A responsive closed dressing system based on multimodal sensing and its control method

Abstract

This invention relates to the field of medical and nursing technology, specifically to a responsive occlusive dressing system and its control method based on multimodal sensing. The system includes: real-time acquisition of tubing displacement data, dressing-skin interface pressure data, and fluid characteristic data within the tubing using multimodal sensors deployed on the occlusive dressing; time synchronization, standardization processing, and feature extraction; inputting the fused feature vectors into a pattern recognition model to generate extubation risk information containing risk level and risk type. Based on this risk information, a graded response strategy library is queried to determine the corresponding response action combination, and control instructions containing the execution object, execution sequence, and execution parameters are generated. The system further drives an audible and visual alarm unit and a mechanical locking unit to respectively issue an alarm and activate the tubing fixation mechanism, and obtains execution status signals through feedback sensors to form intervention completion confirmation information, thereby achieving intelligent identification and proactive intervention of extubation risks.

Description

Technical Field

[0001] This invention relates to the field of medical care technology, and in particular to a responsive closed dressing system based on multimodal sensing and its control method. Background Technology

[0002] In clinical nursing settings, various tubes such as catheters, drainage tubes, and breathing lines are widely used in intensive care, postoperative monitoring, and long-term treatment. To protect the puncture site and exit site, occlusive dressings are often used to cover and isolate the area around the tube exit, supplemented by tape, traditional bandages, or simple plastic fixators to secure the tube, reducing the risk of infection and accidental traction. However, when patients are confused, agitated, in pain, or resisting treatment, they are prone to unconscious traction, attempts to pull out the tube, and loosening of the connector due to turning or moving. If this is not identified and intervened in a timely manner, it may cause serious complications, increase the risk of reintubation, and increase the nursing workload.

[0003] Current technologies for managing extubation risks often rely on a combination of single-parameter monitoring and manual inspection. For example, they may simply use bedside monitoring devices to record vital sign waveform changes, simple tension sensors to detect tubing stress, or rely on nurses to periodically observe the appearance of dressings and the tightness of tubing. These solutions often lack multimodal sensing capabilities for the area covered by the closed dressing, failing to simultaneously acquire tubing displacement data, dressing-skin interface pressure data, and fluid characteristics data within the tubing. This leads to delayed risk identification and missed detections. Furthermore, existing alarm methods are typically based on simple threshold triggers, lacking pattern recognition and risk level classification for different risk types such as unconscious traction, patient deliberate extubation attempts, and connector detachment. This results in a high false alarm rate and an inability to link with tiered response strategies to generate refined control commands. At the intervention execution level, current tubing fixation relies heavily on passive structures, lacking responsive actuators linked to extubation risk information. They cannot achieve proactive intervention through the collaboration of audible and visual alarm modules and mechanical locking modules, and they lack a closed-loop mechanism for collecting execution status signals through feedback sensors to form intervention completion confirmation information. Overall, these approaches fall short of the current needs for refined and intelligent nursing management. Summary of the Invention

[0004] This invention provides a responsive closed dressing system and its control method based on multimodal sensing, which solves the problems in existing nursing scenarios such as the inability to obtain tube displacement and interface force information in real time under closed dressing coverage, the reliance on a single parameter for tube removal risk identification leading to a high false alarm and false alarm rate, and the lack of an active intervention mechanism linked to risk level.

[0005] A responsive occlusive dressing system based on multimodal sensing includes the following modules: Multimodal sensor module: Used for real-time acquisition of multimodal sensing data on pipeline status; Pattern recognition module: used to generate extubation risk information based on the multimodal sensing data through a pattern recognition model; The graded response control module is used to generate control commands based on the tube removal risk information and a graded response strategy. Response execution module: used to execute the control commands to realize the intervention of the dressing system.

[0006] A control method for controlling the above-mentioned multimodal sensing-based responsive occlusive dressing system includes the following steps: S1: Real-time acquisition of multimodal sensing data of pipeline status through multimodal sensors, including pipeline displacement data, dressing-skin interface pressure data, and fluid characteristic data within the pipeline; S2: Based on the multimodal sensing data, generate extubation risk information through a pattern recognition model. The extubation risk information includes risk level and risk type. S3: Based on the extubation risk information, generate control instructions through a graded response strategy, including warning instructions and physical intervention instructions; S4: Execute the control command to intervene in the dressing system through the response actuator. The dressing system intervention includes issuing an alarm and activating the pipe fixing mechanism.

