Pressure injury monitoring system with multimodal real-time sensing and dynamic risk assessment

Abstract

The present application relates to the technical field of intelligent medical monitoring, in particular to a pressure injury monitoring system based on multi-modal real-time sensing and dynamic risk assessment, comprising a multi-modal sensing step: receiving multi-source heterogeneous data streams and performing time sequence synchronization alignment to generate real-time fusion feature data; a digital twin analysis step: constructing a virtual tissue dynamic model to calculate theoretical reference feature data under a healthy benchmark state; a pathological simulation injection step: calling a pathological evolution rule knowledge base to simulate and generate theoretical pathological feature data; a differential topology decision step: performing double-track differential operation to generate a real observation and theoretical pathological residual vector, and generating a risk assessment result by calculating a topological similarity measure; the present application constructs a residual shape matching mechanism through a double reference system, effectively filters out environmental noise and physiological fluctuations, significantly reduces the false positive rate and realizes early warning of deep tissue injury.

Description

Technical Field

[0001] This invention relates to the field of intelligent medical monitoring technology, specifically a stress injury monitoring system with multimodal real-time sensing and dynamic risk assessment. Background Technology

[0002] With the rapid iteration of clinical monitoring technology, the early prevention and treatment of pressure injuries has become a core pain point in the care of long-term bedridden patients. This complication seriously affects the quality of life of patients. How to accurately identify the risk of deep tissue damage in complex clinical environments and effectively filter out environmental interference and physiological fluctuations is also a major challenge. Under the premise of pursuing refinement in the field of medical care, whether non-invasive and accurate early warning can be achieved has also become a major focus of monitoring technology.

[0003] Traditional pressure injury monitoring protocols currently rely primarily on the following methods: periodic artificial skin assessments by healthcare professionals, single-dimensional pressure distribution pad monitoring, and body surface temperature alarms based on simple thresholds.

[0004] However, manual assessment, single pressure monitoring, and temperature threshold alarms all have certain drawbacks. For example, manual assessment relies on visual observation, which often lags behind the occurrence of deep tissue damage and misses the intervention window; single pressure monitoring only reflects the external load and cannot know the actual pathological reaction inside the tissue; threshold-based temperature monitoring is easily interfered with by environmental heat sources such as electric blankets or normal physiological fluctuations such as reactive hyperemia, resulting in a very high false alarm rate and failing to effectively distinguish between physiological compensatory reactions and early pathological damage. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides a multimodal real-time sensing and dynamic risk assessment system for pressure injury monitoring. Specifically, the technical solution of this invention includes:

[0006] The multimodal sensing terminal module is used to receive and process multi-source heterogeneous data streams of the target area. The multi-source heterogeneous data streams contain time-series data of three modes: mechanical, thermal and bioelectric. The module performs time-series synchronization and alignment on the multi-source heterogeneous data streams to generate standardized real-time fused feature data.

[0007] The digital twin analysis module is used to construct a virtual tissue dynamic model based on the prior feature data of an individual, and input the stress-related features in the real-time fused feature data into the virtual tissue dynamic model to calculate and output the theoretical reference feature data corresponding to the health baseline state.

[0008] The pathological simulation injection module is used to call the preset knowledge base of pathological evolution rules of pressure injury, extract the evolution parameter set and inject it into the running logic of the virtual tissue dynamic model, so as to simulate and generate theoretical pathological feature data that characterizes the pathological evolution process.

[0009] The differential topology decision module is used to perform dual-track differential operations based on the real-time fused feature data, the theoretical reference feature data, and the theoretical pathological feature data to generate a real-world observation residual vector and a theoretical pathological residual vector. It also generates a dynamic risk assessment result by calculating the topological similarity measure between the real-world observation residual vector and the theoretical pathological residual vector in the feature space.

[0010] Preferably, the digital twin analysis module includes:

[0011] The parameter initialization unit is used to load the prior characteristic data of an individual, which includes body mass index, Braden score and historical blood perfusion baseline value provided by an external system;

[0012] The model building unit is used to define and initialize the parameters of a dynamic system model with viscoelastic properties based on the prior feature data.

[0013] The benchmark generation unit is used to take the temporal pressure features in the real-time fused feature data as model input, drive the dynamic system model to perform calculations, output the theoretical deformation sequence, theoretical temperature change sequence and theoretical impedance sequence under no pathological disturbance, and combine them into the theoretical reference feature data.

[0014] Preferably, the pathological simulation injection module includes:

[0015] The factor extraction unit is used to query and extract ischemic grade correlation coefficient, inflammatory heat diffusion rate and cell membrane permeability factor from the knowledge base of the pathological evolution rules of the pressure injury to form the evolution parameter set;

[0016] The superimposed simulation unit is used to superimpose the evolution parameter set onto the running logic of the virtual tissue dynamic model to simulate the conduction and diffusion effects defined by ischemia and inflammation rules, thereby generating the theoretical pathological feature data.

