A guide wire motion simulation method, device, equipment and readable storage medium

Abstract

The application discloses a guide wire motion simulation method, device and equipment and a readable storage medium, relates to the technical field of electric digital data processing, and comprises the following steps: according to the geometric shape of the front end of the guide wire, the guide wire is discretized into a plurality of particles based on the Kirchhoff elastic rod theory; the potential energy expression of the particles is constructed according to the position information of the particles at the current moment, the twist angle of the guide wire in the three-dimensional real number domain space and the material coefficient of the guide wire, the twist angle of the guide wire is updated based on the principle of minimum energy according to the potential energy expression of the particles; the differentiable parameterized discrete elastic rod model is iteratively optimized; the potential energy expression of the particles at the current moment is updated according to the updated twist angle and material coefficient, and the speed and position information of the guide wire at the next moment are predicted based on the framework of residual error learning and a deep neural network. The application accurately simulates the motion of the guide wire under the condition of a single sensor, reduces the integral error in the simulation process and improves the simulation accuracy.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of electric digital data processing, and in particular to a guide wire motion simulation method, device, equipment and readable storage medium. BACKGROUND

[0002] The guide wire plays an important role in path guidance in the medical intervention operation process, so the simulation of its motion is very critical during the operation process. The traditional Kirchhoff elastic rod theory can have good characterization ability for linear deformable objects, but because there are many material parameters, the adjustment parameters are too complex, and it is not easy to adjust to the optimal parameters, in addition, in the process of simulating the motion of the guide wire, the traditional physical simulation method uses a discretization method for numerical calculation, and when the inextensible constraint is executed, the momentum conservation is not guaranteed, which will inevitably cause numerical errors, resulting in inconsistent simulation results and actual motion results. SUMMARY

[0003] The present application provides a guide wire motion simulation method, device, equipment and readable storage medium to solve the problem that the existing guide wire motion simulation method is not easy to adjust to the optimal parameters and has a large error.

[0004] The present application achieves the above-mentioned purpose through the following technical solutions:

[0005] The present application provides a guide wire motion simulation method, which comprises:

[0006] Obtaining the position information of the guide wire and the geometric shape of the front end of the guide wire;

[0007] According to the geometric shape of the front end of the guide wire, the guide wire is discretized into a plurality of particles based on the Kirchhoff elastic rod theory, and physical modeling is performed according to the discrete elastic rod;

[0008] According to the position information of the particles at the current time, the twist angle of the guide wire in the three-dimensional real number domain space, and the material coefficient of the guide wire, a potential energy expression of the particles is constructed, and the twist angle of each particle is updated based on the principle of minimum energy according to the potential energy expression of the particles;

[0009] The discretely modeled guide wire is differentiable with respect to any discrete elastic rod model variable, and then the discretely modeled guide wire is differentiable parameterized, and the discretely parameterized discrete elastic rod model is iteratively optimized to obtain the updated material coefficient;

[0010] According to the updated twist angle and the material coefficient, the potential energy expression of the particles at the current time is calculated, the guide wire dynamics equation is constructed according to the potential energy expression at the current time after the update, and the velocity and position information of the guide wire at the next time are predicted based on the framework of residual learning and deep neural network according to the guide wire dynamics equation.

[0011] Further, the position information of the guide wire and the geometric shape of the front end of the guide wire are acquired, including:

[0012] Acquiring the spatial pose information of the guide wire collected by the electromagnetic sensor;

[0013] Acquiring the basic shape of the front end of the guide wire based on a preset sliding window according to the spatial pose information of the guide wire;

[0014] Performing cardinal spline interpolation processing on the basic shape of the front end of the guide wire to obtain the geometric shape of the front end of the guide wire.

[0015] Further, the twist angle is updated based on the principle of minimum energy according to the potential energy expression of the particle, and the expression is:

[0016] ;

[0017] The updated twist angle is represented by , the potential energy expression of the particle, , the position information of each particle at time t, , the twist angle of the particle in the three-dimensional real number space, and c represents the material coefficient.