[0007] Optionally, S1 includes: S11: Activate the multimodal sensor arranged on the occlusive dressing. The multimodal sensor includes a displacement sensor, a distributed pressure sensor array, and a fluid characteristic sensor. The displacement sensor is used to acquire the original displacement signal, the distributed pressure sensor array is used to acquire the original pressure distribution signal, and the fluid characteristic sensor is used to acquire the original fluid impedance signal. S12: The acquired raw displacement signal, raw pressure distribution signal and raw fluid impedance signal are processed respectively to obtain pipeline displacement data, dressing-skin interface pressure data and pipeline fluid characteristic data. S13: Synchronize and package the pipeline displacement data, the dressing-skin interface pressure data, and the fluid characteristic data inside the pipeline according to a unified time base to form primary multimodal sensing data with time tags; S14: Perform filtering and outlier removal preprocessing on the primary multimodal sensing data with time labels to generate standardized multimodal sensing data.

[0008] Optionally, the processing of the acquired raw displacement signal, raw pressure distribution signal, and raw fluid impedance signal includes: The original displacement signal is integrated to convert the acceleration signal into a displacement, and a digital filtering algorithm is used to eliminate high-frequency noise to obtain the pipeline displacement data. Spatial interpolation and region segmentation are performed on the original pressure distribution signal to extract the local pressure matrix surrounding the pipe outlet region, and the pressure gradient and rate of change of the matrix are calculated to obtain the dressing-skin interface pressure data. The original fluid impedance signal is subjected to spectral analysis to extract the characteristic frequency impedance spectrum corresponding to blood, exudate or air, and the impedance spectrum is matched with a preset fluid template library to obtain the fluid characteristic data in the pipeline that characterizes the fluid type and presence.

[0009] Optionally, S2 includes: S21: Perform time window segmentation and feature extraction on the standardized multimodal sensing data, extract displacement trajectory features and velocity features from the pipeline displacement data, extract pressure distribution change features and pressure center migration features from the dressing-skin interface pressure data, and extract fluid impedance transient features from the fluid characteristic data in the pipeline, thereby generating multiple sets of high-dimensional feature vectors. S22: The multiple sets of high-dimensional feature vectors are spatiotemporally aligned and fused to construct a unified fused feature vector. The fused feature vector fully characterizes the state of the pipeline system defined by changes in pipeline displacement, interface pressure and fluid properties within the time window. S23: Input the fused feature vector into the pre-trained pattern recognition model for forward inference calculation, and output an initial recognition result containing the probabilities of multiple risk categories; S24: Post-process and make decisions on the initial identification results. Based on the preset probability threshold and risk mapping rules, the initial identification results are converted into extubation risk information including risk level and risk type, wherein the risk type includes at least one of unconscious traction, patient deliberate extubation attempt and connector detachment.

[0010] Optionally, the pattern recognition model is a hybrid model combining a deep convolutional neural network enhanced by an attention mechanism and a long short-term memory network. After receiving the fused feature vector, it extracts local spatial features through convolutional layers, captures temporal dependencies through long short-term memory network layers, and uses attention layers to weight the features at different time steps, finally outputting the initial recognition result.

[0011] Optionally, S3 includes: S31: Analyze the extubation risk information and extract the specific risk level and risk type from it; S32: Based on the extracted risk level and risk type, query the pre-set graded response strategy library, match and determine the specific response action combination corresponding to the current risk situation; S33: Based on the determined specific combination of response actions, generate control instructions containing a clear execution object, execution sequence and execution parameters. If the combination of response actions includes a warning, generate a warning instruction; if it includes physical intervention, generate a physical intervention instruction.

[0012] Optionally, the preset hierarchical response strategy library is a tree-like decision logic structure, with the risk level and risk type as inputs and the specific response action combination as outputs. The specific response action combination includes at least one or more of "record only", "local sound and light warning", "remote notification" and "initiate physical intervention".

[0013] Optionally, S4 includes: S41: Receive and parse the control command, decompose it into executable independent operation units, the independent operation units including a first operation unit related to issuing an alarm and a second operation unit related to activating the pipe fixing mechanism; S42: Based on the specific parameters of the first operation unit, drive the audible and visual alarm unit in the response actuator to perform an alarm issuing operation, the alarm issuing operation including activating a buzzer of a specific mode and illuminating an LED indicator of a specific color; S43: Synchronously or according to the specific parameters of the second operation unit, drive the mechanical locking unit in the response actuator to perform the operation of activating the pipe fixing mechanism. The operation of activating the pipe fixing mechanism includes applying current to the shape memory alloy actuator to deform and lock it or energizing the electromagnetic lock to make it engage. S44: After performing the alarm issuance operation and the pipeline fixing mechanism activation operation, the execution status signal is collected by the feedback sensor built into the response actuator, and the execution status signal containing the operation completion status is used as confirmation information of system intervention completion.

[0014] Optionally, receiving and parsing the control command and decomposing it into executable independent operation units specifically includes: identifying the operation type and target module by parsing the command header field according to the preset command encoding rules in the control command, and extracting the parameter information of the command body, thereby generating the first operation unit and the second operation unit.

