A method, system, and storage medium for restoring the tactile sensation of guidewire resistance.
By simulating guidewire interventional movement in a vascular model and combining deep learning and physical models, the problem of guidewires being unable to accurately sense resistance during vascular interventional surgery has been solved. This achieves efficient and accurate reproduction of guidewire resistance tactile sensation, reduces reliance on physician experience, and improves surgical safety and success rate.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- THE FIRST AFFILIATED HOSPITAL OF WANNAN MEDICAL COLLEGE (YIJISHAN HOSPITAL OF WANNAN MEDICAL COLLEGE)
- Filing Date
- 2026-03-13
- Publication Date
- 2026-06-02
AI Technical Summary
In existing technologies, doctors cannot accurately perceive the interaction force between the guidewire tip and the blood vessel wall or plaque during vascular interventional surgery. This relies heavily on personal experience, leading to high operational difficulty, easy vascular damage or plaque detachment, and a lack of direct data for training high-quality tactile reconstruction models.
By simulating guidewire intervention movement in a vascular model, data is collected and augmented. Deep learning and physical models are used to establish the mapping relationship between guidewire tip resistance and tactile feedback force. Feedback force calculation models for the slave and master ends are constructed. Combined with data generated by GAN, a convolutional neural network is trained to achieve end-to-end tactile feedback.
It achieves efficient and accurate reproduction of guidewire resistance tactile sensation, reduces reliance on physician experience, ensures the biomechanical rationality and real-time nature of tactile feedback, and improves the safety and success rate of surgery.
Smart Images
Figure CN122123773A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of medical tactile sensing technology, specifically to a method, system, and storage medium for restoring tactile sensation of guidewire resistance. Background Technology
[0002] In interventional vascular surgeries (such as the treatment of coronary heart disease and cerebrovascular diseases), surgeons need to guide the patient to the lesion site using a guidewire. However, current technology suffers from a core bottleneck: a severe lack of tactile information. Operators cannot directly perceive the interaction forces between the guidewire tip and the vessel wall or plaque, relying solely on vague tactile sensation and two-dimensional X-ray images. This inevitably leads to difficulty in precisely controlling the propulsion force, increasing the risk of vascular damage, perforation, or plaque detachment. The procedure is highly dependent on the surgeon's personal experience, lacks the ability to quantify and assess the force applied, and is strongly correlated with the surgeon's experience.
[0003] Currently, although some studies have attempted to integrate force sensors at the tip of the guidewire, they face significant challenges related to size, rigidity, biocompatibility, and cost. Therefore, there is an urgent need for a new method that can efficiently, accurately, and reliably reproduce the tactile sensation of guidewire resistance.
[0004] Furthermore, attempts to construct models that directly map sensor signals to tactile feedback force face a fundamental data acquisition challenge: in real surgery or high-fidelity simulations, the ideal feedback force that the surgeon's hand should perceive—which can be directly and accurately measured and used as a training label for the model—does not exist. This force is a product of the interaction between subjective perception and objective physics, and cannot be directly calculated by a simulator like the resistance at the guidewire tip. This makes it difficult to collect sufficient and reliable end-to-end training data pairs (sensor signals, ideal feedback force), thus hindering the direct construction of high-quality tactile reconstruction models. Summary of the Invention
[0005] The purpose of this invention is to provide a method, system, and storage medium for restoring the tactile sensation of guidewire resistance, in order to solve the technical problem that the sensing of the interaction force between the guidewire tip and the blood vessel wall or plaque is highly dependent on the doctor's personal experience.
