Intelligent burn wound dynamic monitoring and evaluation system

Abstract

The invention relates to the technical field of burn wound treatment, and discloses an intelligent burn wound dynamic monitoring and evaluation system. The system comprises a data acquisition module for acquiring real-time physiological data of a burn wound by using a multi-source sensor; the multi-modal feature fusion module is used for generating wound comprehensive state features based on a deep belief network; the dynamic evaluation module outputs a wound healing level and a risk early warning signal through a pre-trained collaborative evaluation model; the optimization decision generation module is used for constructing a multi-target dynamic optimization model to adjust a treatment scheme according to risk early warning; and the hierarchical execution control module realizes distributed execution of treatment instructions through a hierarchical control architecture. The system can comprehensively monitor the state of the wound surface, accurately evaluate the healing risk, optimize the treatment scheme and accurately execute the treatment scheme, effectively improve the treatment effect of the burn wound surface, reduce the infection risk and shorten the healing period, and has important clinical application value.

Description

Technical Field

[0001] The present invention relates to the technical field of burn wound treatment, and specifically to an intelligent dynamic monitoring and evaluation system for burn wounds. Background Art

[0002] Burn is a common and severely harmful type of trauma, which frequently occurs in industrial accidents, fires, traffic accidents and other scenarios. According to incomplete statistics, the number of burn patients worldwide reaches millions every year. Among them, severely burned patients face many challenges such as difficult wound healing, high infection risk, long treatment cycle, etc. It not only brings great physical pain and psychological trauma to patients, but also causes a heavy burden on social medical resources.

[0003] Traditional burn wound treatment mainly relies on the experience judgment of medical staff and simple detection means. Medical staff evaluate the wound state by regularly observing the appearance of the wound, such as color, exudate, etc., and by means of basic body temperature measurement, blood routine examination, etc. However, this method has many limitations. On the one hand, the change in the appearance of the wound is often a lagging indicator in the wound healing process. By the time obvious changes are observed, adverse physiological changes may have occurred inside the wound, such as overactivation of the inflammatory response but not yet manifested in the appearance, thus missing the best treatment opportunity. On the other hand, simple detection means cannot comprehensively obtain the real-time physiological data of the wound. Key information such as the activity of wound tissue, the temperature and humidity changes of the microenvironment, and bioelectric signals are difficult to accurately measure, making the formulation of treatment plans lack accurate basis and more rely on experience, and it is difficult to achieve personalized and precise treatment.

[0004] With the development of technology, although some advanced detection devices have been introduced into the field of burn treatment, these devices often work independently, and the data lacks integration and collaborative analysis. For example, multi-spectral imaging devices can obtain the spectral information of the wound, and bioelectric signal detection instruments can detect relevant electric signals. However, there is no effective fusion mechanism for these data, and it is impossible to form a comprehensive understanding of the comprehensive state of the wound. In terms of evaluating the risk of wound healing, most of the existing evaluation models analyze based on a single factor or a few factors, without fully considering the complex interaction relationship between the wound and the body, such as the impact of immune function and metabolic level on wound healing, resulting in inaccurate risk assessment and unable to provide reliable early warning for clinical treatment in time.

[0005] In the formulation and adjustment of treatment plans, there is also a lack of scientific and dynamic optimization methods at present. Clinicians usually select physical treatment modes and determine drug doses according to general treatment specifications and personal experience, and it is difficult to find the best balance between the healing rate and the infection risk. Moreover, during the treatment process, when the patient's condition changes, it is impossible to quickly and accurately adjust the treatment plan, making the treatment effect greatly reduced. Summary of the Invention

[0006] The object of the present invention is to provide an intelligent dynamic monitoring and evaluation system for burn wounds to solve the problems raised in the above-mentioned background technology.

[0007] To achieve the above object, the present invention provides the following technical solution: An intelligent dynamic monitoring and evaluation system for burn wounds, the system includes: Data acquisition module: used to collect real-time physiological data of burn wounds through multi-source sensors; Multi-modal feature fusion module: based on a deep belief network, multi-dimensional feature extraction is performed on the real-time physiological data to generate comprehensive state features of the wound surface; Dynamic evaluation module: input the comprehensive state features of the wound surface into a pre-trained collaborative evaluation model to generate wound healing grades and risk warning signals; Optimized decision-making generation module: construct a multi-objective dynamic optimization model according to the risk warning signal, the multi-objective dynamic optimization model takes the maximization of the healing rate and the minimization of the infection risk as optimization objectives, and uses a decomposition and coordination optimization algorithm to globally adjust the treatment plan parameters; Hierarchical execution control module: based on the multi-objective dynamic optimization model, output the optimal control strategy, and realize the distributed execution of treatment instructions through a hierarchical control architecture. The hierarchical control architecture includes a decision-making layer, a coordination layer and an execution layer. Among them, the decision-making layer generates a global control sequence based on the treatment instructions, the coordination layer uses a rolling time window optimization algorithm to dynamically adapt local treatment parameters, and the execution layer realizes the tracking of drug dosage and physical treatment intensity based on a fuzzy adaptive control algorithm.

