Interventional postoperative puncture point pressure self-adaptive regulation and control method based on bimodal feedback

Abstract

The invention provides an interventional postoperative puncture point pressure self-adaptive regulation and control method based on bimodal feedback, and relates to the technical field of pressure regulation and control, and the method comprises the following steps: receiving a pressure safety range and an oxyhemoglobin saturation warning value; a real-time pressure signal and a real-time blood oxygen signal are obtained; generating a real-time prone position type; fitting and predicting a blood oxygen change track; constructing an initial pressure change vector; a pressure-blood oxygen dynamic response model is input for blood oxygen track prediction, and an initial blood oxygen change track and initial blood oxygen change time delay are output; gradient optimization iteration is carried out until a target pressure change vector and a target blood oxygen change time delay are output; incremental PID pressure regulation is performed on the pressure applying device. The technical problems that in the prior art, a manual pressurization bandaging method depends on manual monitoring and pressure adjustment, the pressurization strength cannot be accurately controlled in real time, pressurization is not stable, and the pressurization effect and stability are affected are solved.

Description

Technical Field

[0001] This invention relates to the field of pressure regulation technology, and in particular to a method for adaptive regulation of puncture site pressure after interventional procedures based on bimodal feedback. Background Technology

[0002] Interventional procedures, such as cardiac angiography and cerebrovascular intervention, involve puncture of blood vessels. Post-operatively, effectively managing pressure at the puncture site to ensure its stability during recovery becomes a crucial aspect of nursing care. Traditionally, hemostasis is achieved through pressure bandaging with artificial gauze, typically requiring pressure to be maintained for up to 24 hours. This method has some technical limitations, affecting the stability of pressure application and the efficiency of nursing care.

[0003] Traditional manual pressure bandaging methods cannot monitor the applied pressure in real time, and the pressure intensity cannot be accurately controlled. Due to the lack of real-time feedback, the pressure is often unstable, resulting in situations that are too loose or too tight. Too loose pressure fails to effectively compress the puncture site, while too tight pressure affects blood circulation. The inability to precisely monitor and adjust the pressure intensity makes it impossible to maintain an appropriate pressure level, thus affecting the effectiveness and stability of the compression. Furthermore, current compression methods require frequent manual checks and adjustments of the pressure, which not only increases the workload of medical staff but also increases the risk of errors due to negligence, fatigue, or improper operation.

[0004] It should be noted that the information disclosed in this background section is intended only to enhance the understanding of the overall background of the present invention, and should not be construed as an admission or in any way implying that the information constitutes prior art known to those skilled in the art. Summary of the Invention

[0005] To address the aforementioned deficiencies or improvement needs of existing technologies, this invention provides an adaptive control method for puncture point pressure after interventional procedures based on dual-modal feedback. This method solves the technical problem that existing manual pressure bandaging methods rely on manual monitoring and adjustment of pressure, resulting in unstable pressure due to the inability to control the pressure intensity accurately in real time, thus affecting the effectiveness and stability of the pressure application.

[0006] The specific technical solution is as follows:

[0007] This invention provides a method for adaptive control of puncture site pressure after interventional procedures based on dual-modal feedback, the method comprising:

[0008] The system receives a pre-set pressure safety range and blood oxygen saturation warning value from medical staff; it performs real-time dual-modal signal acquisition on the target user to obtain real-time pressure and blood oxygen signals; it calculates the limb movement state of the target user to generate a real-time supine posture type; using the blood oxygen saturation warning value as a safety boundary, it fits the predicted blood oxygen change trajectory of the real-time blood oxygen signal; within the pressure safety range, it constructs an initial pressure change vector using the real-time pressure signal as a dynamic pressure reference; it inputs the initial pressure change vector, real-time blood oxygen signal, and real-time supine posture type into a pressure-blood oxygen dynamic response model to predict the blood oxygen trajectory, outputting the initial blood oxygen change trajectory and initial blood oxygen change delay; based on the deviation between the initial blood oxygen change trajectory and the predicted blood oxygen change trajectory, it performs gradient optimization iteration on the initial pressure change vector until a target pressure change vector and target blood oxygen change delay are output, wherein the deviation between the target blood oxygen change trajectory corresponding to the target pressure change vector and the predicted blood oxygen change trajectory is less than a preset convergence threshold; based on the target pressure change vector and the target blood oxygen change delay, it performs incremental PID pressure regulation on the pressure application device.

[0009] In one implementation, it further includes:

[0010] During the incremental PID pressure regulation process of the pressure application device, the blood oxygen of the target user is tracked and collected to generate blood oxygen time-series data; the blood oxygen saturation time-series change rate is calculated based on the blood oxygen time-series data; if the blood oxygen saturation time-series change rate is lower than a preset change rate threshold within T consecutive tracking time windows, it is determined that the blood oxygen stabilization stage has been entered, and the pressure application device is switched to step pressure fine-tuning mode; in the step pressure fine-tuning mode, the pressure value is iteratively adjusted with a predefined step size until the blood oxygen saturation approaches the Pareto optimal solution.

[0011] In one implementation, it further includes:

[0012] Using the physiological characteristic parameters of the target user as search criteria, historical clinical data is narrowed down and retrieved to obtain a sample response dataset. The sample response dataset is then reorganized based on the lying posture type to obtain multiple lying posture classification response subsets for various sample lying posture types. Each lying posture classification response subset includes multiple sample pressure change vectors, multiple sample initial blood oxygenation states, multiple sample blood oxygenation change trajectories, and multiple sample blood oxygenation change delays. Independent response association regression analysis is performed on the multiple lying posture classification response subsets to construct multiple posture-dependent blood oxygen dynamics models. These multiple posture-dependent blood oxygen dynamics models are then connected in parallel to complete the construction of the pressure-blood oxygen dynamic response model.

[0013] In one implementation, multiple posture-dependent blood oxygen kinetic models are constructed by performing independent response association regression analysis on the multiple subsets of supine posture classification responses, including:

[0014] An initial blood oxygen dynamics model is constructed based on a conditional variational autoencoder architecture. This initial model includes a conditional encoder, a latent space sampling module, a trajectory decoder, and a time-delay regressor. The conditional encoder and the latent space sampling module are cascaded, and the output of the latent space sampling module is connected in parallel to the trajectory decoder and the time-delay regressor. The conditional encoder extracts spatiotemporal features from multiple pressure change vectors and initial blood oxygen states of multiple samples in the first recumbent posture classification response subset, obtaining multiple encoded feature vectors. The latent space sampling module performs Gaussian reparameterization sampling on these encoded feature vectors, outputting multiple latent variable vectors. These latent variable vectors and the blood oxygen change trajectories of multiple samples are used as training data for variational inference training of the trajectory decoder. The time delays of the multiple latent variable vectors and the blood oxygen change trajectories of multiple samples are used as training data for probability distribution fitting of the time-delay regressor, thus completing the construction of the first posture-dependent blood oxygen dynamics model.