[0017] Preferably, the differential topology decision module includes:

[0018] The first difference unit is used to calculate the difference between the real-time fused feature data and the theoretical reference feature data, and generate the real-world observation residual vector containing composite interference;

[0019] The second difference unit is used to calculate the difference between the theoretical pathological feature data and the theoretical reference feature data, and generate a pure theoretical pathological residual vector.

[0020] The coupled computation unit is used to map the actual observation residual vector to the feature subspace spanned by the theoretical pathological residual vector, and calculate the cosine similarity between the two as the topological similarity measure.

[0021] Preferably, the differential topology decision module further includes:

[0022] A logic determination unit is used to compare the topological similarity metric with a preset risk determination threshold;

[0023] If the topological similarity metric is higher than the risk determination threshold, the target area is determined to have a risk of pressure damage, and a high-risk alarm command is generated.

[0024] If the topological similarity metric is lower than the risk determination threshold, the actual observation residual vector is determined to be an acceptable background fluctuation, and a safety status instruction is generated.

[0025] Preferably, the multimodal sensing terminal module includes:

[0026] The data interface unit is used to receive time-series pressure data, skin surface temperature distribution map and environmental reference temperature data from external sensing units, as well as complex impedance spectrum data of biological tissues.

[0027] The feature synchronization unit is used to perform timestamp alignment and data fusion on the multi-source heterogeneous data streams to form standardized real-time fused feature data.

[0028] Preferably, the system further includes:

[0029] An edge gateway module, connected upstream of the multimodal sensing terminal module, is used to perform protocol parsing, format unification, and noise filtering on the incoming raw multi-source heterogeneous data stream, and encapsulate the cleaned data into standardized messages for transmission to the multimodal sensing terminal module.

[0030] Preferably, the system further includes:

[0031] The feedback adjustment module is used to generate a dynamic adjustment strategy signal for the nursing bed in response to the high-risk alarm command output by the differential topology decision module. The signal can be used to drive an external actuator to change the mechanical environment of the target area.

[0032] Compared with the prior art, the present invention has the following beneficial effects:

[0033] 1. This system introduces a dual reference system of health benchmarks and pathological simulations to construct a dynamic assessment mechanism based on residual morphology matching, effectively solving the problem that existing technologies cannot distinguish between physiological compensation and early pathological damage. Specifically, the digital twin analysis module constructs a virtual tissue dynamic model based on individual prior features and calculates the theoretical reference features in a healthy state in real time. At the same time, the pathological simulation injection module calls the evolutionary rule knowledge base to generate theoretical pathological features that characterize the pathological process. The differential topology decision module calculates the topological similarity between the actual observation residual and the theoretical pathological residual in the feature space by performing dual-track differential operations. This mechanism can effectively filter out environmental noise such as the heating of electric blankets and normal physiological fluctuations such as reactive congestion, and judge the risk only when the actual deviation morphology and the theoretical pathological morphology are highly consistent, thereby significantly reducing the false alarm rate and realizing early warning of deep tissue damage.

[0034] 2. This system addresses the spatiotemporal fragmentation and signal noise issues of multi-source heterogeneous data through the high-frequency main axis alignment strategy of the multimodal sensing terminal module and the preprocessing mechanism of the edge gateway module. Using high-frequency pressure data as the master clock, the system interpolates and aligns low-frequency temperature data and downsamples and fuses high-frequency impedance data to generate unified real-time fused feature data, ensuring temporal consistency in multi-physics coupling analysis. Furthermore, the edge gateway module employs an adaptive Kalman filter algorithm to filter out high-frequency micro-fluctuations caused by respiratory motion and sensor background noise before the data enters the core computation, significantly improving the signal-to-noise ratio and providing a clean and synchronized high-quality input source for the digital twin model, thus ensuring the accuracy of subsequent simulation analysis.

[0035] 3. This system achieves personalized adaptation and improved dynamic response capabilities of the monitoring system through the collaborative work of the parameter initialization unit and the model construction unit. The system can load prior data such as an individual's body mass index, Braden score, and historical blood perfusion baseline values, and execute a zero-pressure calibration procedure to obtain a bioimpedance baseline, thereby defining a dynamic system model with specific viscoelastic properties. This model uses a recursive operator and a difference equation with a discrete time step to iterate in real time the theoretical deformation, temperature change, and impedance sequence of an individual in a completely healthy state. This method of synthesis and analysis based on an individual physical model enables the system to dynamically adjust the baseline according to the patient's specific physiological conditions, avoiding the missed or false alarms caused by individual differences in the traditional general threshold method, and achieving accurate individualized monitoring.