[0018] Further, the differentiable parameterized discrete elastic rod model is iteratively optimized to obtain the updated material coefficient, including:

[0019] A double optimization problem is constructed, and the expression of the double optimization problem is:

[0020] ;

[0021] , the 2-norm, , the predicted value of the particle displacement at time t+1, , the displacement of the particle at time t+1, , the predicted value of the particle displacement at time t, , the updated at time t, used to calculate the material coefficient;

[0022] The above double optimization problem is iteratively optimized to obtain the updated material coefficient.

[0023] Further, the expression of the guide wire dynamics equation is:

[0024]

[0025] , the mass of the particle, , the acceleration of the particle at time t, , the Partial derivative is calculated.

[0026] Further, the velocity and position information of the guide wire at the next time are predicted based on the residual learning framework of the guide wire dynamics equation and the deep neural network, and the expression is as follows:

[0027]

[0028] wherein represents the predicted value of the velocity of the particle at time t, represents the predicted value of the displacement of the particle at time t, represents the deep neural network, represents the predicted value of the velocity of the particle at time t+1, represents the predicted value of the displacement of the particle at time t+1, represents the time step.

[0029] Further, the velocity and position information of the guide wire at the next time are predicted based on the residual learning framework of the guide wire dynamics equation and the deep neural network, and the velocity and position information of the guide wire at the next time are predicted based on the residual learning framework of the guide wire dynamics equation and the deep neural network, and the expression is as follows:

[0030] The velocity and displacement of the particle at the current time t are input into the deep neural network, and the displacement correction amount is output. The displacement correction amount and the predicted value of the displacement of the particle at time t+1 are integrated to obtain the final predicted value of the displacement of the particle at time t+1.

[0031] A constraint function is constructed, and the expression of the constraint function is as follows:

[0032] ;

[0033] ;

[0034] represents the displacement of the i-th segment, represents the displacement of the i+1-th segment, represents the mass of the i+1-th segment, represents the mass of the i-th segment, is a threshold value defined by a user, and are bias terms, is a constraint function of continuous positions;

[0035] The position is iteratively modified through this constraint condition , and finally converges. Finally, the position is obtained, and the updated velocity is calculated using the position difference to obtain the final predicted value of the velocity of the particle at time t+1, and the expression is as follows:

[0036] ;

[0037] represents the final speed prediction value of the particle at t+1 moment, represents a time interval.

[0038] The application further provides a guide wire motion simulation device, comprising:

[0039] An acquisition module is configured to acquire position information of the guide wire and a geometric shape of a front end of the guide wire.

[0040] A discretization module is configured to discretize the guide wire into a plurality of particles based on a Kirchhoff elastic rod theory according to the geometric shape of the front end of the guide wire.

[0041] A first updating module is configured to construct a potential energy expression of the particle according to position information of the particle at a current moment, a twist angle of the guide wire in a three-dimensional real number domain space and a material coefficient of the guide wire, and update the twist angle of each particle based on a principle of minimum energy according to the potential energy expression of the particle.

[0042] A second updating module is configured to make the guide wire after mathematical modeling differentiable for any discrete elastic rod model variable, then perform differentiable parameterization on the discrete elastic rod model of the guide wire, perform iterative optimization on the discrete elastic rod model after the differentiable parameterization, and obtain an updated material coefficient.

[0043] A prediction module is configured to calculate a potential energy expression of the particle at the current moment according to the updated twist angle and the material coefficient, construct a guide wire dynamics equation according to the potential energy expression at the current moment after the update, and predict speed and position information of the guide wire at a next moment based on a framework of residual learning and a deep neural network according to the guide wire dynamics equation.

[0044] The application further provides a guide wire motion simulation device, comprising:

[0045] A memory is configured to store a computer program.

[0046] A processor is configured to implement steps of the guide wire motion simulation method when the computer program is executed.

