Intestinal fluid intermittent return method and system based on pressure self-adaptation

Abstract

The present application relates to the technical field of medical devices, and discloses an intestinal fluid intermittent return method and system based on pressure self-adaptation, wherein multi-modal sensor network is used to collect patient situation state, physiological rhythm and intestinal pressure data to obtain multi-dimensional monitoring data flow; time series signal processing technology is used to extract the corresponding phases of each data and calculate synchronization metrics; a synchronous state classification model is used for collaborative state classification; based on the classification result, a return decision instruction is generated to control the return device and adjust the return parameters according to the intestinal pressure; in addition, the present application also uses an incremental learning algorithm to adjust the model threshold and decision parameters based on individualized data sets, which can accurately control the return timing and parameters, and improve the return effect and safety of intestinal fluid.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical devices, more particularly, it relates to an intermittent intestinal fluid infusion method and system based on pressure adaptation. BACKGROUND

[0002] With the development of medical technology, intestinal fluid infusion technology plays an increasingly important role in clinical treatment. Intestinal fluid infusion refers to the process of reinfusing the intestinal fluid drained during intestinal decompression into the patient's body after treatment. This technology is of great significance for maintaining the patient's water and electrolyte balance, reducing nutrient loss, and promoting the recovery of intestinal function.

[0003] Currently, intestinal fluid infusion technology mainly relies on the experience and judgment of medical personnel and simple pressure monitoring to determine the infusion time and rate. Although this method can meet the basic treatment needs to some extent, due to the complexity of the human physiological system, relying solely on experience and a single indicator for decision-making often fails to achieve optimal treatment results. Especially in the case of varying patient activity states and significant physiological rhythm fluctuations, the existing infusion scheme is difficult to adapt to these changes in a timely manner, which can easily lead to complications.

[0004] In addition, the intestinal function of the human body is coordinated by multiple systems, including the nervous system, endocrine system, and autonomic nervous system. There are complex temporal correlations and interactions between these systems, and the existing technology lacks in-depth consideration of this multi-system coordination feature, making it difficult to achieve precise control of the intestinal fluid infusion process. Therefore, an intelligent intestinal fluid infusion system that takes into account the patient's situational state, physiological rhythm, and pressure state is needed. SUMMARY

[0005] The present application provides an intermittent intestinal fluid infusion method and system based on pressure adaptation, which solves the technical problems of single-dimensional decision-making, binary contradiction balance difficulty, information island, lack of individualized adaptation ability, and multi-scale coordination difficulty in related technologies.

[0006] In one aspect, the present application provides an intermittent intestinal fluid infusion method based on pressure adaptation, comprising the following steps:

[0007] Collecting the situational state data, physiological rhythm data, and intestinal pressure data of the patient through a multi-modal sensor network to obtain a multi-dimensional monitoring data stream;

[0008] According to the multi-dimensional monitoring data stream, using time series signal processing technology to extract the situational state phase corresponding to the situational state data, the physiological rhythm phase corresponding to the physiological rhythm data, and the intestinal pressure phase corresponding to the intestinal pressure data, and calculating the synchronization measure between the situational state phase, the physiological rhythm phase, and the intestinal pressure phase;

[0009] According to the synchronization metric, the synergistic state of the context state, the physiological rhythm and the intestinal pressure is classified by a pre-constructed synchronization state classification model, and the classification of the synergistic state includes an ideal synergistic state, a sub-ideal synergistic state and a non-ideal synergistic state.

[0010] A feedback decision instruction is generated based on the classification result of the synergistic state, and a feedback device is controlled to perform intestinal fluid feedback based on the feedback decision instruction, and a feedback parameter is adaptively adjusted according to the intestinal pressure data.

[0011] In a preferred embodiment, the context state data includes activity state, posture change data, environmental parameters and emotional state data; the physiological rhythm data includes circadian rhythm data, digestive cycle data and hormone secretion cycle data.

[0012] In a preferred embodiment, the specific step of extracting the context state phase is to apply an empirical mode decomposition algorithm to decompose the context state data into a series of intrinsic mode functions, convert each intrinsic mode function to a phase space by using a Hilbert transform to obtain phase information of the intrinsic mode function, and obtain a unified context state phase representation by integrating the phase information through principal component analysis;

[0013] The specific step of extracting the physiological rhythm phase is to apply a wavelet transform to perform multi-time scale decomposition on the physiological rhythm data to obtain wavelet coefficients of different scales, calculate a phase locking value based on the wavelet coefficients, and extract instantaneous phases of each physiological cycle in the phase locking value;

[0014] The specific step of extracting the intestinal pressure phase is to identify a periodic pressure fluctuation pattern from the intestinal pressure data, and convert the fluctuation pattern into a phase representation.

[0015] In a preferred embodiment, the specific step of calculating the synchronization metric between the context state phase, the physiological rhythm phase and the intestinal pressure phase is:

[0016] Based on the context state phase, the physiological rhythm phase and the intestinal pressure phase, a phase difference matrix is constructed; and according to the phase difference matrix, a phase synchronization index formula is used to calculate an overall synchronization metric value between the context state phase, the physiological rhythm phase and the intestinal pressure phase.

