Intelligent stoma chassis intelligent monitoring system

Abstract

The invention discloses an intelligent stoma base plate intelligent monitoring system which dynamically predicts the remaining time of leakage by performing multi-dimensional dynamic fusion on humidity, pH, temperature and leakage sensor signals reflecting the state of a microenvironment below a stoma base plate and real-time activity state data reflecting the physical stress of a user. And further determining whether to trigger an alarm or not based on comparison between the predicted residual time of leakage occurrence and a preset alarm threshold value. By means of the mode, the system can accurately recognize and quantify complex early comprehensive symptoms of leakage which vary from person to person, precious processing time is won for patients, and nursing experience and life quality of the patients are remarkably improved.

Description

Technical Field

[0001] This application relates to the field of intelligent monitoring, and more specifically, to an intelligent monitoring system for an intelligent stoma chassis. Background Technology

[0002] Stoma surgery is a common treatment for certain intestinal or urinary system diseases (such as colorectal cancer and bladder cancer). It involves creating an opening called a "stomach" in the patient's abdominal wall to allow the intestines or urinary tract to be drained from the body. Stoma patients need to wear an ostomy bag and a stoma baseplate to secure it for extended periods. However, accidental leakage of the stoma baseplate is a major problem affecting patients' quality of life. Once leakage occurs, excrement directly contacts and irritates the skin around the stoma, causing complications such as dermatitis and infection, resulting in physical pain and psychological burden for the patient. Therefore, developing an intelligent system capable of real-time monitoring of the stoma baseplate status and accurately predicting and providing early warnings of leakage risks is of crucial practical significance for improving the quality of patient care and experience.

[0003] Existing monitoring solutions often rely on relatively simple sensing technologies, such as using a single humidity sensor to monitor humidity changes at the edge of the stoma base. When the detected humidity value exceeds a fixed, preset threshold, the system triggers an alarm. This approach fails to adequately address two core challenges. First, patients exhibit significant individual variability; the characteristics of excrement, daily activities, and perspiration levels vary considerably among patients, making a uniform leakage assessment standard difficult to apply to all users. Second, the precursors to leakage are complex and insidious. Leakage is not an instantaneous event but a gradual process, often resulting from the combined effects of multiple factors, such as a slow rise in local temperature and pH changes caused by excrement. However, traditional single-sensor solutions cannot capture and comprehensively analyze this multi-dimensional precursor information, often only reacting after leakage has actually occurred and humidity has significantly increased, thus losing their warning significance.

[0004] Therefore, we look forward to an optimized intelligent monitoring system for stoma chassis. Summary of the Invention

[0005] To address the aforementioned technical problems, this application is proposed. Embodiments of this application provide an intelligent stoma chassis monitoring system. This system dynamically fuses multi-dimensional data—including humidity, pH, temperature, and leakage sensor signals reflecting the microenvironment beneath the stoma chassis—with real-time activity data reflecting the user's physical stress. This allows for the dynamic prediction of the remaining time before leakage occurs. Furthermore, based on a comparison between the predicted remaining time and a preset alarm threshold, the system determines whether to trigger an alarm. In this way, the system can accurately identify and quantify complex early leakage syndrome symptoms that vary from person to person, saving patients valuable treatment time and significantly improving their nursing experience and quality of life.

[0006] According to one aspect of this application, an intelligent stoma chassis intelligent monitoring system is provided, comprising: The data acquisition module is used to collect the current humidity value, current pH value, current temperature value, and maximum signal strength of the leakage sensor through the humidity sensor, pH sensor, temperature sensor, and leakage sensor. The patient activity status classification module is used to acquire raw triaxial acceleration data collected by the accelerometer and determine patient activity status classification data based on the raw triaxial acceleration data. The leakage time prediction module is used to input the current humidity value, current pH value, current temperature value, maximum signal strength of leakage sensor and patient activity status classification data as real-time basic features into the locally deployed leakage prediction model to obtain the predicted remaining time of leakage. The warning triggering module is used to determine whether to trigger an alarm based on a comparison between the predicted remaining time before a leak occurs and a preset alarm threshold.

[0007] Compared with existing technologies, the intelligent stoma chassis monitoring system provided in this application dynamically integrates multi-dimensional data such as humidity, pH, temperature, and leakage sensor signals reflecting the microenvironment beneath the stoma chassis with real-time activity data reflecting the user's physical stress. This allows for the dynamic prediction of the remaining time before leakage occurs. Furthermore, based on a comparison between the predicted remaining time and a preset alarm threshold, the system determines whether to trigger an alarm. In this way, the system can accurately identify and quantify complex early leakage syndrome symptoms that vary from person to person, saving patients valuable treatment time and significantly improving their nursing experience and quality of life. Attached Figure Description

[0008] The above and other objects, features, and advantages of this application will become more apparent from the more detailed description of the embodiments of this application in conjunction with the accompanying drawings. The drawings are provided to further illustrate the embodiments of this application and form part of the specification. They are used together with the embodiments of this application to explain this application and do not constitute a limitation thereof. In the drawings, the same reference numerals generally represent the same components or steps.