[0047] The application further provides a readable storage medium, wherein the readable storage medium stores a computer program, and the computer program is configured to implement steps of the guide wire motion simulation method when executed by a processor.

[0048] The application has the following beneficial effects:

[0049] The present invention proposes a guidewire motion simulation method, device, equipment, and readable storage medium that obtains optimal parameters through a gradient descent approach similar to that in deep learning. While keeping the basic physical model unchanged, it corrects the error caused by integration based on residual learning. Under the premise of ensuring momentum conservation, it executes inextensibility constraints, thereby improving the accuracy of guidewire motion simulation. Attached Figure Description

[0050] Figure 1 This is a flowchart of the guidewire motion simulation method based on deep neural networks in this application. Detailed Implementation

[0051] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0052] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0053] The terms “first”, “second”, etc., are used only to distinguish descriptions and should not be interpreted as indicating or implying relative importance.

[0054] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0055] like Figure 1 As shown, a method for simulating guidewire motion includes:

[0056] S1: Obtain the position information of the guidewire and the geometry of the guidewire tip;

[0057] First, the precise spatial pose information of the guidewire is acquired using an electromagnetic sensor. This sensor provides real-time position information of the guidewire. Due to the limitations of single sensors in guidewire applications, a sliding window (ruler method) with a fixed window size is proposed. This method maintains a fixed-size time-preserving window to maintain the guidewire's closest point information, and then utilizes... Spline interpolation preserves the geometry of the guidewire tip.

[0058] S2: Discretize the guidewire into a series of particles based on the geometry of the guidewire tip according to Kirchhoff's elastic rod theory;

[0059] According to the acquired geometry of the guidewire tip, according to the classical Kirchhoff's elastic rod theory, for such a DLO (linearly deformable object) as a guidewire, it can be discretized into a series of particles, and each particle will receive bending force, torsional force and tensile force. Due to the inextensible nature of the guidewire, the tensile force can be ignored.

[0060] S3: Construct the potential energy expression of the particles according to the position information of the particles at the current time, the torsion angle of the guidewire in the three-dimensional real number domain space, and the material coefficient of the guidewire, and update the torsion angle of each particle based on the principle of minimum energy according to the potential energy expression of the particles;

[0061] The potential energy of the DLO can be expressed as , wherein represents the position information of each particle at time t, the scalar represents the torsion in the three-dimensional real number domain space, and c represents the material coefficient (such as mass, bending modulus, and elastic modulus). To obtain the new torsion angle , the DLO can be obtained by reaching an equilibrium state at each time step:

[0062]

[0063] represents the updated torsion angle, , the potential energy expression of the particle, represents the position information of each particle at time t, represents the torsion angle of the particle in the three-dimensional real number domain space, and c represents the material coefficient.

[0064] S4: Make the physically modeled guidewire differentiable with respect to any discrete elastic rod model variable, then perform differentiable parameterization on the discrete elastic rod model of the guidewire, and perform iterative optimization on the differentiable parameterization of the discrete elastic rod model based on the implicit differentiation method to obtain the updated material coefficient;

[0065] Make the guidewire differentiable with respect to any model variable (such as material coefficient, GCN parameter, position, and velocity). Here, the automatic differentiation function of is used to perform differentiable parameterization on the DER, which is called DDER. Based on the differentiability of all variables, all variables are optimized through iterative optimization without manual configuration. Specifically, the following double optimization problem is solved to obtain the material coefficient for calculating the potential energy equation :

[0066] ;

[0067] represents the 2-norm, represents the predicted value of the displacement of the particle at t+1 moment, represents the displacement of the particle at t+1 moment, represents the predicted value of the displacement of the particle at t moment, is based on the principle of using the minimum energy, the gradient of c can be obtained by using the optimization method to perform implicit differentiation, and the value of c is updated to solve the double optimization problem.