[0015] The beneficial effects of this invention are: 1. This invention utilizes multimodal sensors to simultaneously collect data on tube displacement, dressing-skin interface pressure, and fluid characteristics within the tube. After standardization, this data forms high-quality multimodal input data, enabling the system to comprehensively capture subtle dynamic changes before extubation. By combining time window segmentation, trajectory feature extraction, pressure center migration calculation, and fluid impedance transient capture, and then using a hybrid pattern recognition model comprised of a deep convolutional neural network, long short-term memory network, and attention mechanism for inference and judgment, the system can reliably identify various extubation risks, including unconscious traction, patient deliberate extubation attempts, and connector detachment. This mechanism effectively improves the sensitivity and specificity of risk identification, providing an accurate basis for subsequent interventions.

[0016] 2. This invention analyzes extubation risk information and matches the optimal combination of response actions based on a hierarchical response strategy library with a tree-like decision logic structure. This allows the system to record or issue local warnings for low-risk events, and to promptly trigger remote notifications or physical interventions for high-risk events. This strategy, through the joint identification of risk level and risk type, ensures that response actions have clear direction and controllable intensity, avoiding false alarms, over-intervention, and patient discomfort caused by the "uniform high-intensity alarm" in traditional systems. This achieves precise nursing intervention and effective utilization of nursing resources.

[0017] 3. This invention, through the parsing of control commands, the generation of independent operating units, the driving of audible and visual alarm units, and the execution of mechanical locking units, enables the system to automatically trigger local audible and visual warnings or activate fixing mechanisms such as shape memory alloy actuators and electromagnetic locks upon detecting a risk of tube removal, achieving immediate braking of potential tube removal attempts. Simultaneously, the response execution module collects the alarm action status and the execution status of the locking mechanism in real time through feedback sensors, generating confirmation information for the completion of system intervention, ensuring that each intervention has a traceable and verifiable record. This closed-loop execution mechanism significantly improves the reliability of tube fixation and the safety of patient care. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only for this invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a schematic diagram of the system flow according to an embodiment of the present invention.

[0020] Figure 2 This is a schematic diagram of the control method according to an embodiment of the present invention. Detailed Implementation

[0021] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. It should also be noted that, to make the embodiments more comprehensive, the following embodiments are the best and preferred embodiments, and those skilled in the art can use other alternative methods to implement some well-known technologies; moreover, the accompanying drawings are only for more specific description of the embodiments and are not intended to specifically limit the present invention.

[0022] like Figure 1 As shown, a responsive occlusive dressing system based on multimodal sensing includes the following modules: Multimodal sensor module: Used for real-time acquisition of multimodal sensing data on pipeline status; Pattern recognition module: used to generate extubation risk information based on multimodal sensor data and a pattern recognition model; Graded response control module: used to generate control commands based on the extubation risk information and through a graded response strategy; Response execution module: Used to execute control commands and implement intervention in the dressing system.

[0023] like Figure 2 As shown, a control method for controlling the above-mentioned multimodal sensing-based responsive occlusive dressing system includes the following steps: S1: Real-time acquisition of multimodal sensing data on the pipeline status using multimodal sensors. This multimodal sensing data includes pipeline displacement data, dressing-skin interface pressure data, and fluid characteristic data within the pipeline. The specific steps are as follows: S11: After the sealing dressing is applied and the pipeline is in a stable and fixed state, the system control unit activates the multimodal sensors arranged on the sealing dressing. The multimodal sensors include displacement sensors, a distributed pressure sensor array, and fluid characteristic sensors.

[0024] The displacement sensor is a combination of a miniature accelerometer and a gyroscope manufactured using MEMS technology. This displacement sensor is mounted near a fixed point close to the pipe outlet. Upon startup, the system control unit sequentially performs accelerometer zero-bias calibration, gyroscope zero-bias calibration, range calibration, and triaxial alignment calibration. Continuous excitation and static sampling are used to determine if the calibration is complete, ensuring that the output triaxial acceleration and angular velocity signals meet the measurement requirements.

[0025] The distributed pressure sensor array consists of several flexible capacitive pressure sensing units, which are arranged in a ring or matrix around the pipe outlet. The system control unit activates each pressure sensing unit sequentially and measures its initial capacitance value through an internal reference capacitor diaphragm to perform pressure baseline calibration, ensuring that the pressure output of all units is at a uniform reference level.

[0026] The fluid characteristic sensor is a fluid impedance sensor based on multi-frequency excitation impedance measurement technology. The system control unit applies a set of multi-frequency excitation signals with fixed amplitudes to the sensor and records the corresponding impedance response curves to generate impedance baseline data for subsequent comparative analysis. After calibration, the sensor can stably output the original fluid impedance signal.

[0027] After completing the above calibration, the multimodal sensor begins to continuously output the original displacement signal, the original pressure distribution signal, and the original fluid impedance signal.

[0028] S12: Processes the original displacement signal, original pressure distribution signal, and original fluid impedance signal respectively, including: 1. Based on the time sequence of the original displacement signals, the triaxial acceleration signals and angular velocity signals are integrated into a unified data stream. The triaxial acceleration signals are integrated using the built-in digital integration method to convert acceleration into velocity; the velocity is then integrated again to convert the velocity into displacement.