[0015] In one implementation, the initial pressure change vector, real-time blood oxygen signal, and real-time supine posture type are input into a pressure-blood oxygen dynamic response model for blood oxygen trajectory prediction, and the initial blood oxygen change trajectory and initial blood oxygen change delay are output, including:

[0016] Based on the real-time lying posture type, the target-dependent blood oxygen dynamics model is activated in the pressure-blood oxygen dynamic response model; the initial pressure change vector and the real-time blood oxygen signal are input into the target-dependent blood oxygen dynamics model to predict the blood oxygen trajectory, and the initial blood oxygen change trajectory and the initial blood oxygen change delay are output.

[0017] In one implementation, based on the deviation between the initial blood oxygen change trajectory and the predicted blood oxygen change trajectory, gradient optimization iteration of the initial pressure change vector is performed until the target pressure change vector and the target blood oxygen change delay are output, including:

[0018] S1: Calculate the first negative gradient direction of the initial pressure change vector based on the first deviation value between the initial blood oxygen change trajectory and the predicted blood oxygen change trajectory; S2: Update the initial pressure change vector with step-size constrained gradient descent along the first negative gradient direction to generate the first iterative pressure change vector; S3: Input the first iterative pressure change vector and the real-time blood oxygen signal into the target-dependent blood oxygen dynamics model to predict the blood oxygen trajectory, and output the first iterative blood oxygen change trajectory and the first iterative blood oxygen change delay; S4: Calculate the second negative gradient direction of the first iterative pressure change vector based on the second deviation value between the first iterative blood oxygen change trajectory and the predicted blood oxygen change trajectory; S5: Update the first iterative pressure change vector with step-size constrained gradient descent along the second negative gradient direction to generate the second iterative pressure change vector; Iteratively execute steps S1 to S5 to perform gradient optimization iteration of the pressure change vector until the deviation value between the target blood oxygen change trajectory and the predicted blood oxygen change trajectory is less than a preset convergence threshold, and the update amount of the adjacent iterative pressure change vector is less than the step-size threshold, and extract the target pressure change vector and the target blood oxygen change delay.

[0019] In one implementation, the limb movement state of the target user is calculated to generate a real-time lying posture type, including:

[0020] A multi-axis inertial measurement unit (IMU) is used to track the limb motion state of the target user, obtaining raw acceleration and angular velocity data. The IMU is fixed to the proximal end of the target limb of the target user using a medical adhesive package. The raw acceleration data is progressively processed by gravity component elimination and dynamic noise filtering to obtain linear acceleration data. The attitude angles are calculated based on the linear acceleration data, and the initial pitch and roll angles are output. Zero-bias drift compensation of the initial pitch and roll angles is performed using the raw angular velocity data, and the real-time pitch and roll angles are output. The real-time pitch and roll angles are input into a recumbent posture mapping rule base for recumbent posture type matching, and the real-time recumbent posture type is output.

[0021] In one embodiment, incremental PID pressure regulation is performed on the pressure application device based on the target pressure change vector and the target blood oxygen change time delay, including:

[0022] The target pressure change vector is discretized by time shift according to the target blood oxygen change delay, and a time-series pressure setpoint sequence is output. The pressure sensor feedback value of the target user is synchronously read according to the time-series pressure setpoint sequence to dynamically calculate the pressure tracking error. Incremental PID control is calculated based on the pressure tracking error to perform dynamic pressure adjustment on the pressure application device.

[0023] Beneficial effects of the embodiments of the present invention:

[0024] By receiving pre-set pressure safety ranges and blood oxygen saturation warning values ​​from medical staff, a clear safety boundary is provided for the entire control process. This setting ensures that safety standards are followed during the design phase, avoiding damage caused by excessively high or low pressure. Real-time acquisition of pressure and blood oxygen signals, along with calculations of limb movement status, allows for precise monitoring of the patient's physical condition, ensuring pressure is applied within the correct range and preventing pressure discomfort or instability due to patient movement. Combining real-time pressure signals to dynamically construct an initial pressure change vector enables the pressure application device to adjust pressure values ​​in dynamic environments, ensuring appropriate pressure is maintained and preventing complications such as bleeding or thrombosis. By inputting real-time pressure signals, real-time blood oxygen signals, and real-time lying posture type into a pressure-blood oxygen dynamic response model, the trajectory of blood oxygen changes is predicted. This process, through the integration of dual-modal feedback of pressure and blood oxygen, provides a more accurate assessment of the pressure application device. The device employs a precise feedback mechanism; through gradient optimization iterative methods, it achieves accurate adjustment of the initial pressure change vector until the deviation between the blood oxygen change trajectory and the predicted trajectory is less than a preset convergence threshold. This process ensures that pressure regulation is not periodically adjusted, but rather finely adjusted based on the patient's real-time physiological response. This automatically optimizes the pressure application amount under different postures and physiological states, allowing the device to dynamically adjust the pressure output according to the patient's specific condition, eliminating errors or omissions caused by manual intervention, and avoiding the repetitive operations and inaccurate pressure control that may occur in traditional manual pressure regulation. Incremental PID pressure regulation, through real-time pressure tracking error calculation and incremental updates, enables the pressure application device to fine-tune any deviation and quickly respond to changes in blood oxygen. Incremental PID control ensures that pressure changes are not only stable but also optimal, avoiding adverse reactions caused by sudden excessive or insufficient pressure. The device acts on the human body in a non-invasive manner, without puncturing the skin, while ensuring the protection of the patient's health data privacy. This not only protects patient privacy and safety but also allows the device to be widely used within the framework of medical ethics and regulations, enhancing its acceptability and compliance.

[0025] Of course, implementing any product or method of the present invention does not necessarily require achieving all of the advantages described above at the same time. Attached Figure Description

[0026] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0027] Figure 1 The diagram shows a flowchart of the post-interventional puncture point pressure adaptive control method based on dual-modal feedback provided by the present invention.

[0028] Figure 2 The diagram illustrates the process of constructing a pressure-blood oxygen dynamic response model in the post-interventional puncture point pressure adaptive control method based on dual-modal feedback provided by the present invention. Detailed Implementation

[0029] To facilitate understanding of the present invention, a more complete description of the invention will be given below with reference to the accompanying drawings, which illustrate preferred embodiments of the invention. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein; rather, these embodiments are provided so that the disclosure of the invention will be more thorough and complete.

[0030] Furthermore, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0031] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "clockwise," and "counterclockwise," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0032] Unless otherwise expressly stated, throughout the specification and claims, the term "comprising" or its variations such as "including" or "comprises" shall be understood to include the stated elements or components without excluding other elements or other components.

[0033] The present invention provides an adaptive control method for puncture point pressure after interventional surgery based on dual-modal feedback, which solves the technical problem that the existing manual pressure bandaging method relies on manual monitoring and adjustment of pressure, and the pressure intensity cannot be accurately controlled in real time, resulting in unstable pressure and affecting the effect and stability of pressure.

[0034] Example 1: See Figure 1 The present invention provides an adaptive control method for post-interventional puncture site pressure based on dual-modal feedback, the method comprising:

[0035] Y100: Receives the pressure safety range and blood oxygen saturation warning value preset by medical staff.