[0068] S5: calculate the potential energy expression of the particle at the current moment according to the updated torsion angle and material coefficient, construct the guide wire dynamics equation according to the potential energy expression at the current moment after updating, and predict the velocity and position information of the guide wire at the next moment based on the framework of residual learning and deep neural network according to the guide wire dynamics equation.

[0069] The expression of the guide wire dynamics equation is:

[0070]

[0071] where m is the mass of the particle, represents the acceleration of the particle at t moment, represents the derivative of .

[0072] After deriving the dynamics equation, in order to apply it to simulate the movement of the front end of the guide wire and predict the next position, the traditional method usually discretizes the continuous object into one particle after another, and uses the integral method to do the operation, but the discretization will inevitably cause numerical error. In order to make up for the error caused by discrete integration and improve the numerical stability, a framework based on residual learning is integrated into the integral method, as follows:

[0073] ;

[0074] where represents the predicted value of the velocity at t moment, represents the predicted value of the displacement at t moment, and DNN represents a deep neural network, which is preferably implemented by GCN (graph convolutional neural network). GCN can be well applied to the spatial position relationship in DLO and its message passing mechanism, and each node can fully collect information from adjacent positions to learn useful residual information. And it uses a residual learning mechanism, which does not directly predict the final state, but predicts the correction amount of physical simulation, which ensures the accuracy of the prediction under the premise of physical credibility.

[0075] The velocity and displacement of the particle at the current time t are taken as the input of the GCN, the feature dimension is set according to the number of point sets, and finally the displacement correction amount is obtained through a multilayer perception machine composed of two linear layers Integration and predicted , the final .

[0076] However, since the predicted position is simulated only by the integral method, the inextensible property of the guide wire is not considered, so it is necessary to apply an inextensible constraint, and here an improved PBD method is used, which can maintain momentum conservation during simulation. First, define the following constraint function of the continuous position:

[0077] ;

[0078] Where represents the length of the i-th segment, represents the displacement of the particle at t+1, represents the predicted value of the particle displacement at t.

[0079] Since the position changes, this constraint function is not 0, and a pair of bias terms and are added to modify each vertex position and satisfy the following constraints under the condition of momentum conservation:

[0080] ;

[0081] Where, represents the mass of the i+1-th segment, represents the mass of the i-th segment, is a threshold value defined by the user. The position is iteratively modified through this constraint condition, and finally converges.

[0082] Finally, according to the obtained position , the updated velocity is calculated using the position difference:

[0083] ;

[0084] Indicates the final predicted velocity of the guide wire at the next time.

[0085] At this point, the single-step simulation is complete, and if multi-step simulation is required, the predicted position and velocity are simulated again.

[0086] The application also provides a guide wire motion simulation device, comprising:

[0087] an acquisition module, configured to acquire position information of a guide wire and a geometric shape of a front end of the guide wire;

[0088] a discretization module, configured to discretize the guide wire into a plurality of particles based on the geometric shape of the front end of the guide wire according to a Kirchhoff elastic rod theory;

[0089] a first updating module, configured to construct a potential energy expression of the particles according to position information of the particles at a current time, a twist angle of the guide wire in a three-dimensional real number domain space and a material coefficient of the guide wire, and update the twist angle of each particle based on a principle of energy minimization according to the potential energy expression of the particles;

[0090] a second updating module, configured to make the guide wire after mathematical modeling differentiable for any discrete elastic rod model variable, then perform differentiable parameterization on the discrete elastic rod model of the guide wire, perform iterative optimization on the discrete elastic rod model after the differentiable parameterization, and obtain an updated material coefficient;

[0091] a prediction module, configured to calculate a potential energy expression of the particles at the current time according to the updated twist angle and the material coefficient, construct a guide wire dynamics equation according to the potential energy expression at the current time after the update, and predict velocity and position information of the guide wire at a next time based on a framework of residual learning and a deep neural network according to the guide wire dynamics equation.