[0017] In a preferred embodiment, the specific step of pre-constructing the synchronization state classification model includes:

[0018] Based on a pre-set elastic synchronization band threshold range, the synchronization metric value intervals corresponding to the ideal synergistic state, the sub-ideal synergistic state and the non-ideal synergistic state are defined; and according to the interval in which the overall synchronization metric value falls, the synergistic state is classified.

[0019] In a preferred embodiment, the synchronous state classification model further comprises a recurrent neural network based time series prediction model for predicting the future collaborative state classification according to the historical synchronous metric value sequence.

[0020] In a preferred embodiment, the specific steps of generating the feedback decision instruction comprise:

[0021] Based on the classification result and the prediction result of the collaborative state, a time window in which the ideal collaborative state occurs is identified; and according to the time window and the real-time intestinal pressure data, the starting time, the feedback rate and the feedback duration of the feedback are determined.

[0022] In a preferred embodiment, the self-adaptive adjustment of the feedback parameters according to the intestinal pressure data comprises establishing a negative feedback control logic of the feedback rate and the real-time intestinal pressure value, and when the real-time intestinal pressure value exceeds a preset first safety threshold, the feedback rate is reduced or the feedback is paused.

[0023] In a preferred embodiment, the method further comprises:

[0024] The collaborative state classification, the feedback parameters and the intestinal pressure response data before and after each feedback activity are recorded to form an individualized data set; and based on the individualized data set, the incremental learning algorithm is used to adjust the elastic synchronous band threshold range and / or the feedback decision parameters in the synchronous state classification model.

[0025] In another aspect, the present application also provides an intestinal fluid intermittent feedback system based on pressure self-adaptation, which is used to execute the intestinal fluid intermittent feedback method based on pressure self-adaptation.

[0026] A data acquisition module is configured to acquire the context state data, the physiological rhythm data and the intestinal pressure data of a patient through a multi-modal sensor network to obtain a multi-dimensional monitoring data stream.

[0027] A phase extraction and synchronization metric calculation module is configured to extract the context state phase corresponding to the context state data, the physiological rhythm phase corresponding to the physiological rhythm data and the intestinal pressure phase corresponding to the intestinal pressure data by using time series signal processing technology according to the multi-dimensional monitoring data stream, and calculate the synchronization metric between the context state phase, the physiological rhythm phase and the intestinal pressure phase.

[0028] A collaborative state classification module is configured to classify the collaborative state of the context state, the physiological rhythm and the intestinal pressure by using a pre-constructed synchronous state classification model according to the synchronization metric, wherein the classification of the collaborative state comprises an ideal collaborative state, a sub-ideal collaborative state and a non-ideal collaborative state.

[0029] The backflow decision and parameter adjustment module is configured to generate a backflow decision instruction based on the classification result of the coordination state, control a backflow device to perform intestinal fluid backflow based on the backflow decision instruction, and adjust backflow parameters adaptively according to the intestinal pressure data.

[0030] The present application has the following beneficial effects:

[0031] 1) The present application provides a pressure-adaptive intermittent intestinal fluid backflow method and system, which triggers action only when the intestinal tract is in an ideal coordination state that can safely accept backflow by comprehensively analyzing the coordination state of a physiological system composed of situational state, physiological rhythm and intestinal pressure, thereby avoiding backflow during patient activity, stress or intestinal unstable contraction period, and significantly improving safety and comfort through precise backflow decision based on multi-dimensional physiological coordination.

[0032] 2) The present application provides a pressure-adaptive intermittent intestinal fluid backflow method and system, which predicts the development trend of the physiological system coordination state based on a synchronous state timing prediction model, thereby planning backflow activities in advance, avoiding potential risk windows in advance when dealing with physiological rhythm conversion and external situational mutations, ensuring smooth and continuous backflow process, and improving long-term stability of the system.

[0033] 3) The present application provides a pressure-adaptive intermittent intestinal fluid backflow method and system, which records the coordination state classification, backflow parameters and intestinal pressure response data before and after each backflow activity to form an individualized data set, i.e., to build an individualized synchronous parameter learning framework, dynamically optimizes the coordination state judgment threshold and backflow control parameters according to the unique physiological response mode of each patient, and this design enables the system to automatically adapt to individual differences such as the circadian rhythm characteristics and intestinal sensitivity of different patients, thereby improving system adaptability. BRIEF DESCRIPTION OF DRAWINGS

[0034] Figure 1 is a flowchart of a pressure-adaptive intermittent intestinal fluid backflow method of the present application. DETAILED DESCRIPTION

[0035] The subject matter described herein will now be discussed with reference to example implementations. It should be understood that the discussion of these implementations is merely meant to provide a better understanding of the subject matter described herein and can be changed in function and arrangement without departing from the scope of the present description. Various processes or components can be omitted, substituted, or added according to various examples. Additionally, features described in some examples can be combined in other examples.

