Intelligent control method and system for pressure of endotracheal tube cuff

Abstract

This application discloses an intelligent control method and system for endotracheal intubation cuff pressure. The method includes: acquiring multi-dimensional signals such as patient airway pressure and body position; extracting features and quantifying the intensity of physiological events such as coughing and turning over; assessing the impact range of events when the intensity exceeds a threshold, considering hemodynamic fluctuations and tissue perfusion tolerance specific to patients with severe burns and shock; using a machine learning model to predict expected adjustment values ​​by integrating tracheal mucosal vulnerability indicators and historical data; and finally generating commands to drive control through closed-loop feedback and iterative optimization. The solution provided in this application helps achieve scientific and targeted management of cuff pressure in critically ill patients. Through predictive and adaptive closed-loop control, it can effectively respond to emergencies, help reduce the risk of intubation complications, and improve patient safety.

Description

Technical Field

[0001] This application relates to the field of medical monitoring and intelligent control technology, specifically to an intelligent control method and system for endotracheal intubation cuff pressure. Background Technology

[0002] In intensive care medicine, endotracheal intubation is a crucial medical measure for maintaining respiratory function and ensuring vital access to life in critically ill patients. The cuff at the end of the intubation tube, when inflated, seals the gap between the trachea and the tube wall, ensuring effective positive pressure ventilation while preventing aspiration of oropharyngeal secretions or gastric contents into the lower respiratory tract. Cuff pressure control is a core aspect of endotracheal intubation management. Too low a pressure leads to incomplete airway closure, increasing the risk of ventilation leakage and aspiration, while too high a pressure can compress the tracheal mucosa, causing serious complications such as local ischemia, edema, necrosis, and even tracheoesophageal fistula. Therefore, maintaining cuff pressure within an ideal range that effectively seals the airway without damaging the tracheal mucosa is vital for patient treatment and recovery.

[0003] Currently, the clinical regulation of endotracheal intubation cuff pressure mainly relies on intermittent manual measurement and adjustment by healthcare professionals. This approach has significant limitations. First, the frequency of manual adjustment is limited, making it difficult to cope with instantaneous and drastic airway pressure changes caused by factors such as coughing, changes in body position, and adjustments to ventilator parameters, leading to frequent deviations of the cuff pressure from the safe range. Second, adjustment decisions are primarily based on the clinical experience of healthcare professionals and single pressure gauge readings, lacking a comprehensive analysis of the patient's multidimensional physiological signals. For example, changes in patient position may alter airway geometry, thus affecting the ideal cuff pressure value, and pressure readings alone cannot capture such dynamic needs.

[0004] For certain critically ill patient groups, such as those with severe burns complicated by acute respiratory distress syndrome (ARDS) and shock, cuff pressure management is more complex and challenging. These patients often have highly edematous and fragile tracheal mucosa due to systemic inflammation and massive fluid exudation, making them extremely sensitive to pressure injuries. Simultaneously, hemodynamic instability and insufficient tissue perfusion caused by shock further reduce the tracheal mucosa's tolerance to pressure. This makes the "safe window" for cuff pressure regulation very narrow. Furthermore, the prone ventilation therapy commonly used for these ARDS patients directly affects airway geometry due to positional changes, resulting in dynamic variations in cuff pressure requirements. Current techniques cannot incorporate these complex, patient-specific pathophysiological characteristics into regulatory considerations, making real-time, precise, and individualized pressure management difficult, thus increasing the risk of complications and impacting patient prognosis. Summary of the Invention

[0005] The purpose of this application is to provide an intelligent control method and system for endotracheal intubation cuff pressure, in order to solve a series of technical problems in the prior art, namely, the inability of existing control methods to respond to dynamic changes in patients in real time, the lack of comprehensive analysis of multidimensional signals, and the difficulty in adapting to the special physiological needs of critically ill patients with severe burns complicated by ARDS and shock.

[0006] To achieve the above objectives, one aspect of the present invention provides an intelligent control method for endotracheal intubation cuff pressure, the method comprising the following steps:

[0007] Step one involves real-time acquisition of multidimensional signals using sensors positioned around the airway environment of patients with acute respiratory distress syndrome (ARDS). These multidimensional signals include at least airway pressure changes and body position variations while the patient is in a prone ventilation position. The acquired raw signals may contain noise interference from the environment or equipment; therefore, signal filtering methods, such as Kalman filters or bandpass filters, are used to process the raw signals to remove noise components, thereby obtaining a clean multidimensional signal dataset for subsequent analysis.