[0036] All sensors and pressurization devices in this application should be external devices, contacting the skin via patches or strips with pressure sensors, without requiring any invasive surgery. All collected physiological data, including pressure and blood oxygenation signals, are sensitive medical information and will be stored and transmitted in strict accordance with data protection regulations. The device provides access control functions to ensure that only authorized medical personnel can access the relevant data.

[0037] First, medical staff set a safe pressure range and a warning value for blood oxygen saturation based on the target user's health condition, surgical type, and historical data. The safe pressure range refers to the appropriate range of pressure applied at the puncture site. Too high a pressure may lead to distal ischemia or thrombosis, while too low a pressure may fail to effectively stop bleeding and cause hematoma. This range is set based on individual differences, such as weight, age, and health condition. The warning value for blood oxygen saturation is a critical blood oxygen level. A level below this value indicates distal ischemia, requiring immediate warning and adjustment of the pressure application method. Normal blood oxygen saturation for adults should be maintained above 95%.

[0038] Y200: Real-time acquisition of dual-modal signals from the target user to obtain real-time pressure and blood oxygenation signals.

[0039] Dual-modal signal acquisition refers to the simultaneous acquisition of two key physiological signals: real-time pressure signals and real-time blood oxygenation signals. Pressure sensors, installed within a pressurizing device, detect and report the pressure applied to the puncture site in real time via a contact device. These sensors measure the applied pressure using piezoelectric or capacitive principles. Infrared spectroscopy, such as pulse oximeters, is used to non-invasively monitor the target user's blood oxygen saturation. Sensors detect blood oxygen levels by detecting changes in the absorption of red and infrared light; this signal is used to monitor the oxygen content in the target user's blood.

[0040] Y300: Calculates the limb movement state of the target user and generates a real-time lying posture type.

[0041] By dynamically tracking the target user's limb status, the target user's lying posture is determined, as different lying postures have varying effects on the pressure applied by the pressurization device and blood oxygen supply. Specifically, a multi-axis inertial measurement unit is fixed to the proximal end of the target user's limbs, such as the thigh or upper arm, to collect raw acceleration and angular velocity data related to the target user's posture. The acceleration data is used to detect the target user's direction and velocity of movement, and after gravity component elimination and dynamic noise filtering, linear acceleration data is obtained; the angular velocity data is used to measure the rotational speed of various parts of the body, further calculating the body's pitch angle (forward and backward tilt) and roll angle (left and right tilt). Based on this data, the target user's current lying posture type, such as supine or lateral, is calculated. This information is used for subsequent pressure regulation, as different lying postures can affect the pressurization effect or obstruct blood flow.

[0042] Y400: Using the blood oxygen saturation warning value as a safety boundary, fit the predicted blood oxygen change trajectory of the real-time blood oxygen signal.

[0043] It should be understood that the real-time blood oxygen signal is lower than the blood oxygen saturation warning value. In this embodiment, the blood oxygen saturation warning value is used as the short-term blood oxygen improvement target. Starting from the real-time blood oxygen value corresponding to the real-time blood oxygen signal, the blood oxygen change trend is predicted to obtain the predicted blood oxygen change trajectory. This trajectory represents the expected evolution path of blood oxygen saturation returning to the safe threshold after pressure intervention. It avoids the risk of abrupt changes with a steady upward trend, ensuring that the blood oxygen recovery process meets the safety boundary of physiological compensation. The specific trajectory fitting method is to construct a blood oxygen change prediction model based on the real-time collected blood oxygen signal. This model predicts the blood oxygen change trend of the target user in the future period of time through regression analysis, time series prediction, etc., based on historical data or real-time change trends. According to this trend, the time series change of blood oxygen from one value to another in a short period of time is predicted to decrease or increase. The fitted trajectory meets the safety boundary of physiological compensation in the blood oxygen recovery process.

[0044] Y500: Within the pressure safety range, an initial pressure change vector is constructed using the real-time pressure signal as a dynamic pressure reference.

[0045] The real-time pressure signal is the pressure value continuously monitored and fed back to the system by pressure sensors. This data serves as an initial reference, ensuring that the device applies appropriate pressure from the outset. The pressure change vector represents the pressure's trend over time. Without changes in blood oxygenation, the initial pressure change vector is simply a stable, constant pressure value. However, as blood oxygenation signals and patient posture change, the pressure needs to be adjusted accordingly. The initial pressure vector is constrained based on the real-time pressure signal and the set pressure safety range to ensure it remains within a reasonable range.

[0046] Y600: Input the initial pressure change vector, real-time blood oxygen signal, and real-time lying posture type into the pressure-blood oxygen dynamic response model to predict the blood oxygen trajectory, and output the initial blood oxygen change trajectory and the initial blood oxygen change delay.

[0047] The pressure-oxygen dynamic response model is trained using extensive clinical data and can predict future changes in blood oxygenation based on the interrelationships between pressure changes, blood oxygenation changes, and lying posture. The model infers the future blood oxygenation trajectory of the target user based on the initial pressure change vector, real-time blood oxygenation signal, and real-time lying posture. The blood oxygenation trajectory refers to the trend of blood oxygen saturation change over a future period, and the initial blood oxygenation change delay refers to the time lag between changes in blood oxygen level and pressure adjustment. The predicted initial blood oxygenation trajectory and initial blood oxygenation change delay are used for subsequent gradient optimization and pressure regulation calculations.

[0048] Y700: Based on the deviation between the initial blood oxygen change trajectory and the predicted blood oxygen change trajectory, perform gradient optimization iteration of the initial pressure change vector until the target pressure change vector and the target blood oxygen change delay are output, wherein the deviation between the target blood oxygen change trajectory corresponding to the target pressure change vector and the predicted blood oxygen change trajectory is less than a preset convergence threshold.

[0049] The deviation between the initial and predicted blood oxygen saturation trajectory is calculated. A smaller deviation indicates a closer match between the predicted and actual saturation trajectory. Gradient optimization iteration adjusts the pressure change vector by calculating the negative gradient direction of the deviation. The negative gradient direction represents the optimal direction for pressure adjustment; updating the pressure along this direction gradually reduces the deviation of the saturation trajectory. Specifically, based on the deviation between the initial and predicted saturation trajectories, a gradient descent algorithm or similar optimization method is used to optimize the initial pressure change vector. After each optimization iteration, the pressure is updated and the saturation trajectory is predicted again until the deviation is less than a preset convergence threshold. Finally, the target pressure change vector and target saturation change delay are output. These values ​​represent the optimal pressure application method and the time delay of the saturation change after optimization.

[0050] Y800: Based on the target pressure change vector and the target blood oxygen change time delay, perform incremental PID pressure regulation on the pressure application device.

[0051] The optimized target pressure change vector and target blood oxygen change delay are used as inputs to regulate the pressure application device through an incremental PID (Proportional-Integral-Derivative) control algorithm. The PID controller adjusts based on three parts: Proportional (P), adjusting the pressure according to the deviation between the current target pressure and the actual pressure; Integral (I), accumulating past deviations to eliminate persistent small deviations; and Derivative (D), predicting future changes based on the rate of deviation change and making adjustments in advance to reduce overshoot. Unlike conventional PID control, incremental PID primarily adjusts the pressure increment rather than directly adjusting the current pressure value. Incremental PID helps avoid excessive fluctuations in dynamic systems, making pressure changes smoother. Through incremental PID control, the applied pressure can be precisely controlled, ensuring that the target user's blood oxygen level remains within a safe range and reducing complications caused by improper pressure, such as hematoma or distal ischemia.