[0029] To suppress integration amplification noise, a fixed-coefficient FIR digital filtering algorithm is used to filter the raw acceleration data before integration. This ultimately yields pipeline displacement data characterizing the three-dimensional spatial displacement of the pipeline.

[0030] Before integration, the raw acceleration data is filtered using a fixed-coefficient FIR digital filtering algorithm, as shown below: ; The velocity quantity obtained by integrating acceleration is expressed as: ; The displacement is obtained by integrating the velocity, and is expressed as: ; in, The acceleration signal is processed by FIR digital filtering to reduce high-frequency noise. For the FIR filter A fixed filter coefficient is preset by the control unit and does not change over time. The total number of coefficients used in the FIR filter, corresponding to the sliding window length. For velocity quantity, For displacement, This is the system sampling timestamp.

[0031] 2. The instantaneous pressure values ​​of all pressure units are reorganized into a two-dimensional pressure matrix according to the sensor array deployment order. Then, bilinear spatial interpolation is performed on the pressure matrix to improve its resolution, so as to generate a more refined local pressure distribution. Based on the preset geometric region, the local pressure matrix around the pipe outlet is extracted, and the pressure gradient and the rate of pressure change between consecutive frames are calculated point by point for the data, thereby obtaining the dressing-skin interface pressure data.

[0032] Bilinear spatial interpolation is expressed as: ; The calculated pressure gradient is expressed as: ; The calculated rate of change of pressure is expressed as: ; in, , , , , These are the pressure values ​​of the four pressure cells adjacent to the interpolation point in the original pressure matrix. , These are all interpolation coefficients, determined by the relative positions of the interpolation points in the grid. The pressure values ​​output by the pressure unit form a two-dimensional pressure matrix. For pressure gradient, along direction and The rate of change of pressure in the direction is calculated using neighborhood difference. The rate of change of pressure, This represents the local pressure matrix value for the current sampling period. This represents the local pressure matrix value from the previous sampling period. This represents the time interval between adjacent frames.

[0033] 3. Perform a Fast Fourier Transform (FFT) on the raw fluid impedance signal output from the fluid characteristic sensor to obtain the frequency domain impedance spectrum. Then, extract the peak frequency, impedance amplitude, and phase angle from the impedance spectrum and compare them item by item with the characteristic impedance spectra corresponding to blood, exudate, and air in a pre-set fluid template library. Use the Euclidean distance feature matching algorithm to determine the closest template and generate fluid characteristic data representing the type and presence of fluid in the pipeline.

[0034] S13: Based on the internal master clock, add time tags to the pipe displacement data, dressing-skin interface pressure data, and fluid characteristic data in the pipe.

[0035] Subsequently, time alignment processing was performed on the three types of data with millisecond precision, and non-sampling points were filled in by interpolation so that data under the same time label corresponded to the physical state of the same sampling period.

[0036] The three types of aligned data are then combined according to a predefined data structure to form primary multimodal sensing data with time labels.

[0037] S14: Perform preprocessing on the primary multimodal sensing data with time labels.

[0038] Preprocessing includes: applying median filtering to pipe displacement data, dressing-skin interface pressure data, and pipe fluid characteristic data using a fixed-length sliding time window; calculating the statistical range of each data type based on the 3σ principle, and identifying discrete points exceeding the statistical range as outliers.

[0039] Outliers are removed from the sequence, and linear interpolation is performed using adjacent valid data to fill in the missing data positions, ultimately generating standardized multimodal sensing data.

[0040] S2: Based on multimodal sensor data, generate extubation risk information through a pattern recognition model. The extubation risk information includes risk level and risk type. The specific steps are as follows: S21: After data processing, a time axis structure is established based on a unified sampling frequency, and standardized multimodal sensing data is segmented using fixed-length sliding time windows. The length of each sliding time window is preset by the system, and the sliding step size maintains a fixed ratio with the window length. Feature extraction is then performed, including: 1. Arrange the pipeline displacement data in chronological order within each time window, and accumulate the total length of the displacement path based on the Euclidean distance between consecutive displacement points; The change in direction angle is calculated using the three-dimensional direction vectors at adjacent time points, and then the standard deviation of all direction angle change values ​​within the window is calculated to characterize the stability of the change in motion direction. Meanwhile, the composite velocity is obtained by calculating the ratio of displacement increment to time increment for displacement data within each window, and the average value of all velocity values ​​is calculated as the average composite velocity for that window.

[0041] The total length of the displacement path, the standard deviation of the change in direction angle, and the average composite velocity constitute the displacement trajectory characteristics and velocity characteristics.