[0008] Step two involves processing the acquired clean, multidimensional signal dataset using a time-series analysis algorithm to extract characteristic patterns that characterize specific physiological events in the patient. Specifically, for events such as coughing and turning over that may cause drastic fluctuations in airway pressure, unique patterns are extracted from both the time and amplitude dimensions of the signal. For example, a cough event is characterized by a brief and sharp peak in the airway pressure signal. Further, for changes in airway geometry caused by patient positional changes (such as turning over) during prone ventilation, the timing of the event is accurately determined by analyzing the correlation between position sensor data and airway pressure data. The intensity level of the event is quantified by combining the amplitude and duration of pressure fluctuations. In an optional embodiment, the assessment of airway geometry changes can be indirectly evaluated by analyzing changes in the pressure-time and flow-time curves monitored by the ventilator. When changes in patient position cause changes in airway resistance, respiratory mechanics parameters such as peak inspiratory pressure and inspiratory resistance will change accordingly. The system can determine the occurrence and extent of morphological changes in real time using a pre-established correlation model between waveform features and airway geometry changes.

[0009] Step three involves comparing the event intensity level extracted in step two with a preset event intensity threshold. If the extracted event intensity level exceeds the preset threshold, it indicates that the event may have a significant impact on cuff pressure, triggering the interaction effect assessment module. This module performs correlation calculations on various signals in a clean multidimensional signal dataset, such as calculating the correlation coefficient between airway pressure signals and hemodynamic signals like heart rate and blood pressure, to assess the linkage effect of the event. In particular, this invention incorporates the physiological characteristics unique to patients with severe burns and shock, such as severe hemodynamic fluctuations and fragile tissue perfusion tolerance, as key tolerance factors into the assessment model to determine the potential impact range of pressure fluctuations caused by the event on the patient; the potential impact range is defined as a pressure value interval centered on the current pressure value.

[0010] Step four involves retrieving historical control data from the historical database within the potential impact range of pressure fluctuations identified in step three, based on this range. Subsequently, a cuff pressure prediction model is constructed using a Support Vector Machine (SVM) algorithm. A key input feature of this model is an index quantifying the high vulnerability of the tracheal mucosa in severely burned patients. This index is comprehensively assessed based on the patient's total burn area, severity of inhalation injury, and levels of inflammatory markers such as C-reactive protein. By using this vulnerability index, historical control data, and real-time multidimensional signal data as model inputs, the SVM model is trained to learn and predict an expected adjustment value for the cuff pressure that ensures effective airway closure while minimizing mucosal damage. The "expected adjustment value" referred to here is the pressure reference benchmark output by the model, used to guide subsequent control operations.

[0011] Furthermore, in a preferred embodiment, to enhance the ability to handle the uncertainty of the physiological state of critically ill patients, the prediction model can be upgraded to a hybrid prediction model, such as a Gaussian Process Regression-Support Vector Machine (GPR-SVM) hybrid prediction model. Gaussian Process Regression (GPR) is a non-parametric Bayesian regression method. Its characteristic is that it can model the probability distribution of a function, rather than simply fitting a single function. Therefore, its prediction results include not only the prediction mean but also the variance, which quantifies the degree of prediction uncertainty. Support Vector Machine (SVM) is a supervised learning model used for classification or regression analysis. In this invention, it is used as a decision optimization module, and its technical function is to partition or map data points by constructing an optimal hyperplane. This model uses the predicted mean and variance output by the Gaussian process regression model as input features. After training, it can make a more robust final control decision, especially when the variance output by the Gaussian process regression model is large, i.e., when the uncertainty is high, it can choose a more prudent control strategy.

[0012] Specifically, a Gaussian process regression model is first used to perform a preliminary analysis of the multidimensional time-series signal. Gaussian process regression is a nonparametric model based on Bayesian theory that can provide a probability distribution for prediction, that is, it can not only predict the specific pressure adjustment value, but also give the confidence interval of the predicted value (e.g., expressed in the form of mean and variance). Subsequently, the mean and variance of the adjustment value output by the Gaussian process regression model (which represents the degree of uncertainty in the prediction), as well as the aforementioned quantitative index of tracheal mucosal vulnerability, are used as enhancement features and input into the support vector machine model. At this stage, the support vector machine model plays a decision optimization role, making a final, more robust control decision based on the input features. For example, when the variance predicted by the Gaussian process regression model is large, i.e., the uncertainty is high, the support vector machine model will choose a more conservative pressure adjustment instruction, so that the control strategy can better adapt to clinical emergencies and unknown situations.

[0013] Step 5: Based on the expected adjustment value obtained in Step 4, a closed-loop feedback mechanism is used to generate control commands. The system sends an initial command signal to the control device (such as a miniature air pump or solenoid valve) connected to the endotracheal intubation cuff. Simultaneously, the system comprehensively considers the cumulative damage risk that may result from long-term intubation and corrects the command signal to determine a target cuff pressure; this target cuff pressure is a target value for immediate control obtained based on the expected adjustment value and corrected for cumulative damage risk.

