Limb fixing support management system for coronary artery interventional operation

Abstract

The invention relates to the technical field of limb fixing support management, and provides a limb fixing support management system for coronary artery intervention, which comprises a computer vision module, a sensor data fusion module, a dynamic trend analysis module and an operation monitoring module. The computer vision module obtains three-dimensional position information of arms of a patient through various cameras and transmits the three-dimensional position information to the sensor data fusion module; the sensor data fusion module is used for generating three-dimensional position, speed and acceleration information of an arm by integrating linear acceleration and angular velocity data of an acceleration sensor and a gyroscope, and outputting the three-dimensional position, speed and acceleration information to the dynamic trend analysis module; the dynamic trend analysis module analyzes the arm movement trend by using the fused data, determines the adjustment amount of the bracket and controls the bracket to adjust the position and force in real time by predicting the future movement state; and the operation monitoring module is responsible for recording parameters of bracket adjustment.

Description

Technical Field

[0001] The present invention relates to the technical field of limb fixation stent management, and more specifically, to a limb fixation stent management system for coronary artery intervention. Background Art

[0002] In cardiovascular medicine, transradial coronary intervention is a common minimally invasive procedure, which uses the radial artery to enter the coronary artery and use catheters and other instruments to diagnose and treat the diseased area. During this process, the patient's arm needs to remain stable to avoid unnecessary movement interfering with the operation of surgical instruments. During surgery, keeping the patient's limbs fixed is crucial to surgical accuracy, safety, and the stability of intraoperative instrument operation. Therefore, the development of an efficient limb fixation bracket management system that can meet the needs of patient limb positioning and support is of great significance to the success of the operation.

[0003] Document 1 (Colapinto, RF, Harty, PW, & Cook, Inc. (1988). Femoral artery compression device for outpatient angiography. Radiology, 166 (3), 890-891.) introduces the application of a new limb positioning device, which supports the patient's lower limbs through a bracket structure to help doctors maintain a specific position of the limbs during surgery. This fixed bracket device can effectively reduce the workload of assistants, improve the doctor's operating convenience, and ensure the clarity of the surgical field and the stability of the surgical operation. Studies have shown that limb fixation brackets can not only reduce the physical exertion of medical staff during surgery, but also enhance the patient's limb support stability and improve the overall efficiency and safety of the operation. This shows that the application of limb fixation brackets in surgery can not only optimize surgical operations, but also improve the operator's operating comfort.

[0004] Document 2 (Henderson, ER, Prioresch i, B., Mata-Fink, A., & Bell, JE (2014). Use of the spider limb positioner in noncologi c surgery. Expert Review of Medical Devices, 11 (6), 581-585.) describes an existing spider-shaped limb positioner, which has a multi-joint design and dynamically supports the patient's limbs through pneumatic control. The main advantage of this device is that it can achieve flexible positioning in three-dimensional space and support complex limb adjustment requirements. The use of this limb positioner in tumor lower limb surgery provides doctors with greater operating freedom and reduces surgical interruptions caused by changes in limb position. Through automated control, the positioner can achieve more accurate limb adjustment, effectively reducing the need for surgical assistants and optimizing the surgical process.

[0005] However, there are still some limitations in the existing technology. Although the limb positioner mentioned in Document 2 has significant advantages in limb support and positioning, its positioning adjustment lacks the ability to intelligently predict and control the patient's dynamic movements in real time, making it difficult to meet the needs of responding to sudden involuntary limb movements during surgery.

[0006] In order to solve the above problems, a technical solution is now provided. Summary of the invention

[0007] In order to overcome the above-mentioned defects of the prior art, the present invention provides a limb fixation stent management system for coronary artery intervention, which obtains multi-dimensional motion data of the patient's arm through a computer vision module and a sensor data fusion module, and uses a dynamic trend analysis module to analyze and predict the motion trend, thereby realizing real-time intelligent adjustment of the stent, thereby effectively responding to sudden involuntary movements of the patient's arm during the operation, solving the problem in the prior art that limb dynamics cannot be predicted and adjusted in real time, and improving the safety of the operation.