[0009] Figure 1 This is a block diagram of an intelligent stoma chassis intelligent monitoring system according to an embodiment of this application; Figure 2 This is a schematic diagram of the data flow of the intelligent stoma chassis intelligent monitoring system according to an embodiment of this application; Figure 3 This is a block diagram of the patient activity status classification module in the intelligent stoma chassis intelligent monitoring system according to an embodiment of this application. Detailed Implementation

[0010] Hereinafter, exemplary embodiments according to this application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments of this application. It should be understood that this application is not limited to the exemplary embodiments described herein.

[0011] As indicated in this application and claims, unless the context clearly indicates otherwise, the words "a," "an," "an," and / or "the" are not specifically singular and may include plural forms. Generally speaking, the terms "comprising" and "including" only indicate the inclusion of explicitly identified steps and elements, which do not constitute an exclusive list, and the method or apparatus may also include other steps or elements.

[0012] While this application makes various references to certain modules of the systems according to embodiments of this application, any number of different modules can be used and run on user terminals and / or servers. The modules described are merely illustrative, and different aspects of the systems and methods may use different modules.

[0013] Flowcharts are used in this application to illustrate the operations performed by the system according to embodiments of this application. It should be understood that the preceding or following operations are not necessarily performed in exact order. Instead, various steps can be processed in reverse order or simultaneously as needed. Furthermore, other operations can be added to these processes, or one or more steps can be removed from them.

[0014] Hereinafter, exemplary embodiments according to this application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments of this application. It should be understood that this application is not limited to the exemplary embodiments described herein.

[0015] The technical solution of this application proposes an intelligent monitoring system for an intelligent stoma chassis. Figure 1 This is a block diagram of an intelligent stoma chassis intelligent monitoring system according to an embodiment of this application. Figure 2 This is a schematic diagram of the data flow of the intelligent stoma chassis intelligent monitoring system according to an embodiment of this application. Figure 1 and Figure 2 As shown, the intelligent stoma chassis intelligent monitoring system 300 according to an embodiment of this application includes: a data acquisition module 310, used to acquire the current humidity value, current pH value, current temperature value, and maximum signal strength of the leakage sensor through a humidity sensor, a pH sensor, a temperature sensor, and a leakage sensor; a patient activity status classification module 320, used to acquire raw triaxial acceleration data collected by an accelerometer, and determine patient activity status classification data based on the raw triaxial acceleration data; a leakage time prediction module 330, used to input the current humidity value, current pH value, current temperature value, maximum signal strength of the leakage sensor, and patient activity status classification data as real-time basic features into a locally deployed leakage prediction model to obtain the predicted remaining time for leakage; and an alarm triggering module 340, used to determine whether to trigger an alarm based on a comparison between the predicted remaining time for leakage and a preset alarm threshold.

[0016] Specifically, the data acquisition module 310 is used to collect the current humidity value, current pH value, current temperature value, and maximum signal intensity of the leakage sensor through a humidity sensor, pH sensor, temperature sensor, and leakage sensor. It should be understood that humidity, pH value, and temperature are key physiological and chemical indicators for measuring the health status of the skin around the stoma and the corrosiveness of excrement, while the signal from the leakage sensor directly reflects whether the excrement has begun to erode the chassis colloid. Therefore, in the technical solution of this application, for various physical and chemical data regarding the chassis status and excrement condition, the current humidity value, current pH value, current temperature value, and maximum signal intensity of the leakage sensor are collected through a humidity sensor, pH sensor, temperature sensor, and leakage sensor.

[0017] In practice, after the patient has fitted the stoma baseplate, the system initiates real-time monitoring. The data acquisition module periodically and proactively acquires measurement readings from various sensors at a preset frequency. This period can be set according to actual needs, such as once per minute. Within each acquisition cycle, the data acquisition module sequentially or in parallel sends instructions to the humidity sensor, pH sensor, temperature sensor, and leakage sensor, and receives the measurement results returned by each sensor.

[0018] Taking the scheme of this application as an example, after the patient wears the stoma baseplate, the system begins continuous real-time monitoring: Sensor data acquisition: The data acquisition module on the edge device periodically (e.g., once per minute) acquires data from various sensors: Humidity sensor: acquires the current humidity value, such as 65%. pH sensor: acquires the current pH value, such as 6.8. Temperature sensor: acquires the current temperature value, such as 37.2℃. Leakage sensor: acquires and processes the maximum signal strength of the leakage sensor, such as 25μA. This example demonstrates a complete set of real-time basic characteristics obtained by the system from four different sensors through its data acquisition module at a specific point in time. These specific values ​​will serve as direct inputs for the next step of classifying patient activity status and predicting leakage risk.

[0019] Specifically, the patient activity status classification module 320 is used to acquire raw triaxial acceleration data collected by an accelerometer and determine patient activity status classification data based on the raw triaxial acceleration data. It should be understood that activities of varying intensities significantly affect the fit of the stoma baseplate and abdominal pressure, and are one of the important contributing factors to leakage. For example, strenuous activity may cause baseplate displacement or a sudden increase in abdominal pressure, thereby greatly increasing the risk of leakage. Therefore, in the technical solution of this application, patient activity status classification data is input as a key real-time basic feature into a locally deployed leakage prediction model to obtain the predicted remaining time before leakage occurs.