[0014] Step Six: The system continuously monitors the current actual cuff pressure and calculates the deviation between this pressure and the target cuff pressure determined in Step Five. If the absolute value of this deviation exceeds a preset pressure deviation threshold, an optimization iteration loop is activated. In this loop, the system continuously monitors and collects the latest multidimensional signal data to update the prediction model described in Step Four in real time. During the model update process, the model parameters are adjusted in particular, taking into account factors that may lead to decreased pressure tolerance due to changes in the patient's condition, such as increased tissue edema or deterioration of circulatory status. The updated model can generate a more accurate and safer optimized sequence of adjustment instructions.

[0015] Step seven involves using the optimized adjustment command sequence generated in step six to drive the endotracheal intubation cuff control device to perform the final pressure regulation operation. After the operation, the system does not terminate monitoring but continuously analyzes the multidimensional signals after regulation to determine whether each physiological signal has returned to the expected equilibrium state. Specifically, for ARDS patients undergoing prone ventilation, the system focuses on monitoring real-time oxygenation levels and, combined with an assessment of changes in airway geometry, obtains final confirmation data on the regulation effect to verify the effectiveness and safety of the regulation operation and provide a data basis for the next round of regulation.

[0016] Another aspect of the present invention provides an intelligent control system for endotracheal intubation cuff pressure, the system for implementing the above method, comprising:

[0017] The signal acquisition module includes sensing devices such as airway pressure sensors and body position sensors, and integrates signal filtering circuits or corresponding software algorithms. The function of this module is to perform step one of the aforementioned method, namely, to acquire multidimensional signals from the patient's airway environment, perform noise reduction processing, and finally output a clean multidimensional signal dataset.

[0018] The event feature extraction module, typically consisting of a processor executing corresponding software programs, receives a multidimensional signal dataset from the signal acquisition module and executes step two of the aforementioned method. This involves employing a time-series analysis algorithm to identify and extract feature patterns from key physiological events such as coughing and turning over, and to determine the timing and intensity of these events.

[0019] The interaction impact assessment module is configured to perform step three of the aforementioned method. When an event with an intensity exceeding a preset threshold is received from the event feature extraction module, this module is activated to calculate the correlation between multidimensional signals and, in conjunction with a preset patient-specific physiological model (e.g., hemodynamic characteristics and tissue perfusion tolerance of a patient with severe burns and shock), to determine the potential impact range of pressure fluctuations caused by the event.

[0020] The pressure prediction module is configured to perform step four of the aforementioned method. This module embeds a machine learning model, such as a Support Vector Machine (SVM) model or the aforementioned GPR-SVM hybrid prediction model. It receives the influence range output by the interaction impact assessment module, calls historical regulation data, and integrates quantitative indicators characterizing specific patient physiological features (such as tracheal mucosal vulnerability) to predict the optimal expected adjustment value for cuff pressure.

[0021] The control command generation module is responsible for executing steps five and six of the aforementioned method. It receives the expected adjustment value output by the pressure prediction module and generates preliminary control commands based on a closed-loop feedback mechanism. This module continuously compares the command parameters with the current actual cuff pressure. If the deviation exceeds a preset threshold, it initiates an optimization iteration loop, updating the model by calling the pressure prediction module, and ultimately generating an optimized sequence of adjustment commands.

[0022] The pressure execution and monitoring module is configured to perform step seven of the aforementioned method. It includes physical actuators such as a miniature air pump and a solenoid valve, used to receive and execute the instruction sequence from the control instruction generation module to physically regulate the cuff pressure. Simultaneously, this module also includes corresponding pressure sensors to continuously monitor the regulated cuff pressure and other physiological signals, and feeds the monitoring data back to other processing modules to determine whether the system has returned to equilibrium and to provide final confirmation of the control effect.

[0023] In the system implementation, the event feature extraction module, interaction impact assessment module, pressure prediction module, and control command generation module can be integrated into the central data processing unit (such as an embedded system or computer) and communicate via an internal bus. The signal acquisition module and pressure execution and monitoring module are connected to the central data processing unit via wired or wireless communication.

[0024] This invention quantifies the unique physiological characteristics of severely burned patients with ARDS and shock, such as tracheal mucosal fragility, drastic hemodynamic fluctuations, and tissue perfusion tolerance, and incorporates these as key tolerance factors into predictive models and regulatory strategies. This facilitates precise adaptive regulation of cuff pressure in specific critical care scenarios, solving the technical problem of existing technologies that use a uniform regulation strategy and cannot make individualized adjustments for the special physiological conditions of such critically ill patients. Furthermore, this invention actively identifies and quantifies the intensity levels of characteristic patterns of compound events such as coughing and turning over from multidimensional signal datasets using time-series analysis algorithms. This enables the system to proactively respond to pressure fluctuations rather than passively reacting. Combined with subsequent interaction impact assessment, this helps improve the accuracy and predictability of regulation. Furthermore, this invention constructs a predictive model based on machine learning algorithms such as support vector machines and designs an optimization iterative loop that is activated when the deviation exceeds a threshold. This enables the system to not only adjust according to the current state but also learn and self-optimize based on historical data and continuous monitoring. Compared with traditional reactive control based on fixed thresholds or simple proportional-integral-derivative (PID) control, this predictive and adaptive closed-loop control mechanism helps to safely and effectively manage the endotracheal intubation cuff pressure of critically ill patients, thereby helping to reduce the risk of long-term intubation-related complications and improve patient safety. Attached Figure Description