[0008] To achieve the above object, the present invention provides the following technical solutions:

[0009] A limb fixation support management system for coronary artery intervention includes a computer vision module, a sensor data fusion module, a dynamic trend analysis module and an operation monitoring module, wherein the computer vision module is connected to the sensor data fusion module, the sensor data fusion module is connected to the dynamic trend analysis module, and the dynamic trend analysis module is connected to the operation monitoring module. The dynamic trend analysis module is used to process and analyze the motion state data of the patient's arm, predict the future motion trend of the arm, and adjust and control the limb fixation support. The management system includes a motion trend prediction unit and a control strategy generation unit. The motion trend prediction unit is used to generate time-dependent trend features based on the current and historical motion state data of the arm using the self-attention mechanism in the converter model, and calculate the time series motion trend data of the future time point, wherein the formula for calculating the predicted value of the arm position at the future time point is:

[0010]

[0011] Where P t+Δt is the predicted value of the arm position at the future time t+Δt, P t is the position coordinate of the arm at the current time point, V t is the velocity vector of the arm at the current time point, A t is the current arm acceleration vector, Δt is the time step, T t Time-dependent trend features generated for the converter model;

[0012] The control strategy generation unit is used to generate the position adjustment amount and strength adjustment amount of the bracket through reinforcement learning based on the arm movement trend prediction data at a future time point and the current arm movement state data, and adjust the position and strength of the bracket in real time.

[0013] As a further solution of the present invention, the motion trend prediction unit is used to generate a future motion trend prediction based on the motion data of the patient's arm, comprising the following steps:

[0014] Step A1, acquiring visual data and inertial sensor data of the arm from the sensor data fusion module, wherein the visual data of the arm includes three-dimensional position information of the patient's arm; and the inertial sensor data of the arm includes linear acceleration and angular velocity data of the arm;

[0015] Step A2, calculating velocity and acceleration information from visual data by finite difference method, and deriving position information from inertial sensor data by numerical integration method;

[0016] Step A3, preprocessing the visual data and the inertial sensor data in space and time respectively, wherein the three-dimensional position information in the visual data is smoothed by a spatial filter; and the acceleration and angular velocity in the inertial sensor data are smoothed by a temporal filter;

[0017] Step A4, weighting the smoothed visual data and inertial sensor data respectively, and the sum of the weight coefficients is equal to 1;

[0018] Step A5, calculating the position change rate of the visual data in a certain direction and the acceleration change rate of the inertial sensor data in the same direction at two consecutive time points, and calculating the difference value of the change rate between the two;

[0019] Step A6, comparing the difference value with a preset threshold value, when the difference value is less than or equal to the preset threshold value, it is considered that the change trends of the visual data and the inertial data are consistent, the weight coefficient of the visual data is increased by a fixed value, and the weight coefficient of the inertial data is reduced accordingly; when the difference value is greater than the preset threshold value, it is considered that the change trends of the visual data and the inertial data are inconsistent, the weight coefficient of the visual data is reduced by a fixed value, and the weight coefficient of the inertial data is increased accordingly;

[0020] Step A7, calculating the acquired visual data and inertial sensor data to generate weighted motion state data of the arm; wherein the weighted motion state data of the arm is generated by weighted combination of the visual data and the inertial sensor data, and the formula is:

[0021]

[0022] In the formula, V t y and are the components of the three-dimensional position information provided by the visual data, V t vy and The velocity component provided for the visual data, V t ay and The acceleration component provided for the visual data, and are the components of the three-dimensional position information provided by the inertial sensor data, and The velocity component provided by the inertial sensor data, and The acceleration component provided by the inertial sensor data;

[0023] Step A8, inputting the weighted motion state data of the arm at consecutive time points into the converter model, and extracting the motion trend features reflecting time dependence in the weighted motion state data;

[0024] Step A9, after obtaining the time-dependent trend characteristics of the arm, the predicted value of the arm position at the future time point is calculated by combining the arm position, speed and acceleration at the current time point.

[0025] As a further solution of the present invention, the converter model uses a self-attention mechanism to perform time-dependency analysis on the weighted motion state data of the arm at different time points, and extracts motion trend features reflecting time dependence in the weighted motion state data, specifically including the following steps:

[0026] Step B1, converting the weighted motion state data of the arm at each time point into a query vector, a key vector and a value vector respectively;

[0027] Step B2, calculating the similarity score between each query vector and the key vectors at all other time points;

[0028] Step B3, normalizing the similarity score at each time point to obtain a normalized similarity score;

[0029] Step B4, weighted summing of the similarity scores of all time points and the corresponding value vectors, and performing sequence analysis on the weighted sums of the time points to generate the time-dependent motion trend characteristics of the arm at the current time point.