[0006] To solve the above-mentioned technical problems, the present invention specifically provides the following technical solution: A method for restoring the tactile sensation of resistance in a dielectric guidewire includes the following steps: The guidewire intervention process is simulated in a vascular model. Blood flow signals, displacement values, and resistance values at the tip of the guidewire are collected to form a raw dataset. Data augmentation is then performed on the raw dataset to obtain an augmented dataset. By enhancing the dataset, a mapping relationship is established between the blood flow signal and displacement value at the tip of the guidewire and the resistance value at the tip of the guidewire, thus obtaining a model for calculating the resistance at the end. The resistance value output from the end resistance measurement model is converted into the main end feedback force for the doctor's tactile perception through the mass-spring-damping model. A mapping relationship between the blood flow signal and displacement value at the guidewire tip and the feedback force at the main end is established to obtain the calculation model of the feedback force at the main end.
[0007] As a preferred embodiment of the present invention, the data augmentation method for the original dataset includes: New data is generated using a generative adversarial network (GAN) based on blood flow signals, displacement values, resistance values, and the speed at which the operator moves the handle in the original dataset. Adding new data to the original dataset expands the data and yields an enhanced dataset.
[0008] As a preferred embodiment of the present invention, the method for constructing the end resistance calculation model includes: On the augmented dataset, a convolutional neural network (CNN) is trained to construct a slave-end resistance measurement model that takes blood flow signals and displacement values as inputs and resistance values as outputs. ,in, The resistance value output by the end resistance calculation model is used to calculate the resistance value. This is a blood flow signal. This is the displacement value; The training loss of the end resistance measurement model ,in, The collected resistance values.
[0009] As a preferred embodiment of the present invention, the method for converting the main end feedback force includes: Establish a dynamic model of guidewire movement during intervention in a vascular model. ,in, Main feedback force, A spring dynamics model representing the contact between the guidewire and the blood vessel. This represents a viscous resistance dynamic model of the guidewire in contact with blood vessels. A dynamic model representing the frictional forces in contact between the guidewire and the blood vessel wall; ,in, The elastic modulus in contact with blood vessels. , They are respectively The three components on the x, y, and z axes; ,in, This represents the damping coefficient of the vascular environment. The force feedback control lever's movement speed; ,in, The coefficient of friction for blood vessel contact. , To extract The symbolic function.
[0010] As a preferred embodiment of the present invention, the method for constructing the main-end feedback force calculation model includes: The main feedback force, calculated by converting the resistance value output from the end resistance measurement model with the speed at which the operator moves the handle in the corresponding enhanced dataset, is added to the enhanced dataset to form a new dataset. On the new dataset, a convolutional neural network (CNN) was trained to construct a master-end feedback force calculation model that takes blood flow signals and displacement values as inputs and master-end feedback force as output. ,in, The main feedback force output by the main feedback force calculation model. This is a blood flow signal. This is the displacement value; The training loss of the master feedback force measurement model .
[0011] As a preferred embodiment of the present invention, the present invention provides a dielectric guidewire resistance tactile feedback restoration system, applied to a dielectric guidewire resistance tactile feedback restoration method, the system comprising: The main operation module includes a force feedback joystick and a tactile feedback interface. The force feedback joystick is used to receive the action signals of the doctor's control over the guidewire, and at the same time provides the main feedback force for the doctor to perceive. The tactile feedback interface is used to transmit the main feedback force to the force feedback joystick. The controller module includes an input shaper and a PID controller, which are used to shape the motion signal and transmit it to the slave actuator. The slave execution module includes a guidewire driver and an information sensor. The guidewire driver is used to drive the guidewire to perform interventional movements according to the action signal, and the information sensor is used to collect blood flow signals and displacement values at the tip of the guidewire. The numerical conversion module has a built-in master feedback force calculation model, which is used to calculate the master feedback force based on the blood flow signal and displacement signal at the tip of the guidewire and transmit it to the tactile feedback interface.
[0012] As a preferred embodiment of the present invention, the main feedback force calculation model in the numerical conversion module ,in, The main feedback force output by the main feedback force calculation model. This is a blood flow signal. This represents the displacement value.
[0013] As a preferred embodiment of the present invention, the present invention provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement a method for restoring the resistance tactile sensation of a guide wire.