[0042] 2. Within each time window, organize the dressing-skin interface pressure data into a sequence of local pressure matrices of consecutive frames. Based on the pressure values ​​of each pressure cell in the matrix, calculate the total pressure, the location of the maximum pressure cell, and the location of the pressure center for each frame. Subsequently, the displacement of the pressure center position of all frames within the window is calculated in chronological order to obtain the pressure center migration characteristics; and the absolute values ​​of the difference matrices between continuous pressure matrices are accumulated to obtain the magnitude of pressure distribution change over time, forming the pressure distribution change characteristics.

[0043] 3. Within each time window, perform point-by-point characteristic analysis on the continuous impedance spectrum sequence, including calculating the changes in peak frequency, peak impedance amplitude, and phase angle. The maximum change within the window is used as a characteristic representing fluid state fluctuations, forming the transient characteristics of fluid impedance.

[0044] Finally, within each time window, multiple sets of high-dimensional feature vectors are generated, including displacement trajectory features, velocity features, pressure distribution change features, pressure center migration features, and fluid impedance transient features.

[0045] S22: The multiple sets of high-dimensional feature vectors extracted in S21 are compared point by point according to their respective attached time labels, and the alignment is achieved based on the time labels.

[0046] After alignment, the high-dimensional feature vectors from three sources within the same time window—pipe displacement, dressing-skin interface pressure, and fluid properties within the pipe—are concatenated according to a preset feature concatenation order to form a unified vector with a higher dimension.

[0047] This unified vector fully characterizes the state of the pipeline system within the time window, defined by changes in pipeline displacement, dressing-skin interface pressure, and fluid properties within the pipeline, thereby constructing a fused feature vector.

[0048] S23: Input the fused feature vector constructed in S22 into the pre-trained pattern recognition model. The pattern recognition model structure includes: 1. Deep Convolutional Neural Network Structure: The input layer of the pattern recognition model receives the fused feature vector and reshapes it into a fixed-dimensional feature tensor. The deep convolutional neural network consists of multiple convolutional layers, activation layers, and pooling layers in sequence. Convolutional layer: Performs one-dimensional convolution operations to extract local feature patterns from the fused feature vector. The number of convolution kernels is preset by the system, and the convolution stride is 1.

[0049] Activation layer: The ReLU activation function is used to improve nonlinear expressive power.

[0050] Pooling layer: Max pooling is used to reduce dimensionality while retaining the most salient features.

[0051] The output of a deep convolutional neural network is a multi-channel local spatial feature map, which is used to describe the local variation structure in the fused features.

[0052] 2. Long Short-Term Memory Network Structure: The feature maps output by the deep convolutional neural network are input into the Long Short-Term Memory (LSTM) network sequentially within a time window. The LSM network comprises an input gate, a forget gate, an output gate, and internal memory units. Input gate: controls which parts of the current feature map are written into the internal memory unit; Forget Gate: Determines the amount of memory retained from the previous time step; Output gate: Generates the corresponding timing output based on the current memory state.

[0053] Long Short-Term Memory (LSTM) networks capture feature dependencies between consecutive time windows through their gating mechanism, enabling the model to distinguish between stable traction, short-term shocks, and gradual pressure trends, thereby reflecting the dynamic patterns of extubation risk formation.

[0054] 3. Attention Mechanism Layer: The output sequence of the Long Short-Term Memory network is input into the attention mechanism layer: The attention layer calculates the similarity between the output at each time step and the global context; Weights are assigned based on similarity, which mathematically highlights the importance of each time step. High-weight time steps correspond to moments with a higher risk of potential extubation events.

[0055] The output after processing by the attention layer is a set of weighted feature vectors, which can highlight key signal segments.

[0056] 4. Output layer: The output of the attention layer is mapped to the probability values ​​of multiple risk categories through a fully connected layer to obtain the initial recognition result.

[0057] The pattern recognition model was trained using supervised learning. During training, multimodal sensor data from multiple subjects under different clinical conditions were first collected and segmented according to fixed time windows. At the same time, each time window was labeled with a corresponding risk type, including "unconscious traction," "patient deliberate extubation attempt," and "connector detachment," and the corresponding risk level label was recorded.

[0058] Subsequently, the fused feature vectors consistent with the system operation phase are used as model inputs. These feature vectors are composed of a combination of displacement trajectory features, velocity features, pressure distribution change features, pressure center migration features, and fluid impedance transient features.

[0059] During training, the model extracts local spatial features through a deep convolutional neural network, captures temporal dependencies through a long short-term memory network, and weights features at different time steps through an attention mechanism layer to enhance the impact of key time segments on the model output. The difference between the predicted probabilities of each risk category output by the model and the true labels is measured using the cross-entropy loss function. The Adam optimization algorithm is used to iteratively update the network parameters during training, and the learning rate is decayed according to the training epochs to improve training stability.

[0060] After all training rounds are completed and the convergence condition is met, the trained network weights and structural parameters are solidified and stored in the system control unit for forward inference calculations during the runtime phase.