[0030] As a further solution of the present invention, the time series motion trend data at the future time point includes the arm position prediction value at the future time point and the arm speed prediction value and acceleration prediction value at the future time point; the arm speed prediction value at the future time point is calculated by the finite difference method based on the arm position at the current time point and the arm position prediction value at the future time point, and the arm acceleration prediction value at the future time point is calculated by the finite difference method based on the arm speed at the current time point and the arm speed prediction value at the future time point.

[0031] As a further solution of the present invention, the control strategy generation unit is used to generate the position adjustment amount and force adjustment amount of the bracket at the current moment through a reinforcement learning algorithm based on future motion trend prediction data and current motion state data.

[0032] As a further solution of the present invention, the control strategy generation unit generates the position adjustment amount and the force adjustment amount of the bracket through the following steps:

[0033] Step C1, inputting the future motion trend prediction data generated by the motion trend prediction unit, including the predicted values ​​of the arm position, speed and acceleration at the future time point, and the motion state data at the current moment, including the arm position, speed and acceleration data at the current moment, into the state input layer of the reinforcement learning algorithm;

[0034] Step C2, through the value function in the deep deterministic policy gradient model, combined with the actual position and force parameters of the stent in the current surgical environment, the current state data and the hypothetical action are input into the value function to generate the policy evaluation value at the current moment; the state data includes the motion state of the arm and the current actual parameters of the stent, and the hypothetical action includes the adjustment suggestions of the stent position and force;

[0035] Step C3, the reinforcement learning algorithm generates the target bracket position adjustment amount and force adjustment amount according to the strategy evaluation value, and the position adjustment formula is:

[0036]

[0037] In the formula, ΔL is the instant adjustment of the bracket position, Q is the strategy evaluation value, and P t+Δt is the predicted value of the arm position at the future time t+Δt, L t is the current position of the bracket, γ is the adjustment factor for position adjustment;

[0038] The intensity adjustment formula is:

[0039]

[0040] In the formula, ΔF is the instant adjustment of the support strength, A t+Δt is the predicted value of arm acceleration at future time t+Δt, F t is the strength of the current support, m is the mass of the patient's arm, which is used to derive the required support strength of the support, λ is the adjustment factor for strength adjustment, and Q is the strategy evaluation value;

[0041] Step C4, adjusting the position and strength of the bracket in real time according to the generated bracket position adjustment amount and strength adjustment amount.

[0042] As a further solution of the present invention, the computer vision module images the patient's arm through depth cameras and infrared cameras installed around the operating table, obtains the three-dimensional position information of the arm, pre-processes the data through low-pass filtering and Kalman filtering, and combines the convolutional neural network model to identify the position of specified points of the arm posture, generate the three-dimensional position information of the arm and transmit it to the sensor data fusion module.

[0043] As a further solution of the present invention, the sensor data fusion module includes an acceleration sensor and a gyroscope, which are used to collect linear acceleration and angular velocity data of the arm, and fuse the smoothed visual data with the inertial sensor data through particle filtering to generate three-dimensional position, velocity and acceleration information of the arm, and pass the fused motion state data to the dynamic trend analysis module.

[0044] As a further solution of the present invention, the operation monitoring module is used to record the specific time, adjustment range and number of adjustments of the support position and strength, and generate an operation log of the surgical process.

[0045] Compared with the prior art, the beneficial effects of the limb fixation stent management system for coronary artery intervention of the present invention are:

[0046] The limb fixation support management system in the present invention realizes real-time monitoring and analysis of the patient's arm movement state through computer vision and sensor data fusion. The system can not only obtain the three-dimensional position of the arm, but also extract its speed and acceleration data, and predict future movement trends through the dynamic trend analysis module, so as to realize real-time adjustment of the support. In contrast, although the fixing device in Document 1 can provide stable support, it lacks tracking and prediction of the dynamic state of the patient's arm, and cannot achieve flexible response to sudden movements. The present invention improves the response speed of the support and the real-time operation, and reduces the risk of involuntary movement of the patient during surgery.