[0036] 4. This system introduces a risk-based proportional unloading algorithm through a feedback adjustment module, achieving closed-loop control from precise monitoring to automated intervention. When the differential topology decision module issues a high-risk alarm, it dynamically calculates the target pressure value based on topological similarity measurement. While reducing the airbag pressure in the risk area, it increases the pressure of neighboring airbags to provide compensatory support, following the principle of support force conservation. This mechanism can dynamically adjust the decompression amplitude according to the severity of the risk, which can both immediately block the pathological evolution of pressure injury and maintain the necessary body support stability. It effectively solves the problem of delayed manual nursing intervention and improves the therapeutic value and automated nursing level of the monitoring system. Attached Figure Description

[0037] The present invention will be further explained below with reference to the accompanying drawings and embodiments:

[0038] Figure 1 This is a structural diagram of the system of the present invention. Detailed Implementation

[0039] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0040] Example 1:

[0041] Please see Figure 1 A multimodal real-time sensing and dynamic risk assessment system for pressure injury monitoring, comprising:

[0042] The multimodal sensing terminal module is used to receive and process multi-source heterogeneous data streams of the target area. The multi-source heterogeneous data streams contain time-series data of three modes: mechanical, thermal and bioelectric. The module performs time-series synchronization and alignment of the multi-source heterogeneous data streams to generate standardized real-time fused feature data.

[0043] The digital twin analysis module is used to build a virtual tissue dynamic model based on the prior feature data of an individual, and input the stress-related features in the real-time fused feature data into the virtual tissue dynamic model to calculate and output the theoretical reference feature data corresponding to the health baseline state.

[0044] The pathological simulation injection module is used to call the preset knowledge base of pathological evolution rules for pressure injury, extract the evolution parameter set and inject it into the running logic of the virtual tissue dynamic model, so as to simulate and generate theoretical pathological feature data that characterize the pathological evolution process.

[0045] The differential topology decision module is used to perform dual-track differential operations based on real-time fused feature data, theoretical reference feature data, and theoretical pathological feature data to generate actual observation residual vectors and theoretical pathological residual vectors. By calculating the topological similarity measure between the actual observation residual vectors and theoretical pathological residual vectors in the feature space, dynamic risk assessment results are generated.

[0046] This embodiment details the overall architecture and core processing flow of the system, which aims to solve the technical challenge of effectively distinguishing between physiological compensatory responses and early pathological damage in existing monitoring technologies. The multimodal sensing terminal module, as the system's data entry point, executes spatiotemporal alignment logic. It not only receives data but also addresses the issue of asynchronous sampling rates for different physical quantities through a high-precision clock synchronization protocol. It interpolates and aligns high-frequency bioimpedance data with low-frequency temperature data to generate unified real-time fused feature data. The digital twin analysis module, acting as a health benchmark generator, employs synthetic analysis, not directly analyzing abnormalities but rather based on the patient's individual characteristics. The system utilizes a physical model to calculate the physical response of a patient in a completely healthy state in real time, i.e., theoretical reference feature data; the pathological simulation injection module acts as a virtual lesion generator, calling a preset knowledge base of pressure injury pathological evolution rules, transforming medical pathological mechanisms into mathematical operators and injecting them into the digital twin model, simulating the sensor values ​​when the patient is currently experiencing stage 1 pressure injury, i.e., theoretical pathological feature data; the differential topology decision module acts as the core decision unit, performing dual-track differential operations to calculate the deviation between real data and the healthy baseline, as well as the deviation between simulated pathological data and the healthy baseline, and calculating the topological similarity of these two residual vectors in the feature space;

[0047] This embodiment introduces a dual reference system of health benchmarks and pathological simulations to construct a dynamic assessment mechanism based on residual morphology matching. In complex clinical environments, this mechanism can effectively filter out environmental noise and normal physiological fluctuations. The system only determines the risk when the deviation morphology of the actual data and the deviation morphology of the theoretical pathology are highly consistent in topology, thereby significantly reducing the false alarm rate and realizing early warning of deep tissue damage.

[0048] The digital twin analytics module includes:

[0049] The parameter initialization unit is used to load the individual's prior characteristic data, which includes body mass index, Braden score, and historical blood perfusion baseline values ​​provided by an external system.

[0050] The model building unit is used to define and initialize the parameters of a dynamic system model with viscoelastic properties based on prior feature data.

[0051] The benchmark generation unit is used to take the time-series pressure features in the real-time fused feature data as model input, drive the dynamic system model to perform calculations, and output the theoretical deformation sequence, theoretical temperature change sequence and theoretical impedance sequence under no pathological disturbance, and combine them into theoretical reference feature data.