[0061] S24: Post-process the initial identification results based on preset probability thresholds. For the probability of the "unconscious traction" category in the initial identification results, when its value exceeds the first threshold but does not reach the second threshold, the system determines the risk level in the extubation risk information of that time window as "low risk" and the risk type as "unconscious traction".

[0062] When the probability of the "patient deliberate extubation attempt" category in the initial identification results exceeds the second threshold, the system determines the extubation risk information in that time window as "high risk" and classifies the risk type as "patient deliberate extubation attempt".

[0063] After the judgment is completed according to the risk mapping rules, the extubation risk information containing the risk level and risk type is output for subsequent graded response control.

[0064] S3: Based on the extubation risk information, control instructions are generated through a tiered response strategy. These control instructions include warning instructions and physical intervention instructions. The specific steps are as follows: S31: Upon receiving the tube removal risk information output by S2, the system first parses and processes it. The tube removal risk information is encapsulated in a predefined encoding format, which includes a risk level field and a risk type field. According to preset data parsing rules, the information packet is decoded field by field, and the risk level code and risk type code encapsulated in different byte segments are read and converted into recognizable text or number forms respectively.

[0065] The decoding process includes three operations: 1. Locate the risk level field from the tube removal risk information according to the field's starting position and length; 2. According to the field coding standard, convert the code of the risk level field into the corresponding risk level value; 3. Locate the risk type field in the same way and encode it into the corresponding risk type value.

[0066] Through the above analysis, a single and clear risk level and risk type can be obtained, providing input basis for subsequent matching of graded response strategies.

[0067] S32: After obtaining the risk level and risk type, these are sent as input parameters to a pre-defined hierarchical response strategy library. The hierarchical response strategy library is constructed using a tree-like decision logic structure. Its root node defines the entry point for the combination of risk level and risk type, and each level node forms a decision path according to different branches of the risk level range and risk type category.

[0068] The risk level and risk type are matched layer by layer down the tree structure from the root node until a specific leaf node is reached. This leaf node records the specific combination of response actions corresponding to that risk state. The combination of response actions includes at least one or more of the following: "Record Only," "Local Audible and Visual Warning," "Remote Notification," and "Initiate Physical Intervention."

[0069] The entire sequence of actions is read to ensure that all response actions are recorded for use in generating corresponding control commands. This guarantees that the output sequence of response actions is completely consistent with the nature and severity of the current extubation risk.

[0070] S33: After obtaining the response action combination determined in S32, analyze each response action in the combination one by one, and generate the corresponding control instruction according to the predefined action parameter template. The generated control instruction includes three dimensions: execution object, execution sequence, and execution parameters, each of which corresponds to a clearly defined field within the system.

[0071] 1. When the action combination includes "local audible and visual alarm", the audible and visual alarm is selected as the execution target, immediate execution is selected as the execution sequence, and the execution parameters are assigned values ​​according to the alarm parameter table. This parameter table includes fixed flashing frequency, light intensity level, and sound tone mode. These parameter combinations are encoded into control fields, and finally, an alarm command containing the above execution parameters is generated, targeting the audible and visual alarm.

[0072] 2. When the action combination includes "initiate physical intervention", the intelligent fixing structure is used as the execution object, the execution sequence is set to immediate execution, and the tightening torque, tightening speed, and target fixing position are written into the control field according to the preset mechanical intervention parameters. The above parameters are combined into a physical intervention command, causing the intelligent fixing structure to perform the fixing action according to the command, thereby reducing the risk of impending tube removal.

[0073] 3. If the action combination includes "record only", the tube removal risk information will be recorded to the local log file according to the preset format, and a recording instruction will be generated, specifying that the storage unit is the execution object, immediate writing is the execution sequence, and the current risk information is the execution parameter.

[0074] If the action combination includes "remote notification", the remote monitoring terminal is used as the execution target, the execution timing is set to send immediately, and the execution parameters consist of the fields required in the remote notification protocol, including risk level, risk type and current timestamp, thereby generating a remote notification instruction.

[0075] Finally, all the generated instructions are combined in sequence to form a complete set of control instructions, which is used to schedule subsequent response execution modules to perform actual intervention.

[0076] S4: Execute control commands to intervene in the dressing system via the response actuator. Dressing system intervention includes issuing alarms and activating the pipe fixing mechanism. Specific steps are as follows: S41: After generating the S3 control command, the control command is sent to the response execution module. Upon receiving the control command, the response execution module first parses the control command according to the command encoding rules. The parsing process includes reading the command header fields to identify the operation type and target module of the control command; then reading the command body fields to extract the execution object, execution timing, and execution parameters.

[0077] Based on the different content of the instructions, the parsed control instructions are split into two independent operation units: 1. A first operating unit related to issuing an alarm, used to drive the audible and visual alarm unit to perform buzzer and LED indicator light illumination operations; 2. A second operating unit associated with activating the pipe fixing mechanism, used to drive the mechanical locking unit to perform the locking action of the shape memory alloy actuator or electromagnetic lock.