[0020] In the embodiments of this application, firstly, raw triaxial acceleration data collected by the accelerometer is acquired. Specifically, when the patient wears the stoma baseplate integrated with the sensor, the microcontroller inside the system initializes and activates the accelerometer. An accelerometer is a sensor that can measure the acceleration of an object itself, typically measuring along three mutually perpendicular axes: X, Y, and Z. The system software, through a corresponding driver, continuously reads data from the accelerometer's hardware registers at a very high sampling frequency (e.g., tens or hundreds of times per second). Each read yields a data point containing three values, representing the acceleration values ​​on the X, Y, and Z axes at that instant. This continuously generated data stream constitutes the raw triaxial acceleration data.

[0021] An accelerometer is a microelectromechanical system (MEMS) sensor used to measure the acceleration of a device itself, including static gravitational acceleration and dynamic motion acceleration. It is a core piece of hardware for capturing user body posture and motion information.

[0022] Next, in a specific example of this application, such as Figure 3As shown, the patient activity state classification module 320 includes: a temporal encoding unit 321, used to temporally encode the original triaxial acceleration data to obtain X-axis acceleration temporal pattern feature encoding vectors, Y-axis acceleration temporal pattern feature encoding vectors, and Z-axis acceleration temporal pattern feature encoding vectors; a temporal co-coding unit 322, used to calculate the XY-axis acceleration temporal co-coding vector by dividing the XY-axis acceleration temporal pattern feature encoding vector by the position point; a multidimensional state co-response inference unit 323, used to perform multidimensional state co-response inference on the XY-axis acceleration temporal co-coding vector and the Z-axis acceleration temporal pattern feature encoding vector to obtain a multidimensional acceleration temporal co-coding vector; and a state classification unit 324, used to generate patient activity state classification data based on the multidimensional acceleration temporal co-coding vector.

[0023] Specifically, the timing encoding unit 321 is used to perform timing encoding on the original triaxial acceleration data to obtain X-axis acceleration timing pattern feature encoding vectors, Y-axis acceleration timing pattern feature encoding vectors, and Z-axis acceleration timing pattern feature encoding vectors. In the embodiments of this application, firstly, the original triaxial acceleration data is processed to obtain X-axis acceleration time series, Y-axis acceleration time series, and Z-axis acceleration time series. It should be understood that the original triaxial acceleration data is a set of discrete data points generated continuously at a very high frequency, with each data point containing components of the three axes. This form is too fragmented for directly revealing motion patterns. Therefore, in the technical solution of this application, the original triaxial acceleration data is processed to organize the data into three independent time series representing specific time windows. In this way, the system can focus on the motion change patterns in the three orthogonal directions of X, Y, and Z, laying a solid data foundation for the subsequent accurate extraction of timing pattern features in each axis.

[0024] In practice, the system first performs data caching and window segmentation. At the hardware level, the accelerometer continuously generates triaxial acceleration readings at a fixed sampling rate (e.g., 50Hz) and uses a sliding window mechanism to capture data. That is, a fixed time length is defined, for example, the last 30 seconds. When the amount of data accumulated in the buffer reaches the length of this time window, the system treats all data points within this window as a complete processing batch. Next, the data is separated along the axes. For a set of all raw triaxial acceleration data points collected within a defined time window, for example, if the sampling rate is 50Hz and the window is 30 seconds, 1500 data tuples will be obtained. The data processing program iterates through each tuple in this set, separating its three components, and finally outputting the X-axis acceleration time series, Y-axis acceleration time series, and Z-axis acceleration time series. These three time series have the same length, and their internal elements are strictly ordered by time, together forming a complete and structured raw description of the patient's motion state within the time window, which serves as input for the next step of the time-series coding model.

[0025] Furthermore, the X-axis, Y-axis, and Z-axis acceleration time series are respectively encoded using an LSTM-based acceleration time series coding model to obtain X-axis, Y-axis, and Z-axis acceleration time series pattern feature encoding vectors. It is understandable that directly analyzing raw time series data presents numerous challenges: the data points are numerous, contain a large amount of noise and redundant information, and it is difficult to directly reveal complex dynamic patterns such as walking rhythm and body swaying frequency. Therefore, in the technical solution of this application, the X-axis, Y-axis, and Z-axis acceleration time series are encoded using LSTM-based time series coding to extract deep features that effectively represent core motion patterns from the raw, noisy, and high-dimensional acceleration time series. LSTM, as a special type of recurrent neural network, is designed with an internal structure that makes it highly adept at capturing and learning long-term dependencies in time series data, effectively identifying representative motion patterns from continuous acceleration changes. This encoding method can compress and extract the complex dynamic information contained in the original data into a feature vector with extremely high information density, which greatly facilitates the subsequent model processing and improves the accuracy of the final activity state classification.

[0026] Specifically, the temporal co-coding unit 322 is used to calculate the XY-axis acceleration temporal pattern feature encoding vector by dividing the X-axis acceleration temporal pattern feature encoding vector by its position point. It should be understood that mining and quantifying the intrinsic coupling relationship of a patient's motion on a horizontal plane (i.e., the plane formed by the X and Y axes) is crucial for distinguishing different types of activities. For example, in actions such as steady walking, running, turning, or lateral movement, there is a specific and stable proportional or co-cooperative relationship between the motion patterns on the X-axis (usually representing the forward-backward direction) and Y-axis (usually representing the left-right direction). Traditional feature fusion methods, such as simple vector concatenation or addition, while integrating information, often overlook this deep-seated proportional relationship. Therefore, in the technical solution of this application, this fine-grained co-cooperative pattern is captured by calculating the ratio of feature vectors in each dimension. This approach generates a new feature that is more sensitive to changes in horizontal plane motion posture, thus providing more discriminative information for subsequent activity state classification than single-axis features, significantly improving the accuracy and robustness of classification.