[0047] The dynamic trend analysis module in the present invention adopts a self-attention mechanism, which enables the system to generate time-dependent trend features based on the current and historical movement data of the arm, thereby predicting future movements. This mechanism has advantages in dealing with sudden involuntary movements of patients, and can effectively avoid the delayed response of the spider-shaped limb locator in Document 2 to sudden movements during operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 This is a diagram of the new limb fixation device in document 1.

[0049] Figure 2 Illustration of a leg immobilizer device used in knee replacement surgery.

[0050] Figure 3 An example of a leg immobilizer device used in knee replacement surgery.

[0051] Figure 4 The present invention is a schematic structural diagram of a limb fixation stent management system for coronary artery intervention.

[0052] In the figure, control lever (A); thigh support (B); pointer (C); Alignment point: alignment pointer; Pelvic reaction pad: pelvic reaction pad. DETAILED DESCRIPTION

[0053] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0054] Example 1

[0055] A limb fixation bracket management system for coronary artery intervention includes a computer vision module, a sensor data fusion module, a dynamic trend analysis module and an operation monitoring module.

[0056] The computer vision module in the embodiment of the present invention is connected to the sensor data fusion module, the sensor data fusion module is connected to the dynamic trend analysis module, and the dynamic trend analysis module is connected to the operation monitoring module.

[0057] The computer vision module images the patient's arm through depth cameras and infrared cameras installed around the operating table to obtain the three-dimensional position information of the arm, pre-processes the data through low-pass filtering and Kalman filtering, and combines the convolutional neural network model to identify the position of the specified points of the arm posture, generate the three-dimensional position information of the arm and transmit it to the sensor data fusion module.

[0058] The sensor data fusion module includes an acceleration sensor and a gyroscope, which are used to collect the linear acceleration and angular velocity data of the arm, fuse the smoothed visual data with the inertial sensor data through particle filtering, generate the three-dimensional position, velocity and acceleration information of the arm, and transmit the fused motion state data to the dynamic trend analysis module.

[0059] The dynamic trend analysis module is used to process and analyze the motion state data of the patient's arm, predict the future motion trend of the arm, and adjust and control the limb fixation support, and includes a motion trend prediction unit and a control strategy generation unit.

[0060] The motion trend prediction unit in the embodiment of the present invention is used to generate time-dependent trend features based on the current and historical motion state data of the arm, using the self-attention mechanism in the converter model, and calculate the time series motion trend data of future time points to provide accurate forward-looking information support for the dynamic adjustment of the bracket.

[0061] The motion trend prediction unit in the embodiment of the present invention is used to generate a future motion trend prediction based on the motion data of the patient's arm, comprising the following steps:

[0062] Step A1, acquiring visual data and inertial sensor data of the arm from the sensor data fusion module, wherein the visual data of the arm includes three-dimensional position information of the patient's arm; the inertial sensor data of the arm includes linear acceleration and angular velocity data of the arm, so as to ensure that the dynamic motion state of the arm is captured in real time during the operation;

[0063] Step A2, calculating velocity and acceleration information from visual data by finite difference method, and deriving position information from inertial sensor data by numerical integration method;

[0064] Step A3, preprocessing the visual data and the inertial sensor data in space and time respectively, wherein the three-dimensional position information in the visual data is smoothed by a spatial filter to eliminate high-frequency noise caused by ambient lighting and occlusion; and the acceleration and angular velocity in the inertial sensor data are smoothed by a temporal filter to reduce data jitter caused by instantaneous motion;

[0065] Step A4, weighting the smoothed visual data and inertial sensor data respectively, the visual data is assigned an initial weight coefficient to reflect the importance of spatial position information, the inertial sensor data is assigned an initial weight coefficient to emphasize acceleration and angular velocity information, and the sum of the weight coefficients is equal to 1;

[0066] Step A5, calculate the position change rate of the visual data in a certain direction and the acceleration change rate of the inertial sensor data in the same direction at two consecutive time points, and calculate the difference between the two change rates, the formula is:

[0067] D = |ΔV-ΔA|;

[0068] Where ΔV is the rate of change of position of visual data in a certain direction, and ΔA is the rate of change of acceleration of inertial sensor data in the same direction;