[0078] After the split is completed, the two operation units are stored in the task queue to be executed, for subsequent scheduling and execution.

[0079] S42: Read the execution parameters from the first operation unit, including alarm level, buzzer operating frequency, buzzer duty cycle, LED indicator color, LED flashing frequency and flashing sequence. The system control unit then sends a control signal to the audible and visual alarm unit.

[0080] After receiving the control signal, the buzzer in the audible and visual alarm unit performs a buzzing action according to the working frequency and duty cycle provided by the first operation unit, generating a continuous or intermittent sound signal; after receiving the control signal, the LED indicator light performs light emission according to the color and flashing frequency provided by the first operation unit, presenting a light signal mode corresponding to the current risk level.

[0081] The buzzer and LED indicator light operate simultaneously or sequentially under the scheduling of the system control unit, forming a complete alarm operation.

[0082] S43: When the second operation unit contains a physical intervention initiation command, the execution parameters are read, including locking force parameters, locking timing parameters, pulse current intensity, and pulse duration. Depending on the type of fixing mechanism, the system control unit executes the following drive process: 1. When the second operating unit specifies the use of a shape memory alloy actuator as the fixing mechanism, the system control unit calculates the pulse current intensity required for actuation based on the locking force parameters, and controls the drive circuit to apply a pulse current of that intensity and duration to the shape memory alloy actuator. Under the action of current heating, the shape memory alloy undergoes a crystal phase transformation, producing shrinkage deformation, which drives the mechanical latch to perform a locking action, thus stably fixing the pipe within the sealing dressing.

[0083] (2) When the second operating unit specifies the use of an electromagnetic lock as a fixing mechanism, the system control unit energizes the electromagnetic lock coil according to the execution parameters, so that the electromagnetic lock generates an attraction force, drives the latch to make tight contact with the fixing surface, thereby completing the locking action.

[0084] The system control unit ensures that the locking force and execution timing strictly conform to the parameter requirements in the second operation unit throughout the entire execution process, so as to ensure that the intervention action has an immediate and effective suppression effect on potential extubation risks.

[0085] S44: After the operations in S42 and S43 are completed, the response execution module collects the execution status signal through the built-in feedback sensors. The feedback sensors include a photosensitive sensing unit and a current detection unit for detecting the operating status of the audible and visual alarm unit, and a displacement sensing unit or strain detection unit for detecting the action status of the mechanical locking unit.

[0086] The feedback sensor transmits the collected execution status signal to the system control unit. Based on this signal, the system control unit determines whether the alarm issuance and pipeline fixing mechanism activation operations were completed as expected, and generates confirmation information containing the operation completion status. This confirmation information serves as the final marker of system intervention completion, providing a status basis for subsequent system operation.

[0087] This invention encompasses any substitutions, modifications, equivalent methods, and solutions made within the spirit and scope of this invention. To provide the public with a thorough understanding of this invention, specific details are described in detail in the following preferred embodiments; however, those skilled in the art will fully understand the invention even without these details. Furthermore, to avoid unnecessary misunderstanding of the essence of this invention, well-known methods, processes, procedures, components, and circuits are not described in detail.

[0088] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A multi-modal sensor based responsive occlusive dressing system, characterized in that, include: Multimodal sensor module: Used for real-time acquisition of multimodal sensing data on pipeline status; Pattern recognition module: used to generate extubation risk information based on the multimodal sensing data through a pattern recognition model; The graded response control module is used to generate control commands based on the tube removal risk information and a graded response strategy. Response execution module: used to execute the control commands to realize the intervention of the dressing system.

2. A control method for controlling the responsive occlusive dressing system based on multimodal sensing as described in claim 1, characterized in that, Includes the following steps: Multimodal sensing data of the pipeline status are collected in real time by multimodal sensors. The multimodal sensing data includes pipeline displacement data, dressing-skin interface pressure data, and fluid characteristic data inside the pipeline. Based on the multimodal sensing data, extubation risk information is generated through a pattern recognition model. The extubation risk information includes risk level and risk type. Based on the extubation risk information, control instructions are generated through a tiered response strategy, including early warning instructions and physical intervention instructions. The control commands are executed to intervene in the dressing system via a response actuator. The dressing system intervention includes issuing an alarm and activating the pipe fixing mechanism.

3. The responsive closed dressing system and its control method based on multimodal sensing according to claim 2, characterized in that, S1 includes: A multimodal sensor arranged on a closure dressing is activated. The multimodal sensor includes a displacement sensor, a distributed pressure sensor array, and a fluid characteristic sensor. The displacement sensor is used to acquire raw displacement signals, the distributed pressure sensor array is used to acquire raw pressure distribution signals, and the fluid characteristic sensor is used to acquire raw fluid impedance signals. The original displacement signal, original pressure distribution signal, and original fluid impedance signal were processed to obtain pipeline displacement data, dressing-skin interface pressure data, and fluid characteristic data within the pipeline. The pipeline displacement data, the dressing-skin interface pressure data, and the fluid characteristic data inside the pipeline are synchronized and packaged according to a unified time base to form primary multimodal sensing data with time tags. The primary multimodal sensing data with time tags is preprocessed by filtering and outlier removal to generate standardized multimodal sensing data.