[0008] Preferably, the multi-source sensors include multi-spectral imaging sensors, bioelectric signal sensors, temperature sensors, humidity sensors and pressure sensors; the collection of real-time physiological data of burn wounds through multi-source sensors includes: Perform spatio-temporal registration on multi-spectral imaging data and bioelectric signal data to construct a three-dimensional wound tissue activity map; perform multi-scale wavelet decomposition on temperature sensor data and humidity sensor data to generate a wound microenvironment feature sequence; Construct a two-channel feature fusion network. The first channel uses a convolutional autoencoder to extract the spatial distribution features of the three-dimensional wound tissue activity map, and the second channel uses a long short-term memory network to extract the temporal change features of the microenvironment feature sequence; Fuse the spatial distribution features and temporal change features through a cross-modal attention mechanism to generate a joint feature matrix; perform dynamic correlation modeling on the joint feature matrix based on a gated recurrent unit, and output comprehensive state features including tissue activity, infection probability and healing stage.

[0009] Preferably, the collaborative evaluation model adopts a hierarchical graph attention network structure to generate risk assessment signals in multiple dimensions based on a dynamic weight allocation mechanism; the hierarchical graph attention network structure includes: Construct a wound-body interaction graph, where the nodes in the graph include wound area nodes, immune nodes, metabolic nodes, and treatment nodes, and the node attributes include inflammatory factor concentration, cell proliferation rate, and drug penetration rate; Adopt a two-stage attention mechanism. In the first stage, calculate the interaction weights between wound area nodes and adjacent nodes through a spatial graph attention layer, and in the second stage, screen the importance of historical healing states through a temporal graph attention layer; Iteratively update the node features based on a multi-head graph attention module. Each attention head fuses node attributes and external treatment parameters; stabilize the training process through residual connection and batch normalization mechanisms, and finally output an evaluation signal including immunosuppression risk and healing stagnation warning.

[0010] Preferably, the decomposition and coordination optimization algorithm includes: Model the treatment plan optimization problem as a multi-objective mixed-integer nonlinear programming problem, where the decision variables include discrete physical treatment mode variables and continuous drug dose adjustment variables; Initialize the relaxed sub-problem and calculate the approximate solution of the Pareto front. Adopt a dynamic target priority adjustment mechanism to update the weight coefficients of the objective function according to the real-time infection risk; In the decomposition stage, divide the global problem into multiple single-objective sub-problems based on the target space decomposition technique; in the coordination stage, introduce slack variables to balance the conflict constraints between sub-problems; Use a parallel solver to iteratively optimize the sub-problems, update the local solution each time, and synchronize the global solution through coordination variables.

[0011] Preferably, the rolling time window optimization algorithm dynamically adapts local treatment parameters, including: Construct a time-varying prediction model, discretize the wound healing kinetic equation into a state space model, and the state space model includes inflammatory factor prediction error, healing rate tracking error, and drug metabolism differential terms; Design a sliding window optimization objective function, including an inflammatory fluctuation suppression term, a tissue repair smoothing term, and an immune response balance term.

[0012] Preferably, the execution layer realizes the tracking of drug dose and physical treatment intensity based on a fuzzy adaptive control algorithm, including: Design a two-layer fuzzy inference system to map drug dose error and treatment intensity error to fuzzy rule activation degrees; construct an adaptive membership function adjustment mechanism to dynamically scale the coverage range of the fuzzy set according to the error change rate.

[0013] Preferably, the convolutional autoencoder adopts a multi-scale pooling structure to accelerate feature extraction, including: Divide the three-dimensional wound tissue activity map into multi-resolution grids, and each grid cell stores tissue activity gradients and inflammation intensity statistics; In the encoding stage, dilated convolutional kernels are used to expand the receptive field, and in the decoding stage, spatial details are restored through transposed convolutional layers; Introduce a spatial attention module to re-weight the feature maps.

[0014] Preferably, the spatial graph attention layer adopts a relative metabolic distance encoding mechanism, including: Define the relative metabolic distance vector between the wound area nodes and adjacent nodes, including the diffusion rate of inflammatory factors, the cell migration efficiency, and the drug concentration gradient; Map the relative metabolic distance vector to the bias parameter of the graph attention weight through a non-linear transformation layer; Superimpose the bias parameter on the standard graph attention operation.

[0015] Preferably, the dynamic target priority adjustment mechanism includes: Collect the objective function values and constraint violation degrees during the historical optimization process as training samples, construct a radial basis function network to fit the mapping relationship between the target priority coefficient and the system state; online update the network parameters through the incremental gradient descent method, and adjust the weight allocation of the objective function in real time. When a sudden change in the infection risk is detected, trigger the priority emergency reset operation.

[0016] Preferably, the time-varying prediction model realizes uncertainty modeling through orthogonal decomposition, including: Model the uncertainty of inflammatory factors as interval variables, and the upper and lower bounds are determined by the standard deviation of the prediction error; perform orthogonal decomposition on the interval variables to separate the deterministic and random components.