[0027] Specifically, the multidimensional state-coordinated response inference unit 323 is used to perform multidimensional state-coordinated response inference on the XY-axis acceleration temporal co-encoding vector and the Z-axis acceleration temporal pattern feature encoding vector to obtain a multidimensional acceleration temporal co-encoding vector. It should be understood that instantaneous actions such as the patient's posture shift from lying down to walking will trigger a dynamic reconstruction of the stoma floor's stress state, thereby affecting the migration rate and direction of excrement in the floor gap. Traditional solutions simply splice or weighted average the three-axis acceleration data, ignoring the physical nature of the cross-axis coupling effect in kinematics, such as the nonlinear phase difference between XY-plane rotational motion and Z-axis linear acceleration. This coupling relationship precisely characterizes the potential impact of body center of gravity shift on the floor edge sealing. Therefore, in the technical solution of this application, by modeling the interaction between the XY-axis co-encoding vector and the Z-axis feature vector as a progressive inference chain, the actual path of motion posture on the floor microenvironment is reproduced at the algorithm level. This allows the leakage time prediction module to proactively adapt to patient behavior patterns, avoiding misjudging normal activities as leakage risks. At the same time, it dynamically adjusts the warning threshold when high-risk actions occur, significantly improving the system's robustness in real-life scenarios.

[0028] Specifically, firstly, the global temporal fine-grained response matrix of acceleration is calculated between the XY-axis acceleration temporal co-encoding vector and the Z-axis acceleration temporal pattern feature encoding vector. It should be understood that when a patient performs a complex movement (such as bending over to pick up an object), the change in trunk rotational angular momentum represented by the XY-plane acceleration co-encoding vector dynamically modulates the center of gravity migration trajectory reflected by the Z-axis linear acceleration. Traditional methods process the three-axis data independently or simply superimpose them, failing to quantify this cross-dimensional dynamic stress transmission relationship. Therefore, in the technical solution of this application, by calculating the global response matrix of the co-encoding vector relative to the Z-axis feature vector, a transmission coefficient field of planar motion patterns on vertical forces is constructed mathematically, thereby revealing the mechanical transmission path of micro-deformation at the chassis edge during changes in body posture. This fine-grained mapping breaks through the superficial limitations of traditional motion recognition: each element in the matrix represents the physical rules of how a specific planar motion pattern excites a vertical dynamic response at a specific moment, providing an interpretable physical rule carrier for subsequent LSTM inference.

[0029] In other words, the specific process of calculating the acceleration global temporal fine-grained response matrix of the XY-axis acceleration temporal co-coding vector relative to the Z-axis acceleration temporal pattern feature coding vector in the technical solution of this application includes: First, the acceleration temporal interaction matrix of the XY-axis acceleration temporal co-coding vector relative to the Z-axis acceleration temporal pattern feature coding vector is calculated. The process is expressed by the formula: , in, It is the XY-axis acceleration temporal co-coding vector. It is the transpose of the Z-axis acceleration time-series pattern feature encoding vector. This represents vector multiplication. This indicates dot product by position. express function, It is a dynamic weight matrix. It is the acceleration time-series interaction matrix; Next, the acceleration timing gating matrix of the XY-axis acceleration timing co-coding vector relative to the Z-axis acceleration timing pattern feature coding vector is calculated. The process is expressed by the formula: , in, Let be the projection matrix. express function, It is the acceleration timing gating matrix; Furthermore, the temporal response matrix of the acceleration temporal interaction matrix and the acceleration temporal gating matrix is ​​calculated to obtain the global temporal fine-grained response matrix of acceleration. The process is expressed by the following formula: , in, This indicates addition by position. It is the global temporal fine-grained response matrix of acceleration.

[0030] Next, the global temporal fine-grained response matrix of acceleration is decomposed into column vectors to obtain the sequence distribution of local temporal response encoding vectors of acceleration. It can be understood that when a patient completes continuous movements, the XY-plane co-encoding vector first captures the initial angular acceleration of trunk rotation, while the Z-axis eigenvector delays the response to the vertical acceleration abrupt change caused by the upward shift of the center of gravity. This asynchronicity of the cross-dimensional response precisely maps the mechanical delay effect of abdominal pressure transmission to the stoma floor. Although the global response matrix fully records the correlation strength between planar motion and vertical dynamics, its two-dimensional static structure cannot characterize the causal temporal sequence of dynamic transmission. Therefore, in the technical solution of this application, by decomposing the matrix into local response encoding sequences into column vectors, the static interaction map is reconstructed into a time-driven causal event chain, providing a mechanical script with temporal resolution for subsequent inference.

[0031] In a specific example of this application, the acceleration global temporal fine-grained response matrix is ​​decomposed into column vectors using the following formula to obtain the sequence distribution of the acceleration local temporal response encoding vector; wherein, the formula is: , in, This represents the matrix decomposition operation. and These are the 1st, 2nd, and 3rd elements in the sequence distribution of the local temporal response encoding vector of the acceleration. The and the first An acceleration local timing response encoding vector.