[0014] Compared with the prior art, the present invention has the following advantages: This invention employs a two-stage strategy: deep learning to estimate physical quantities (resistance) and a physical model to generate high-fidelity feedback force labels that conform to biomechanical principles. This approach bypasses the obstacle of not being able to directly obtain the true value of ideal tactile force, providing a feasible data foundation for training high-quality tactile reconstruction models.
[0015] This invention introduces the classical mass-spring-damping model as a physical prior, ensuring that the tactile feedback conforms to operational intuition and biomechanical principles. Finally, through knowledge distillation technology, this complex process, which incorporates physical knowledge, is compressed into a lightweight end-to-end neural network, enabling the system to achieve millisecond-level real-time response during deployment, meeting the stringent requirements of interventional surgery simulation.
[0016] This invention uses GAN to augment simulated vascular scenes and combines it with the constraints of physical models, so that the final model can still produce reasonable and stable tactile feedback when faced with complex vascular morphologies (such as rare lesions) not covered by the training data, thereby reducing the risk of model overfitting. Attached Figure Description
[0017] To more clearly illustrate the embodiments of the present invention or the technical solutions in 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 merely exemplary, and those skilled in the art can derive other embodiments based on the provided drawings without creative effort.
[0018] Figure 1 A flowchart of the dielectric guide wire resistance tactile feedback restoration method provided in an embodiment of the present invention; Figure 2 A block diagram of a dielectric guidewire resistance tactile feedback restoration system provided in an embodiment of the present invention; Figure 3 The vascular model provided in the embodiments of the present invention; Figure 4 A data augmentation structure block diagram provided for embodiments of the present invention; Figure 5 The following are block diagrams of the various models provided in the embodiments of the present invention. Detailed Implementation
[0019] 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 embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0020] like Figure 1 As shown, the present invention provides a method for restoring the tactile sensation of dielectric guidewire resistance, comprising the following steps: In vascular models (e.g.) Figure 3 As shown, the vascular model used is built based on a publicly available, de-identified medical image database. All simulation experiments are conducted in a closed scientific research simulation environment and do not involve any illegal collection or use of real patient data. The simulation process of guidewire intervention is carried out, and the blood flow signal, displacement value and resistance value of the guidewire tip are collected to form the original dataset. The original dataset is then augmented to obtain the augmented dataset. In the simulation environment, the three-dimensional coordinates (displacement s) of the guidewire tip over time are accurately recorded. The blood flow velocity vector and pressure value (synthetic blood flow signal Q) at this location are obtained through the computational fluid dynamics module, as well as the operator's speed in moving the handle. The force (resistance) exerted on the guidewire by the blood vessel wall is calculated using a mechanical solver. The original dataset contains approximately 500 groups (Q, s, ...). , ) Quadruple samples.
[0021] By enhancing the dataset, a mapping relationship is established between the blood flow signal and displacement value at the tip of the guidewire and the resistance value at the tip of the guidewire, thus obtaining a model for calculating the resistance at the end. The resistance value output from the end resistance measurement model is converted into the main end feedback force for the doctor's tactile perception through the mass-spring-damping model. A mapping relationship between the blood flow signal and displacement value at the guidewire tip and the feedback force at the main end is established to obtain the calculation model of the feedback force at the main end.
[0022] like Figure 4 As shown, this invention utilizes advanced data augmentation techniques such as GANs to expand the dataset and trains a CNN network to establish a data structure from blood flow signal Q and displacement s to resistance. mapping relationship .
[0023] Since traditional physical modeling methods are difficult to accurately describe the dynamic behavior of guidewires in complex vascular environments, this invention uses deep learning, which does not rely on complex physical formulas, but allows the network to learn the optimal mapping relationship directly from the data, thus capturing subtle features that traditional methods cannot describe.