[0052] One implementation also includes:

[0053] Y810: During the incremental PID pressure regulation process of the pressure application device, the blood oxygen of the target user is tracked and collected to generate blood oxygen time-series data; Y820: The blood oxygen saturation time-series change rate is calculated based on the blood oxygen time-series data; Y830: If the blood oxygen saturation time-series change rate is lower than the preset change rate threshold within T consecutive tracking time windows, it is determined that the blood oxygen stabilization stage has been entered, and the pressure application device is switched to the step-by-step pressure fine-tuning mode; Y840: In the step-by-step pressure fine-tuning mode, the pressure value is iteratively adjusted with a predefined step size until the blood oxygen saturation approaches the Pareto optimal solution.

[0054] During incremental PID control, not only is pressure continuously monitored and adjusted, but blood oxygen data is also continuously collected. This means periodically or continuously tracking the target user's blood oxygen saturation and generating time-series blood oxygen data—a record of how blood oxygen saturation changes over time. This data reflects the fluctuations in the target user's blood oxygen saturation over time, allowing for assessment of whether blood oxygen levels are within the normal range or showing a downward trend, enabling timely adjustments. Throughout this process, blood oxygen acquisition is performed non-invasively, using a pulse oximeter, ensuring continuous tracking without interfering with the target user's experience.

[0055] The rate of change of blood oxygen saturation refers to the rate at which blood oxygen saturation changes over time, that is, the amount of change in blood oxygen saturation per unit of time. By analyzing time-series blood oxygen data, the rate of change of blood oxygen at each time point can be calculated. Specifically, based on time-series blood oxygen data, such as blood oxygen values ​​per minute, the rate of change of blood oxygen saturation is calculated progressively. For example, the change in blood oxygen saturation between adjacent time points is calculated, and then divided by the time interval to obtain the rate of change of blood oxygen saturation. The time-series rate of change of blood oxygen saturation is used to determine whether the current blood oxygen is in a stable state. If the rate of change of blood oxygen is large, it indicates that the target user's blood oxygen is in a fluctuating state and pressure needs to be further adjusted; if the rate of change of blood oxygen is small, it indicates that blood oxygen has entered a stable stage.

[0056] A preset rate of change threshold is established. When the rate of change in blood oxygen is less than this threshold, it indicates that the blood oxygen change has become gradual, and the target user's blood oxygen status is gradually stabilizing. This threshold is set based on clinical experience or historical data to ensure accurate judgment of when blood oxygen saturation stabilizes. T consecutive tracking time windows represent a fixed time period, such as 1 minute or 5 minutes. If the rate of change in blood oxygen is less than the preset rate of change threshold within these T consecutive tracking time windows, it indicates that the blood oxygen status is stable and no longer fluctuates significantly. Once blood oxygen enters the stable phase, the pressure application device is switched from incremental PID control mode to step-by-step pressure fine-tuning mode. In step-by-step pressure fine-tuning mode, the pressure is not adjusted drastically like in PID mode, but rather in small, precise steps to ensure that blood oxygen saturation gradually approaches the ideal level.

[0057] Pareto optimality refers to the state where, under given conditions, it is impossible to improve a goal without sacrificing other objectives. In this scenario, Pareto optimality means adjusting pressure to achieve the ideal blood oxygen saturation while avoiding excessive pressure or other complications. A small pressure adjustment step, such as 0.1 mmHg, is set to ensure that each adjustment does not cause a sudden impact on the target user, but rather gradually reaches the optimal value through meticulous adjustments. Based on real-time blood oxygen data, the pressure continues to be adjusted until blood oxygen saturation stabilizes within the optimal range and Pareto optimality is detected. At this point, the pressure application device stops adjusting, maintaining a stable state.

[0058] In one implementation, see Figure 2 It also includes:

[0059] Y610: Using the physiological characteristic parameters of the target user as search criteria, historical clinical data is narrowed down and retrieved to obtain a sample response dataset; Y620: The sample response dataset is reorganized based on the lying posture type to obtain multiple lying posture classification response subsets for various sample lying posture types. Each lying posture classification response subset includes multiple sample pressure change vectors, multiple sample initial blood oxygenation states, multiple sample blood oxygenation change trajectories, and multiple sample blood oxygenation change delays; Y630: Multiple posture-dependent blood oxygen dynamics models are constructed by performing independent response association regression analysis on the multiple lying posture classification response subsets; Y640: The multiple posture-dependent blood oxygen dynamics models are connected in parallel to complete the construction of the pressure-blood oxygen dynamic response model.

[0060] Historical clinical data is retrieved based on the target user's physiological characteristics to obtain a relevant sample dataset. These physiological characteristics include the target user's age, weight, height, blood pressure, heart rate, surgical type, and medical history. Query conditions are set based on these physiological characteristics to filter historical case data similar to the target user from the database. This data includes, but is not limited to, information on the target user's post-interventional procedure pressurization process and blood oxygenation changes. The data retrieval process is narrowed to ensure that only data most relevant to the target user's physiological characteristics is retrieved, thereby improving the model's prediction accuracy and reducing interference from irrelevant data. The sample response dataset extracted from historical data contains a large amount of clinical data, specifically including data on pressure and blood oxygenation changes during post-operative puncture site pressurization. This data will support subsequent model training and optimization.

[0061] Posture type refers to the target user's posture during postoperative rest, such as lying flat, lying on their side, or lying prone. Different postures have different effects on blood oxygenation changes and pressure distribution, therefore, data needs to be reorganized and classified according to posture type. The sample dataset is regrouped according to the target user's posture type, resulting in multiple posture classification response subsets. Each subset contains data collected under that specific posture and includes the following: multiple sample pressure change vectors, each containing the current pressure value P. t (Unit: mmHg), and pressure change trend ΔP / Δt (unit: mmHg / s), representing the pressure distribution and changes at the puncture point under different lying positions; initial blood oxygen status of multiple samples refers to the initial blood oxygen saturation of each sample, reflecting the blood oxygen level of the target user at the beginning of the procedure; blood oxygen change trajectory of multiple samples represents the curve of blood oxygen saturation of the target user changing over time during pressurization, and these changes are affected by factors such as lying position and pressure adjustment; blood oxygen change delay of multiple samples refers to the delay time of blood oxygen change, that is, the time difference from the application of pressure to the change in blood oxygen. Different lying positions result in different delays in blood oxygen change, so it is necessary to analyze according to the type of lying position.

[0062] Regression analysis is a statistical method used to study the relationships between variables. In this step, independent response association regression analysis is performed on each subset of responses for each supine position category, analyzing the relationship between pressure changes, blood oxygenation trajectory, and supine position type. For each supine position type, the relationship between pressure changes and blood oxygenation changes is analyzed independently, determining how these factors collectively influence blood oxygenation changes and their time delay through regression analysis. Multiple posture-dependent blood oxygen kinetic models are established through regression analysis—specific models for each supine position type—to predict how pressure application affects blood oxygenation changes in a given supine position.