[0017] Compared with the prior art, the beneficial effects of the present invention are: The intelligent burn wound dynamic monitoring and evaluation system of the present invention brings significant positive impacts to burn wound treatment from multiple dimensions. In terms of data collection, the use of multi-source sensors enables the comprehensive acquisition of real-time physiological data of burn wounds. The multi-spectral imaging sensor, bioelectrical signal sensor, temperature sensor, humidity sensor, and pressure sensor work together, not only capable of collecting the appearance information of the wound surface, but also deeply obtaining key data such as tissue viability and microenvironment changes. By performing spatio-temporal registration on multi-spectral imaging data and bioelectrical signal data to construct a three-dimensional wound tissue viability map, and performing multi-scale wavelet decomposition on temperature and humidity data to generate microenvironment feature sequences, it provides a rich and accurate data basis for the subsequent precise evaluation of the wound state, changing the limitations of relying solely on simple observation and basic detection methods in the past, enabling medical staff to understand the real-time condition of the wound surface more comprehensively and deeply.

[0018] The multi-modal feature fusion module, based on a deep belief network, performs multi-dimensional feature extraction and fusion on the collected multi-source data. This fusion method fully explores the potential connections between different types of data and can more accurately reflect the comprehensive state of the wound surface compared with traditional single-data analysis methods. Through a dual-channel feature fusion network and a cross-modal attention mechanism, it generates comprehensive state features including tissue viability, infection probability, and healing stage, providing more valuable information for subsequent evaluation and decision-making, greatly improving the accuracy of judging the wound healing situation, helping to detect potential problems early, and intervening in advance.

[0019] The collaborative evaluation model adopted by the dynamic evaluation module, based on a hierarchical graph attention network structure and a dynamic weight allocation mechanism, fully considers the complex interaction relationship between the wound surface and the body. The constructed wound surface-body interaction graph covers wound surface area nodes, immune nodes, metabolic nodes, and treatment nodes, comprehensively analyzes various attributes such as inflammatory factor concentration, cell proliferation rate, and drug penetration rate, and realizes the multi-dimensional generation of risk assessment signals. This evaluation method can more accurately predict the risk of immunosuppression and the warning of healing stagnation, provide a reliable basis for clinicians to adjust treatment plans in a timely manner, effectively reduce the treatment risk of patients, and avoid the deterioration of the condition.

[0020] The optimization decision-making module constructs a multi-objective dynamic optimization model based on risk warning signals. Aiming at maximizing the healing rate and minimizing the infection risk, it globally adjusts the treatment plan parameters using a decomposition and coordination optimization algorithm. This module models the treatment plan optimization problem as a multi-objective mixed-integer non-linear programming problem, considering discrete physical therapy mode variables and continuous drug dosage adjustment variables. Through a dynamic target priority adjustment mechanism, it flexibly updates the weight coefficients of the objective function according to the real-time infection risk. This optimization method can find the best combination of treatment plans at different treatment stages, effectively balance the healing rate and infection risk, improve the treatment effect, shorten the patient's treatment cycle, and relieve the patient's pain and medical burden.

[0021] The hierarchical execution control module realizes the distributed execution of treatment instructions through a hierarchical control architecture. The decision-making layer generates a global regulation sequence to provide macro guidance for the entire treatment process; the coordination layer uses a rolling time window optimization algorithm to adjust local treatment parameters in real time according to the dynamic changes of wound healing to ensure the adaptability of the treatment plan; the execution layer realizes the precise tracking of drug dosage and physical therapy intensity based on a fuzzy adaptive control algorithm and dynamically adjusts treatment parameters according to error changes. This hierarchical cooperation method makes the treatment process more precise and efficient, improves the execution effect of the treatment plan, and ensures that patients can receive the most appropriate treatment. Brief Description of the Drawings

[0022] Figure 1 It is the working principle diagram of the intelligent burn wound dynamic monitoring and evaluation system described in the present invention; Figure 2 It is the schematic diagram of data acquisition and comprehensive state feature generation; Figure 3 It is the working principle diagram of the collaborative evaluation model; Figure 4 It is the working principle diagram of the fuzzy adaptive control algorithm of the execution layer. Detailed Embodiments

[0023] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0024] Please refer to Figures 1-4 , the present invention provides an intelligent burn wound dynamic monitoring and evaluation system, and its specific implementation method will be elaborated in detail below.

[0025] The data acquisition module uses multi-source sensors to collect real-time physiological data of burn wounds. The multi-source sensors include multi-spectral imaging sensors, bioelectrical signal sensors, temperature sensors, humidity sensors, and pressure sensors. These sensors obtain wound information from different dimensions. The multi-spectral imaging sensor can capture the spectral characteristics of the wound, the bioelectrical signal sensor can detect bioelectrical signals, the temperature sensor measures the wound temperature, the humidity sensor obtains humidity data, and the pressure sensor senses the pressure situation.

[0026] The multi-modal feature fusion module performs multi-dimensional feature extraction on the collected real-time physiological data based on a deep belief network. This module first performs spatio-temporal registration on the multi-spectral imaging data and bioelectrical signal data to construct a three-dimensional wound tissue activity map. At the same time, it performs multi-scale wavelet decomposition on the temperature sensor data and humidity sensor data to generate a wound microenvironment feature sequence. Then, a two-channel feature fusion network is constructed. The first channel uses a convolutional autoencoder to extract the spatial distribution features of the three-dimensional wound tissue activity map, and the second channel uses a long short-term memory network to extract the temporal variation features of the microenvironment feature sequence. After that, the spatial distribution features and temporal variation features are fused through a cross-modal attention mechanism to generate a joint feature matrix, and then a dynamic association model is built for the joint feature matrix based on a gated recurrent unit, and the comprehensive state features including tissue activity, infection probability, and healing stage are output.