[0024] Furthermore, due to the high cost and difficulty in annotating medical data, generating realistic simulation data through generative adversarial networks (GANs) greatly expands the scale and diversity of the training set, enabling the model to remain robust when faced with rare vascular morphologies or lesions, effectively preventing overfitting and improving the model's generalization ability.
[0025] Generative Adversarial Networks (GANs) are primarily used to learn the distribution characteristics of complex vascular morphologies. They generate paired data of blood flow signals Q and displacement values s corresponding to diverse vascular geometric models that conform to anatomical variations, thereby greatly expanding the scale and scenario coverage of training data. The guidewire tip resistance values, which serve as training labels corresponding to these generated data, are still calculated using high-precision computational fluid dynamics and finite element method (FEM) simulation solvers. This ensures that the data augmentation process does not introduce physical distortion and guarantees the accuracy of the core labels.
[0026] Specifically, the GAN architecture employs a deep convolutional generative adversarial network. The generator G uses an encoder-decoder structure, containing four downsampled convolutional layers (4x4 filter size, stride 2), eight residual blocks, and four upsampled transposed convolutional layers. The discriminator D uses a PatchGAN structure, containing five convolutional layers.
[0027] During training, Wasserstein distance was used as the loss function, coupled with gradient penalty. The optimizer used was Adam (β1=0.5, β2=0.999). The batch size was 16, and the training epochs were 200. This GAN takes random noise vectors and blood vessel morphology category labels as input and learns to generate corresponding, diverse pairs of blood flow signals Q and displacement values s. Ultimately, this GAN generates approximately 2000 high-quality (Q, s) simulation data pairs, significantly increasing the sample coverage of complex morphologies such as vascular stenosis, bifurcation, and aneurysm, effectively improving the robustness of subsequent models. The generated data corresponds to resistance value labels. The result is still obtained through calculation using a high-precision mechanical simulation solver.
[0028] Data augmentation methods for the original dataset include: New data is generated using a generative adversarial network (GAN) based on blood flow signals, displacement values, resistance values, and the speed of moving the handle in the original dataset. Adding new data to the original dataset expands the data and yields an enhanced dataset.
[0029] like Figure 5 As shown, the method for constructing the end resistance calculation model includes: On the augmented dataset, a convolutional neural network (CNN) is trained to construct a slave-end resistance measurement model that takes blood flow signals and displacement values as inputs and resistance values as outputs. ,in, The resistance value output by the end resistance calculation model is used to calculate the resistance value. This is a blood flow signal. This is the displacement value; Training loss of the model is calculated from end resistance. ,in, The collected resistance values.
[0030] The CNN architecture of the end-resistance measurement model employs a spatiotemporal convolution-based neural network. The input is a concatenated sequence of aligned multi-frame blood flow signals Q and displacements s.
[0031] The first module consists of two 3D convolutional layers (3x3x3 filters, 64 and 128 filters in total), followed by a 3D batch normalization layer and a ReLU activation function, used to extract local spatiotemporal features.
[0032] The second module consists of three one-dimensional convolutional layers (filter size 3, number of filters 256, 512, and 256) used to fuse features in the temporal dimension.
[0033] The third module consists of three fully connected layers (512, 256, and 3 neurons respectively), with the final output layer using linear activation to output a three-dimensional resistance vector. .
[0034] During training, a smoothed L1 loss function was used. The optimizer was Adam, with an initial learning rate of 0.001, and cosine annealing was employed for scheduling. The training and validation sets were split in an 8:2 ratio, and early stopping was used to prevent overfitting. The final model's mean absolute error (MAE) on the validation set was less than 0.05N.
[0035] This invention simplifies the movement of the entire guidewire within the blood vessel into a virtual spring-damped mass system connecting the operator's handle and the guidewire tip. One key component is the virtual stiffness (spring) model. The contact between the guidewire and the blood vessel is simulated as a spring. The greater the resistance experienced by the guidewire, the more the virtual spring is compressed, and the greater the force fed back to the hand.