[0063] These independent posture-dependent blood oxygen dynamics models are connected in parallel to form a comprehensive pressure-blood oxygen dynamic response model. This comprehensive model can select the appropriate model according to the target user's current lying posture, and realize the prediction of the blood oxygen change trajectory and change delay under pressure regulation based on real-time blood oxygen and the set pressure regulation vector. Parallel connection means that each posture-dependent blood oxygen dynamics model works in parallel in the model, and can dynamically select the corresponding model for prediction and adjustment according to the change of lying posture.

[0064] In one implementation, multiple posture-dependent blood oxygenation dynamics models are constructed by performing independent response correlation regression analysis on the multiple subsets of supine posture classification responses, including:

[0065] Y631: Construct an initial blood oxygen dynamics model based on a conditional variational autoencoder architecture. This initial blood oxygen dynamics model includes a conditional encoder, a latent space sampling module, a trajectory decoder, and a time-delay regressor. The conditional encoder and the latent space sampling module are cascaded, and the output of the latent space sampling module is connected in parallel to the trajectory decoder and the time-delay regressor. Y632: Use the conditional encoder to extract spatiotemporal features from multiple sample pressure change vectors and multiple sample initial blood oxygen states in the first recumbent posture classification response subset, obtaining multiple encoded feature vectors. Y633: Use the latent space sampling module to perform Gaussian reparameterization sampling on the multiple encoded feature vectors, outputting multiple latent variable vectors. Y634: Use the multiple latent variable vectors and multiple sample blood oxygen change trajectories as training data to perform variational inference training on the trajectory decoder. Y635: Use the multiple latent variable vectors and multiple sample blood oxygen change time delays as training data to perform probability distribution fitting on the time-delay regressor, completing the construction of the first posture-dependent blood oxygen dynamics model.

[0066] An initial blood oxygen dynamics model is constructed based on a conditional variational autoencoder (CVA) architecture. A CVA is a deep learning model used for generative and probabilistic modeling. It learns a latent spatial representation from input data and generates data related to input conditions, such as the target user's initial blood oxygenation state and pressure regulation changes. The conditional encoder extracts spatiotemporal features from the input data, which includes pressure change vectors and initial blood oxygenation states from multiple samples. The CVA transforms this input data into a low-dimensional latent spatial representation, capturing key features. In the latent space, the model learns a probability distribution, typically a Gaussian distribution. The latent space sampling module samples from this distribution to generate latent variable vectors, enabling the model to model and generate blood oxygenation changes using these latent variables. A trajectory decoder and a time-delay regressor decode the blood oxygenation trajectory from the latent space representation and predict the time delay of blood oxygenation changes. The trajectory decoder generates the blood oxygenation trajectory, and the time-delay regressor predicts the time delay of blood oxygenation changes. The conditional encoder and the latent space sampling module are cascaded; the conditional encoder processes the input data and passes the results to the latent space sampling module. The output of the latent space sampling module is passed not only to the trajectory decoder but also to the time delay regressor. These two decoders generate predictions of blood oxygen trajectory and time delay, respectively.

[0067] The conditional encoder processes sample data from the first recumbent posture classification response subset, including sample pressure change vectors and initial blood oxygenation states. The aim is to extract spatiotemporal features from this input data, including pressure variation patterns over time and blood oxygenation response patterns under pressure changes. By encoding these features, the conditional encoder can capture the combined influence of these factors on blood oxygenation changes. The spatiotemporal features extracted from pressure changes and initial blood oxygenation states are ultimately transformed into multiple encoded feature vectors. These vectors compress crucial information from the original data, describing the relationship between pressure and blood oxygenation, and providing a foundation for subsequent generation and prediction.

[0068] In variational autoencoders, Gaussian reparameterization sampling is used to sample latent variables from the latent space. Specifically, the conditional encoder maps the spatiotemporal features of the input to the mean and standard deviation of the latent space, which define the distribution of the latent space. The latent space is typically assumed to be Gaussian. Through reparameterization techniques, the model can sample from this distribution to obtain latent variable vectors. This sampling process ensures that the model can learn effectively in the latent space without being affected by gradient blocking during backpropagation. The latent variable vectors obtained through reparameterization sampling represent the latent space, capturing the core features of blood oxygen dynamics. The latent variable vectors are low-dimensional, meaning they compress important information from the original data while preserving key spatiotemporal dynamic features.

[0069] Using multiple latent variable vectors and multiple sample blood oxygen change trajectories as training data, the trajectory decoder aims to decode the blood oxygen change trajectory from the latent variable vectors to generate an accurate blood oxygen change curve. In this process, variational inference methods are used to train the trajectory decoder, ensuring that the model can predict results close to the actual blood oxygen change trajectory from the representation in the latent space. By maximizing the variational lower bound, the difference between the decoder output and the actual blood oxygen change trajectory can be minimized. This process uses optimization algorithms, such as gradient descent, to adjust the model parameters, enabling the trajectory decoder to accurately generate blood oxygen change trajectories. Through this training process, the trajectory decoder learns how to predict blood oxygen change trajectories based on latent variable vectors, thus providing a reliable trajectory generation model for subsequent prediction and regulation.

[0070] A time-delay regressor is used to predict the time delay of blood oxygen saturation changes, i.e., the time difference between the application of pressure and the onset of a change in blood oxygen saturation. By using a latent variable vector and the time delay of blood oxygen change as training data, the model can learn the relationship between the time delay of blood oxygen change and pressure changes and lying posture. The time-delay regressor trains a regression model that can output the time delay of blood oxygen change through probability distribution fitting. For example, treating the time delay as a continuous variable, it fits a suitable distribution by minimizing regression errors, such as mean squared error. Through this training process, the time-delay regressor can provide an accurate prediction of the time delay of blood oxygen change for each lying posture and pressure state.

[0071] By training a trajectory decoder and a time-delay regressor, a first posture-dependent blood oxygen dynamics model is constructed. This model can predict the trajectory and time delay of blood oxygen changes based on the type of lying posture, changes in applied pressure, and initial blood oxygen state.

[0072] In one implementation, the initial pressure change vector, real-time blood oxygen signal, and real-time supine posture type are input into a pressure-blood oxygen dynamic response model for blood oxygen trajectory prediction, and the initial blood oxygen change trajectory and initial blood oxygen change delay are output, including:

[0073] Y650: Based on the real-time lying posture type, the target-dependent blood oxygen dynamics model is activated in the pressure-blood oxygen dynamic response model; Y660: The initial pressure change vector and the real-time blood oxygen signal are input into the target-dependent blood oxygen dynamics model to predict the blood oxygen trajectory, and the initial blood oxygen change trajectory and the initial blood oxygen change delay are output.

[0074] In practical applications, the target user's lying posture type will change over time, such as from supine to lateral. The appropriate target-dependent blood oxygen dynamics model is dynamically selected based on the target user's current real-time lying posture type. This selection process is accomplished through search and matching operations. For example, if the current lying posture type is supine, the model trained for supine posture is activated; if it is lateral, the lateral model is selected.