[0027] The dynamic assessment module inputs the comprehensive state features of the wound into a pre-trained collaborative assessment model. The collaborative assessment model adopts a hierarchical graph attention network structure and generates multi-dimensional risk assessment signals based on a dynamic weight allocation mechanism. It first constructs a wound-body interaction graph, where the nodes in the graph include wound area nodes, immune nodes, metabolic nodes, and treatment nodes, and the node attributes include inflammatory factor concentration, cell proliferation rate, and drug permeability. Then, a two-stage attention mechanism is used. In the first stage, the interaction weights between the wound area nodes and adjacent nodes are calculated through a spatial graph attention layer, and in the second stage, the importance of the historical healing state is screened through a temporal graph attention layer. Finally, the node features are iteratively updated based on a multi-head graph attention module. Each attention head fuses the node attributes and external treatment parameters, and the training process is stabilized through a residual connection and batch normalization mechanism. Finally, an assessment signal including immunosuppression risk and healing stagnation warning is output.

[0028] The optimization decision-making generation module constructs a multi-objective dynamic optimization model based on the risk warning signal. This model takes the maximization of the healing rate and the minimization of the infection risk as the optimization objectives, and uses the decomposition and coordination optimization algorithm to globally adjust the treatment plan parameters. The decomposition and coordination optimization algorithm models the treatment plan optimization problem as a multi-objective mixed-integer nonlinear programming problem. The decision variables include discrete physical therapy mode variables and continuous drug dose adjustment variables. First, the relaxed sub-problem is initialized and the approximate solution of the Pareto front is calculated. The dynamic target priority adjustment mechanism is used to update the weight coefficients of the objective function according to the real-time infection risk. In the decomposition stage, the global problem is divided into multiple single-objective sub-problems based on the objective space decomposition technology; in the coordination stage, slack variables are introduced to balance the conflict constraints between sub-problems. Finally, a parallel solver is used to iteratively optimize the sub-problems. The local solution is updated in each iteration and the global solution is synchronized through the coordination variables.

[0029] The hierarchical execution control module outputs the optimal regulation strategy based on the multi-objective dynamic optimization model, and realizes the distributed execution of the treatment instructions through the hierarchical control architecture. The hierarchical control architecture includes a decision-making layer, a coordination layer and an execution layer. The decision-making layer generates a global regulation sequence based on the treatment instructions. The coordination layer uses the rolling time window optimization algorithm to dynamically adapt the local treatment parameters. The execution layer realizes the tracking of the drug dose and the physical therapy intensity based on the fuzzy adaptive control algorithm.

[0030] The technical solution of the present invention will be further described in detail below through specific embodiments.

[0031] Embodiment 1: In this embodiment, the wound monitoring of a certain burn patient is taken as an example. The multi-spectral imaging sensor and the bioelectric signal sensor work simultaneously to collect data on the wound. The multi-spectral imaging sensor obtains image data of the wound under different spectral bands, and the bioelectric signal sensor collects bioelectric signal data of the wound. These data are subjected to spatio-temporal registration. The process of spatio-temporal registration is to unify the data collected at different times and spaces into the same spatio-temporal coordinate system for subsequent processing. By constructing a three-dimensional wound tissue activity map, the activity distribution of the wound tissue in space can be intuitively reflected. At the same time, the temperature sensor and the humidity sensor continuously collect data, and these data are subjected to multi-scale wavelet decomposition. Multi-scale wavelet decomposition is a signal processing method that can decompose the signal at different scales to obtain information on different frequency components. Through this decomposition, a wound microenvironment feature sequence is generated, which contains the information on the change of the wound microenvironment over time.

[0032] Construct a dual-channel feature fusion network. The first channel uses a convolutional autoencoder. The working principle of the convolutional autoencoder is to extract features from the input data through convolutional operations. In this embodiment, a three-dimensional wound tissue activity map is input into the convolutional autoencoder. The convolutional autoencoder adopts a multi-scale pooling structure, divides the three-dimensional wound tissue activity map into multi-resolution grids, and each grid cell stores tissue activity gradients and inflammation intensity statistics. In the encoding stage, dilated convolutional kernels are used, which can expand the receptive field and enable the model to obtain more extensive information. In the decoding stage, the spatial details are restored through the transposed convolution layer, so that the extracted features can be restored to the spatial dimension of the original data. At the same time, a spatial attention module is introduced, which re-weights the feature map according to the importance of the features and highlights the important features. The second channel uses a long short-term memory network, which is good at processing time series data and can effectively extract the temporal variation features of the microenvironment feature sequence. Through the cross-modal attention mechanism, the spatial distribution features and temporal variation features extracted by the two channels are fused to generate a joint feature matrix. Finally, dynamic correlation modeling is performed on the joint feature matrix based on the gated recurrent unit, and comprehensive state features including tissue activity, infection probability, and healing stage are output, providing a basis for subsequent evaluation and decision-making.