[0052] This embodiment details the internal structure and operational logic of the digital twin analysis module, which aims to construct a virtual soft tissue model capable of dynamically responding to changes in external pressure. The parameter initialization unit loads the individual's prior characteristic data through the hospital information system interface. To address the boundary conditions required for formula calculations, the system executes a zero-pressure calibration procedure before monitoring begins: instructing the nursing bed airbag to completely depressurize for 10 seconds, and measuring the bioimpedance value at this time as... The average value of the patient's most recent Doppler flow imaging was extracted from the electronic medical record as... ;

[0053] Based on prior data, the model building unit defines and initializes a Kelvin-Voigt viscoelastic dynamic model, which treats the skin and subcutaneous tissue as a system composed of springs and dampers connected in parallel. Its state equation is as follows:

[0054]

[0055] To ensure that the above differential equations can be solved in real time on a discrete-time digital processor, the model building unit adopts a backward Euler discretization scheme and constructs the following recursive calculation formula to solve the theoretical deformation sequence. :

[0056]

[0057] This recursive operator enables the system to utilize the pressure sample value at the current moment. Deformation state at the previous moment The current theoretical value of deformation is obtained through rapid iteration; among which, This is the theoretical stress response, driven by real-time pressure data, and its physical meaning is the stress borne by the tissue, with the unit being Pa. This is a theoretical deformation sequence, derived from recursive formula calculations, and its physical meaning is the degree of tissue compression, with a unit of 1. The effective elastic modulus is derived from the BMI index nonlinear mapping function, which is set as follows: ,in, As the reference modulus, The hardening coefficient, in physical terms, represents the elastic stiffness of the tissue. The effective viscosity coefficient is derived from the Braden score correction, and the formula is: ;in, As the reference viscosity coefficient, For risk sensitivity coefficient, This represents the theoretical maximum value of the Braden rating scale. This represents the patient's current actual assessment score; its physical meaning is the viscous resistance of the tissue.

[0058] The baseline generation unit uses the temporal pressure features from the real-time fused feature data as model input to drive the above model, and uses a recursive difference equation with discrete time steps to calculate the theoretical temperature change sequence in real time:

[0059]

[0060] in, For the present The theoretical reference temperature at time t, the initial value for iteration is set to ; This is a numerically stable minimum value used to prevent the denominator from being zero; The instantaneous deep heat conduction time constant is derived from the lumped parameterized form of the Pennes biothermal equation; the original Pennes equation includes a heat diffusion term. With blood perfusion This model reduces partial differential equations to ordinary differential equations by introducing geometric characteristic constants. The space thermal conductivity term is equivalent to a linear decay term. ,in, Physically, this corresponds to the eigenvalue of the Laplace operator under a specific tissue geometry, used to construct a heat dissipation rate with correct dimensions W / (m³·K); in this lumped model, the time constant... The ratio of the system's thermal inertia to its total heat dissipation capacity is determined by the system's thermal inertia, where the total heat dissipation capacity is composed of both blood flow convection heat transfer and tissue heat conduction, while arterial temperature... The driving potential is already included in the steady-state objective value of the main equation. In this process, it does not participate in the construction of the time constant; therefore, the calculation formula is as follows:

[0061]

[0062] in, and These are the density and specific heat capacity of biological soft tissue, respectively, to ensure that molecules have the dimension of energy density; Perfusion rate per unit volume; and These are the density and specific heat capacity of blood, respectively. Thermal conductivity of biological tissue; This is a geometric characteristic constant, the value of which depends on the effective contact radius of the sensing probe. The calculation formula is: This formula is used to characterize the thermal diffusion boundary conditions at a specific geometric scale; its construction makes... With time dimension It accurately reflects the combined effect of blood perfusion and tissue heat conduction on the temperature response rate; To estimate the blood perfusion rate in real time, a negative exponential decay model was used for calculation: ;

[0063] For the generation of the theoretical impedance sequence, this unit calculates based on the piezoresistive effect principle:

[0064]

[0065] in, The reference impedance obtained during the calibration phase. This is the geometric resistivity coefficient, whose values ​​are derived from the Gabriel biological tissue dielectric properties database. It characterizes the rate of resistivity change in soft tissue due to intercellular compression under pressure. In this embodiment, a statistically typical value is used. .

[0066] The pathology simulation injection module includes:

[0067] The factor extraction unit is used to query and extract ischemic grade connection number, inflammatory heat diffusion rate and cell membrane permeability factor from the knowledge base of pathological evolution rules of pressure injury to form an evolution parameter set;

[0068] The superimposed simulation unit is used to superimpose the set of evolutionary parameters onto the running logic of the virtual tissue dynamic model to simulate the transmission and diffusion effects defined by the rules of ischemia and inflammation, thereby generating theoretical pathological feature data.