[0036] In at least one embodiment of the present application, a pressure-adaptive intermittent intestinal fluid backflow method is disclosed, as shown in Figure 1As shown, comprising the following steps:

[0037] Step 1, collecting the patient's situational state data, physiological rhythm data and intestinal pressure data through a multi-modal sensor network, obtaining a multi-dimensional monitoring data stream;

[0038] Specifically comprising the following steps:

[0039] Step 1.1, collecting the patient's activity state, posture change data, environmental parameters and emotional state data using wearable devices and environmental sensors; further, the activity state and posture change data includes the activity type, intensity and location information of the patient obtained through sensors such as accelerometers, gyroscopes and GPS; the environmental parameters include environmental parameters such as ambient temperature, humidity and illumination obtained through environmental sensors; the emotional state data includes physiological indicators reflecting emotional state such as heart rate, skin conductance obtained through smartwatches or other wearable devices.

[0040] Preferably, the situational state data can also include social interaction monitoring data, which evaluates the patient's social activity intensity through the communication records and social application usage of the smart phone as a supplementary indicator for emotional state assessment.

[0041] Step 1.2, collecting the patient's circadian rhythm data, digestive cycle data and hormone secretion cycle data using physiological parameter sensors, further, the circadian rhythm data includes monitoring circadian rhythm through indicators such as body temperature curve and melatonin level; the digestive cycle data includes obtaining digestive cycle information through methods such as electrogastrogram, intestinal sound monitoring; the hormone secretion cycle data includes collecting hormone secretion cycle data through hormone level detection devices in saliva or sweat.

[0042] Preferably, the patient's sleep stage can also be monitored through a portable electroencephalogram (EEG) device as an important supplementary data source for circadian rhythm assessment.

[0043] Step 1.3, continuously monitoring the intestinal pressure value and its change characteristics as intestinal pressure data through implanted or connected micro pressure sensors, including collecting the absolute value, change rate and fluctuation pattern of intestinal pressure; monitoring the intestinal peristalsis waveform characteristics and frequency; recording the time correlation of pressure changes with drug, diet and other intervention factors.

[0044] Preferably, a multi-point pressure sensing array can also be used to simultaneously monitor the pressure distribution at different parts of the intestine, constructing a more comprehensive pressure state atlas.

[0045] Step 1.4, filtering, denoising and time alignment processing are performed on the data collected in steps 1.1-1.3 to ensure that each data stream is synchronized in the time dimension, and multi-dimensional monitoring data streams are obtained. The filtering, denoising and time alignment processing specifically include:

[0046] Applying an adaptive filter to eliminate motion artifacts and environmental interference;

[0047] Performing data interpolation and resampling to make all data streams reach a uniform sampling frequency;

[0048] Establishing a unified timestamp system to ensure the time synchronization of each data source.

[0049] Preferably, the embodiment adopts a multi-resolution analysis method based on wavelet transform to perform hierarchical denoising processing on physiological signals with different frequency characteristics, thereby improving signal quality.

[0050] Step 2, according to the multi-dimensional monitoring data stream, using time series signal processing technology to extract the context state phase corresponding to the context state data, the physiological rhythm phase corresponding to the physiological rhythm data and the intestinal pressure phase corresponding to the intestinal pressure data, and calculate the synchronization measure between the context state phase, the physiological rhythm phase and the intestinal pressure phase;

[0051] Specifically includes the following steps:

[0052] Step 2.1, convert the context state data into a phase representation to capture the periodic characteristics of context changes. Specifically includes:

[0053] Applying the Empirical Mode Decomposition (EMD) algorithm to process the context state data, i.e. decomposing the context state data into a series of Intrinsic Mode Functions (IMF); performing Hilbert transform on each intrinsic mode function obtained by decomposition to convert each intrinsic mode function from time domain to phase space, obtaining the phase information of each intrinsic mode function, taking the phase information as input, and using Principal Component Analysis (PCA) method for processing. Principal Component Analysis can convert multiple correlated variables into a set of uncorrelated principal components. In this process, by calculating the weight and contribution of each principal component, the principal components representing most of the data information are selected, and these principal components are linearly combined to finally obtain a unified context state phase representation .

[0054] Preferably, the embodiment adopts an ensemble empirical mode decomposition algorithm instead of a standard empirical mode decomposition algorithm to improve the processing capability of non-stationary signals. In addition, the embodiment also introduces an adaptive noise threshold mechanism to reduce mode aliasing.

[0055] Step 2.2, analyze the periodic patterns of the circadian data and extract the phase information. Specifically, it includes:

[0056] Applying wavelet transform to process the circadian data, the circadian data is decomposed into different frequency components on different time scales, obtaining a series of wavelet coefficients of different scales, which reflect the characteristics of the circadian data in different frequencies and times; using the wavelet coefficients obtained by decomposition, calculate the phase locking value (PLV), which is an index for measuring the degree of phase synchronization between two signals, in the analysis of the circadian data in this embodiment, the phase relationship of each physiological cycle is determined by calculating the phase locking value between different wavelet coefficients. For each wavelet component corresponding to a physiological cycle, according to the calculation result of the phase locking value, the instantaneous phase is extracted. These instantaneous phases reflect the dynamic changes of the circadian rhythm in different cycles, that is, the circadian rhythm phase representation .