[0025] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0026] Figure 1 This is a flowchart of an intelligent control method for endotracheal intubation cuff pressure provided in an embodiment of the present invention.

[0027] Figure 2 This is a structural block diagram of an intelligent control system for endotracheal intubation cuff pressure provided in an embodiment of the present invention.

[0028] Figure 3 This is a schematic diagram of the data processing flow of the hybrid prediction model in an embodiment of the present invention.

[0029] Figure 4 This is a waveform diagram used in this embodiment of the invention to schematically demonstrate the extraction of cough event feature patterns from airway pressure signals.

[0030] Figure 5 This is a schematic diagram of the decision-making process of the hybrid prediction model in handling uncertain events in an embodiment of the present invention.

[0031] Figure 6 This is a comparison chart of the expected adjustment values ​​output by the prediction model under different tracheal mucosal vulnerability indices in this embodiment of the invention. Detailed Implementation

[0032] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0033] Example 1

[0034] This embodiment provides an intelligent control method for the pressure of the endotracheal intubation cuff. (Refer to...) Figure 1 This method is mainly used for cuff pressure management in patients with acute respiratory distress syndrome (ARDS) receiving prone ventilation, especially critically ill patients with severe burns and shock. The specific steps may include the following:

[0035] Step S100 involves real-time acquisition of multidimensional signals using sensors positioned around the airway environment of the patient with acute respiratory distress syndrome (ARDS). These multidimensional signals include at least airway pressure change data and body position change data during prone ventilation. Specifically, airway pressure change data can be acquired by a miniature pressure sensor connected to the endotracheal tube, and body position change data can be acquired by a triaxial accelerometer fixed to the patient's chest or back.

[0036] In one specific embodiment, the multidimensional signal may also include hemodynamic signals acquired through other monitoring devices, such as continuous heart rate and arterial blood pressure. Since the acquired raw signal inevitably contains noise components caused by environmental electromagnetic interference, device noise, or unconscious minor movements of the patient, preprocessing of the raw signal is necessary to ensure the accuracy of subsequent analysis. This embodiment preferably employs a signal filtering method, such as a Kalman filter. The Kalman filter is a highly efficient recursive filter capable of optimally estimating measurements based on the dynamic model of the signal, and is particularly suitable for processing dynamic system signals containing random noise. Alternatively, a bandpass filter can be used, setting an appropriate passband range, such as 0.1 Hz to 50 Hz, to filter out high-frequency noise and baseline drift, thereby obtaining a clean multidimensional signal dataset for subsequent analysis.

[0037] Step S200: Based on the acquired clean multidimensional signal dataset, a time-series analysis algorithm is used to process the signal sequence to extract feature patterns that can characterize specific physiological events in the patient. (Refer to...) Figure 4 The figure schematically illustrates the process of identifying and extracting cough event feature patterns 402 from airway pressure signals 401. Figure 4 In this context, the airway pressure signal 401 represents the raw signal waveform of continuous monitoring, while the cough event feature pattern 402 is a schematic representation of the data points in the raw signal that match the characteristics of a cough event, clearly showing the identified event, which is not an independent physiological signal.

[0038] Specifically, for events such as coughing and turning over that may cause drastic fluctuations in airway pressure, the system extracts unique patterns from both the temporal and amplitude dimensions of the signal. For example, a cough event 402 is characterized by one or more consecutive, brief, and abrupt pressure peaks in the airway pressure signal, with amplitudes far exceeding the range of normal respiratory fluctuations. Postural changes, such as turning over, may manifest as a step change in the position sensor data, accompanied by a persistent shift in the airway pressure baseline. Further, this step determines the intensity level of the event by quantifying key parameters of these patterns. For example, the intensity level of a cough event can be quantified as: Intensity = Peak Pressure × Duration. Similarly, the intensity level of a postural change event can be comprehensively assessed based on the magnitude of the change in positional angle and the amplitude of the airway pressure baseline shift.