[0069] This embodiment specifies the pathological simulation injection module; the factor extraction unit extracts the evolutionary parameter set from a preset knowledge base; the knowledge base is constructed as a multidimensional lookup table, with injury stage and anatomical location as the joint index key; when the system is configured to monitor early risk in the sacrococcygeal region, it queries Key=[StageI,Sacrum] and returns the corresponding parameter value: ischemia-grade correlation coefficient. Inflammatory thermal diffusivity and cell membrane permeability factors ;

[0070] The superimposed simulation unit injects the above parameters into the runtime logic of the virtual tissue dynamic model; for the pathological thermal response, the calculation is as follows:

[0071]

[0072] in, This is a theoretical pathological temperature sequence; The activation function for vascular occlusion is the hyperbolic tangent function. ;in, The critical capillary occlusion pressure is set at 4.0 kPa, corresponding to approximately 30 mmHg, derived from the capillary collapse threshold defined in microcirculation physiology; this function ensures that the ischemic effect only occurs when the pressure approaches the critical value. Smooth intervention in time avoids erroneous triggering of cooling logic under low pressure conditions; The ischemic-order contact number is dimensionless and physically represents the combined attenuation ratio of tissue heat production and convective heat transfer capacity caused by microcirculation blockage. Its value directly determines the magnitude of the ischemic cooling effect. For the inflammatory thermal diffusivity, in the mathematical model of this embodiment, this parameter is specifically defined as a dimensionless gain coefficient of the intensity of the inflammatory response, used to quantify the diffusion factor of the thermal effect caused by enhanced metabolism in the inflammatory phase relative to the baseline state.

[0073] Although its name uses the term thermal diffusivity to correspond to the concept of pathological conduction, in the physical calculation logic, it removes the dimensional constraint of m2 / s to adapt to the normalization operation in the superimposed simulation unit, thereby avoiding calculation errors caused by inconsistency of dimensions. This represents the maximum temperature rise associated with inflammation. The effective ischemia accumulation time variable is maintained by the system state machine. The amount is accumulated over time; otherwise, it decreases according to the reperfusion rate.

[0074] For pathological impedance response, a first-order inertial link model is adopted:

[0075]

[0076] in, The degree of impedance attenuation caused by increased ion permeability after cell membrane damage was quantified; The characteristic time constant is not arbitrarily set, but is based on the cell membrane potential decay half-life measured in in vitro tissue ischemia experiments. It characterizes the typical metabolic tolerance time of cells from the onset of ischemia to irreversible ion pump failure. In this embodiment, the statistical median value of 3600 seconds is used to adapt to the physical condition of most adult patients. Through the above-defined parameter mapping and formula calculation, this module realizes the transformation from medical qualitative rules to mathematical quantitative simulation.

[0077] The differential topology decision module includes:

[0078] The first difference unit is used to calculate the difference between the real-time fused feature data and the theoretical reference feature data, and generate the real-world observation residual vector containing composite interference.

[0079] The second difference unit is used to calculate the difference between theoretical pathological feature data and theoretical reference feature data, and generate a pure theoretical pathological residual vector.

[0080] The coupled computation unit is used to map the real-world observation residual vector to the feature subspace spanned by the theoretical pathological residual vector, and to calculate the cosine similarity between the two as a measure of topological similarity.

[0081] This embodiment specifies the differential topology decision module; the differential calculation unit performs normalization and weighted construction of multimodal features; and calculates the original difference sequences of real-time data and reference data respectively. Furthermore, dynamic dispersion scaling is performed using the statistical properties of the theoretical reference data: to avoid division overflow caused by the standard deviation approaching zero when the reference data fluctuation is extremely small, the normalization formula is modified as follows:

[0082]

[0083] in, This serves as a baseline value for the sensor's noise level, ensuring that the denominator is always meaningful.

[0084] Construct a unified high-dimensional residual vector; considering the pressure in this system... As the driving input, its real-time observation With input into the digital twin model Since they are from the same data source, theoretically the mechanical residuals To focus on pathological changes in tissue response, the vector construction emphasizes thermal and electrical components while retaining the pressure component as a verification gating for the loading state.

[0085]

[0086] The weights are set as follows: , , The aforementioned weights are established based on a comprehensive score of the sensitivity and signal-to-noise ratio of each modality feature to early injury, and are specifically determined by the feature projection coefficients obtained through linear discriminant analysis of historical case data. Bioimpedance changes most significantly at the cellular level of injury and is least affected by environmental interference, therefore it is assigned the highest weight. Temperature changes have a lag effect and are therefore given the second highest weight; while pressure changes mainly reflect external loads rather than the state of the tissue, so they are given the lowest weight.