[0032] Furthermore, the sequence distribution of the local temporal response encoding vector is input into an acceleration temporal response inference generation engine based on a forward LSTM model to obtain a multidimensional acceleration temporal co-encoding vector. It should be understood that although the local response sequence has been decomposed into the stage-specific characteristics of motion transmission, the dispersed encoding vectors cannot autonomously reveal the causal accumulation patterns between stages. Therefore, in the technical solution of this application, a forward LSTM model is introduced to model the nonlinear evolution logic of local response segments in the time dimension. That is, key mechanical events are dynamically filtered through a gating mechanism (forget gate / input gate), and stress transmission paths spanning multiple time steps are integrated to ultimately condense a transaxial dynamic fingerprint spanning the entire motion cycle.

[0033] In a specific example of this application, the sequence distribution of the acceleration local temporal response encoding vector is input into an acceleration temporal response inference generation engine based on a feedforward LSTM model using the following formula to obtain a multidimensional acceleration temporal co-encoding vector; wherein, the formula is: , in, Represents temporal response reasoning, It is the multidimensional acceleration temporal co-coding vector.

[0034] Specifically, the state classification unit 324 is used to generate patient activity state classification data based on the multidimensional acceleration temporal co-coding vector. In the technical solution of this application, the multidimensional acceleration temporal co-coding vector is processed through a multi-classifier-based patient activity state detection model to obtain patient activity state classification data, which includes static, mild activity, and vigorous activity. It should be understood that although the multidimensional acceleration temporal co-coding vector contains motion pattern information, its high-dimensional numerical form cannot be directly understood by the prediction model and cannot be directly used as a basis for judging leakage risk. Therefore, in the technical solution of this application, the key task is to transform the multidimensional acceleration temporal co-coding vector into an intuitive, interpretable classification label with direct application value for subsequent leakage prediction through a multi-classifier-based patient activity state detection model.

[0035] In a specific example of this application, the multi-classifier is a random forest model. Taking the scheme of this application as an example, firstly, the unit receives the multi-dimensional acceleration temporal co-coding vector (denoted as...) output from the preceding steps. , dimension Next, Input a random forest model, where the decision-making process of the random forest (taking three-class classification as an example) is as follows: Single-tree decision-making: Each decision tree starts from the root node and is indexed by feature. and threshold Split node (decision rule: if) If the result is not found, proceed to the left subtree; otherwise, proceed to the right subtree, until a leaf node is reached to output the category label. {Stationary, slight activity, vigorous activity}; Integrated Voting: Suppose the forest contains T trees, and each tree outputs a category. The final classification result is the category with the most votes: , in, This indicates exponentiation. This indicates taking the maximum value. For category labels, Classify patient activity status data.

[0036] Taking the scheme of this application as an example, the state classification unit inputs the multidimensional acceleration temporal co-encoded vector into a pre-set random forest model (containing 50 decision trees). Each tree is classified according to the feature vector component values ​​(e.g., , , , The model executed a split decision, resulting in 32 trees outputting "mild activity," 15 trees outputting "still," and 3 trees outputting "vigorous activity." Based on the majority voting rule, the model classified the patient's activity status as "mild activity."

[0037] Specifically, the leakage time prediction module 330 is used to input the current humidity value, current pH value, current temperature value, maximum signal strength of the leakage sensor, and patient activity status classification data as real-time basic features into a locally deployed leakage prediction model to obtain the predicted remaining time before leakage occurs. It should be understood that the prediction of leakage risk depends not only on the physicochemical indicators (humidity, pH, temperature, and leakage signal strength) of the environment surrounding the stoma, but also more closely on the patient's activity status. Humidity reflects the water accumulation in the stoma area, pH value indicates the acidity or alkalinity of excrement and its potential impact on the skin, temperature is related to local blood circulation and infection risk, and the intensity of the leakage signal directly reflects the dampness or damage to the stoma chassis. Simultaneously, the patient's physical activity significantly affects the stability of the stoma chassis; strenuous exercise may accelerate leakage. Therefore, to effectively combine these real-time collected basic feature information, in the technical solution of this application, the current humidity value, current pH value, current temperature value, maximum signal strength of the leakage sensor, and patient activity status classification data are input as real-time basic features into a locally deployed leakage prediction model to obtain the predicted remaining time before leakage occurs. The leakage prediction model is trained in the cloud and transmitted and deployed on the edge device of the stoma chassis through a secure channel.

[0038] In a specific example of this application, firstly, recent historical basic features from the local cache are extracted from the stoma chassis edge device. It should be understood that leakage risk depends not only on the current physiological and environmental state, but also closely on recent historical changes in the patient's characteristics. For example, dynamic indicators such as the rate of change in humidity, the deviation trend of pH value, and the duration of activity are all sensitive variables for predicting leakage. Therefore, in the technical solution of this application, by "extracting recent historical basic features from the local cache of the stoma chassis edge device," the system can obtain environmental and behavioral information within past time windows, becoming the basis for deriving complex features (such as the rate of change in humidity, the time elapsed since the last intense activity, etc.), greatly improving the predictive model's ability to perceive trends and fluctuations, and avoiding the random influence of isolated moment data.