[0039] In an optional embodiment, the assessment of airway geometry changes caused by changes in body position can be performed indirectly, in addition to directly analyzing body position sensor data, by analyzing changes in the pressure-time and flow-time curves monitored by the ventilator. When changes in patient position cause airway compression or tortuosity, leading to changes in airway resistance, respiratory mechanics parameters such as peak inspiratory pressure, plateau pressure, and inspiratory resistance monitored by the ventilator will change accordingly. The system can use a pre-established correlation model between waveform characteristics and airway geometry changes, such as a decision tree model that maps changes in inspiratory resistance to levels of airway geometry change (e.g., mild, moderate, significant), to determine the occurrence and extent of morphological changes in real time, providing a basis for subsequent adjustments.

[0040] In step S300, the event intensity level extracted in step S200 is compared with a preset event intensity threshold. This threshold is statistically derived from clinical experience data and is used to distinguish between events that significantly affect cuff pressure and negligible physiological fluctuations. If the extracted event intensity level exceeds the preset threshold, it indicates that the event may have a significant impact on cuff pressure, at which point the interaction effect assessment module is triggered. This module performs correlation calculations on various signals in a clean multidimensional signal dataset to assess the linkage effect of the event. For example, the Pearson correlation coefficient between the airway pressure signal sequence and hemodynamic signal sequences such as heart rate and blood pressure can be calculated within the event occurrence time window.

[0041] Specifically, this invention incorporates physiological characteristics specific to patients with severe burns complicated by shock as key tolerance factors into the assessment model. These characteristics include dramatic hemodynamic fluctuations (manifested as large fluctuations in blood pressure and heart rate) and tissue perfusion tolerance (e.g., assessed based on indicators such as mean arterial pressure and lactate levels). The assessment model quantifies these factors into a comprehensive tolerance score, which, for example, can be calculated by weighting the amplitude of hemodynamic fluctuations and tissue perfusion-related indicators. When the calculated signal correlation is high and the patient's comprehensive tolerance score is low, the system determines that the pressure fluctuations caused by the event may have a wide potential impact on the patient, requiring more cautious and precise intervention.

[0042] In step S400, based on the potential impact range of pressure fluctuations determined in step S300, historical regulatory data within this impact range is retrieved from the historical database as a reference. Subsequently, a cuff pressure prediction model is constructed using the Support Vector Machine (SVM) algorithm. A key input feature of this model is an index quantifying the high vulnerability of the tracheal mucosa in severely burned patients. This index can be obtained through comprehensive assessment based on clinical data, and for example, it can be defined as: Vulnerability Index = w1 × Total Burn Area (%) + w2 × Severity Grade of Inhalation Injury + w3 × C-Reactive Protein Level, where w1, w2, and w3 are weighting coefficients. The SVM model is trained by using this vulnerability index, the retrieved historical regulatory data, and real-time multidimensional signal data (such as event intensity, hemodynamic parameters, etc.) as multidimensional input features of the model. Support Vector Machines (SVMs) learn and predict an optimal cuff pressure adjustment value that ensures effective airway sealing while minimizing mucosal damage by finding an optimal hyperplane to divide data points with different control outcomes (such as "safe and effective," "risk of leakage," and "risk of excessive pressure"). Figure 6 This figure illustrates the changing trends of the predicted adjustment values ​​output by the predictive model under different tracheal mucosal vulnerability indices (low, medium, and high) based on varying cough event intensity levels. It can be seen that when the vulnerability index is high, even when facing cough events of equal intensity, the model outputs a more conservative (i.e., lower) pressure adjustment value, thus reflecting the model's adaptability to individualized patient characteristics. The predicted adjustment value mentioned here serves as a pressure reference benchmark output by the model to guide subsequent regulatory actions.

[0043] Furthermore, in a preferred embodiment, to enhance the ability to handle uncertainties in the physiological state of critically ill patients, the prediction model can be upgraded to a hybrid prediction model, namely a Gaussian process regression-support vector machine (GPR-SVM) hybrid prediction model. (Refer to...) Figure 3The data processing flow of this hybrid model is as follows: First, the multidimensional time-series signal 301 is input into the Gaussian process regression (GPR) model 302. Gaussian process regression is a nonparametric model based on Bayesian theory, which can provide a probability distribution for prediction. That is, it not only predicts the specific pressure adjustment value mean 303, but also gives the confidence interval of the predicted value, usually expressed in the form of variance 304. This variance represents the degree of uncertainty of the prediction. Subsequently, the adjustment value mean 303, variance 304 output by the Gaussian process regression model, as well as the aforementioned tracheal mucosal vulnerability quantification index 305, are used as enhancement features and input together into the support vector machine (SVM) model 306. The support vector machine model plays a decision optimization function at this stage, making a final, more robust control decision based on these enhancement features, and outputting the final expected adjustment value 307. For example, when the variance predicted by the Gaussian process regression model is large (304), indicating high uncertainty, the support vector machine model is trained to select a more conservative stress adjustment instruction, such as reducing the adjustment magnitude or maintaining the current stress, thereby enabling the regulatory strategy to better adapt to clinical emergencies and unknown situations. (Refer to...) Figure 5 The figure illustrates the working mechanism of this hybrid prediction model. When a "postural change event" occurs, the external disturbance leads to an "increased uncertainty" in the prediction, manifested as a significant expansion of the GPR confidence interval (light gray area). At this time, although the mean of the GPR prediction (solid black line) fluctuates considerably, the SVM model, acting as the decision optimizer, integrates the information of high uncertainty and outputs a more conservative adjustment value (black star) located in the lower half of the confidence interval. This effectively avoids aggressive regulation under uncertain conditions, ensuring patient safety.