[0087] Coupled computational units compute topological similarity metrics: This addresses the potential product trap that can occur with simple cosine similarity at low modulus lengths, specifically when both real and pathological residuals are extremely weak but their directions coincidentally align. False alarms approaching 1 are addressed by defining the following before performing similarity calculations in this unit: The actual observation residual vector obtained from the aforementioned calculation, The theoretical pathological residual vector obtained from the aforementioned calculation is used; a modulus-length gating logic is introduced: first, it is judged... Is it greater than the preset valid signal threshold? ;like Then directly determine Ignore the directionality of weak background noise; if Then, the following normalized inner product operation is performed:

[0088]

[0089] in, As an amplitude stability constant, this value is determined based on the floating-point machine precision of the computing platform to prevent denominator underflow. Through this weighted, gating, and regularization process, the system can accurately capture weak pathological trend signals while effectively shielding low-amplitude co-directional noise interference.

[0090] The differential topology decision module also includes:

[0091] The logical judgment unit is used to compare the topological similarity measure with a preset risk judgment threshold;

[0092] If the topological similarity metric is higher than the risk assessment threshold, the target area is determined to be at risk of pressure damage, and a high-risk alarm command is generated.

[0093] If the topological similarity metric is below the risk assessment threshold, the observed residual vector is determined to be an acceptable background fluctuation, and a safe state instruction is generated.

[0094] This embodiment specifies the logic judgment unit and its risk response mechanism. The unit has a preset risk judgment threshold, which is set based on a retrospective analysis of a large amount of clinical data. The basis for this threshold is that when the cosine similarity between the actual residual and the pathological residual exceeds this value, the probability of deep tissue damage increases significantly. The system calculates the topological similarity metric in real time. In response to the topological similarity metric being higher than the risk judgment threshold, the system determines it as high-risk. This means that the actual skin change trend is highly parallel to the damage evolution trend simulated by the computer, thus generating a high-risk alarm command. This command not only includes the alarm signal but also the feature component with the largest contribution in the observed residual vector. In response to the topological similarity metric being lower than the risk judgment threshold, the system determines it as background fluctuation and generates a safe status command, and the system continues monitoring.

[0095] This embodiment achieves standardization and forward-looking risk assessment through quantitative similarity threshold determination; in particular, for early indicators such as abnormal decrease in impedance that are not visible to the naked eye, by comparing with the threshold, it can issue an alarm several hours earlier than the traditional observation of skin redness, thus gaining a valuable intervention window for clinical care.

[0096] The multimodal sensing terminal module includes:

[0097] The data interface unit is used to receive time-series pressure data, skin surface temperature distribution map and environmental reference temperature data from external sensing units, as well as complex impedance spectrum data of biological tissues.

[0098] The feature synchronization unit is used to perform timestamp alignment and data fusion on multi-source heterogeneous data streams to form standardized real-time fused feature data.

[0099] This embodiment details the hardware interface and data fusion of the multimodal sensing terminal module. The data interface unit is equipped with multiple physical interfaces for connecting to a flexible pressure sensing array to provide time-series pressure data, connecting to an infrared thermal imaging probe to provide skin surface temperature distribution maps, and connecting to a complex impedance spectroscopy measurement electrode to provide complex impedance spectroscopy data of biological tissues. Due to the significant differences in the sampling rates of the above three types of data, direct fusion would cause phase misalignment and information loss. To adapt to the computational requirements of the digital twin model for 0.1s high-frequency dynamic evolution, the feature synchronization unit adopts a high-frequency principal axis alignment strategy. This unit uses a 10Hz pressure data timestamp as the system master clock. For the 1Hz low-frequency skin surface temperature data, a cubic spline interpolation algorithm is used to upsample it to 10Hz to fill the smooth trajectory of inter-frame thermal field changes. For the 1kHz complex impedance spectroscopy data, downsampling and averaging processing is used to align it to the 10Hz principal axis. Finally, the aligned modal data is encapsulated into a standardized real-time fusion feature data packet with a time interval of 0.1s.

[0100] This embodiment solves the problem of spatiotemporal fragmentation of multi-source heterogeneous data by using high-frequency pressure signals as the main axis for interpolation synchronization. This scheme not only preserves the fast response characteristics of mechanical data to drive high-precision viscoelastic simulation, but also smooths the temporal distribution of thermal data, providing a high-quality, time-synchronized input source for subsequent digital twin models and ensuring the temporal consistency of multi-physics coupling analysis.

[0101] Example 2:

[0102] The system also includes:

[0103] The edge gateway module is connected upstream of the multimodal sensing terminal module. It is used to perform protocol parsing, format unification, and noise filtering on the incoming raw multi-source heterogeneous data stream, and encapsulate the cleaned data into standardized messages for transmission to the multimodal sensing terminal module.