[0033] Example 2: This embodiment details the construction of a collaborative evaluation model and its application in actual wound evaluation. Construct a wound-body interaction graph. Taking a patient with large-area burns as an example, wound area nodes, immune nodes, metabolic nodes, and treatment nodes are set in the graph. The attributes of the wound area nodes are set as the concentration of inflammatory factors, the cell proliferation rate, and the drug penetration rate, and these attribute values are obtained through the previous detection and analysis of the patient's wound. The immune node reflects the relevant indicators of the patient's immune system, the metabolic node reflects the patient's body metabolism, and the treatment node includes the relevant parameters of the current treatment methods.

[0034] Adopt a two-stage attention mechanism. In the first stage, the spatial graph attention layer adopts a relative metabolic distance encoding mechanism. Define the relative metabolic distance vector between the wound area node and its adjacent nodes. Let the relative metabolic distance vector be , which includes the inflammatory factor diffusion rate , the cell migration efficiency , and the drug concentration gradient , that is, . Map the relative metabolic distance vector to the bias parameter of the graph attention weight through a non-linear transformation layer. Assume the non-linear transformation function is , and obtain the bias parameter . The bias parameter Overlay it to the standard graph attention operation, so that the interaction weights between the wound area nodes and adjacent nodes can be calculated more accurately, highlighting the important node relationships related to wound healing. In the second stage, the temporal graph attention layer performs importance screening on the historical healing states, considering the wound healing data of the patient at different time points and analyzing which stages are more important for the current evaluation.

[0035] Iteratively update the node features based on the multi-head graph attention module, and each attention head fuses the node attributes and external treatment parameters. During the training process, the training process is stabilized through the residual connection and batch normalization mechanisms. Finally, input the comprehensive wound state features output by the multi-modal feature fusion module into this collaborative evaluation model to obtain an evaluation signal including the immunosuppression risk and the warning of healing stagnation. For example, when the evaluation signal shows a high immunosuppression risk, it indicates that the patient's immune system may be suppressed and the treatment plan needs to be adjusted to enhance the immune function.

[0036] Example 3: This example takes the optimization of the treatment plan for a certain burn patient as an example to elaborate in detail the implementation process of the decomposition and coordination optimization algorithm. Model the treatment plan optimization problem as a multi-objective mixed-integer nonlinear programming problem, and the decision variables include discrete physical therapy mode variables (for example it can take the value of 1 to represent laser treatment and 0 to represent non-use) and continuous drug dose adjustment variables (in milligrams). The objective function is where represents the healing rate function, and represents the infection risk function.

[0037] Initialize the relaxed sub-problem and calculate the approximate Pareto front solution. At initialization, set the initial values according to the patient's basic conditions and previous treatment experience. Adopt a dynamic target priority adjustment mechanism, and collect the objective function values and constraint violation degrees during the historical optimization process as training samples. Construct a radial basis function network to fit the mapping relationship between the target priority coefficients and the system state. Let the output of the radial basis function network be representing the objective function weight coefficients, the input is the system state vector and the expression of the network is where is the mapping function of the radial basis function network. Update the network parameters online through the incremental gradient descent method and adjust the objective function weight allocation in real time. When a mutation in the infection risk is detected, such as the infection risk index exceeding the preset threshold, trigger the priority emergency reset operation and readjust the objective function weights to pay more attention to reducing the infection risk.

[0038] In the decomposition phase, the global problem is divided into multiple single-objective sub-problems based on the goal space decomposition technique. For example, maximizing the healing rate is one sub-problem, and minimizing the infection risk is another sub-problem. In the coordination phase, slack variables are introduced to balance the conflict constraints between sub-problems. Assuming that there are conflicts such as resource allocation between sub-problems, the slack variables are used to balance the conflict constraints between sub-problems. To achieve a certain degree of coordinated solution to each subproblem, the authors used a parallel solver to iteratively optimize the subproblems. Each iteration updated the local solution and synchronized the global solution through the coordination variables. Through continuous iteration, the optimized treatment plan parameters, such as the appropriate physical therapy mode and drug dosage, were ultimately obtained.

[0039] Example 4: During burn wound treatment, the wound state is constantly changing. To ensure that local treatment parameters accurately match wound healing needs, this embodiment uses a rolling time window optimization algorithm. The core of this algorithm is to continuously adjust treatment parameters based on the real-time status of wound healing to ensure optimal treatment results.

[0040] When building a time-varying prediction model, the wound healing dynamics equation needs to be discretized into a state-space model. In this process, wound healing is considered a dynamic system, and its state changes at different times are described by a mathematical model. The state-space model expression is:

[0041] Where, represent The state vector at the moment, which contains the prediction error of inflammatory factors , healing rate tracking error and drug metabolism differentials ,Right now . This reflects the difference between the current predicted level of inflammatory factors and the actual level. It reflects the deviation between the actual healing rate and the expected healing rate. It indicates the dynamic changes of drug metabolism in the body. 、 、 are the coefficient matrices of the state-space model, which determine the transition relationship of the system state and the connection between the control input and the observed output and the state. express The control input at each moment is the local treatment parameters that need to be adjusted, such as the dosage of the drug, the frequency of physical therapy, etc. Represents process noise, which reflects random interference factors in the system that cannot be accurately modeled, such as individual differences among patients and slight effects of environmental factors on wound healing. is The observed output at a moment, that is, the wound - related data actually measured by the sensor. is the observation noise, which is the measurement error caused by the sensor accuracy limitation or the interference during the measurement process.