[0044] In step S500, based on the expected adjustment value obtained in step S400, a closed-loop feedback mechanism is used to generate a control command. The system sends an initial command signal to the control device (such as a miniature air pump or solenoid valve) connected to the endotracheal intubation cuff. Simultaneously, the system comprehensively considers the cumulative risk of injury that may result from long-term intubation and corrects the command signal accordingly. For example, the system can maintain a cumulative pressure-time integral value. When this integral value exceeds a preset long-term risk threshold, the system will make a small negative correction to the expected adjustment value output by the model, thereby determining a target cuff pressure.

[0045] In step S600, the system continuously monitors the current actual cuff pressure and calculates the deviation between this pressure and the target cuff pressure determined in step S500. If the absolute value of this deviation is greater than a preset pressure deviation threshold (e.g., 2 cmH2O), an optimization iteration loop is activated. In this loop, the system continuously monitors and collects the latest multidimensional signal data to update the prediction model described in step S400 in real time. During the model update process, factors that may lead to decreased pressure tolerance due to changes in the patient's condition are specifically considered, such as by monitoring indicators like peripheral tissue oxygen saturation or central venous pressure to determine whether the patient's tissue edema has worsened or circulatory status has deteriorated. If tolerance is determined to be decreased, the weight of the tracheal mucosal vulnerability index is increased accordingly during model update. The updated model can generate a more accurate and safer optimized adjustment instruction sequence.

[0046] In step S700, the optimized adjustment command sequence generated in step S600 is used to drive the control device of the endotracheal intubation cuff to perform the final pressure control operation. The control device can be a component consisting of a miniature air pump and a solenoid valve, achieving fine adjustment of the cuff pressure by precisely controlling the inflation volume of the air pump and the deflation volume of the solenoid valve. After the operation is executed, the system does not terminate monitoring but continuously analyzes the multidimensional signals after control to determine whether the physiological signals have returned to the expected equilibrium state. Specifically, for ARDS patients undergoing prone ventilation, the system will focus on monitoring real-time oxygenation level indicators (such as pulse oxygen saturation, SpO2) after control, and combine this with continuous assessment of changes in airway geometry to obtain final confirmation data of the control effect. For example, confirming that the cuff pressure remains stable at the target value in the new position and that there is no air leakage (which can be determined by auscultation or monitoring of exhaled tidal volume), while the patient's oxygenation level remains stable or improves, verifies the effectiveness and safety of the control operation and provides a data basis for the next round of control cycle.

[0047] Example 2

[0048] This embodiment provides an intelligent control system for the pressure of the endotracheal intubation cuff, which is used to implement the method described in Embodiment 1. (Refer to...) Figure 2 The system may include: a signal acquisition module 201, an event feature extraction module 202, an interaction impact assessment module 203, a pressure prediction module 204, a control command generation module 205, and a pressure execution and monitoring module 206.

[0049] The signal acquisition module 201 includes sensing devices such as airway pressure sensors and body position sensors, and integrates signal filtering circuits or corresponding software algorithms. The function of this module is to perform step S100 in the aforementioned method, namely, acquiring multidimensional signals from the patient's airway environment, performing noise reduction processing, and finally outputting a clean multidimensional signal dataset. Physically, the sensors can be wired to a data acquisition unit, which incorporates an analog-to-digital converter and a basic digital signal processor.

[0050] The event feature extraction module 202 is typically composed of a processor executing a corresponding software program. It receives a multidimensional signal dataset from the signal acquisition module 201 and executes step S200 of the aforementioned method, that is, using a time-series analysis algorithm to identify and extract feature patterns of key physiological events such as coughing and turning over, and determine the time point and intensity level of their occurrence.

[0051] The interaction impact assessment module 203 is configured to perform step S300 in the aforementioned method. When an event with an intensity exceeding a preset threshold is received from the event feature extraction module 202, this module is activated to calculate the correlation between multidimensional signals and, in conjunction with a preset patient-specific physiological model (e.g., hemodynamic characteristics and tissue perfusion tolerance of a patient with severe burns complicated by shock), to determine the potential impact range of pressure fluctuations caused by the event.