[0075] The initial pressure change vector and real-time blood oxygenation signal are input into a selected target-dependent oxidynamic model. Based on these inputs, the model predicts the future blood oxygenation trajectory, representing the change in blood oxygenation over time and providing a basis for pressure regulation. Based on the predicted blood oxygenation trajectory, the model outputs the initial blood oxygenation trajectory and the initial blood oxygenation change delay. The blood oxygenation trajectory is the predicted trend of blood oxygenation change based on the applied pressure and real-time blood oxygenation signal, while the blood oxygenation change delay refers to the time delay after pressure application before blood oxygenation begins to change. This process not only optimizes the accuracy of pressure regulation but also dynamically adapts to different target user postures, ensuring that blood oxygenation remains within the optimal range during postoperative pressure application.

[0076] In one implementation, based on the deviation between the initial blood oxygen change trajectory and the predicted blood oxygen change trajectory, gradient optimization iteration of the initial pressure change vector is performed until the target pressure change vector and the target blood oxygen change delay are output, including:

[0077] S1: Calculate the first negative gradient direction of the initial pressure change vector based on the first deviation value between the initial blood oxygen change trajectory and the predicted blood oxygen change trajectory; S2: Update the initial pressure change vector with step-size constrained gradient descent along the first negative gradient direction to generate the first iterative pressure change vector; S3: Input the first iterative pressure change vector and the real-time blood oxygen signal into the target-dependent blood oxygen dynamics model to predict the blood oxygen trajectory, and output the first iterative blood oxygen change trajectory and the first iterative blood oxygen change delay; S4: Calculate the second negative gradient direction of the first iterative pressure change vector based on the second deviation value between the first iterative blood oxygen change trajectory and the predicted blood oxygen change trajectory; S5: Update the first iterative pressure change vector with step-size constrained gradient descent along the second negative gradient direction to generate the second iterative pressure change vector; Y710: Iteratively execute steps S1 to S5 to perform gradient optimization iteration of the pressure change vector until the deviation value between the target blood oxygen change trajectory and the predicted blood oxygen change trajectory is less than a preset convergence threshold, and the update amount of the adjacent iterative pressure change vector is less than the step-size threshold, and extract the target pressure change vector and the target blood oxygen change delay.

[0078] The deviation value is the difference between the initial blood oxygenation trajectory and the predicted blood oxygenation trajectory. It is obtained by calculating the distance between the two trajectories, such as the mean squared error. To reduce the error, the initial pressure change vector is optimized so that the blood oxygenation trajectory predicted by the model is as close as possible to the desired blood oxygenation trajectory fitted in step Y400. The negative gradient direction represents the direction in which the error function decreases the fastest; therefore, calculating the negative gradient direction guides how to update the initial pressure change vector. The gradient of the first deviation value relative to the initial pressure change vector is calculated to obtain the direction of pressure adjustment; the negative gradient direction points in the direction of error reduction.

[0079] Gradient descent is a classic optimization algorithm that updates parameters (in this case, the pressure change vector) along the negative gradient direction. To ensure stability during the update process, a step size constraint is set, such as using an adaptive learning rate or limiting the maximum step size for each update. The step size constraint effectively controls the stability of the optimization process. The step size represents the size of each update and determines the update magnitude of gradient descent. An excessively large step size can lead to oscillations, while an excessively small step size can result in slow convergence. After the gradient descent update, a new pressure change vector is obtained, namely the first iteration pressure change vector. This vector is closer to the target blood oxygen trajectory and represents the pressure application state adjusted based on the current optimization results.

[0080] The first-iteration pressure change vector and real-time blood oxygen signal are input into the target-dependent blood oxygen dynamics model. The model then re-predicts the blood oxygen trajectory, and these predictions represent the trend of blood oxygen change after the pressure update. The first-iteration blood oxygen change trajectory shows how blood oxygen changes over time; this trajectory is an estimate of the target user's blood oxygen status after the pressure update. The first-iteration blood oxygen change delay, i.e., the time delay at which blood oxygen saturation begins to change, reflects the speed at which blood oxygen responds to applied pressure.

[0081] By comparing the difference between the first iteration blood oxygen change trajectory and the predicted blood oxygen change trajectory, i.e. the second deviation value, the error is calculated again. The gradient is calculated based on the second deviation value to obtain the second negative gradient direction. This direction represents the need for further adjustment of the first iteration pressure change vector in order to reduce the error. The direction of adjustment is obtained by calculating the gradient of the deviation value relative to the first iteration pressure change vector.

[0082] To further optimize the first iteration pressure change vector, it is updated to the second iteration pressure change vector. Through this update step, a new pressure change vector is obtained, namely the second iteration pressure change vector. This vector is the result of further improvement in the first optimization process. The second iteration pressure change vector can make the predicted blood oxygen trajectory closer to the actual trajectory.

[0083] Iterative optimization is performed, with each iteration updating the pressure change vector according to steps S1 to S5. The goal is to continuously adjust the initial pressure change vector until the deviation between the predicted and target blood oxygenation trajectories is less than a set convergence threshold. Furthermore, the update amount of the pressure change vector between two adjacent iterations must be less than a step size threshold; that is, the change between two iterations must be very small, indicating that convergence is near. The step size threshold ensures that the adjustment amplitude during optimization is not too large, avoiding over-adjustment and system instability. When the convergence condition is met, the target pressure change vector and the target blood oxygenation delay are extracted. These final results will be used as the optimal pressure intensification scheme to ensure that blood oxygenation changes are within the optimal range and to avoid complications such as hematoma or ischemia.

[0084] In one implementation, the limb movement state of the target user is calculated to generate a real-time lying posture type, including:

[0085] Y310: A multi-axis inertial measurement unit (IMU) is used to track the limb motion state of the target user, obtaining raw acceleration data and raw angular velocity data. The IMU is fixed to the proximal end of the target limb of the target user using a medical adhesive package. Y320: Gravity component elimination and dynamic noise filtering are progressively applied to the raw acceleration data to obtain linear acceleration data. Y330: Attitude angles are calculated based on the linear acceleration data, and initial pitch and roll angles are output. Y340: Zero-bias drift compensation is performed on the initial pitch and roll angles using the raw angular velocity data, and real-time pitch and roll angles are output. Y350: The real-time pitch and roll angles are input into a prone position mapping rule base for prone position type matching, and the real-time prone position type is output.

[0086] A multi-axis inertial measurement unit (IMU) is a sensor used to monitor and track the motion of an object. It consists of an accelerometer, a gyroscope, and a magnetometer. It can accurately measure acceleration, angular velocity, and changes in direction. In this step, it is used to track the limb movement of the target user in order to determine the changes in the target user's posture while lying in bed.

[0087] The multi-axis inertial measurement unit (IMU) uses accelerometers to detect the linear acceleration of the limbs and obtain raw acceleration data. This data can help determine the speed and direction of limb movement. In addition, it uses gyroscopes to measure angular velocity and provide information about limb rotation. This helps to accurately capture changes in the target user's posture, such as prone, supine, or lateral positions.

[0088] Multi-axis inertial measurement units are fixed to the target user's body via medical-grade adhesive packaging, typically on the proximal end of the target limb, such as the arm, leg, or chest. This packaging design ensures the stability and accuracy of the sensor, enabling it to continuously monitor and record motion data during the target user's activities.