[0057] Optionally, when processing physiological signals with strong noise interference, empirical mode decomposition can be applied for signal denoising before wavelet transform is performed to obtain more stable phase extraction results.

[0058] Step 2.3, identify the periodic pattern from the intestinal pressure data and convert it to a phase representation. Specifically, it includes:

[0059] Using threshold detection to filter out fluctuations in the pressure data that are below a preset pressure threshold, highlighting possible periodic signals; using a wave peak identification algorithm to find the wave peak positions in the pressure data, and preliminarily judging whether there is periodicity according to the time interval between adjacent wave peaks; if there is periodicity, calculate the mean and standard deviation of the time interval between multiple wave peaks as periodicity characteristics;

[0060] Since the intestinal pressure fluctuations may be affected by various factors, leading to certain non-uniformity of the cycle length, this embodiment uses the dynamic time warping (DTW) algorithm to standardize the non-uniform cycles, and adjusts each non-uniform pressure cycle to a uniform standard length with reference to the mean value in the periodicity characteristics, eliminating the influence of cycle length difference on subsequent analysis;

[0061] After completing the cycle standardization, select a specific feature point (such as a wave peak or a wave trough) in each standardized cycle as a reference point, and divide a complete cycle into a phase interval of 0-2π (or 0-360°). According to the time position of the intestinal pressure data relative to the reference point in each cycle, it is mapped to the corresponding phase value, obtaining the intestinal pressure phase representation . For example, when the pressure reaches the wave peak, the phase is 0; as the pressure drops and rises, the phase value continuously changes between 0 and 2π.

[0062] Further, for patients with irregular pressure waveforms, an adaptive threshold algorithm can be used to dynamically adjust the peak recognition threshold based on the local pressure fluctuation characteristics, improving the recognition accuracy.

[0063] Step 2.4, based on the context state phase, the physiological rhythm phase and the intestinal pressure phase, a phase difference matrix is constructed; according to the phase difference matrix, the overall synchronization metric value between the context state phase, the physiological rhythm phase and the intestinal pressure phase is calculated by using the phase synchronization index formula. Specifically, it includes:

[0064] Based on the context state phase, the physiological rhythm phase and the intestinal pressure phase, the pairwise phase difference at each time point is calculated, which is expressed by the formula as:

[0065] ;

[0066] ;

[0067] ;

[0068] In the formula, , and are the phase differences between the context state phase and the physiological rhythm phase, the context state phase and the intestinal pressure phase, and the physiological rhythm phase and the intestinal pressure phase at time point ;

[0069] Based on the above pairwise phase difference, a phase difference matrix at time point is constructed , which is expressed by the formula as:

[0070] ;

[0071] In order to extract a stable overall synchronization metric value from these dynamic phase relationships, this embodiment introduces a coupling strength weight reflecting the strength of the internal connection between each subsystem (context state, physiological rhythm and intestinal pressure). The coupling strength weight is learned from historical data or determined based on expert knowledge. The coupling strength weight is normalized to obtain a normalized coupling strength weight.

[0072] Based on the phase difference matrix and the normalized coupling strength weight, the overall synchronization metric value within the time window is calculated by using the phase synchronization index formula, which is expressed by the formula as:

[0073] ;

[0074] ;

[0075] wherein, is the overall synchrony metric value; , and are the coupling strength between normalized context state and circadian rhythm, between normalized context state and gut pressure, and between normalized circadian rhythm and gut pressure, respectively;

[0076] Further, the embodiment also provides a step of calculating the synchrony metric value under different perspectives. The synchrony metric values under different perspectives may have different names, but it is worth noting that their core goal is consistent with that of the embodiment, which is to generate a quantitative synchronization level indicator to comprehensively judge the overall coordination state of the system. Therefore, in step 3, the synchrony metric values under different perspectives are uniformly defined as the synchrony metric, which is not different due to the specific definition name in step 2.4.

[0077] Preferably, the embodiment also provides a step of calculating the synchrony metric considering the non-instantaneous interaction between subsystems, which reveals the potential causal relationship through time-delayed synchronization analysis. The step includes introducing time delay and calculating the corresponding synchronization index , which is expressed in the formula as:

[0078] ;

[0079] wherein, is the phase representation of the circadian rhythm at time point ;

[0080] finding the optimal time delay that maximizes the synchrony, the maximum synchronization strength at this time is the synchrony metric of the pair of subsystems considering the time delay effect. After applying this step to the three pairs of combinations, the overall time-delayed synchronization metric value of the system is obtained by calculating the geometric mean value, which is expressed in the formula as:

[0081] ;

[0082] wherein, is the overall time-delayed synchronization metric value; , and are the maximum synchronization strengths of the context state and the circadian rhythm, the context state and the gut pressure, and the circadian rhythm and the gut pressure, respectively, under their respective optimal time delays;

[0083] Furthermore, the embodiment also provides a step of directly evaluating the cooperative behavior of the group by using a multi-element phase synchronization framework beyond pairwise analysis, i.e., calculating a multi-variable synchronization index based on group sequence parameters , which is expressed by a formula as