[0039] This process is implemented by a software module running on the edge device of the stoma chassis. Specifically, the edge device stores a circular cache or time-series database to continuously record and store periodically collected basic sensor data and classification tags, including humidity, pH, temperature, leakage signal strength, and activity status classification data. Here, the edge device can respond to historical data queries with low latency through an efficient cache management mechanism, avoiding frequent access to the cloud, thus balancing response speed and data security. At the same time, to avoid excessive consumption of storage resources, the caching mechanism typically implements a data aging and cleanup strategy to ensure optimal cache size.

[0040] In practice, first, determine the length and end point of the query time window (usually the current time t). For example, to query cached data from the most recent hour, then filter all timestamps in the cache structure based on the timestamp index. The system first extracts the data records within the system; then, it extracts the time series set composed of these multi-dimensional data fields and performs interpolation correction for missing values ​​or measurement errors to ensure data continuity and accuracy; finally, it outputs the extracted multi-channel historical data series as the basic features of recent history.

[0041] Next, derived features are calculated based on real-time basic features and recent historical basic features. It should be understood that relying solely on instantaneous real-time basic features can easily overlook the dynamic trends and contextual relationships of physiological and environmental states, potentially leading to a one-sided or delayed leakage risk assessment. Historical basic features provide the system with continuous information in the time dimension. By combining real-time data with analysis of their rate of change, volatility, and temporal patterns, the risk evolution of the chassis environment and the potential impact of patient activities can be captured more accurately. In the technical solution of this application, by comprehensively utilizing currently collected real-time data and recent historical data, more expressive and predictive comprehensive indicators, namely derived features, are extracted, thereby optimizing the input structure and predictive performance of subsequent models. These derived features include the humidity change rate over the past hour, the deviation between the current pH value and the user's personalized pH limit, the time since the last strenuous activity, the average osmotic signal intensity over the past 30 minutes, the interaction term between temperature and pH, and whether it is nighttime.

[0042] In a specific example of this application, the rate of change of humidity over the past hour can be calculated using the following formula: , in, The humidity value for the past hour was obtained through precise lookup or interpolation from historical data. More specifically, the deviation between the current pH value and the user's personalized pH limit is calculated using the following formula: , in, This is dynamically adjusted based on individual user characteristics and historical data. More specifically, the time elapsed since the last strenuous activity is calculated using the following formula: , in, The time point at which the patient was most recently classified as having strenuous activity was obtained by scanning a sequence of recent activity states. This refers to the time since the last strenuous activity; More specifically, the average penetration signal intensity over the past 30 minutes is calculated using the following formula: , , in, This represents the average penetration signal intensity over the past 30 minutes. The number of penetration signals in the past 30 minutes. For the past Permeation signal strength per minute; More specifically, the interaction term between temperature and pH is calculated using the following formula: , in, This is the interaction term between temperature and pH; More specifically, whether it is nighttime is calculated using the following formula: , That is, the response to the current time t belongs to If the range is within this range, then it is nighttime; otherwise, it is not.

[0043] Then, the real-time basic features and derived features are merged to obtain the local prediction feature vector. It should be understood that, to improve the accuracy and robustness of the stoma chassis leakage prediction model, the system requires not only single-moment environmental and activity state data (i.e., real-time basic features), but also deeper dynamic trend information and complex relationship descriptions (i.e., derived features). In the technical solution of this application, the effective merging of the two fully integrates instantaneous information and temporal evolution features, giving the model a more comprehensive input perspective and avoiding prediction bias and errors caused by isolated data. Furthermore, the unified construction of the local prediction feature vector helps edge devices efficiently process and manage input data streams, enabling real-time and rapid model inference and meeting the system requirements for real-time leakage risk early warning of stoma chassis.

[0044] In practice, the system first reads the collected real-time basic feature vector and the derived feature vector generated by the feature calculation module, respectively, as follows: , , in, This is the current humidity value. The current pH value. The current temperature. This represents the maximum signal strength of the leakage sensor. Classify patient activity status data. This represents the rate of change in humidity over the past hour. This represents the deviation between the current pH value and the user's personalized pH limit. This refers to the time since the last strenuous activity. This represents the average penetration signal intensity over the past 30 minutes. The interaction term between temperature and pH, Indicate whether it is nighttime; Subsequently, the feature fusion module concatenates the two vectors using the following formula to obtain the locally predicted feature vector: , in, This is the locally predicted feature vector.

[0045] Furthermore, the locally predicted feature vector is input into the locally deployed leakage prediction model to obtain the predicted remaining time before leakage occurs. In the technical solution of this application, the locally deployed leakage prediction model uses a regression decoding matrix to decode and regress the locally predicted feature vector to obtain the predicted remaining time before leakage occurs.

[0046] In practice, the module first receives the locally predicted feature vector generated by the fusion process. and to The eigenvalues ​​in the vector are normalized to obtain the normalized local prediction feature vector, denoted as: : , in, Represents a one-dimensional normalized real-time or derived feature; Next, a linear regression-based prediction algorithm is used to decode and regress the normalized local prediction feature vector using the following formula to obtain the predicted remaining time of leakage: , in, This is the weight matrix. , For bias terms, The remaining time before the predicted leak occurs.

