dynamic conductivity.

[0068] To address the lack of physiological relevance in training data in existing technologies, this invention innovatively constructs a finite element simulation model that simulates muscle changes in the human lower limb thigh during specific dynamic activities. Based on physiological knowledge, this model divides the gait cycle into key phases (e.g., resting state, early stance phase, late stance phase, and swing phase). The finite element simulation model explicitly simulates the morphological changes (e.g., increases and decreases in cross-sectional area, thickness, width, and shape changes) and possible electrical conductivity changes caused by the physiological behaviors (e.g., concentric contraction, eccentric contraction, and relaxation) of major muscle groups (e.g., quadriceps and hamstrings) during these phases. Model parameters (e.g., muscle region geometry and electrical conductivity values) are adaptively adjusted parametrically with each gait phase to reflect the dynamic contraction-stretching cycle of muscles. The model also includes other relevant tissues (e.g., bone and fat layer) and considers their relative positions and electrical properties. Furthermore, the data generated by the model includes physiological variations (e.g., individual differences in muscle size, random fluctuations in tissue conductivity within the normal range). This dynamic, physiologically relevant simulation model allows for the generation of large amounts of training data using EIDORS simulation software or other software. The generated simulation data sequences containing spatiotemporal dynamic information (voltage sequences as input and conductivity distribution sequences as target output) are then used to train the RNN-LSTM neural network model.

[0069] In one embodiment, based on the advantages of the RNN-LSTM model in capturing temporal patterns and processing serialized continuous data, the dynamic changes of thigh muscles during human walking are innovatively simplified into four states, and training data is generated by modeling based on these states, thereby obtaining a neural network imaging algorithm specifically for EIT dynamic imaging of lower limb muscle tissue.

[0070] Specifically, the gait cycle of walking comprises two main phases: the stance phase (foot touches the ground) and the swing phase (foot leaves the ground). The thigh muscles perform their function through alternating concentric contractions (muscle fiber shortening for force), eccentric contractions (muscle fiber lengthening for braking), and relaxation. The quadriceps femoris muscle group (represented by the rectus femoris and vastus lateralis) is primarily responsible for knee extension in the early stance phase to support body weight; while the hamstring muscles (such as the biceps femoris) dominate knee flexion control and lower limb forward swing braking from the end of the swing phase to the beginning of the stance phase. This functional antagonism and spatiotemporal synergy result in the two muscle groups forming opposing anterior-posterior biomechanical structures in cross-section. In a two-dimensional cross-section perpendicular to the femur, these biomechanical behaviors manifest as real-time changes in muscle thickness, width, and overall morphology. The cross-sectional area (CSA) of the muscle reflects the deformation characteristics of a specific muscle during contraction or stretching. Although the adductor muscles are involved in hip joint stability, their mechanical contribution in the two-dimensional cross-sectional model is manifested as static support with a small dynamic deformation range. Therefore, they can be classified as secondary muscle groups and simplified.

[0071] During the gait cycle, the cross-sectional area and morphological changes of the thigh muscles exhibit significant temporal characteristics. Based on a two-dimensional cross-sectional model (centered on the femur, with the anterior quadriceps crescent region and the posterior hamstring symmetrical structure), the mechanical behavior of the muscles can be divided into the following four stages:

[0072] (1) Resting state

[0073] The resting state serves as the baseline, characterized by a relaxed balance between the quadriceps and hamstring muscles. At this point, the two muscle groups form a symmetrical crescent-shaped structure around the femur, with the CSA at baseline level (without significant bulging or depression), and its cross-sectional profile is a flattened ellipse. This state provides a geometric reference for subsequent morphological changes during contraction and stretching.

[0074] (2) Early support phase (from heel strike to middle support phase)

[0075] The anterior quadriceps femoris drives knee extension through concentric contraction to support body weight. During this contraction, muscle fibers shorten along their long axis. Due to the near-incompressible nature of muscle tissue (the principle of volume conservation), its cross-section bulges outward, with thickness and width increasing simultaneously. Ultrasound studies show that the rectus femoris's center-arm span (CSA) can increase by 5-15% during isometric contraction. Dynamic contraction during walking may induce similar morphological changes: the cross-section changes from a flat crescent shape at rest to a thickened, convex shape, with the region near the medial femur remaining fixed and the lateral belly significantly expanding. Simultaneously, the posterior hamstrings control knee extension speed through eccentric contraction, and the passive elongation of muscle fibers leads to a decrease in CSA of approximately 3-8%. Studies have shown that muscle fiber elongation under tension promotes lateral extension of the muscle contour, forming a thinner, conical crescent structure.