[0036] It is an adjustable parameter that determines the gain of the force feedback. It can amplify tiny guidewire resistance into a force that is easily perceived by the operator.
[0037] Second: a virtual damping model to simulate the viscous resistance of the guidewire moving in the blood. It is proportional to speed, providing a smooth, viscous feel to the motion.
[0038] It is the speed at which the operator moves the handle to prevent sudden changes in force feedback and make the feel smoother and more natural. When the operator stops moving, the damping force is zero.
[0039] Third: Virtual friction model (Coulomb friction model) simulates the static and dynamic friction between the guidewire and the blood vessel wall, which is key to generating tactile texture (such as the grainy feeling when passing through plaque).
[0040] It is a sign function of the controller's velocity direction, representing the velocity vector of the operator moving the controller. It is a directional value. For example, pushing forward on the X-axis... It is a positive value; pull it back. It is a negative value.
[0041] Its result is a... Vectors of the same dimension, but each component retains only its sign (+1, 0, or -1). This ensures that friction is always opposite to the direction of motion. If the operator pushes the handle forward (…), If the friction is greater than 0), then the friction force is... The direction should be negative. If the value is positive, multiplying it by the entire formula results in the force direction being opposite to the velocity direction. The output of +1 or -1 directly determines the direction of the final force vector.
[0042] Meanwhile, the classic Coulomb friction model states that the magnitude of kinetic friction is independent of the magnitude of velocity, and depends only on the normal force (here). It is related to the friction coefficient μ.
[0043] use Instead This means that whether you move the controller slowly or quickly, the simulated friction force is always μ. It doesn't increase with speed. It only provides a constant resistance in the opposite direction of your movement.
[0044] Final Synthetic Total Virtual Feedback Force By combining all the above component forces, we obtain the final force acting on the tactile device. It is a 3D vector that contains size and orientation, directly instructing the output of haptic devices (such as force feedback joysticks).
[0045] The conversion methods for master feedback force include: Establish a dynamic model of guidewire movement during intervention in a vascular model. ,in, Main feedback force, A spring dynamics model representing the contact between the guidewire and the blood vessel. This represents a viscous resistance dynamic model of the guidewire in contact with blood vessels. A dynamic model representing the frictional forces in contact between the guidewire and the blood vessel wall; ,in, The elastic modulus in contact with blood vessels. , They are respectively The three components on the x, y, and z axes; ,in, This represents the damping coefficient of the vascular environment. The force feedback control lever's movement speed; ,in, The coefficient of friction for blood vessel contact. , To extract The symbolic function.
[0046] In this embodiment, (3x3 diagonal stiffness matrix) is obtained by mapping the Young's modulus of different regions of the blood vessel wall obtained in the simulation environment to the guidewire-blood vessel contact point according to Hertz contact theory and performing regression fitting. The damping coefficient is calculated based on blood viscosity and the equivalent hydrodynamic diameter of the guidewire in the blood vessel, and is set to 0.1 N·s / m to provide a smooth feel. (Friction coefficient) is set to 0.05, referencing tribological experimental data on the interaction between the vascular endothelium and guidewire materials, to simulate a slight scraping sensation. These parameters can all be fine-tuned via the user interface to suit different physician preferences.
[0047] The construction methods for the master-end feedback force calculation model include: The main feedback force, calculated by converting the resistance value output from the end resistance measurement model with the speed at which the operator moves the handle (the two correspond to the same Q and s) in the augmented dataset, is added to the augmented dataset to form a new dataset. On the new dataset, a convolutional neural network (CNN) was trained to construct a master-end feedback force calculation model that takes blood flow signals and displacement values as inputs and master-end feedback force as output. ,in, The main feedback force output by the main feedback force calculation model. This is a blood flow signal. This is the displacement value; Training loss of the master feedback force measurement model .