[0069] Step A6, comparing the difference value with a preset threshold value, when the difference value is less than or equal to the preset threshold value, it is considered that the change trends of the visual data and the inertial data are consistent, the weight coefficient of the visual data is increased by a fixed value, and the weight coefficient of the inertial data is correspondingly reduced to strengthen the influence of the spatial position information in the visual data; when the difference value is greater than the preset threshold value, it is considered that the change trends of the visual data and the inertial data are inconsistent, the weight coefficient of the visual data is reduced by a fixed value, and the weight coefficient of the inertial data is correspondingly increased to enhance the acceleration and angular velocity information;

[0070] Step A7, calculating the acquired visual data and inertial sensor data to generate weighted motion state data of the arm; wherein the weighted motion state data of the arm is generated by weighted combination of the visual data and the inertial sensor data, and the formula is:

[0071]

[0072] In the formula, V t y and are the components of the three-dimensional position information provided by the visual data, V t vy and The velocity component provided for the visual data, V t ay and The acceleration component provided for the visual data, and are the components of the three-dimensional position information provided by the inertial sensor data, and The velocity component provided by the inertial sensor data, and The acceleration component provided by the inertial sensor data;

[0073] Step A8, inputting the weighted motion state data of the arm at consecutive time points into the converter model to analyze the time dependency of the arm motion, extracting the motion trend features reflecting the time dependency in the weighted motion state data, for further motion trend prediction;

[0074] Step A9, after obtaining the time-dependent trend characteristics of the arm, the predicted value of the arm position at the future time point is calculated by combining the arm position, speed and acceleration at the current time point, wherein the formula for calculating the predicted value of the arm position at the future time point is:

[0075]

[0076] Where P t+Δt is the predicted value of the arm position at the future time t+Δt, P t is the position coordinate of the arm at the current time point, P t is the velocity vector of the arm at the current time point, A t is the current arm acceleration vector, Δt is the time step, T t Time-dependent trend features generated for the converter model.

[0077] The converter model in the embodiment of the present invention uses a self-attention mechanism to perform time-dependency analysis on the weighted motion state data of the arm at different time points, and extracts motion trend features reflecting time dependence in the weighted motion state data, specifically including the following steps:

[0078] Step B1, converting the weighted motion state data of the arm at each time point into a query vector, a key vector and a value vector respectively; the query vector is used to represent the demand of the arm motion state at the current time point for the motion state information at other time points; the key vector is used to represent the degree of response of the arm motion state at other time points to the demand at the current time point; the value vector is used to represent the contribution of the arm motion state at other time points in generating the motion trend at the current time point;

[0079] Step B2, calculate the similarity score between each query vector and the key vectors of all other time points, and obtain the influence strength of the arm movement state at each time point on the movement trend at the current time point. The higher the similarity score, the greater the influence strength of the arm movement state at that time point on the current time point. These similarity scores are used to quantify the time dependency strength of the arm movement state, thereby expressing the long-term dependency relationship of the arm in the time dimension;

[0080] Step B3, normalizing the similarity score at each time point to obtain a normalized similarity score;

[0081] Step B4, weighted sum of the similarity scores of all time points and the corresponding value vectors, and sequence analysis of the weighted sums of each time point to generate the time-dependent motion trend characteristics of the arm at the current time point. The trend characteristics reflect the movement pattern of the arm in a long time series, and can provide time series dependency feature data for subsequent motion prediction, thereby ensuring that the system can accurately predict the dynamic changes of the patient's arm movement during coronary artery intervention surgery, and provide accurate data support for the real-time control of the stent.

[0082] In the embodiment of the present invention, the dependency of the current arm state on other time points in predicting future trends is calculated, that is, the similarity score of the query vector at the current time point to the key vector at each historical time point is used to evaluate the dependency of the arm motion state at the current time point on different historical time points. The similarity score is calculated using the dot product attention formula, which is obtained by the following formula:

[0083]

[0084] In the formula, sim t,i is the similarity score between the query vector at time point t and the key vector at time point i, Q t ·K i is the dot product of the query vector and the key vector; Qt is the query vector at time point t, K i is the key vector at time point i, d k is the dimension of the key vector, used to scale the similarity to prevent the value from being too large.

[0085] The higher the similarity score, the more dependent the arm movement state at the current time point t is on the time point i.