[0092] The application further provides a guide wire motion simulation device, comprising:

[0093] a memory, configured to store a computer program;

[0094] a processor, configured to implement steps of the guide wire motion simulation method when the computer program is executed.

[0095] The application further provides a readable storage medium, wherein the readable storage medium stores a computer program, and the computer program is executed by a processor to implement steps of the guide wire motion simulation method.

[0096] The guide wire motion simulation method, device, equipment and readable storage medium provided by the application obtain the best parameters in a manner similar to gradient descent in deep learning, correct errors caused by integration based on a residual learning manner under the condition that a basic physical model is unchanged, and perform inextensible constraint under the premise of momentum conservation, so that the accuracy of guide wire motion simulation is improved.

[0097] The above merely describes the preferred embodiments of the present application, and it should be pointed out that, for those skilled in the art, several improvements and refinements can be made without departing from the technical principles of the present application, and these improvements and refinements should also be considered as the protection scope of the present application.

Claims

1. A method for simulating guidewire motion, characterized in that, include: Obtain the position information of the guidewire and the geometry of the guidewire tip; Based on the geometry of the guidewire tip, the guidewire is discretized into several mass points according to Kirchhoff's elastic rod theory, and a physical model is performed based on the discrete elastic rods. Based on the current position of the particle, the twist angle of the guide wire in the three-dimensional real domain space, and the material coefficient of the guide wire, a potential energy expression for the particle is constructed. The twist angle of each particle is then updated based on the principle of minimizing energy according to the potential energy expression of the particle. The physical model of the guide wire is made differentiable for any discrete elastic rod model variables. Then, the discrete elastic rod model of the guide wire is parameterized to be differentiable. The parameterized discrete elastic rod model is then iteratively optimized to obtain the updated material coefficients. The potential energy expression of the particle at the current moment is calculated based on the updated torsion angle and material coefficients. The guide wire dynamics equation is constructed based on the updated potential energy expression at the current moment. Based on the guide wire dynamics equation, the velocity and position information of the guide wire at the next moment are predicted using a residual learning framework and a deep neural network. Based on the residual learning framework and deep neural network of the guidewire dynamics equation, the velocity and position information of the guidewire at the next moment are predicted, as shown in the following expression: ， in This represents the predicted velocity of the particle at time t. The value represents the predicted displacement of the particle at time t, and DNN represents a deep neural network. This represents the predicted velocity of the particle at time t+1. This represents the predicted displacement of the particle at time t+1. It represents the time step; Based on the residual learning framework and deep neural network, the system predicts the velocity and position information of the guidewire at the next moment according to the guidewire dynamics equation, and also includes: The velocity and displacement of the particle at the current time t are input into the deep neural network, which outputs the displacement correction. The displacement correction and the displacement prediction value of the particle at time t+1 are integrated to obtain the final displacement prediction value of the particle at time t+1. Construct constraint functions, the expression of which is: ； ； This represents the displacement of the i-th segment. This represents the displacement of the (i+1)th segment. This represents the quality of the (i+1)th segment. This represents the quality of the i-th segment. It is a user-defined threshold. and All are bias terms. The constraint function is for continuous positions; The position is iteratively modified using this constraint. Finally, convergence is achieved, and the final result is based on the obtained position. The updated velocity is calculated using position difference, yielding the final predicted velocity of the particle at time t+1, as shown in the following expression: ； This represents the predicted final velocity of the particle at time t+1. Indicates a time interval.

2. The guidewire motion simulation method according to claim 1, characterized in that, Obtain the position information and geometry of the guidewire tip, including: Acquire the spatial orientation information of the guidewire collected by the electromagnetic sensor; Based on the spatial pose information of the guidewire, the basic shape of the guidewire tip is obtained according to a preset sliding window; Cardinal spline interpolation is performed on the basic shape of the guidewire tip to obtain its geometric shape.