[0042] When designing the sliding - window optimized objective function, several key factors such as inflammation fluctuation, tissue repair, and immune response are fully considered. The optimized objective function has the following expression:

[0043] where, represents the length of the sliding window, which determines the time range considered by the algorithm during the optimization process. The value of needs to comprehensively balance the computational complexity and the tracking ability of the dynamic changes of the wound surface, and is generally determined according to clinical experience and a large amount of experimental data. , , are weight coefficients, which are used to adjust the relative importance of inflammation - fluctuation suppression, tissue - repair smoothing, and immune - response balance in the optimization objective respectively. By adjusting these weight coefficients, according to the specific condition and treatment stage of the patient, different optimization objectives can be flexibly balanced. For example, in the initial stage with a high risk of wound infection, the weight of can be appropriately increased to focus on suppressing inflammation fluctuation; while in the later stage of wound healing, the weight of can be increased to promote the smooth progress of tissue repair.

[0044] represents the amount of inflammation - factor fluctuation at moment, which is obtained by calculating the difference in the concentration of inflammation factors at adjacent moments and reflects the severity of the inflammatory reaction. is the change amount of tissue - repair smoothness at moment, which is used to measure the stability of the tissue - repair process and can be determined by analyzing the changes of tissue - growth indicators at different moments. represents the change amount of immune - response balance degree at moment, which reflects the dynamic balance situation of the immune system during the wound - healing process, such as the changes of indicators such as immune - cell activity and immune - factor secretion amount.

[0045] In practical applications, taking a burn patient as an example. In the initial stage of treatment, the relevant data of the wound surface are obtained through the sensor, and the initial state vector is calculated. According to the time - varying prediction model, the state changes of the wound surface in the future period are predicted. At the same time, combined with the sliding - window optimized objective function, the optimized objective value within the current sliding window is calculated. Then, through the optimization algorithm, the local treatment parameters Make adjustments so that is minimized, that is, to achieve the optimal combination of suppressing inflammatory fluctuations, smoothing tissue repair, and balancing the immune response. As time goes by, the sliding window rolls forward, continuously incorporating new wound data, recalculating the state vector, predicting the future state, optimizing the objective function, and adjusting the local treatment parameters again. This cycle repeats, continuously adapting the local treatment parameters dynamically according to the real-time situation of wound healing, ensuring that the treatment plan always fits the actual needs of the wound, improving the treatment effect, and promoting the rapid and stable healing of the wound.

[0046] Example 5: In the execution stage of burn wound treatment, accurately controlling the drug dosage and the intensity of physical therapy plays a key role in the wound healing effect. In this example, a fuzzy adaptive control algorithm is used to achieve precise tracking of the drug dosage and the intensity of physical therapy to better meet the needs of wound treatment.

[0047] When designing a double-layer fuzzy inference system, first determine the input variables. Take the drug dosage error and the treatment intensity error as the system inputs. The drug dosage error is calculated by the difference between the actually used drug dosage and the ideal drug dosage determined according to the patient's wound condition and treatment plan, that is ; the treatment intensity error is the difference between the actual physical therapy intensity and the ideal physical therapy intensity , .

[0048] To effectively process these errors, fuzzy sets are defined. The drug dosage error is divided into fuzzy sets such as "Negative Big (NB)", "Negative Small (NS)", "Zero (ZO)", "Positive Small (PS)", "Positive Big (PB)", etc., and the treatment intensity error is similarly divided. A fuzzy rule base is established based on a large amount of clinical experience and actual data. For example, when the drug dosage error is "Positive Big" and the treatment intensity error is "Positive Small", according to expert experience, it may be necessary to significantly reduce the drug dosage and appropriately reduce the treatment intensity at the same time, and the activation degree of the fuzzy rule is a specific value at this time. Another example is that if the drug dosage error is "Negative Small" and the treatment intensity error is "Zero", then a small amount of drug dosage can be appropriately increased to maintain the current treatment intensity, corresponding to another fuzzy rule activation degree. These rules cover various possible combinations of errors and provide a basis for subsequent control decisions.

[0049] In the double-layer fuzzy inference system, in the first layer, based on the input drug dose error and treatment intensity error, combined with the fuzzy rule base, the preliminary control output is calculated. In the second layer, the output of the first layer is further optimized and adjusted, comprehensively considering more factors such as the error change trend and patient individual differences, to obtain the final fuzzy rule activation degree, thereby determining the adjustment direction and amplitude of the drug dose and treatment intensity.

[0050] Constructing an adaptive membership function adjustment mechanism is a key part of this algorithm. Dynamically scale the coverage range of the fuzzy set according to the error change rate to improve the flexibility and accuracy of control. Let the drug dose error change rate be , which is obtained by calculating the ratio of the difference in drug dose error between adjacent time instants to the time interval, that is , where and are the drug dose errors at time instant and time instant respectively, is the time interval; the treatment intensity error change rate is , and the calculation method is similar to that of the drug dose error change rate, , and are the treatment intensity errors at time instant and time instant respectively.