[0084]

[0085] , wherein is a phase index; is a complex exponential representation, for example is to represent the phase of the situational state at the time point with a unit complex number;

[0086] directly reflects the aggregation degree of the three phases in the complex plane, and provides a whole synchronization perspective independent of pairwise matching;

[0087] In addition, for scenarios aiming to capture high-order cooperative patterns, the embodiment also provides a system synchronization state analysis based on tensor product, which represents the joint state by constructing a system synchronization state tensor, wherein the system synchronization state tensor , which is expressed by a formula as

[0088]

[0089] extracts its core features by using tensor decomposition (such as HOSVD); and quantifies the high-order synchronization by calculating the normalized Frobenius norm of the core features, which is expressed by a formula as

[0090]

[0091] , wherein is a whole synchronization measurement value of the high-order cooperative effect among the situational state, the physiological rhythm and the intestinal pressure three subsystems; is the norm of the core feature; , and respectively correspond to the dimensions of the situational state phase vector, the physiological rhythm phase vector and the intestinal pressure phase vector;

[0092] The greater the value is, the more concentrated the state distribution of the system in the joint phase space is.

[0093] Step 3, according to the synchronization measurement, classifying the cooperative state of the situational state, the physiological rhythm and the intestinal pressure by using a pre-constructed synchronization state classification model, wherein the classification of the cooperative state includes an ideal cooperative state, a sub-ideal cooperative state and a non-ideal cooperative state.

[0094] ​​​​Specifically comprising the following steps:

[0095] Step 3.1, presetting the threshold range of the elastic synchronization band defining three coordination states based on historical clinical data and expert knowledge. Specifically comprising: collecting a large number of patient synchronization metric data in stable state, determining the baseline range of ideal coordination state through statistical distribution analysis; combining with clinical efficacy feedback, defining the transition interval of sub-ideal coordination state; according to the synchronization characteristics in pathological state, determining the threshold boundary of non-ideal coordination state. The elastic synchronization band adopts dynamic boundary design, and the synchronization metric value in the range of 0.75-1.00 is defined as ideal coordination state, reflecting that the three subsystems are highly coordinated; the synchronization metric value in the range of 0.45-0.74 is defined as sub-ideal coordination state, reflecting that the system has mild disorder but still has basic functions; the synchronization metric value in the range of 0-0.44 is defined as non-ideal coordination state, reflecting that the system has obvious disorder and needs intervention.

[0096] Preferably, the embodiment adopts a personalized threshold adjustment mechanism, which adjusts the above general threshold range according to the patient's age, underlying disease and treatment stage, to improve the accuracy of classification.

[0097] Step 3.2, according to the calculated overall synchronization metric value, real-time state classification is carried out based on the preset elastic synchronization band threshold range. Specifically comprising: real-time monitoring of the synchronization metric value, matching it with the preset threshold interval; when the synchronization metric value falls into the ideal coordination state interval, the system is marked as green state, indicating that the current is in the best return opportunity; when the synchronization metric value falls into the sub-ideal coordination state interval, the system is marked as yellow state, prompting to pay attention to the system trend; when the synchronization metric value falls into the non-ideal coordination state interval, the system is marked as red state, triggering the synchronization disorder warning mechanism. The classification process adopts a sliding window mechanism, and the classification results of continuous multiple time points are consistent before confirming the state conversion, avoiding misclassification caused by instantaneous fluctuations.

[0098] Optionally, fuzzy logic processing is introduced to handle the uncertainty near the threshold boundary, to calculate the probability of the synchronization metric value belonging to the adjacent state, and to realize smooth state transition.

[0099] Step 3.3, build a time series prediction model based on recurrent neural network, which is used to predict the future collaborative state classification according to the historical synchronization metric value sequence. Specifically, it includes: adopting long short-term memory network (LSTM) architecture to build the prediction model, the input layer receives the time series of synchronization metric values in the past 60 minutes, the hidden layer contains 64 LSTM units, which can effectively capture the time series dependence of synchronization state; the output layer predicts the trend of synchronization metric value in the future 15-30 minutes, and accordingly predicts the imminent collaborative state classification. The model training adopts mean square error loss function and Adam optimizer to learn the evolution law of synchronization state from historical data. In the deployment stage, the model receives the latest synchronization metric sequence in real time, and predicts the state classification probability at multiple future time points.

[0100] Preferably, the embodiment introduces attention mechanism to enhance model performance, so that the LSTM network can automatically focus on the key time points in the historical sequence that are most important for predicting the future state, improving the accuracy of prediction. At the same time, the model also outputs the uncertainty estimate of the prediction, when the prediction confidence is lower than the threshold, the system will rely on the current rule classification result to ensure the reliability of the decision.

[0101] Step 3.4, integrate the real-time classification results based on threshold rule and the future state trend predicted based on neural network to form a comprehensive classification decision. Specifically, when the rule classification and the prediction classification are consistent, the classification result is directly adopted; when there is a disagreement, a weighted decision is made according to the prediction time span and confidence, short-term high-confidence prediction has higher weight; at the same time, a conflict resolution mechanism is established, when the rule shows a non-ideal state but the prediction shows that it will quickly improve to an ideal state, the intervention decision is appropriately delayed, and the system self-regulation ability is observed.