[0076] (3) End of the support phase (toes off the ground)

[0077] As the center of gravity shifts forward, the quadriceps gradually relaxes, and the CSA (contraction subtraction anterior segment) returns to its baseline level due to passive lengthening, reverting to a flattened crescent shape, although the degree of bulging is less pronounced than in the early stance phase. At this time, the hamstrings switch to concentric contraction mode, with muscle fiber shortening driving knee flexion and providing power for the forward swing of the lower limb. During contraction, the CSA increases by approximately 8-12%, and the cross-sectional shape thickens into a rounded and compact crescent shape, with the muscle belly bulging outward from the posterior side of the femur. This phenomenon is consistent with the predictions of Blemker et al.'s computational model of muscle deformation: during concentric contraction, muscle fiber aggregation leads to local stress concentration, causing the cross-sectional profile to tend towards a more circular shape.

[0078] (4) Swing phase (from foot off the ground to touch the ground again)

[0079] During the early to middle swing phase, both hamstring muscles are in a state of low activation, with the CSA maintaining at baseline and exhibiting a symmetrical crescent-shaped structure in cross-section. Towards the end of the swing phase, the hamstrings undergo eccentric contraction to decelerate the forward swing of the lower limb, resulting in a brief 2-5% decrease in CSA. After a brief bulge, the CSA quickly returns to a flattened shape, preparing for foot contact in the next cycle. The quadriceps, on the other hand, undergo a slight eccentric activity before contact with the ground, with the CSA decreasing by 2-5% due to pre-activation and lengthening, exhibiting a slightly stretched and flattened shape.

[0080] This rapid cycle of contraction and stretching causes the CSA (Cuticle-Stasis Acuity) to oscillate continuously with the gait rhythm, and the muscle morphology dynamically transforms between "thickening and widening" and "thinning and narrowing." Based on the above gait cycle division, a dynamic EIT (Extended Intact Time) model of the human thigh during the walking cycle is constructed, for example, using the EIT-specific simulation software EIDORS to build a finite element model. The model consists of 2500 finite element meshes, with the skeleton as the geometric center. The quadriceps femoris on the front and the hamstrings on the back evolve dynamically in crescent shapes, and the morphological parameters of muscles, fat, and bones are adaptively adjusted according to the gait stage.

[0081] Specifically, an automated modeling process was developed for establishing the simulation model. Using the resting state as the baseline model, 4000 sets of models were generated in batches in EIDORS by writing customized scripts. Each set of models consisted of four parts, representing the resting state, early stance phase, late stance phase, and swing phase. For the dynamic phases (early stance phase, late stance phase, and swing phase), the area of ​​the quadriceps and hamstring muscles was parametrically adjusted to simulate the continuous process of muscle morphology evolution with gait. The cross-sectional area of ​​the muscles in each set of models exhibited a physiological difference of ±50% in the resting state. Simultaneously, the conductivity values ​​of each tissue fluctuated randomly within the normal physiological range (muscle: 0.2-0.5 S / m, fat: 0.02-0.06 S / m, bone: 0.008-0.02 S / m) to enhance data diversity. The surface voltage distribution of each model under the 16-electrode excitation mode was calculated using finite element simulation, ultimately obtaining a training set containing 4000 sets of data, each set corresponding to conductivity-voltage pairing data for the four phases. See the process of constructing the training set. Figure 3 As shown.

[0082] After obtaining the training set, the RNN-LSTM model is trained until the set loss criterion is met, such as minimizing the loss function as the training optimization objective.

[0083] Specifically, the RNN-LSTM model training process fully integrates the temporal characteristics of gait cycle data with deep learning techniques. The model uses LSTM (Long Short-Term Memory) network as its core architecture, capturing the dynamic relationships between the four gait phases (early and late support phase, swing phase, and resting state) through its unique "memory unit" mechanism. See also... Figure 4As shown, during training, data is input into the model in batches, with each batch containing complete gait sequences of 32 model groups. Each sequence contains paired voltage input and conductivity output data from four models. The voltage characteristics of the input data come from the surface voltage distribution under 16-electrode adjacent excitation and adjacent measurement modes. That is, by exciting one pair of electrodes, the remaining electrodes are measured in pairs, resulting in a total of 208 voltage data points. The output target is the conductivity distribution matrix of muscle, fat, and bone at the corresponding stage.