3. The guidewire motion simulation method according to claim 1, characterized in that, Based on the potential energy expression of a particle and the principle of minimizing energy, the torsion angle of each particle is updated, and the expression is: ； This indicates the updated torsion angle. Here is the expression for the potential energy of a point mass. This represents the position information of each particle at time t. The angle of twist of the particle in the three-dimensional real number domain is represented by c, which represents the material coefficient.

4. The guidewire motion simulation method according to claim 3, characterized in that, Iterative optimization is performed on the discrete elastic rod model after differentiability parameterization to obtain updated material coefficients, including: Construct a dual optimization problem, the expression of which is: ； It represents the 2-norm. This represents the predicted displacement of the particle at time t+1. This represents the displacement of the particle at time t+1. The predicted value representing the displacement of the particle at time t. This represents the value updated at time t. , used to calculate material coefficients; The above dual optimization problem is iteratively optimized using the implicit differential method to obtain the updated material coefficients.

5. The guidewire motion simulation method according to claim 4, characterized in that, The expression for the guidewire dynamics equation is: ， Where m is the mass of the particle. This represents the acceleration of the particle at time t. Representative to Find the partial derivative.

6. A guidewire motion simulation device, characterized in that, include: The acquisition module is used to acquire the position information of the guidewire and the geometry of the guidewire tip. Discretization module, which is used to discretize the guidewire into several mass points based on the geometry of the guidewire tip and Kirchhoff's elastic rod theory; The first update module is used to construct the potential energy expression of the particle based on the current position information of the particle, the twist angle of the guide wire in the three-dimensional real number domain space, and the material coefficient of the guide wire, and update the twist angle of each particle based on the principle of minimum energy according to the potential energy expression of the particle. The second update module is used to make the physical modeled guide wire differentiable for any discrete elastic rod model variables. Then, the discrete elastic rod model of the guide wire is parameterized to be differentiable, and the parameterized discrete elastic rod model is iteratively optimized to obtain the updated material coefficients. The prediction module is used to update the potential energy expression of each mass point at the current moment based on the updated torsion angle and material coefficient, construct the guide wire dynamic equation based on the updated potential energy expression at the current moment, and predict the velocity and position information of the guide wire at the next moment based on the guide wire dynamic equation using a residual learning framework and a deep neural network. Based on the residual learning framework and deep neural network of the guidewire dynamics equation, the velocity and position information of the guidewire at the next moment are predicted, as shown in the following expression: ， in This represents the predicted velocity of the particle at time t. The value represents the predicted displacement of the particle at time t, and DNN represents a deep neural network. This represents the predicted velocity of the particle at time t+1. This represents the predicted displacement of the particle at time t+1. It represents the time step; Based on the residual learning framework and deep neural network, the system predicts the velocity and position information of the guidewire at the next moment according to the guidewire dynamics equation, and also includes: The velocity and displacement of the particle at the current time t are input into the deep neural network, which outputs the displacement correction. The displacement correction and the displacement prediction value of the particle at time t+1 are integrated to obtain the final displacement prediction value of the particle at time t+1. Construct constraint functions, the expression of which is: ； ； This represents the displacement of the i-th segment. This represents the displacement of the (i+1)th segment. This represents the quality of the (i+1)th segment. This represents the quality of the i-th segment. It is a user-defined threshold. and All are bias terms. The constraint function for continuous positions; The position is iteratively modified using this constraint. Finally, convergence is achieved, and the final result is based on the obtained position. The updated velocity is calculated using position difference, yielding the final predicted velocity of the particle at time t+1, as shown in the following expression: ； This represents the predicted final velocity of the particle at time t+1. Indicates a time interval.

7. A guidewire motion simulation device, characterized in that, include: Memory, used to store computer programs; A processor, configured to implement the steps of the guidewire motion simulation method as described in any one of claims 1 to 5 when executing the computer program.

8. A readable storage medium, characterized in that, The readable storage medium stores a computer program that, when executed by a processor, implements the steps of the guidewire motion simulation method as described in any one of claims 1 to 5.

Images