[0102] Optionally, an ensemble learning framework can be introduced to weight and integrate the outputs of threshold rule model, LSTM prediction model and additional time series classification algorithms (such as time series forest), to further improve the robustness and accuracy of the classification system by compensating for the limitations of single model through diversity.

[0103] Step 4, generate feedback decision instructions based on the classification results of the collaborative state, control the intestinal fluid feedback device based on the feedback decision instructions, and adaptively adjust the feedback parameters according to the intestinal pressure data.

[0104] Specifically, it includes the following steps:

[0105] Step 4.1, based on the collaborative state classification results (ideal, sub-ideal, non-ideal) and future collaborative state prediction results, identify the time window of ideal collaborative state and generate feedback decision instructions. Specifically, it includes: using a sliding window method combined with an improved peak detection algorithm to automatically analyze the synchronization metric time series, identifying the continuous period with high and stable synchronization degree as the candidate time window by finding the local extreme points and evaluating their persistence; Introducing an exponential weighted moving average based trend analysis method to predict the rising edge and falling edge of the synchronization metric, to identify the upcoming window in advance and estimate its duration; According to the identification results, generate decision instructions containing start time, expected duration, for ideal collaborative state, generate allow feedback instructions and label confidence, for non-ideal state, generate pause instructions and record abnormal types, for sub-ideal state and good trend, generate cautious feedback instructions and suggest parameter adjustment range.

[0106] Preferably, the embodiment adopts an intelligent feedback timing planning framework based on model predictive control (MPC) for more accurate window identification, which takes the prediction results of the aforementioned LSTM network as a dynamic system model, constructs a multi-objective optimization function with the main goal of maximizing the synchronization degree within the next 5-8 prediction steps and the secondary goal of minimizing feedback interference, and solves the optimal feedback start time point and expected operation time length through a sequence quadratic programming algorithm. At each control period, update the optimization results based on the latest observation data.

[0107] Step 4.2, under the determined feedback timing, optimize the feedback parameters combined with real-time intestinal pressure data. Specifically, it includes: establishing a parameter mapping system based on fuzzy reasoning, taking the synchronization window quality (including synchronization degree mean, variance, duration, etc.) and real-time intestinal pressure value (including current value, trend, fluctuation characteristics, etc.) as input, and obtaining the optimized feedback rate, duration and mode parameters through the pre-set fuzzy rule base reasoning; The rule base of this mapping system can optimize the initial parameters through genetic algorithm, and continuously improve it through case-based reasoning method; Realize the parameter smooth transition mechanism, process the parameter change request through a first-order low-pass filter to avoid patient discomfort or system oscillation caused by parameter mutation; Design a multi-objective optimization function considering multiple conflicting objectives to seek a Pareto optimal solution between feedback efficiency, system stability and patient comfort.

[0108] For example, in the identified high-quality synchronization window, if the real-time intestinal pressure is at a low level and the fluctuation is stable, a higher feedback rate (such as 120% of the reference value) and a longer feedback duration (such as 150% of the standard duration) are adopted; If the real-time pressure value is high but in a downward trend, adopt a gradual parameter adjustment strategy, initially use a lower rate, and gradually increase to an appropriate level according to the pressure response. Optionally, a reinforcement learning algorithm can also be introduced to learn the optimal parameter adjustment strategy through interaction with the environment.

[0109] Step 4.3, set a three-level safety threshold system, including a warning threshold (first safety threshold), an intervention threshold, and an emergency stop threshold (second safety threshold), each level of threshold is equipped with a corresponding response strategy; when the pressure exceeds the warning threshold, the system starts a mild preventive adjustment, slowly reduces the return rate according to the fuzzy PID control law; when the pressure exceeds the intervention threshold, a proportional-integral-derivative (PID) control strategy is adopted to quickly adjust the return rate to pull the pressure back into the safety range; when the pressure exceeds the emergency stop threshold or continues to be higher than the intervention threshold for a certain period of time and the adjustment is invalid, the return is immediately suspended and an alarm is triggered.

[0110] Preferably, the embodiment adopts a model-based adaptive PID controller, which adjusts the controller parameters in real time by online identification of the pressure-rate response model, improving the control performance under different working conditions. At the same time, a feedforward-feedback composite control architecture is established to adjust the return parameters in advance according to the pressure prediction value, enhancing the anti-interference ability of the system. In addition, the embodiment also designs a pressure change rate monitoring module to respond quickly to sharp changes in pressure and prevent accidents from happening.