[0084] The RNN-LSTM model can be trained using standard deep learning training procedures, such as employing the Adam optimizer and the mean squared error (MSE) loss function. Backpropagation (including backpropagation along the LSTM time steps) adjusts the network weights, enabling the model to learn the complex nonlinear mapping from dynamic voltage input to dynamic conductivity output. Specifically, the initial learning rate is set to 0.001, and the step size is updated adaptively by adjusting parameters to balance convergence speed and stability. The mean squared error loss function is chosen to directly measure the difference between the model's predicted conductivity distribution and the simulated true value. The entire training process consists of 125 iterations. In each iteration, the model sequentially performs forward propagation, loss calculation, and backpropagation. First, the voltage data is processed by convolutional layers to extract spatial features. Then, max pooling is used for feature dimensionality reduction. The data is then organized into sequences at four time steps and input into the LSTM layer to gradually generate conductivity predictions. Subsequently, the mean squared error loss function is calculated based on the prediction results and the actual conductivity values. Finally, the error signal is propagated backward along the model topology through the backpropagation mechanism (including gradient backpropagation along the time step of the LSTM layer) to update the network model weight parameters and obtain optimized model parameters.

[0085] Step S130: For the actual acquired muscle dynamic boundary voltage data sequence, the imaging results are obtained using a trained RNN-LSTM neural network model.

[0086] After the RNN-LSTM neural network model is trained, it can be directly applied to process real-time or offline muscle dynamic boundary voltage data sequences collected from EIT hardware systems (such as wearable devices), quickly reconstructing the corresponding dynamic image sequence of muscle internal conductivity, and realizing high-speed, high-accuracy and physiologically relevant muscle dynamic visualization.

[0087] It should be noted that the training process of the neural network model involved in this invention can be performed offline on a server or in the cloud. The trained model can be embedded into an electronic device to achieve real-time muscle electrical impedance imaging. This electronic device can be a terminal device or a server. Terminal devices include any terminal device such as mobile phones, tablets, personal digital assistants (PDAs), point-of-sale (POS) terminals, in-vehicle computers, and smart wearable devices (smartwatches, virtual reality glasses, virtual reality headsets, etc.). Servers include, but are not limited to, application servers or web servers, and can be independent servers, cluster servers, or cloud servers.

[0088] To further verify the effectiveness of this invention, measurements of a human thigh were performed under static conditions using a commercial EIT device, and the data were processed using this invention. The verification results show that this invention can successfully distinguish major tissues within the thigh (such as bone, fat, quadriceps, and hamstrings), possessing basic tissue resolution capabilities, laying the foundation for subsequent application to complete dynamic imaging. This invention achieves fast, accurate, and physiologically relevant electrical impedance tomography (EIT) dynamic muscle imaging by combining a specially designed RNN-LSTM network architecture with high-quality training data generated from a physiologically based dynamic muscle activity simulation model.

[0089] In summary, compared with the prior art, the present invention has the following advantages:

[0090] 1) This invention designs a specific hybrid neural network architecture (RNN-LSTM), employing convolutional neural network layers to extract the spatial features of the EIT boundary voltage at each time point, and specifically selecting Long Short-Term Memory (LSTM) units from recurrent neural networks (RNNs) to process the temporal series of these spatial features. LSTM can effectively process sequential data, capture temporal dependencies (such as muscle state evolution within the gait cycle), and overcome the vanishing / exploding problem of standard RNNs. The RNN-LSTM hybrid architecture can effectively fuse the spatial and temporal information of voltage data, achieving end-to-end fast mapping from dynamic voltage measurement sequences to internal conductivity image sequences, significantly improving the speed of dynamic imaging.

[0091] 2) Unlike existing static or simplified models, this invention constructs a physiologically based dynamic simulation model, designing a finite element simulation model to simulate the physiological changes of specific muscle groups (such as the quadriceps femoris and hamstrings) during specific dynamic activities (such as the walking gait cycle). This simulation model divides the target dynamic activity (walking gait) into multiple key physiological stages (such as the resting state, early / late stance phase, and swing phase), and parameterizes and adaptively adjusts the geometry (such as CSA, thickness, and shape) and / or conductivity values ​​of the muscle regions in the model based on the known muscle physiological behaviors (concentric / eccentric contraction, relaxation, etc.) of each stage, thereby highly realistically simulating the dynamic changes of muscles.