[0052] The pressure prediction module 204 is configured to perform step S400 of the aforementioned method. This module embeds a machine learning model, such as a support vector machine (SVM) model or the aforementioned GPR-SVM hybrid prediction model. It receives the influence range output by the interaction impact assessment module 203, calls historical regulation data, and integrates quantitative indicators characterizing specific patient physiological characteristics (such as tracheal mucosal vulnerability) to predict the optimal expected adjustment value of the cuff pressure.

[0053] The control command generation module 205 is responsible for executing steps S500 and S600 in the aforementioned method. It receives the expected adjustment value output by the pressure prediction module 204 and generates preliminary control commands based on a closed-loop feedback mechanism. This module continuously compares the command parameters with the current actual cuff pressure. If the deviation exceeds a preset threshold, it initiates an optimization iteration loop, updating the model by calling the pressure prediction module 204, and ultimately generating an optimized sequence of adjustment commands.

[0054] The pressure execution and monitoring module 206 is configured to execute step S700 of the aforementioned method. It includes physical actuators such as a miniature air pump and a solenoid valve, used to receive and execute the instruction sequence from the control instruction generation module 205 to physically regulate the cuff pressure. Simultaneously, this module also includes corresponding pressure sensors to continuously monitor the regulated cuff pressure and other physiological signals, and feeds the monitoring data back to other processing modules, such as the signal acquisition module 201 or the control instruction generation module 205, to determine whether the system has returned to equilibrium and to finally confirm the control effect, forming a complete closed-loop control.

[0055] In the system implementation, the event feature extraction module 202, the interaction impact assessment module 203, the pressure prediction module 204, and the control command generation module 205 can be integrated into a central data processing unit (such as an embedded system or a medical-grade tablet computer), exchanging data through an internal bus or inter-process communication. The signal acquisition module 201 and the pressure execution and monitoring module 206 serve as peripheral devices, connected to the central data processing unit via wired (such as USB) or wireless communication (such as Bluetooth Low Energy), forming a complete, automated intelligent control system.

[0056] Example 3

[0057] This embodiment provides a specific clinical application of the method of the present invention. The application scenario involves a severely burned patient with acute respiratory distress syndrome (ARDS) and septic shock, who is receiving prone ventilation. Given the highly fragile tracheal mucosa and poor tissue perfusion in such patients, resulting in a narrow window for safe cuff pressure control, the method of the present invention enables dynamic and precise control.

[0058] The specific implementation process of the method of this invention is as follows: First, the signal acquisition module continuously acquires multidimensional signals from the patient, including airway pressure, data from a triaxial accelerometer fixed to the patient's back, and continuous arterial blood pressure and heart rate data acquired through a monitor. During the patient's repositioning operation from supine to prone, the event feature extraction module identifies the "repositioning change" event by analyzing the continuous step changes in the accelerometer data. Simultaneously, a sharp pressure peak appears in the airway pressure signal, and the time-series analysis algorithm extracts the characteristic pattern of the "cough" event accordingly. The system quantifies the amplitude of the repositioning change and the peak pressure of the cough, determining that the intensity levels of both events exceed a preset threshold, constituting a composite event.

[0059] Subsequently, the interaction impact assessment module was triggered. The module's analysis process was as follows: First, it detected a transient and significant drop in the patient's arterial blood pressure simultaneously with the coughing event, indicating hemodynamic instability. Second, considering the patient's pre-defined physiological characteristics of a large burn area and fragile tissue perfusion, the assessment module determined that this combined event posed a high risk to the stability of the cuff pressure and the safety of the tracheal mucosa, and accordingly identified a potential impact range requiring careful adjustment.

[0060] Based on the above assessment results, the pressure prediction module is invoked. This module employs a hybrid prediction model of GPR-SVM, taking quantitative parameters of postural changes, cough event intensity, blood pressure fluctuation amplitude, and quantitative indicators representing tracheal mucosal vulnerability as model inputs. The Gaussian process regression (GPR) model predicts the trend of pressure changes; due to uncertainties such as blood pressure fluctuations, its output prediction results contain significant variance. After receiving this high-variance signal and the high vulnerability indicator, the support vector machine (SVM) model performs decision optimization and outputs a prudent expected adjustment value sequence. The specific instructions of this sequence are: first, perform a small-amplitude rapid pressure replenishment to address the potential leakage risk during coughing; then, after the patient's position stabilizes in the prone position, adjust the target pressure value to a new baseline, which is higher than the pressure value in the supine position but still within the preset protective pressure range.

[0061] Finally, the control command generation module and the pressure execution and monitoring module work together. The control command generation module generates the final control command based on the command sequence output by the prediction module. The pressure execution and monitoring module drives the micro-pump and solenoid valve to execute this command sequence, completing the precise control operation. After the operation is completed, the system continues to perform the monitoring function, continuously monitoring the cuff pressure, ventilator delivery parameters, and the patient's pulse oximetry, and uses the monitoring data as input for the next control cycle to confirm that the system has reached the expected equilibrium state after control and to provide a data basis for subsequent control.