[0111] Step 4.4, the embodiment also provides a multi-level decision conflict resolution mechanism when the macro-optimization decision based on the synchronization state conflicts with the micro-safety decision based on the real-time pressure. Specifically, it includes: establishing a decision priority model based on a clinical knowledge graph, placing patient safety at the highest priority, followed by synchronization state optimization goals, and then return efficiency and other goals; designing a retreat strategy library containing three modes of active avoidance, compromise negotiation, and forced intervention, and intelligently selecting the appropriate solution mode according to the severity of the conflict; implementing an instruction safety execution chain, including instruction syntax checking, semantic verification, logical reasonableness evaluation, and multiple review mechanisms, to ensure that the instructions sent to the return device are accurate and error-free.

[0112] Optionally, a distributed decision-making architecture based on a multi-agent system is adopted, which allocates tasks such as synchronization state analysis, pressure safety monitoring, and patient comfort evaluation to specialized agents, and solves decision-making conflicts through negotiation and cooperation between agents. In addition, digital twin technology can also be introduced to simulate and test various decision-making schemes in a virtual space before implementing real return, to predict the effects and risks of each scheme.

[0113] Step 5, build a comprehensive safety monitoring network during the execution of the return. Specifically includes: real-time acquisition and fusion of intestinal pressure data, flow sensor readings, device operating state parameters, patient subjective feedback and other multi-source information, using D-S evidence theory for information fusion to improve the accuracy of abnormal identification; Establish safety monitoring logic based on state machine, define normal, warning, intervention, and emergency four system states, and design corresponding monitoring indicators and response strategies for each state according to user needs; Realize the dynamic adjustment ability of safety boundary, adjust each safety threshold value adaptively according to the current physiological state and historical data of the patient.

[0114] Preferably, the embodiment uses a deep time series anomaly detection model based on a long short-term memory network (LSTM) combined with an autoencoder architecture to learn normal patterns from multi-parameter time series and provide early warning for abnormal situations that deviate from normal patterns. At the same time, a blockchain storage system for safety data is constructed to ensure the non-tamperability and complete traceability of all safety-related operations, providing reliable basis for subsequent analysis and responsibility identification.

[0115] Step 5.2, implement multi-dimensional effect evaluation during and after the return process. Specifically includes: define a multi-dimensional evaluation index system including synchronization improvement rate, pressure stability index, patient comfort score, etc.; Build a short-term effect prediction model to predict the final effect of this intervention based on the data of the first half of the return; Design a dashboard-based visual feedback interface to provide intuitive and comprehensive effect display for clinical personnel to support quick decision-making.

[0116] Optionally, the embodiment uses dynamic time warping (DTW) algorithm combined with k-nearest neighbor classification method to match the effect pattern of the current return with historical successful cases, quickly evaluate the return effect and recommend improvement measures. At the same time, an intelligent labeling system for feedback data is established, combined with an active learning strategy, to preferentially label samples that are most valuable for model improvement, continuously improving the quality of subsequent machine learning models.

[0117] Step 6, record the synchronization state classification, return parameters and intestinal pressure response data before and after each return activity to form an individualized data set; based on the individualized data set, adjust the elastic synchronization belt threshold range and / or return decision parameters in the synchronization state classification model using incremental learning algorithm.

[0118] Specifically includes the following steps:

[0119] Step 6.1, systematically record the entire process data of each return activity, and build an individualized data lake rich in spatiotemporal features. Specifically, it includes: designing a theme-oriented data model, standardizing the storage of multi-granularity data from raw sensor data to high-level decision information; implementing an automated data quality control pipeline, including outlier detection, missing value interpolation, data consistency verification, etc.; using deep feature extraction technology based on autoencoder to learn individual feature representation with discriminability from high-dimensional data; using modern dimensionality reduction techniques such as t-SNE and UMAP to visualize individual feature distribution, assisting clinical staff in understanding patient characteristics.

[0120] Preferably, the embodiment uses knowledge graph technology to build individual health records, not only storing numerical data, but also capturing complex semantic relationships and temporal evolution patterns between data. At the same time, a secure computing framework based on differential privacy and federated learning is implemented to fully exploit data value while strictly protecting patient privacy.

[0121] Step 6.2, use an elastic synchronous band threshold online learning algorithm based on Bayesian update to adaptively adjust the state classification boundary according to the dynamic changes of patient synchronization metric data, and give the uncertainty estimate of the adjustment; design a return decision optimization framework based on deep reinforcement learning, model the return decision as a partially observable Markov decision process (POMDP), and use advanced algorithms such as proximal policy optimization (PPO) to train the decision agent; establish a model performance continuous monitoring mechanism to timely detect changes in data distribution through concept drift detection technology, triggering model updates.

[0122] For example, use streaming stochastic gradient descent (SGD) combined with adaptive learning rate adjustment strategy to achieve smooth online update of classification model threshold; use a distributed reinforcement learning architecture to train multiple exploration agents in parallel, accelerating the policy optimization process. Optionally, meta-learning techniques can also be introduced to enable the system to quickly adapt to the characteristics of new patients, reducing adverse experiences during the initial learning phase.

[0123] Step 6.3, the embodiment also designs a model evaluation scheme based on time series cross-validation, accurately evaluates the system performance after parameter adjustment by considering the temporal dependence of data; implements a multi-armed bandit (Multi-armed Bandit) experiment framework to intelligently balance the relationship between model exploration (trying new parameters) and utilization (using known effective parameters); builds a patient multi-dimensional feature similarity calculation model, considering the similarity of multiple dimensions such as physiological parameters, behavior patterns, and treatment responses, to realize knowledge transfer between similar patients.