[0092] 3) This invention utilizes dynamic and physiologically relevant simulation models to generate large-scale training datasets containing spatiotemporal dynamic information and physiological variability, and uses them to train the designed RNN-LSTM neural network model, which can learn the real dynamic laws of muscles and improve imaging accuracy and physiological relevance.

[0093] 4) This invention employs end-to-end deep learning mapping. After training, image reconstruction primarily involves forward propagation computation of the neural network, which is inherently faster than the EKF method in terms of computational speed. Although EKF is designed for dynamic systems, it still requires matrix operations (Jacobi, covariance update, Kalman gain calculation), limiting its maximum achievable temporal resolution. This invention, however, can more easily achieve near real-time or even real-time muscle dynamic imaging, better meeting the need to capture rapid muscle activity. This speed advantage mainly stems from the end-to-end deep learning architecture itself and its design specifically for dynamic imaging. Furthermore, the architecture of this invention can be independent of accuracy selection and can also apply hybrid accuracy techniques according to deployment requirements. This invention is completely end-to-end, directly from voltage data to dynamic images, avoiding any time-consuming traditional reconstruction preprocessing steps, thus possessing a fundamental advantage in imaging speed.

[0094] 5) The LSTM network used in this invention, through its unique gating mechanism and recurrent structure, can directly learn complex long-term and short-term temporal dependencies from data without the need for pre-defined simplified physical models or linearization. This allows for more effective modeling and prediction of the true evolution of muscle states over time. By introducing LSTM, it gains structural advantages in processing muscle dynamic EIT data with strong temporal continuity. Through its progressively updated hidden states and memory cells, it can better capture the correlations and dynamic trends between frames, more naturally simulating the temporal evolution process and state memory. This is more effective for understanding muscle activities with clear sequential and phased characteristics, such as gait cycles.

[0095] 6) This invention innovatively constructs a dynamic simulation model that reflects the physical changes (CSA, morphology, and potential electrical conductivity) caused by muscle contraction (concentric / eccentric) and relaxation behaviors during walking. By dividing the gait cycle into physiological stages and adjusting the model according to muscle function parameters, training data that highly approximates the real physiological process is generated. Using this unprecedented, high-fidelity dynamic physiological simulation data for training enables the neural network to learn the real spatiotemporal electrical behavior patterns of muscles in actual movement. This directly translates into significantly improved accuracy of reconstructed images, enabling clearer differentiation of different muscle groups (such as the quadriceps and hamstrings) and their states at different stages of movement (bulging, thinning, etc.), resulting in more physiologically meaningful and clinically valuable dynamic imaging results. This is difficult to achieve with existing methods that rely on general, static, or simplified model training.

[0096] This invention can be a system, method, and / or computer program product. A computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for causing a processor to implement various aspects of the invention.

[0097] Computer-readable storage media can be tangible devices capable of holding and storing instructions for use by an instruction execution device. Computer-readable storage media can be, for example, but not limited to, electrical storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any suitable combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable compact disc read-only memory (CD-ROM), digital multifunction disc (DVD), memory sticks, floppy disks, mechanical encoding devices, such as punch cards or recessed protrusions storing instructions thereon, and any suitable combination thereof. The computer-readable storage media used herein are not to be construed as transient signals themselves, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., light pulses through fiber optic cables), or electrical signals transmitted through wires.

[0098] The computer-readable program instructions described herein can be downloaded from computer-readable storage media to various computing / processing devices, or downloaded via a network, such as the Internet, local area network, wide area network, and / or wireless network, to an external computer or external storage device. The network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards them to the computer-readable storage media in the respective computing / processing device.

[0099] The computer program instructions used to perform the operations of this invention may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk, C++, Python, etc., and conventional procedural programming languages ​​such as "C" or similar languages. The computer-readable program instructions may be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, electronic circuitry, such as programmable logic circuitry, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs), is personalized by utilizing state information from the computer-readable program instructions. This electronic circuitry can execute the computer-readable program instructions to implement various aspects of the invention.

[0100] Various aspects of the present invention are described herein with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer-readable program instructions.

[0101] These computer-readable program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that, when executed by the processor of the computer or other programmable data processing apparatus, they create means for implementing the functions / actions specified in one or more blocks of the flowchart and / or block diagram. These computer-readable program instructions can also be stored in a computer-readable storage medium that causes a computer, programmable data processing apparatus, and / or other device to operate in a particular manner; thus, the computer-readable medium storing the instructions comprises an article of manufacture that includes instructions for implementing aspects of the functions / actions specified in one or more blocks of the flowchart and / or block diagram.