[0051] When is large, it means that the drug dose error changes rapidly. At this time, appropriately expand the coverage range of the drug dose error fuzzy set. For example, originally the error range corresponding to the "negative large" fuzzy set is less than -5mg. When is large, its range can be expanded to less than -8mg, so as to more comprehensively cover the error situation, enabling the control algorithm to respond more timely to large error changes and avoiding poor treatment effects caused by error accumulation. On the contrary, when is small, narrow the coverage range of the drug dose error fuzzy set to improve the control accuracy and make the drug dose adjustment more refined. For the treatment intensity error change rate , adopt the same adaptive adjustment strategy.

[0052] In the actual treatment process, taking a burn patient as an example. In the initial stage of treatment, the sensor monitors the drug dose and physical treatment intensity in real time, calculates the errors and by comparing with the ideal values, and simultaneously calculates the error change rates and . The double-layer fuzzy inference system uses this data and pre-set fuzzy rules to obtain the activation degrees of the fuzzy rules, and determine the adjustment directions and amplitudes of the drug dosage and treatment intensity. The adaptive membership function adjustment mechanism adjusts the coverage range of the fuzzy sets in real time according to the error change rate to further optimize the control effect. As the treatment progresses, the system continuously monitors and adjusts to ensure that the drug dosage and physical treatment intensity always match the actual needs of the patient's wound surface, promoting good wound healing.

[0053] Example 6: The overall working process and comprehensive application of the system. In this example, the entire treatment process of a severely burned patient from admission to wound healing is taken as an example to demonstrate the overall working process and comprehensive application of the system. After the patient is admitted to the hospital, the data acquisition module immediately starts working. The multi-spectral imaging sensor, bioelectric signal sensor, temperature sensor, humidity sensor, and pressure sensor continuously collect real-time physiological data of the burned wound surface. These data are transmitted to the multi-modal feature fusion module in real time. This module processes the data, constructs a three-dimensional wound tissue activity map, generates a sequence of wound microenvironment features, and then outputs a comprehensive state feature including tissue activity, infection probability, and healing stage through a dual-channel feature fusion network and a gated recurrent unit.

[0054] The comprehensive state feature is input into the dynamic evaluation module, and the collaborative evaluation model analyzes it and outputs an evaluation signal including the immunosuppression risk and the warning of healing stagnation. If the evaluation signal shows that the patient has an immunosuppression risk, the optimization decision-making generation module will construct a multi-objective dynamic optimization model based on this risk warning signal. Using the decomposition and coordination optimization algorithm, considering the patient's specific conditions such as age and physical condition, the treatment plan parameters are globally adjusted to determine the appropriate physical treatment mode and drug dosage.

[0055] The hierarchical execution control module realizes the distributed execution of treatment instructions through a hierarchical control architecture according to the optimized treatment plan. The decision-making layer generates a global regulation sequence. The coordination layer uses a rolling time window optimization algorithm to continuously adjust the local treatment parameters according to the dynamic changes of the patient's wound healing. The execution layer is based on a fuzzy adaptive control algorithm to precisely control the drug dosage and physical treatment intensity. During the treatment process, the system continuously monitors various data of the patient's wound surface, continuously evaluates and adjusts until the patient's wound surface is completely healed. Through such a complete working process, the system realizes the intelligent dynamic monitoring and evaluation of the burned wound surface and provides a precise and effective treatment plan for the patient.

[0056] It should be noted that in this text, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements not only includes those elements, but also includes other elements not expressly listed, or further includes elements inherent to such process, method, article or device.

[0057] Although the embodiments of the present invention have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention, and the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An intelligent dynamic monitoring and evaluation system for burn wounds, characterized in that, Including: Data acquisition module: used to collect real-time physiological data of burn wounds through multi-source sensors; Multi-modal feature fusion module: based on a deep belief network, perform multi-dimensional feature extraction on the real-time physiological data to generate comprehensive wound state features; Dynamic evaluation module: input the comprehensive wound state features into a pre-trained collaborative evaluation model to generate wound healing grades and risk warning signals; Optimization decision generation module: construct a multi-objective dynamic optimization model according to the risk warning signals. The multi-objective dynamic optimization model takes maximizing the healing rate and minimizing the infection risk as optimization objectives, and uses a decomposition and coordination optimization algorithm to globally adjust the treatment plan parameters; Hierarchical execution control module: based on the multi-objective dynamic optimization model, output the optimal regulation strategy, and realize the distributed execution of treatment instructions through a hierarchical control architecture. The hierarchical control architecture includes a decision-making layer, a coordination layer, and an execution layer. The decision-making layer generates a global regulation sequence based on the treatment instructions, the coordination layer dynamically adapts local treatment parameters using a rolling time window optimization algorithm, and the execution layer realizes the tracking of drug dosage and physical therapy intensity based on a fuzzy adaptive control algorithm.