[0086] In the embodiment of the present invention, the similarity score at each time point is normalized by Softmax so as to be used for subsequent weighted calculation. The normalization formula is as follows:

[0087]

[0088] In the formula, α t,i is the normalized similarity score of time point t to time point i, sim t,i is the similarity score between the query vector at time point t and the key vector at time point i, ∑ j exp(sim t,j ) is the normalization term of all similarity scores.

[0089] In the embodiment of the present invention, the representativeness of the arm movement state in the entire time series is calculated, that is, the average similarity score between the key vector of each historical time point and each query vector in the entire time series. By calculating the average value of the matching degree of each key vector with multiple query vectors, the influence of the arm movement state at the time point on the overall trend is quantified. The calculation formula is as follows:

[0090]

[0091] In the formula, rep i is the average similarity score between the key vector at time point i and each query vector in the entire sequence, T is the total number of time points in the time series, is the sum of similarity scores between time point i and query vectors at all time points.

[0092] In the embodiment of the present invention, the weighted sum of the current time point is calculated to generate the time-dependent motion trend feature of the current time point. The system performs weighted summation on the corresponding value vectors according to the normalized similarity score to generate the time-dependent motion trend feature of the arm at the current time point, which is expressed by the following formula:

[0093] T t =∑ i α t,i ·V i ;

[0094] Where, T tis the time-dependent motion trend feature at the current time point t, α t,i is the dependence strength of time point t on time point i, used as a weighting coefficient, V i is the value vector at time point i, representing the characteristic value of the arm movement state at that time point.

[0095] The time series motion trend data at the future time point in the embodiment of the present invention includes the arm position prediction value at the future time point and the arm speed prediction value and acceleration prediction value at the future time point; the arm speed prediction value at the future time point is calculated by the finite difference method based on the arm position at the current time point and the arm position prediction value at the future time point, and the arm acceleration prediction value at the future time point is calculated by the finite difference method based on the arm speed at the current time point and the arm speed prediction value at the future time point, wherein the formula for calculating the arm speed prediction value at the future time point is:

[0096]

[0097] Where V t+Δt The predicted value of arm speed at the future time point t+Δt, P t+Δt is the predicted value of the arm position at the future time point t+Δt, P t is the arm position at the current time point t, Δt is the time step;

[0098] The formula for calculating the predicted value of arm acceleration at a future time point is:

[0099]

[0100] Where V t+Δt The predicted arm speed at the future time point t+Δt, A t+Δt is the predicted value of arm acceleration at the future time point t+Δt, V t is the arm speed at the current time point t, and Δt is the time step.

[0101] The control strategy generation unit is used to generate the position adjustment amount and force adjustment amount of the bracket based on the arm movement trend prediction data at future time points and the current arm movement state data through reinforcement learning, and adjust the position and force of the bracket in real time to achieve adaptive control of the arm movement by the bracket.

[0102] The control strategy generation unit in the embodiment of the present invention is used to generate the position adjustment amount and force adjustment amount of the bracket at the current moment through a reinforcement learning algorithm based on future motion trend prediction data and current motion state data, so as to ensure that the bracket can proactively adapt to the dynamic motion state of the arm and achieve real-time control.

[0103] The control strategy generation unit in the embodiment of the present invention generates the position adjustment amount and the force adjustment amount of the bracket through the following steps:

[0104] Step C1, inputting the future motion trend prediction data generated by the motion trend prediction unit, including the predicted values ​​of the arm position, speed and acceleration at the future time point, and the motion state data at the current moment, including the arm position, speed and acceleration data at the current moment, into the state input layer of the reinforcement learning algorithm;

[0105] Step C2, through the value function in the deep deterministic policy gradient model, combined with the actual position and force parameters of the stent in the current surgical environment, the current state data and the hypothetical action are input into the value function to generate the policy evaluation value at the current moment; the state data includes the motion state of the arm and the current actual parameters of the stent, and the hypothetical action includes the adjustment suggestions of the stent position and force;

[0106] Step C3, the reinforcement learning algorithm generates the target bracket position adjustment amount and force adjustment amount according to the strategy evaluation value, and the position adjustment formula is:

[0107]

[0108] In the formula, ΔL is the instant adjustment of the bracket position, Q is the strategy evaluation value, and P t+Δt is the predicted value of the arm position at the future time t+Δt, L t is the current position of the bracket, γ is the adjustment factor for position adjustment;

[0109] The intensity adjustment formula is:

[0110]

[0111] In the formula, ΔF is the instant adjustment of the support strength, A t+Δt is the predicted value of arm acceleration at future time t+Δt, F t is the strength of the current support, m is the mass of the patient's arm, which is used to derive the required support strength of the support, λ is the adjustment factor for strength adjustment, and Q is the strategy evaluation value;

[0112] Step C4, adjusting the position and strength of the bracket in real time according to the generated bracket position adjustment amount and strength adjustment amount.