[0062] The above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An intelligent regulating device for cuff pressure of a tracheal tube, characterized in that, include: The system comprises a signal acquisition module (201), an event feature extraction module (202), an interaction impact assessment module (203), a pressure prediction module (204), a control command generation module (205), and a pressure execution and monitoring module (206). The output of the signal acquisition module (201) is connected to the input of the event feature extraction module (202), the output of the event feature extraction module (202) is connected to the input of the interaction impact assessment module (203), the output of the interaction impact assessment module (203) is connected to the input of the pressure prediction module (204), the output of the pressure prediction module (204) is connected to the input of the control command generation module (205), the output of the control command generation module (205) is connected to the input of the pressure execution and monitoring module (206), and the monitoring data from the pressure execution and monitoring module (206) is fed back to the control command generation module (205) to form a closed-loop control. The signal acquisition module (201) is used to acquire multidimensional signals from patients with acute respiratory distress syndrome in real time through sensors, and to filter the multidimensional signals to obtain a clean multidimensional signal dataset; the multidimensional signals include at least airway pressure change data and body position change data of the patient in prone ventilation; wherein, the filtering of the multidimensional signals is specifically performed by using a Kalman filter or a bandpass filter to process the multidimensional signals to remove noise components; The event feature extraction module (202) is used to process the clean multidimensional signal dataset using a time-series analysis algorithm to extract feature patterns and event intensity levels that can characterize specific physiological events of the patient; specifically, it includes: identifying cough events or postural change events from the clean multidimensional signal dataset, and extracting the event patterns from the time and amplitude dimensions of the signal; and determining the event intensity level by quantifying the key parameters of the event patterns, wherein the intensity level of the cough event is quantified as the product of peak pressure and duration; The interaction impact assessment module (203) is used to compare the event intensity level with a preset event intensity threshold. When the event intensity level exceeds the preset event intensity threshold, the interaction impact assessment is triggered to determine the potential impact range of the pressure fluctuation. The interaction impact assessment specifically includes: performing correlation calculation on the airway pressure signal and hemodynamic signal in the clean multidimensional signal dataset; and incorporating the hemodynamic fluctuation characteristics and tissue perfusion tolerance of patients with severe burns and shock as key tolerance factors into the assessment model, and combining the results of the correlation calculation to determine the potential impact range of the pressure fluctuation caused by the event. The pressure prediction module (204) is used to construct and employ a machine learning prediction model, which integrates a pre-quantified index characterizing the vulnerability of the patient's tracheal mucosa with real-time multidimensional signal data to generate an expected adjustment value for the cuff pressure; wherein, the index characterizing the vulnerability of the patient's tracheal mucosa is a quantitative index obtained by comprehensively assessing the patient's total burn area, the severity of inhalation injury and C-reactive protein level. The regulation instruction generation module (205) is used to maintain a cumulative pressure-time integral value. When the integral value exceeds a preset long-term risk threshold, the expected adjustment value is negatively corrected to generate a target cuff pressure for immediate regulation. An optimization iteration loop is activated based on the deviation between the target cuff pressure and the actual cuff pressure to generate an optimized adjustment instruction sequence. The prediction model is updated in real time during the optimization iteration loop, specifically including: adjusting the model parameters of the prediction model in combination with factors that reduce the patient's pressure tolerance due to changes in the patient's condition. The factors that reduce pressure tolerance include aggravated tissue edema or deterioration of circulatory status. The pressure execution and monitoring module (206) is used to drive the control device to perform the final pressure control operation using the optimized adjustment command sequence.

2. The apparatus of claim 1, wherein, The event feature extraction module (202) is also used to indirectly assess the changes in airway geometry caused by the body position change event by analyzing the changes in pressure-time curves and flow-time curves monitored by the ventilator and based on a pre-established correlation model between waveform features and changes in airway geometry.

3. The apparatus of claim 1, wherein, The machine learning prediction model is a support vector machine model; or the machine learning prediction model is a Gaussian process regression-support vector machine hybrid prediction model. This hybrid prediction model first uses a Gaussian process regression model to analyze the clean multidimensional signal dataset to predict the mean adjusted value of the cuff pressure and the variance of the adjusted value, which represents the degree of prediction uncertainty. Then, the mean adjusted value, the variance of the adjusted value, and the index representing the fragility of the patient's tracheal mucosa are used as enhancement features and input together into the support vector machine model for decision optimization to generate the expected adjusted value.

4. The apparatus of claim 1, wherein, The patient was a severely burned patient with complications of acute respiratory distress syndrome and shock.

Images