[0047] It's worth noting that, in terms of software implementation, the model runs on low-power chips on edge devices, with extremely short execution times (e.g., milliseconds), ensuring real-time performance and continuity. Input data undergoes unified preprocessing (normalization, missing value handling) before model execution to ensure stable and reliable predictions.

[0048] Specifically, the warning triggering module 340 is used to determine whether to trigger an alarm based on a comparison between the predicted remaining time before leakage and a preset alarm threshold. In a specific example of this application, an alarm is triggered if the predicted remaining time before leakage is less than the preset alarm threshold; otherwise, no alarm is triggered. That is, triggering an alarm when the predicted remaining time is greater than the threshold indicates that the leakage risk is approaching the user-defined safety boundary and requires immediate attention.

[0049] In practical implementation, this judgment process is typically accomplished through conditional statements. First, it checks whether the input model prediction value and threshold are valid and have undergone standardized unit conversion to ensure data type and format consistency. Next, it performs a numerical comparison to determine if the trigger threshold has been reached. When the trigger condition is met, the system activates a multi-channel alarm mechanism, including audible and visual alerts, mobile app push notifications, and even synchronization with remote medical platforms, ensuring timely patient response. Furthermore, after an alarm is triggered, the system performs a logging function, saving the trigger timestamp and current feature status, providing data support for subsequent clinical analysis and personalized user tuning.

[0050] In particular, due to the individual variability and complexity of leakage in stoma chassis, an initial model alone cannot cover all scenarios. User feedback, especially actual leakage events and their timing, can provide the model with authentic labeled data, helping it learn more accurate leakage warning characteristics and improve the accuracy and universality of predictions. Therefore, to effectively upload and utilize user feedback data, thereby enabling continuous model optimization and self-iteration, user feedback can be further collected and uploaded in real time.

[0051] In practice, firstly, the user device (smartphone app or edge device) needs to collect user feedback data, including feedback timestamp, user unique identifier, device identifier, and the feature vector currently used for local leakage prediction, i.e. basic real-time feature data. Next, the above data is combined into an input package to ensure the integrity of the uploaded information and its relevance to the prediction; Subsequently, user feedback data is securely and reliably transmitted to the cloud platform via the data upload function. To ensure data privacy and integrity, the upload process may use a TLS encrypted channel and undergo authentication. After being uploaded to the cloud, the system performs data preprocessing, including noise reduction, format verification, and time synchronization, to ensure data quality. Subsequently, the feedback data is integrated into the user dataset stored in the cloud, especially by associating user feedback results (such as labels for whether leakage occurred) with corresponding feature data to form a labeled training sample set. Then, the system triggers the model training module, continuously optimizing the leakage prediction model parameters using the newly uploaded labeled data. The training process includes selecting a suitable machine learning algorithm (such as random forest, XGBoost, or neural networks), dividing the training / validation datasets, and updating the model weights using optimization methods such as gradient descent. After training is complete, the latest version of the prediction model is stored in the cloud and synchronized to the user's edge device, enabling iterative model upgrades. Finally, the cloud platform sends back a confirmation message to ensure that the data has been received correctly.

[0052] Furthermore, in order to transform intelligent leakage prediction data into medical decision support and realize dynamic health monitoring and timely intervention for ostomy patients, patient data can be centrally managed through a cloud platform. Medical staff can remotely and in real time grasp the monitoring trends, prediction results and user feedback of patients' ostomy status, thereby making scientific clinical judgments and personalized care plans, minimizing the risk of complications, and improving patients' quality of life and medical service efficiency.

[0053] In practice, the system first reads the patient’s comprehensive data input from the cloud database, including user datasets with sensor features and tags, historical leakage prediction results, user feedback from time to time, and the patient’s personal file information. These inputs together constitute the data foundation for remote monitoring. Next, the system uses visualization and analysis modules to comprehensively process the above data. Various data visualization tools (such as time-series line charts, heatmaps, and risk trend charts) dynamically display the real-time monitoring trends and leakage risk evolution of patients. Statistical analysis methods, such as trend detection and outlier identification, are used to help medical staff understand changes in patients' health conditions. Meanwhile, based on machine learning model output and historical feedback, the system evaluates model performance and provides accuracy feedback, assisting medical staff in judging the confidence and applicability of prediction results. Furthermore, based on patient monitoring data, the system intelligently pushes personalized nursing suggestions, such as adjusting the ostomy bag replacement cycle and implementing preventative measures for skin inflammation. Ultimately, healthcare professionals receive specialized clinical advice through a cloud-based portal system, which supports intervention plans based on data analysis results and existing nursing guidelines.

[0054] As described above, the intelligent stoma chassis monitoring system 300 according to the embodiments of this application can be implemented in various wireless terminals, such as servers with intelligent stoma chassis monitoring algorithms. In one possible implementation, the intelligent stoma chassis monitoring system 300 according to the embodiments of this application can be integrated into the wireless terminal as a software module and / or hardware module. For example, the intelligent stoma chassis monitoring system 300 can be a software module in the operating system of the wireless terminal, or it can be an application developed for the wireless terminal; of course, the intelligent stoma chassis monitoring system 300 can also be one of many hardware modules of the wireless terminal.

[0055] Alternatively, in another example, the intelligent stoma chassis intelligent monitoring system 300 and the wireless terminal can also be separate devices, and the intelligent stoma chassis intelligent monitoring system 300 can be connected to the wireless terminal via wired and / or wireless networks, and transmit interactive information in accordance with the agreed data format.