[0089] Raw acceleration data contains two components: gravitational acceleration and dynamic acceleration. Dynamic acceleration reflects the patient's movement, while gravitational acceleration is caused by Earth's gravity. The gravitational component introduces bias into the acceleration data; therefore, it's crucial to remove its influence from the raw acceleration signal. This can be achieved through low-pass filtering or Gaussian filtering, preserving the trend of dynamic acceleration. Gravity component elimination involves comparing the acceleration data with the direction of Earth's gravity, removing the static gravitational influence and retaining only the dynamic component related to limb movement. Raw acceleration data also contains dynamic noise, particularly sensor noise or interference from external factors. Dynamic noise filtering can employ algorithms such as Kalman filtering or median filtering to remove high-frequency noise, resulting in smoother and more reliable data. The data after gravity component elimination and noise filtering is called linear acceleration data, containing only acceleration information related to limb movement, facilitating subsequent posture calculations.

[0090] Attitude angles are used to represent the orientation of an object in three-dimensional space. Commonly used attitude angles include pitch, roll, and yaw. In this step, the focus is on calculating pitch and roll angles. Pitch angle describes the angle of rotation of an object around its horizontal axis, representing tilt in the vertical direction. For example, the difference between supine and prone positions is mainly reflected in the pitch angle. Roll angle describes the angle of rotation of an object around its vertical axis, representing tilt in the horizontal direction. For example, the side-lying posture is mainly reflected in the roll angle. By calculating the attitude angles, the initial pitch and initial roll angles are output. These two angles are used for subsequent attitude calculations and prone position matching.

[0091] Multi-axis inertial measurement units, especially gyroscopes, can experience zero-bias drift during prolonged use. This means that the angular velocity signal output by the sensor may drift slightly even without movement, leading to inaccurate attitude angle calculations. To eliminate this drift, a gyroscope integral correction method is used. This method combines the initial pitch and roll angles with the raw angular velocity data to compensate for the zero-bias drift. Specifically, by performing integral correction over continuous time steps—calculating the angle change based on the gyroscope's raw angular velocity data and then correcting for it—real-time pitch and roll angles are obtained. These angles, now free from drift, accurately reflect the patient's real-time attitude.

[0092] The angle-based recumbent posture mapping rule base contains the mapping relationship between different recumbent posture types and their corresponding pitch and roll angles. Each recumbent posture type corresponds to a specific range of pitch and roll angles. For example, the pitch angle and roll angle of a supine patient are close to 0°, and the roll angle of a lateral recumbent patient is close to 90°. The calculated real-time pitch and roll angles are input into this mapping rule base. By matching these angle values ​​with the predefined recumbent posture type ranges, the current patient's recumbent posture type is determined. For example, if the pitch angle is less than 10° and the roll angle is greater than 80°, the patient is determined to be in a supine posture; if the roll angle is close to 90°, the patient is determined to be in a lateral recumbent posture. Finally, the real-time recumbent posture type is output, indicating the patient's current body posture.

[0093] In one implementation, incremental PID pressure regulation is performed on the pressure application device based on the target pressure change vector and the target blood oxygen change time delay, including:

[0094] Y850: Discretize the target pressure change vector according to the time delay of the target blood oxygen change, and output the time-series pressure setpoint sequence; Y860: Synchronously read the pressure sensor feedback value of the target user according to the time-series pressure setpoint sequence to dynamically calculate the pressure tracking error; Y870: Calculate the incremental PID control quantity based on the pressure tracking error, and perform dynamic pressure adjustment on the pressure application device.

[0095] Based on the time delay of target blood oxygenation changes, the target pressure change vector is time-shifted. This means that the predicted value of the target pressure change vector is delayed along the time axis, ensuring that pressure adjustments are synchronized with blood oxygenation changes. This is because the effect of pressure changes on blood oxygenation is time-delayed, requiring pressure to be applied at appropriate times. Through time shifting, a series of time-discrete pressure setpoints are obtained. Each setpoint represents the pressure that should be applied at a certain moment. These setpoints serve as reference values ​​for pressure regulation, ensuring that the pressure application device adjusts the pressure according to the correct time sequence.

[0096] The pressure application device is equipped with pressure sensors for real-time monitoring and feedback of the applied pressure value. These sensors measure the actual applied pressure and feed the results back to the control system. Synchronous readings are performed according to a time-series pressure setpoint sequence to ensure that the pressure sensor feedback value is read at each time point. In other words, at each time-series setpoint, the difference between the actual applied pressure and the target set pressure is compared, and real-time feedback is provided. Based on the difference between the measured pressure value and the target set pressure value, the pressure tracking error is calculated. This error value represents the accuracy of the current pressure application; a larger error indicates a greater deviation between the actual applied pressure and the target value, and vice versa. The pressure tracking error is a dynamic process that changes continuously over time. By continuously reading the pressure sensor data, the error can be continuously calculated and adjustments made accordingly.

[0097] PID control is a classic feedback control method used to dynamically adjust system parameters, enabling the system to stably and accurately achieve a predetermined target. Here, PID is used to regulate the pressure application device, ensuring that the applied pressure remains within the target range. Unlike traditional PID control, incremental PID control calculates the increment of the adjustment rather than its absolute value. That is, each adjustment is calculated based on the changes in the current error and historical errors. Through incremental control, the pressure can be gradually adjusted after each feedback, making pressure changes smoother and more stable. Based on the pressure adjustment calculated by PID control, the pressure output of the pressure application device is dynamically adjusted. According to the current error value and the adjustment amount, the pressure of the pressure application device will increase or decrease until the actual applied pressure approaches the target setpoint.

[0098] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

[0099] The foregoing description of specific exemplary embodiments of the invention is for illustrative and explanatory purposes. These descriptions are not intended to limit the invention to the precise forms disclosed, and it will be apparent that many changes and variations can be made in accordance with the foregoing teachings. The exemplary embodiments were chosen and described in order to explain the specific principles of the invention and its practical application, thereby enabling those skilled in the art to implement and utilize various different exemplary embodiments of the invention, as well as various different choices and variations. The scope of the invention is intended to be defined by the claims and their equivalents.

Claims

1. A method for adaptive control of puncture site pressure after interventional procedures based on dual-modal feedback, characterized in that, The method includes: Receive the pressure safety range and blood oxygen saturation warning value preset by medical staff; Real-time dual-modal signal acquisition is performed on the target user to obtain real-time pressure signal and real-time blood oxygen signal; The target user's limb movement state is calculated to generate a real-time lying posture type; Using the blood oxygen saturation warning value as a safety boundary, the predicted blood oxygen change trajectory of the real-time blood oxygen signal is fitted. Within the pressure safety range, an initial pressure change vector is constructed using the real-time pressure signal as a dynamic pressure reference. The initial pressure change vector, real-time blood oxygen signal and real-time lying posture type are input into the pressure-blood oxygen dynamic response model to predict the blood oxygen trajectory, and the initial blood oxygen change trajectory and the initial blood oxygen change time delay are output. Based on the deviation between the initial blood oxygen change trajectory and the predicted blood oxygen change trajectory, gradient optimization iteration of the initial pressure change vector is performed until the target pressure change vector and the target blood oxygen change delay are output, wherein the deviation between the target blood oxygen change trajectory corresponding to the target pressure change vector and the predicted blood oxygen change trajectory is less than a preset convergence threshold. Incremental PID pressure regulation is performed on the pressure application device based on the target pressure change vector and the target blood oxygen change time delay.