[0102] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, thereby causing the instructions executed on the computer, other programmable data processing apparatus, or other device to perform the functions / actions specified in one or more boxes of a flowchart and / or block diagram.

[0103] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of an instruction containing one or more executable instructions for implementing a specified logical function. In some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions. It will be known to those skilled in the art that implementation in hardware, implementation in software, and implementation using a combination of software and hardware are equivalent.

[0104] The various embodiments of the present invention have been described above. These descriptions are exemplary and not exhaustive, and are not limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or technical improvements to the embodiments in the market, or to enable others skilled in the art to understand the embodiments disclosed herein. The scope of the invention is defined by the appended claims.

Claims

1. A power impedance imaging method for dynamic muscle imaging, characterized in that, Includes the following steps: For the target, collect dynamic boundary voltage data sequences of muscles during walking; The muscle dynamic boundary voltage data sequence is input into a trained neural network model to reconstruct the corresponding dynamic image sequence of muscle internal conductivity, thereby obtaining electrical impedance imaging results. The neural network model uses convolutional layers to extract the spatial features of the muscle dynamic boundary voltage data sequence and recurrent neural network layers to predict the temporal evolution pattern of conductivity distribution caused by muscle activity. The training set for training the neural network model is obtained according to the following steps: Based on the dynamic changes of the thigh muscles during human walking, the gait cycle is divided into resting state, early support phase, late support phase and swing phase. The resting state is characterized by the relaxed balance of the quadriceps and hamstring muscles. The early support phase refers to the heel strike to the middle stage. The late support phase refers to the toes leaving the ground. The swing phase refers to the foot leaving the ground and touching the ground again. Using the resting state as the baseline model, multiple sets of models are obtained by constructing a finite element model. The finite element model consists of multiple finite element meshes, with the skeleton as the geometric center. The quadriceps femoris on the front and the hamstrings on the back evolve dynamically in a crescent shape. Each set of models is divided into four groups, representing the resting state, the early support phase, the late support phase, and the oscillating phase, respectively. For the dynamic phases of the early support phase, the late support phase, and the swing phase, the area of ​​the quadriceps and hamstring muscles is adjusted parametrically to simulate the continuous process of muscle morphology evolving with gait, and the electrical conductivity values ​​of each tissue fluctuate randomly within a set normal physiological range. The surface voltage distribution of the multiple models under the electrode excitation mode is calculated by finite element simulation, thereby obtaining a training set containing multiple sets of data. Each set of data corresponds to the conductivity and voltage pairing data of the four gait cycles of the resting state, the early stage of the support phase, the late stage of the support phase, and the oscillating phase.

2. The method according to claim 1, characterized in that, The neural network model includes multiple convolutional layers, recurrent neural network layers, and fully connected layers. Each convolutional layer is followed by a nonlinear activation function and a pooling layer. The convolutional layers generate convolutional features by performing convolution operations on the input data and the convolutional kernel. The pooling layers generate a dimensionality reduction matrix by extracting set values ​​from the convolutional features. The recurrent neural network layer captures the dynamic correlation between the four gait cycles—the resting state, the early support phase, the late support phase, and the oscillation phase—through a memory unit mechanism.

3. The method according to claim 1, characterized in that, The recurrent neural network layer is a long short-term memory network layer.

4. The method according to claim 1, characterized in that, During the training of the neural network model, data is input in batches, each batch containing complete gait sequences of multiple models. Each sequence contains paired data of voltage characteristics and conductivity output for four gait cycles: the resting state, the early support phase, the late support phase, and the swing phase. The voltage characteristics are derived from the surface voltage distribution under multiple adjacent electrode excitation and measurement modes. The output target of the neural network model is the conductivity distribution matrix of muscle, fat, and bone corresponding to the gait cycle.

5. The method according to claim 1, characterized in that, The loss function used to train the neural network model is the mean squared error loss, which measures the difference between the predicted conductivity distribution and the true value.

6. The method according to claim 2, characterized in that, The pooling layer uses max pooling operation for feature dimensionality reduction.

7. The method according to claim 1, characterized in that, The finite element model was constructed using the simulation software EIDORS.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

9. A computer device comprising a memory and a processor, wherein a computer program capable of running on the processor is stored in the memory, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

Images