[0113] The operation monitoring module is used to record the specific time, adjustment range and number of adjustments of the bracket position and strength, and generate an operation log of the surgical process.

[0114] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0115] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A limb fixation stent management system for coronary artery intervention, comprising a computer vision module, a sensor data fusion module and an operation monitoring module, characterized in that: It also includes a dynamic trend analysis module, wherein the computer vision module is connected to the sensor data fusion module, the sensor data fusion module is connected to the dynamic trend analysis module, and the dynamic trend analysis module is connected to the operation monitoring module. The dynamic trend analysis module is used to process and analyze the motion state data of the patient's arm, predict the future motion trend of the arm, and adjust and control the limb fixation bracket. It includes a motion trend prediction unit and a control strategy generation unit. The motion trend prediction unit is used to generate time-dependent trend features based on the current and historical motion state data of the arm using the self-attention mechanism in the converter model, and calculate the time series motion trend data of the future time point, wherein the formula for calculating the predicted value of the arm position at the future time point is: Where P t+Δt is the predicted value of the arm position at the future time t+Δt, P t is the position coordinate of the arm at the current time point, V t is the velocity vector of the arm at the current time point, A t is the current arm acceleration vector, Δt is the time step, T t Time-dependent trend features generated for the converter model; The control strategy generation unit is used to generate the position adjustment amount and strength adjustment amount of the bracket through reinforcement learning based on the arm movement trend prediction data at a future time point and the current arm movement state data, and adjust the position and strength of the bracket in real time.

2. A limb fixation stent management system for coronary artery intervention according to claim 1, characterized in that: The motion trend prediction unit is used to generate a future motion trend prediction based on the motion data of the patient's arm, comprising the following steps: Step A1, acquiring visual data and inertial sensor data of the arm from the sensor data fusion module, wherein the visual data of the arm includes three-dimensional position information of the patient's arm; and the inertial sensor data of the arm includes linear acceleration and angular velocity data of the arm; Step A2, calculating velocity and acceleration information from visual data by finite difference method, and deriving position information from inertial sensor data by numerical integration method; Step A3, preprocessing the visual data and the inertial sensor data in space and time respectively, wherein the three-dimensional position information in the visual data is smoothed by a spatial filter; and the acceleration and angular velocity in the inertial sensor data are smoothed by a temporal filter; Step A4, weighting the smoothed visual data and inertial sensor data respectively, and the sum of the weight coefficients is equal to 1; Step A5, calculating the position change rate of the visual data in a certain direction and the acceleration change rate of the inertial sensor data in the same direction at two consecutive time points, and calculating the difference value of the change rate between the two; Step A6, comparing the difference value with a preset threshold value, when the difference value is less than or equal to the preset threshold value, it is considered that the change trends of the visual data and the inertial data are consistent, the weight coefficient of the visual data is increased by a fixed value, and the weight coefficient of the inertial data is reduced accordingly; when the difference value is greater than the preset threshold value, it is considered that the change trends of the visual data and the inertial data are inconsistent, the weight coefficient of the visual data is reduced by a fixed value, and the weight coefficient of the inertial data is increased accordingly; Step A7, calculating the acquired visual data and inertial sensor data to generate weighted motion state data of the arm; wherein the weighted motion state data of the arm is generated by weighted combination of the visual data and the inertial sensor data, and the formula is: In the formula, and are the components of the three-dimensional position information provided by the visual data, and The velocity component provided for the visual data, and The acceleration component provided for the visual data, and are the components of the three-dimensional position information provided by the inertial sensor data, and The velocity component provided by the inertial sensor data, and The acceleration component provided by the inertial sensor data; Step A8, inputting the weighted motion state data of the arm at consecutive time points into the converter model, and extracting the motion trend features reflecting time dependence in the weighted motion state data; Step A9, after obtaining the time-dependent trend characteristics of the arm, the predicted value of the arm position at the future time point is calculated by combining the arm position, speed and acceleration at the current time point.