[0048] This invention, by introducing a classical mechanics model, ensures that the feedback force is highly consistent with the doctor's operational intuition. For example, F2 simulates the viscous feeling of movement in blood, F3 simulates the frictional feeling of scraping against the blood vessel wall, while F1 is the primary resistance feedback. This decomposition makes the tactile sensation rich, natural, and believable. The direction of friction is precisely expressed through a sign function: sign(V) is introduced into the F3 model. b This ensures that the direction of friction is always opposite to the direction of motion, a key detail for achieving a realistic sense of dynamic friction. This allows the master-end feedback force calculation model to obtain training data that conforms to physical constraints.
[0049] Furthermore, we first train a slave-end resistance calculation model, then use its output and the physical model to construct a dataset, and finally train an end-to-end master-end feedback force calculation model. This step-by-step optimization ensures system performance: the first stage focuses on solving the most critical and challenging problem of resistance estimation. The second stage trains a more efficient direct mapping model based on physically plausible pseudo-labels (generated by the physical model). This strategy leverages the plausibility of the physical model to constrain the learning process while ultimately achieving the efficiency and flexibility of the deep learning model.
[0050] Thus, during final deployment, the system can use a lightweight master-end feedback force calculation model for real-time inference, which is equivalent to distilling the complex physical calculation process into a highly efficient feedforward neural network, meeting the stringent low-latency requirements of interventional surgery.
[0051] Specifically, using the previously trained end-resistance measurement model and the aforementioned physical model, to enhance all (Q, s, ...) in the dataset. , ) The corresponding sample is calculated (Depend on , (After conversion), a new training set {(Q, s)} is formed. }
[0052] The model architecture and training employ a lighter CNN architecture than the end-resistance measurement model (containing four convolutional layers with filter sizes of 5x5, 3x3, 3x3, and 3x3, and the number of filters being 16, 32, 64, and 128 respectively, all using ReLU activation and batch normalization, followed by a global average pooling layer, and two fully connected layers with 64 and 3 neurons respectively. The final output layer uses a linear activation function to output the three-dimensional master-end feedback force vector). This design significantly reduces the number of parameters and computational load while maintaining accuracy, thereby reducing real-time inference latency. The training process is similar to that of the end-resistance measurement model, with the loss function being MSE. The final model achieves a force prediction error of less than 0.1N and an inference time of less than 5ms on the test set.
[0053] After the master feedback force calculation model is trained, it is integrated into the numerical conversion module of the haptic feedback system to achieve end-to-end millisecond-level real-time calculation from sensor signals to haptic commands.
[0054] Therefore, this invention proposes a physics-guided, phased knowledge distillation framework. This framework first solves the estimation problem of core physical quantities through deep learning, then uses an interpretable physical model to convert the estimated values into feedback force labels that conform to the laws of human tactile perception, and finally obtains an efficient end-to-end inference model through distillation training. This overall solution creatively solves the two coexisting technical challenges of high-quality label shortage and stringent real-time requirements in tactile reconstruction, achieving a complete technical closed loop from data generation and physical constraints to model deployment. This enables high-fidelity, low-latency tactile reconstruction in medical training simulations, effectively reducing the reliance on personal experience in surgical training.
[0055] like Figure 2 As shown, this invention provides a dielectric guidewire resistance tactile feedback restoration system, applied to a dielectric guidewire resistance tactile feedback restoration method. The system includes: The main operation module includes a force feedback joystick and a tactile feedback interface. The force feedback joystick is used to receive the doctor's action signals for controlling the guidewire, and at the same time provides the main feedback force for the doctor to perceive. The tactile feedback interface is used to transmit the main feedback force to the force feedback joystick. The controller module includes an input shaper and a PID controller, which are used to shape the motion signal and transmit it to the slave actuator. The slave execution module includes a guidewire driver and an information sensor. The guidewire driver is used to drive the guidewire to perform interventional movements according to the action signal, and the information sensor is used to collect blood flow signals and displacement values at the tip of the guidewire. The numerical conversion module has a built-in master feedback force calculation model, which is used to calculate the master feedback force based on the blood flow signal and displacement signal at the tip of the guidewire and transmit it to the tactile feedback interface.