[0056] The various embodiments of this disclosure have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or improvement of the technology in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. An intelligent monitoring system for an ostomy chassis, characterized in that, The method comprises the following steps: a data acquisition module for acquiring current humidity value, current pH value, current temperature value and maximum signal strength of the leakage sensor collected by the humidity sensor, the pH sensor, the temperature sensor and the leakage sensor; a patient activity state classification module for obtaining original three-axis acceleration data collected by the accelerometer and determining patient activity state classification data based on the original three-axis acceleration data; a leakage time prediction module for inputting the current humidity value, the current pH value, the current temperature value, the maximum signal strength of the leakage sensor and the patient activity state classification data as real-time basic features into a locally deployed leakage prediction model to obtain a predicted leakage occurrence remaining time; a warning triggering module for determining whether to trigger an alarm based on a comparison between the predicted leakage occurrence remaining time and a preset alarm threshold.

2. The intelligent monitoring system of the intelligent ostomy base plate according to claim 1, characterized in that, The leakage prediction model is trained in the cloud, transmitted through a secure channel and deployed on the ostomy base plate edge device.

3. The intelligent monitoring system of the intelligent ostomy base plate according to claim 2, characterized in that, The leakage time prediction module comprises: a historical feature extraction unit for extracting recent historical basic features cached locally from the ostomy base plate edge device; a derived feature calculation unit for calculating derived features based on real-time basic features and recent historical basic features; a feature merging unit for merging real-time basic features and derived features to obtain a local prediction feature vector; a leakage prediction unit for inputting the local prediction feature vector into a locally deployed leakage prediction model to obtain a predicted leakage occurrence remaining time.

4. The intelligent monitoring system of the intelligent ostomy base plate according to claim 3, characterized in that, The derived features include a humidity change rate in the past 1 hour, a deviation of the current pH value from the user's personalized extreme pH, a time since the last vigorous activity, an average permeation signal strength in the past 30 minutes, an interaction term of temperature and pH, and whether it is nighttime.

5. The intelligent monitoring system of the intelligent ostomy base plate according to claim 3, characterized in that, The leakage prediction unit is configured to: The locally deployed leakage prediction model uses a regression decoding matrix to decode and regress the local prediction feature vector to obtain the predicted leakage occurrence remaining time.

6. The intelligent monitoring system of the intelligent ostomy base plate according to claim 1, characterized in that, The patient activity state classification module comprises: a time series encoding unit for time series encoding of the original three-axis acceleration data to obtain an X-axis acceleration time series pattern feature encoding vector, a Y-axis acceleration time series pattern feature encoding vector and a Z-axis acceleration time series pattern feature encoding vector; a time series collaborative encoding unit for calculating an X-Y axis acceleration time series collaborative encoding vector by dividing the X-axis acceleration time series pattern feature encoding vector by the Y-axis acceleration time series pattern feature encoding vector; a multi-dimensional state collaborative response reasoning unit for multi-dimensional state collaborative response reasoning of the X-Y axis acceleration time series collaborative encoding vector and the Z-axis acceleration time series pattern feature encoding vector to obtain a multi-dimensional acceleration time series collaborative encoding vector; a state classification unit for generating patient activity state classification data based on the multi-dimensional acceleration time series collaborative encoding vector.

7. The intelligent monitoring system of the intelligent ostomy base plate according to claim 6, characterized in that, The time series encoding unit is configured to: perform data arrangement on the original three-axis acceleration data to obtain an X-axis acceleration time series, a Y-axis acceleration time series and a Z-axis acceleration time series; The X-axis acceleration time sequence, the Y-axis acceleration time sequence and the Z-axis acceleration time sequence are respectively input into the LSTM-based acceleration time sequence encoding model to obtain an X-axis acceleration time sequence pattern feature encoding vector, a Y-axis acceleration time sequence pattern feature encoding vector and a Z-axis acceleration time sequence pattern feature encoding vector.

8. The intelligent monitoring system of the intelligent ostomy base plate according to claim 6, characterized in that, The multi-dimensional state collaborative response inference unit is configured to: calculate an acceleration global time sequence fine-grained response matrix of the X-Y-axis acceleration time sequence collaborative encoding vector relative to the Z-axis acceleration time sequence pattern feature encoding vector; perform matrix decomposition on the acceleration global time sequence fine-grained response matrix according to column vectors to obtain a sequence distribution of acceleration local time sequence response encoding vectors; input the sequence distribution of the acceleration local time sequence response encoding vectors into the acceleration time sequence response inference generation engine based on the forward LSTM model to obtain a multi-dimensional acceleration time sequence collaborative encoding vector.

9. The intelligent monitoring system of the intelligent ostomy base plate according to claim 6, characterized in that, The state classification unit is configured to: input the multi-dimensional acceleration time sequence collaborative encoding vector into the patient activity state detection model based on the multi-classifier to obtain patient activity state classification data, the patient activity state classification data including stillness, slight activity and intense activity.

10. The intelligent monitoring system of the intelligent ostomy base plate according to claim 1, characterized in that, The warning triggering module is configured to: trigger a warning in response to the predicted remaining time to occurrence of a leak being less than a preset warning threshold; otherwise, do not trigger the warning.

Images