[0104] This embodiment specifies the edge gateway module, which is deployed between the sensing terminal and the cloud or local server. This module performs protocol parsing, unifying proprietary protocols from sensors from different manufacturers into a standard JSON message body. It also performs noise filtering, employing an adaptive Kalman filter algorithm to filter out high-frequency, minute fluctuations caused by respiratory movements in pressure data. This algorithm defines state variables. State transition matrix ,in, The sampling time interval of the pressure sensor is; the observation equation is... Meanwhile, to ensure the recursive stability of the filter, a pre-defined process noise covariance matrix is ​​used. To achieve adaptive characteristics, the algorithm calculates the variance of the observed residuals in real time. In specific calculations, the observation residual scalar at the current moment is obtained. The variance is estimated using the sliding window method, and the formula is as follows: ,in, The length of the sliding window is used to dynamically adjust the observation noise covariance matrix. The specific adaptive adjustment function is defined as follows:

[0105]

[0106] in, This is the sensor's noise floor covariance, and its value is derived from the static measurement error variance specified in the sensor's datasheet. The variance threshold for pressure fluctuations caused by normal breathing is obtained by calculating the upper limit of variance statistics from 100 normal breathing pressure data in a resting state. This is the penalty gain coefficient, used to adjust the filter's suppression response speed to non-respiratory mutations; when the residual variance exceeds the respiratory threshold, Rapidly increase to suppress the impact of non-physiological mutations on state estimation, and conversely maintain high-sensitivity tracking; smooth transient changes in temperature data caused by air convection, and finally encapsulate and transmit the cleaned data;

[0107] This embodiment effectively reduces the computational load of the subsequent digital twin module by performing edge-side preprocessing before the data enters the core algorithm; at the same time, targeted filtering algorithms eliminate the inherent high-frequency electronic noise and non-injurious physiological fluctuations of the sensor, significantly improving the signal-to-noise ratio and ensuring the purity of the input data.

[0108] Example 3:

[0109] The system also includes:

[0110] The feedback adjustment module is used to generate a dynamic adjustment strategy signal for the nursing bed in response to the high-risk alarm command output by the differential topology decision module. The signal can be used to drive the external actuator to change the mechanical environment of the target area.

[0111] This embodiment specifies the feedback adjustment module, which is connected to the intelligent nursing bed or air mattress. When a high-risk alarm command is received, the module locates the pressure risk point based on the spatial coordinates of the skin surface temperature distribution map. The module generates a drive signal and sends a command to the air mattress controller. To achieve precise closed-loop control, the module uses a proportional unloading algorithm based on risk level to calculate the target pressure value.

[0112]

[0113] It should be clarified that the application of this formula has strict logical gating conditions: that is, it only applies when... The decompression calculation is only activated at this time; if The system will force Or make This is to prevent the target pressure from increasing in the opposite direction due to directly substituting negative values ​​into the formula, thereby ensuring that the intervention strategy always conforms to the physical logic of decompression protection; among which, The set pressure for the airbag in the target area, in kPa; This is the current airbag pressure, in kPa. This is the currently calculated topological similarity measure; This is the threshold for risk assessment; To adjust the gain coefficient, its value is set to... It is calibrated based on the pressure-volume characteristic curve of the air cushion airbag and is used to match the mechanical response sensitivity of the actuator to ensure that the higher the risk, the greater the unloading amplitude.

[0114] Based on this instruction, the module reduces the inflation pressure of the airbag in the corresponding coordinate area to... Simultaneously, the pressure of the airbags in the adjacent 8-neighborhood area is increased to provide compensating support. The calculation of the compensating pressure follows the principle of conservation of support force, and the calculation formula is as follows:

[0115]

[0116] in, New pressure settings for neighboring airbags, For the original pressure, This is the mattress force transmission efficiency coefficient, thereby changing the mechanical environment of the target area.

[0117] This embodiment achieves refined closed-loop control from monitoring to intervention by introducing a quantitative feedback control law based on similarity over-limit amplitude. This mechanism can dynamically adjust the decompression amplitude according to the severity of the risk, which can not only block the pathological evolution of pressure injury in time, but also maintain the necessary body support stability, demonstrating the therapeutic value and automated nursing capabilities of the monitoring system.