3. A limb fixation stent management system for coronary artery intervention according to claim 1 or 2, characterized in that: The converter model uses the self-attention mechanism to perform time-dependency analysis on the weighted motion state data of the arm at different time points, and extracts the motion trend features reflecting time dependence in the weighted motion state data. Specifically, the following steps are included: Step B1, converting the weighted motion state data of the arm at each time point into a query vector, a key vector and a value vector respectively; Step B2, calculating the similarity score between each query vector and the key vectors at all other time points; Step B3, normalizing the similarity score at each time point to obtain a normalized similarity score; Step B4, weighted summing of the similarity scores of all time points and the corresponding value vectors, and performing sequence analysis on the weighted sums of the time points to generate the time-dependent motion trend characteristics of the arm at the current time point.

4. The limb fixation stent management system for coronary artery intervention according to claim 1, characterized in that: The time series motion trend data at the future time point includes the arm position prediction value at the future time point and the arm speed prediction value and acceleration prediction value at the future time point; the arm speed prediction value at the future time point is calculated by the finite difference method based on the arm position at the current time point and the arm position prediction value at the future time point, and the arm acceleration prediction value at the future time point is calculated by the finite difference method based on the arm speed at the current time point and the arm speed prediction value at the future time point.

5. The limb fixation stent management system for coronary artery intervention according to claim 1, characterized in that: The control strategy generation unit is used to generate the position adjustment amount and the force adjustment amount of the bracket at the current moment through a reinforcement learning algorithm based on the future motion trend prediction data and the current motion state data.

6. A limb fixation stent management system for coronary artery intervention according to claim 1 or 5, characterized in that: The control strategy generation unit generates the position adjustment amount and the force adjustment amount of the bracket through the following steps: Step C1, inputting the future motion trend prediction data generated by the motion trend prediction unit, including the predicted values ​​of the arm position, speed and acceleration at the future time point, and the motion state data at the current moment, including the arm position, speed and acceleration data at the current moment, into the state input layer of the reinforcement learning algorithm; Step C2, through the value function in the deep deterministic policy gradient model, combined with the actual position and force parameters of the stent in the current surgical environment, the current state data and the hypothetical action are input into the value function to generate the policy evaluation value at the current moment; the state data includes the motion state of the arm and the current actual parameters of the stent, and the hypothetical action includes the adjustment suggestions of the stent position and force; Step C3, the reinforcement learning algorithm generates the target bracket position adjustment amount and force adjustment amount according to the strategy evaluation value, and the position adjustment formula is: In the formula, ΔL is the instant adjustment of the bracket position, Q is the strategy evaluation value, and P t+Δt is the predicted value of the arm position at the future time t+Δt, L t is the current position of the bracket, γ is the adjustment factor for position adjustment; The intensity adjustment formula is: In the formula, ΔF is the instant adjustment of the support strength, A t+Δt is the predicted value of arm acceleration at future time t+Δt, F t is the strength of the current support, m is the mass of the patient's arm, which is used to derive the required support strength of the support, λ is the adjustment factor for strength adjustment, and Q is the strategy evaluation value; Step C4, adjusting the position and strength of the bracket in real time according to the generated bracket position adjustment amount and strength adjustment amount.

7. The limb fixation stent management system for coronary artery intervention according to claim 1, characterized in that: The computer vision module images the patient's arm through depth cameras and infrared cameras installed around the operating table to obtain the three-dimensional position information of the arm, pre-processes the data through low-pass filtering and Kalman filtering, and combines the convolutional neural network model to identify the position of the specified points of the arm posture, generate the three-dimensional position information of the arm and transmit it to the sensor data fusion module.

8. The limb fixation stent management system for coronary artery intervention according to claim 1, characterized in that: The sensor data fusion module includes an acceleration sensor and a gyroscope, which are used to collect the linear acceleration and angular velocity data of the arm, fuse the smoothed visual data with the inertial sensor data through particle filtering, generate the three-dimensional position, velocity and acceleration information of the arm, and transmit the fused motion state data to the dynamic trend analysis module.

9. The limb fixation stent management system for coronary artery intervention according to claim 1, characterized in that: The operation monitoring module is used to record the specific time, adjustment range and number of adjustment of the support position and strength, and generate an operation log of the surgical process.

Images