2. The method for adaptive control of puncture site pressure after interventional surgery based on dual-modal feedback as described in claim 1, characterized in that, Also includes: During the incremental PID pressure regulation process of the pressure application device, the blood oxygen of the target user is tracked and collected to generate blood oxygen time series data. Calculate the time-series change rate of blood oxygen saturation based on the aforementioned blood oxygen time-series data; If the time-series change rate of blood oxygen saturation is lower than the preset change rate threshold within T consecutive tracking time windows, it is determined that the blood oxygen saturation has entered the stable stage, and the pressure application device is switched to the step-by-step pressure fine-tuning mode. In the step-by-step pressure fine-tuning mode, the pressure value is iteratively adjusted with a predefined step size until the blood oxygen saturation approaches the Pareto optimal solution.

3. The method for adaptive control of puncture site pressure after interventional surgery based on dual-modal feedback as described in claim 1, characterized in that, Also includes: Using the physiological characteristic parameters of the target user as search criteria, historical clinical data is retrieved in a narrowed manner to obtain a sample response dataset; Based on the lying posture type, the sample response dataset is reorganized to obtain multiple lying posture classification response subsets for various sample lying posture types. Each lying posture classification response subset includes multiple sample pressure change vectors, multiple sample initial blood oxygenation states, multiple sample blood oxygenation change trajectories, and multiple sample blood oxygenation change time delays. By performing independent response correlation regression analysis on the multiple supine posture classification response subsets, multiple posture-dependent blood oxygen dynamics models are constructed. By connecting the multiple attitude-dependent blood oxygen dynamics models in parallel, the pressure-blood oxygen dynamic response model is constructed.

4. The method for adaptive control of puncture site pressure after interventional surgery based on dual-modal feedback as described in claim 3, characterized in that, By performing independent response association regression analysis on the multiple supine posture classification response subsets, several posture-dependent blood oxygen dynamics models were constructed, including: An initial blood oxygen dynamics model is constructed based on a conditional variational autoencoder architecture. The initial blood oxygen dynamics model includes a conditional encoder, a latent space sampling module, a trajectory decoder, and a time delay regressor. The conditional encoder and the latent space sampling module are cascaded, and the output of the latent space sampling module is connected in parallel with the trajectory decoder and the time delay regressor. The conditional encoder is used to extract spatiotemporal features from multiple samples in the first supine posture classification response subset, including pressure change vectors and initial blood oxygenation states, to obtain multiple encoded feature vectors. The latent space sampling module is used to perform Gaussian reparameterization sampling of the multiple encoded feature vectors, and outputs multiple latent variable vectors. The multiple latent variable vectors and multiple sample blood oxygen change trajectories are used as training data to perform variational inference training for the trajectory decoder; Using the multiple latent variable vectors and the time delay of blood oxygen change in multiple samples as training data, the probability distribution of the time delay regressor is fitted to complete the construction of the first posture-dependent blood oxygen dynamics model.

5. The method for adaptive control of puncture site pressure after interventional surgery based on dual-modal feedback as described in claim 3, characterized in that, The initial pressure change vector, real-time blood oxygen signal, and real-time supine posture type are input into the pressure-blood oxygen dynamic response model to predict the blood oxygen trajectory. The output is the initial blood oxygen change trajectory and the initial blood oxygen change delay, including: Based on the real-time lying posture type, the target-dependent blood oxygen dynamics model is activated in the pressure-blood oxygen dynamic response model. The initial pressure change vector and real-time blood oxygen signal are input into the target-dependent blood oxygen dynamics model to predict the blood oxygen trajectory, and the initial blood oxygen change trajectory and the initial blood oxygen change time delay are output.

6. The method for adaptive control of puncture site pressure after interventional surgery based on dual-modal feedback as described in claim 5, characterized in that, Based on the deviation between the initial blood oxygen change trajectory and the predicted blood oxygen change trajectory, gradient optimization iteration of the initial pressure change vector is performed until the target pressure change vector and the target blood oxygen change delay are output, including: S1: Calculate the first negative gradient direction of the initial pressure change vector based on the first deviation value between the initial blood oxygen change trajectory and the predicted blood oxygen change trajectory; S2: Update the initial pressure change vector with step-size constrained gradient descent along the first negative gradient direction to generate the first iterative pressure change vector; S3: Input the first iterative pressure change vector and the real-time blood oxygen signal into the target-dependent blood oxygen dynamics model to predict the blood oxygen trajectory, and output the first iterative blood oxygen change trajectory and the first iterative blood oxygen change time delay; S4: Calculate the second negative gradient direction of the first iterative pressure change vector based on the second deviation value of the first iterative blood oxygen change trajectory and the predicted blood oxygen change trajectory; S5: Update the first iterative pressure change vector with step-size constraint gradient descent along the second negative gradient direction to generate the second iterative pressure change vector; Iteratively execute steps S1 to S5, perform gradient optimization iterations of the pressure change vector until the deviation between the target blood oxygen change trajectory and the predicted blood oxygen change trajectory is less than a preset convergence threshold, and the update amount of the pressure change vector in adjacent iterations is less than the step size threshold, and extract the target pressure change vector and the target blood oxygen change time delay.

7. The method for adaptive control of puncture site pressure after interventional surgery based on dual-modal feedback as described in claim 1, characterized in that, The target user's limb movement state is calculated to generate a real-time lying posture type, including: A multi-axis inertial measurement unit is used to track the limb motion state of the target user to obtain raw acceleration data and raw angular velocity data. The multi-axis inertial measurement unit is fixed to the proximal end of the target limb of the target user through a medical adhesive package. The raw acceleration data is progressively processed by gravity component elimination and dynamic noise filtering to obtain linear acceleration data. Based on the linear acceleration data, the attitude angles are calculated, and the initial pitch angle and initial roll angle are output. The original angular velocity data is used to perform zero-bias drift compensation for the initial pitch angle and initial roll angle, and the real-time pitch angle and real-time roll angle are output. The real-time pitch angle and real-time roll angle are input into a recumbent posture mapping rule library for recumbent posture type matching, and the real-time recumbent posture type is output.

8. The method for adaptive control of puncture site pressure after interventional surgery based on dual-modal feedback as described in claim 1, characterized in that, Based on the target pressure change vector and the target blood oxygen change time delay, incremental PID pressure regulation is performed on the pressure application device, including: The target pressure change vector is discretized by time shift according to the target blood oxygen change time delay, and the time-series pressure setpoint sequence is output. The pressure sensor feedback value of the target user is synchronously read according to the time-series pressure setpoint sequence in order to dynamically calculate the pressure tracking error. Based on the pressure tracking error, incremental PID control quantity is calculated to dynamically adjust the pressure of the pressure application device.

Images