[0118] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A multimodal real-time sensing and dynamic risk assessment system for pressure injury monitoring, characterized in that, The system includes: The multimodal sensing terminal module is used to receive and process multi-source heterogeneous data streams of the target area. The multi-source heterogeneous data streams contain time-series data of three modes: mechanical, thermal and bioelectric. The module performs time-series synchronization and alignment on the multi-source heterogeneous data streams to generate standardized real-time fused feature data. The digital twin analysis module is used to construct a virtual tissue dynamic model based on the prior feature data of an individual, and input the stress-related features in the real-time fused feature data into the virtual tissue dynamic model to calculate and output the theoretical reference feature data corresponding to the health baseline state. The pathological simulation injection module is used to call the preset knowledge base of pathological evolution rules of pressure injury, extract the evolution parameter set and inject it into the running logic of the virtual tissue dynamic model, so as to simulate and generate theoretical pathological feature data that characterizes the pathological evolution process. The differential topology decision module is used to perform dual-track differential operation based on the real-time fused feature data, the theoretical reference feature data, and the theoretical pathological feature data to generate a real observation residual vector and a theoretical pathological residual vector, and to generate a dynamic risk assessment result by calculating the topological similarity measure between the real observation residual vector and the theoretical pathological residual vector in the feature space. The digital twin analysis module includes: The parameter initialization unit is used to load the prior characteristic data of an individual, which includes body mass index, Braden score and historical blood perfusion baseline value provided by an external system; The model building unit is used to define and initialize the parameters of a dynamic system model with viscoelastic properties based on the prior feature data. The benchmark generation unit is used to take the time-series pressure features in the real-time fused feature data as model input, drive the dynamic system model to perform calculations, output the theoretical deformation sequence, theoretical temperature change sequence and theoretical impedance sequence under no pathological disturbance, and combine them into the theoretical reference feature data. The model building unit adopts the backward Euler discretization scheme, and the following recursive calculation formula is constructed to solve the theoretical deformation sequence. : ； in, This is the theoretical stress response, driven by real-time pressure data, and its physical meaning is the stress borne by the tissue, with the unit being Pa. This is a theoretical deformation sequence, derived from recursive formula calculations, and its physical meaning is the degree of tissue compression, with a unit of 1. The effective elastic modulus is derived from the BMI index nonlinear mapping function, which is set as follows: ,in, As the reference modulus, The hardening coefficient, in physical terms, represents the elastic stiffness of the tissue. The effective viscosity coefficient is derived from the Braden score correction, and the formula is: ;in, As the reference viscosity coefficient, For risk sensitivity coefficient, This represents the theoretical maximum value of the Braden rating scale. This represents the patient's current actual assessment score.

2. The multimodal real-time sensing and dynamic risk assessment pressure injury monitoring system according to claim 1, characterized in that, The pathological simulation injection module includes: The factor extraction unit is used to query and extract ischemic grade correlation coefficient, inflammatory heat diffusion rate and cell membrane permeability factor from the knowledge base of the pathological evolution rules of the pressure injury to form the evolution parameter set; The superimposed simulation unit is used to superimpose the evolution parameter set onto the running logic of the virtual tissue dynamic model to simulate the conduction and diffusion effects defined by ischemia and inflammation rules, thereby generating the theoretical pathological feature data.

3. The multimodal real-time sensing and dynamic risk assessment pressure injury monitoring system according to claim 1, characterized in that, The differential topology decision module includes: The first difference unit is used to calculate the difference between the real-time fused feature data and the theoretical reference feature data, and generate the real-world observation residual vector containing composite interference; The second difference unit is used to calculate the difference between the theoretical pathological feature data and the theoretical reference feature data, and generate a pure theoretical pathological residual vector. The coupled computation unit is used to map the actual observation residual vector to the feature subspace spanned by the theoretical pathological residual vector, and calculate the cosine similarity between the two as the topological similarity measure.

4. The multimodal real-time sensing and dynamic risk assessment pressure injury monitoring system according to claim 3, characterized in that, The differential topology decision module also includes: A logic determination unit is used to compare the topological similarity metric with a preset risk determination threshold; If the topological similarity metric is higher than the risk determination threshold, the target area is determined to have a risk of pressure damage, and a high-risk alarm command is generated. If the topological similarity metric is lower than the risk determination threshold, the actual observation residual vector is determined to be an acceptable background fluctuation, and a safety status instruction is generated.

5. The multimodal real-time sensing and dynamic risk assessment pressure injury monitoring system according to claim 1, characterized in that, The multimodal sensing terminal module includes: The data interface unit is used to receive time-series pressure data, skin surface temperature distribution map and environmental reference temperature data from external sensing units, as well as complex impedance spectrum data of biological tissues. The feature synchronization unit is used to perform timestamp alignment and data fusion on the multi-source heterogeneous data streams to form standardized real-time fused feature data.

6. The multimodal real-time sensing and dynamic risk assessment pressure injury monitoring system according to claim 1, characterized in that, The system also includes: An edge gateway module, connected upstream of the multimodal sensing terminal module, is used to perform protocol parsing, format unification, and noise filtering on the incoming raw multi-source heterogeneous data stream, and encapsulate the cleaned data into standardized messages for transmission to the multimodal sensing terminal module.

7. The multimodal real-time sensing and dynamic risk assessment pressure injury monitoring system according to claim 4, characterized in that, The system also includes: The feedback adjustment module is used to generate a dynamic adjustment strategy signal for the nursing bed in response to the high-risk alarm command output by the differential topology decision module. The signal can be used to drive an external actuator to change the mechanical environment of the target area.

Images