[0124] Preferably, the embodiment adopts a domain adaptive method based on optimal transport theory, accurately quantifies the distribution difference between the source patient and the target patient, and learns the optimal feature transformation matrix to improve the accuracy of knowledge transfer. At the same time, a lifelong learning architecture is designed to enable the system to continuously accumulate and integrate experience from different patients, forming an evolving medical knowledge base.

[0125] The above describes the embodiments of the present application, but the embodiments are not limited to the specific implementation described above, which is only illustrative but not restrictive. Those skilled in the art can make more forms of equivalent embodiments under the inspiration of the embodiments, which are all within the protection of the embodiments.

Claims

1. A pressure-adaptive based intermittent intestinal fluid return system, comprising: The method comprises the following steps: a data acquisition module is used to collect the context state data, the physiological rhythm data and the intestinal pressure data of a patient through a multi-modal sensor network to obtain a multi-dimensional monitoring data stream; the context state data includes activity state, posture change data, environmental parameters and emotional state data; the physiological rhythm data includes circadian rhythm data, digestive cycle data and hormone secretion cycle data; a phase extraction and synchronization metric calculation module is used to extract the context state phase corresponding to the context state data, the physiological rhythm phase corresponding to the physiological rhythm data and the intestinal pressure phase corresponding to the intestinal pressure data by using time series signal processing technology according to the multi-dimensional monitoring data stream, and to calculate the synchronization metric between the context state phase, the physiological rhythm phase and the intestinal pressure phase; a coordinated state classification module is used to classify the coordinated state of the context state, the physiological rhythm and the intestinal pressure by using a pre-constructed synchronization state classification model according to the synchronization metric, wherein the classification of the coordinated state includes an ideal coordinated state, a sub-ideal coordinated state and a non-ideal coordinated state; a feedback decision and parameter adjustment module is used to generate a feedback decision instruction based on the classification result of the coordinated state, to control a feedback device to perform intestinal fluid feedback based on the feedback decision instruction, and to adjust the feedback parameter adaptively according to the intestinal pressure data; the specific steps of calculating the synchronization metric between the context state phase, the physiological rhythm phase and the intestinal pressure phase are as follows: a phase difference matrix is constructed based on the context state phase, the physiological rhythm phase and the intestinal pressure phase; and the overall synchronization metric value between the context state phase, the physiological rhythm phase and the intestinal pressure phase is calculated by using a phase synchronization index formula according to the phase difference matrix; the specific steps of pre-constructing the synchronization state classification model include: based on a pre-set elastic synchronization belt threshold range, the synchronization metric value intervals corresponding to the ideal coordinated state, the sub-ideal coordinated state and the non-ideal coordinated state are defined; and the coordinated state is classified according to the interval in which the overall synchronization metric value falls; the specific steps of generating the feedback decision instruction include: based on the classification result and the prediction result of the coordinated state, a time window in which the ideal coordinated state appears is identified; and the start time, the rate and the duration of the feedback are determined according to the time window and the real-time intestinal pressure data.

2. The pressure-adaptive, intermittent intestinal liquid return system of claim 1, wherein, The specific steps of extracting the context state phase are as follows: the context state data is decomposed into a series of intrinsic mode functions by using an empirical mode decomposition algorithm; the phase information of the intrinsic mode functions is obtained by converting each intrinsic mode function to a phase space by using a Hilbert transform; and a unified context state phase representation is obtained by integrating the phase information by using principal component analysis; The specific steps of extracting the physiological rhythm phase are as follows: the physiological rhythm data is decomposed into different scales by using a wavelet transform to obtain wavelet coefficients of different scales; the phase locking value is calculated based on the wavelet coefficients; and the instantaneous phase of each physiological cycle in the phase locking value is extracted; The specific steps of extracting the intestinal pressure phase are as follows: the periodic pressure fluctuation pattern is identified from the intestinal pressure data; and the fluctuation pattern is converted into a phase representation.

3. The pressure-adaptive, intermittent intestinal liquid return system of claim 1, wherein, The synchronous state classification model further comprises a time series prediction model based on a recurrent neural network, configured to predict future collaborative state classification according to a historical sequence of synchronization metric values.

4. The pressure-adaptive, intermittent intestinal liquid return system of claim 1, wherein, Adaptive adjustment of the return parameters according to the intestinal pressure data comprises establishing a negative feedback control logic between the return rate and the real-time intestinal pressure value, and reducing the return rate or suspending the return when the real-time intestinal pressure value exceeds a preset first safety threshold.

5. The pressure-adaptive, intermittent enteral liquid return system of claim 1, wherein, The system further comprises: The collaborative state classification, return parameters and intestinal pressure response data before and after each return activity are recorded to form an individualized data set; and the elastic synchronization band threshold range and / or return decision parameters in the synchronous state classification model are adjusted based on the individualized data set using an incremental learning algorithm.

Images