4. The responsive occlusive dressing system and its control method based on multimodal sensing according to claim 3, characterized in that, The processing of the acquired raw displacement signal, raw pressure distribution signal, and raw fluid impedance signal includes: The original displacement signal is integrated to convert the acceleration signal into a displacement, and a digital filtering algorithm is used to eliminate high-frequency noise to obtain the pipeline displacement data. Spatial interpolation and region segmentation are performed on the original pressure distribution signal to extract the local pressure matrix surrounding the pipe outlet region, and the pressure gradient and rate of change of the matrix are calculated to obtain the dressing-skin interface pressure data. The original fluid impedance signal is subjected to spectral analysis to extract the characteristic frequency impedance spectrum corresponding to blood, exudate or air, and the impedance spectrum is matched with a preset fluid template library to obtain the fluid characteristic data in the pipeline that characterizes the fluid type and presence.

5. The responsive occlusive dressing system and its control method based on multimodal sensing according to claim 4, characterized in that, S2 includes: The standardized multimodal sensing data is segmented into time windows and features are extracted. Displacement trajectory features and velocity features are extracted from the pipeline displacement data. Pressure distribution change features and pressure center migration features are extracted from the dressing-skin interface pressure data. Fluid impedance transient features are extracted from the fluid characteristic data in the pipeline, thereby generating multiple sets of high-dimensional feature vectors. The multiple sets of high-dimensional feature vectors are spatiotemporally aligned and fused to construct a unified fused feature vector; The fused feature vector is input into a pre-trained pattern recognition model for forward inference calculation, and an initial recognition result containing the probabilities of multiple risk categories is output. The initial identification results are post-processed and decision-making is performed. Based on the preset probability threshold and risk mapping rules, the initial identification results are converted into extubation risk information including risk level and risk type, wherein the risk type includes at least one of unconscious traction, patient deliberate extubation attempt and connector detachment.

6. The responsive closed dressing system and its control method based on multimodal sensing according to claim 5, characterized in that, The pattern recognition model is a hybrid model combining a deep convolutional neural network enhanced by an attention mechanism and a long short-term memory network. After receiving the fused feature vector, it extracts local spatial features through convolutional layers, captures temporal dependencies through long short-term memory network layers, and uses attention layers to weight the features at different time steps, finally outputting the initial recognition result.

7. The responsive occlusive dressing system and its control method based on multimodal sensing according to claim 6, characterized in that, S3 includes: Analyze the extubation risk information to extract the specific risk level and risk type; Based on the extracted risk level and risk type, query the pre-set graded response strategy library, match and determine the specific response action combination corresponding to the current risk situation; Based on the determined specific combination of response actions, control instructions containing a clear execution object, execution sequence, and execution parameters are generated. If the combination of response actions includes a warning, a warning instruction is generated; if it includes physical intervention, a physical intervention instruction is generated.

8. The responsive occlusive dressing system and its control method based on multimodal sensing according to claim 7, characterized in that, The pre-set hierarchical response strategy library has a tree-like decision logic structure. Its input is the risk level and the risk type, and its output is the specific response action combination. The specific response action combination includes at least one or more of the following: "record only", "local sound and light warning", "remote notification" and "initiate physical intervention".

9. A responsive occlusive dressing system and its control method based on multimodal sensing according to claim 8, characterized in that, S4 includes: The control command is received and parsed, and decomposed into executable independent operation units, including a first operation unit related to issuing an alarm and a second operation unit related to activating the pipeline fixing mechanism. Based on the specific parameters of the first operation unit, the audible and visual alarm unit in the response actuator is driven to perform an alarm operation, which includes activating a buzzer in a specific mode and illuminating an LED indicator of a specific color. Synchronously or according to the specific parameters of the second operation unit, drive the mechanical locking unit in the response actuator to perform the operation of activating the pipe fixing mechanism. The operation of activating the pipe fixing mechanism includes applying current to the shape memory alloy actuator to deform and lock it, or energizing the electromagnetic lock to make it attract. After performing the alarm issuance operation and the pipeline fixing mechanism activation operation, the execution status signal is collected by the feedback sensor built into the response actuator, and the execution status signal containing the operation completion status is used as confirmation information of system intervention completion.

10. A responsive occlusive dressing system and its control method based on multimodal sensing according to claim 9, characterized in that, The step of receiving and parsing the control command and decomposing it into executable independent operation units specifically includes: identifying the operation type and target module by parsing the command header field according to the preset command encoding rules in the control command, and extracting the parameter information of the command body, thereby generating the first operation unit and the second operation unit.

Images