2. The intelligent burn wound dynamic monitoring and evaluation system according to claim 1, wherein The multi-source sensors include multi-spectral imaging sensors, bioelectric signal sensors, temperature sensors, humidity sensors, and pressure sensors; the collection of real-time physiological data of burn wounds through multi-source sensors includes: Perform spatio-temporal registration on multi-spectral imaging data and bioelectric signal data to construct a three-dimensional wound tissue activity map; perform multi-scale wavelet decomposition on temperature sensor data and humidity sensor data to generate a wound microenvironment feature sequence; Construct a two-channel feature fusion network. The first channel uses a convolutional autoencoder to extract the spatial distribution features of the three-dimensional wound tissue activity map, and the second channel uses a long short-term memory network to extract the temporal change features of the microenvironment feature sequence; Fuse the spatial distribution features and temporal change features through a cross-modal attention mechanism to generate a joint feature matrix; perform dynamic correlation modeling on the joint feature matrix based on a gated recurrent unit, and output comprehensive state features including tissue activity, infection probability, and healing stage.

3. The intelligent burn wound dynamic monitoring and evaluation system according to claim 1, characterized in that, The collaborative evaluation model adopts a hierarchical graph attention network structure and generates risk assessment signals multi-dimensionally based on a dynamic weight assignment mechanism; The hierarchical graph attention network structure includes: Construct a wound-body interaction graph. The nodes in the graph include wound area nodes, immune nodes, metabolic nodes, and treatment nodes. The node attributes include inflammatory factor concentration, cell proliferation rate, and drug permeability; Adopt a two-stage attention mechanism. In the first stage, calculate the interaction weights between wound area nodes and adjacent nodes through a spatial graph attention layer, and in the second stage, screen the importance of historical healing states through a temporal graph attention layer; Iteratively update node features based on a multi-head graph attention module. Each attention head fuses node attributes and external treatment parameters; stabilize the training process through a residual connection and batch normalization mechanism, and finally output an evaluation signal including immunosuppression risk and healing stagnation warning.

4. The intelligent burn wound dynamic monitoring and evaluation system according to claim 1, wherein The decomposition and coordination optimization algorithm includes: Model the treatment plan optimization problem as a multi-objective mixed-integer non-linear programming problem, where the decision variables include discrete physical therapy mode variables and continuous drug dose adjustment variables; Initialize the relaxation sub-problem and calculate the approximate Pareto front solution, and use a dynamic objective priority adjustment mechanism to update the objective function weight coefficients according to the real-time infection risk; In the decomposition stage, divide the global problem into multiple single-objective sub-problems based on the objective space decomposition technique; in the coordination stage, introduce relaxation variables to balance the conflicting constraints between sub-problems; Use a parallel solver to iteratively optimize the sub-problems, update the local solution at each iteration, and synchronize the global solution through the coordination variables.

5. The intelligent burn wound dynamic monitoring and evaluation system according to claim 1, wherein The rolling time window optimization algorithm dynamically adapts the local treatment parameters, including: Construct a time-varying prediction model, discretize the wound healing kinetic equation into a state space model, and the state space model includes the prediction error of inflammatory factors, the tracking error of the healing rate, and the differential term of drug metabolism; Design a sliding window optimization objective function, including an inflammatory fluctuation suppression term, a tissue repair smoothing term, and an immune response balance term.

6. The intelligent burn wound dynamic monitoring and evaluation system according to claim 1, wherein The execution layer realizes the tracking of drug dose and physical therapy intensity based on the fuzzy adaptive control algorithm, including: Design a two-layer fuzzy inference system to map the drug dose error and the treatment intensity error to the activation degree of fuzzy rules; construct an adaptive membership function adjustment mechanism to dynamically scale the coverage range of the fuzzy set according to the error change rate.

7. The intelligent burn wound dynamic monitoring and evaluation system according to claim 2, characterized in that, The convolutional autoencoder adopts a multi-scale pooling structure to accelerate feature extraction, including: Divide the three-dimensional wound tissue activity map into multi-resolution grids, and each grid cell stores the tissue activity gradient and the inflammatory intensity statistic; Use dilated convolutional kernels in the encoding stage to expand the receptive field, and restore the spatial details through the transposed convolutional layer in the decoding stage; Introduce a spatial attention module to re-weight the feature map.

8. The intelligent burn wound dynamic monitoring and evaluation system according to claim 3, wherein The spatial graph attention layer adopts a relative metabolic distance encoding mechanism, including: Define the relative metabolic distance vector between the wound area nodes and adjacent nodes, including the diffusion rate of inflammatory factors, the cell migration efficiency, and the drug concentration gradient; Map the relative metabolic distance vector to the bias parameter of the graph attention weight through a non-linear transformation layer; Superimpose the bias parameter on the standard graph attention operation.

9. The intelligent burn wound dynamic monitoring and evaluation system according to claim 4, wherein The dynamic objective priority adjustment mechanism includes: Collect the objective function values and constraint violation degrees in the historical optimization process as training samples, construct a radial basis function network to fit the mapping relationship between the objective priority coefficient and the system state; update the network parameters online through the incremental gradient descent method, and adjust the objective function weight allocation in real time. When a sudden change in the infection risk is detected, trigger the priority emergency reset operation.

10. The intelligent burn wound dynamic monitoring and evaluation system according to claim 5, wherein The time-varying prediction model realizes uncertainty modeling through orthogonal decomposition, including: Model the uncertainty of inflammatory factors as an interval variable, and the upper and lower bounds are determined by the standard deviation of the prediction error; perform orthogonal decomposition on the interval variable to separate the deterministic and random components.

Images