[0056] The numerical conversion module integrates the master-end feedback force calculation model, trained and solidified according to the embodiment. This module receives real-time Q and S signals from the sensor and directly outputs... The system features a force feedback joystick. It is designed entirely for medical training and simulation, not for the direct diagnosis or treatment of real patients, and all data processing is performed locally, complying with data security regulations.
[0057] This invention constructs a complete system comprising a master-end operation module, a controller module, a slave-end execution module, and a numerical conversion module. The system achieves a complete closed loop from information perception (sensors), intelligent decision-making (deep learning model), to tactile execution (force feedback joystick). The numerical conversion module, as the system's AI module, breaks down the barrier between slave-end physical signals and master-end tactile perception, creating an immersive operating environment.
[0058] Moreover, each module has a clear division of labor: the controller module ensures precise actions, the slave execution module is responsible for physical world interaction, and the numerical conversion module is dedicated to intelligent computing. This design facilitates system iteration, upgrades, and maintenance; for example, when new sensors or algorithms are available, the numerical conversion module can be updated separately.
[0059] The main feedback force calculation model in the numerical conversion module ,in, The main feedback force output by the main feedback force calculation model. This is a blood flow signal. This represents the displacement value.
[0060] This invention provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement a method for restoring the tactile sensation of a guide wire resistance.
[0061] This invention establishes a machine learning model for feedback force measurement, which can accurately quantify feedback force without relying on the doctor's personal experience, thus enhancing objectivity. At the same time, the end-to-end feedback force measurement can directly generate feedback force based on sensor information, thereby improving the efficiency of tactile restoration.
[0062] The above embodiments are merely exemplary embodiments of this application and are not intended to limit this application. The scope of protection of this application is defined by the claims. Those skilled in the art can make various modifications or equivalent substitutions to this application within its substance and scope of protection, and such modifications or equivalent substitutions should also be considered to fall within the scope of protection of this application.
Claims
1. A method for restoring the tactile sensation of resistance in a dielectric guidewire, characterized in that, Includes the following steps: The guidewire intervention process is simulated in a vascular model. Blood flow signals, displacement values, and resistance values at the tip of the guidewire are collected to form a raw dataset. The raw dataset is then augmented to obtain an augmented dataset. By enhancing the dataset, a mapping relationship is established between the blood flow signal and displacement value at the tip of the guidewire and the resistance value at the tip of the guidewire, thus obtaining a model for calculating the resistance at the end. The resistance value output from the end resistance measurement model is converted into the main end feedback force for the doctor's tactile perception through the mass-spring-damping model. A mapping relationship between the blood flow signal and displacement value at the guidewire tip and the feedback force at the main end is established to obtain the calculation model of the feedback force at the main end.
2. The method for restoring the tactile sensation of resistance in a dielectric guidewire according to claim 1, characterized in that: The data augmentation methods for the original dataset include: New data is generated using a generative adversarial network (GAN) based on blood flow signals, displacement values, resistance values, and the speed at which the operator moves the handle in the original dataset. Adding new data to the original dataset expands the dataset, resulting in an enhanced dataset.
3. The method for restoring the tactile sensation of dielectric guidewire resistance according to claim 2, characterized in that: The method for constructing the end resistance calculation model includes: On the augmented dataset, a convolutional neural network (CNN) is trained to construct a slave-end resistance measurement model that takes blood flow signals and displacement values as inputs and resistance values as outputs. ,in, The resistance value output by the end resistance calculation model is used to calculate the resistance value. This is a blood flow signal. This is the displacement value; The training loss of the end resistance measurement model ,in, The collected resistance values.
4. The method for restoring the tactile sensation of dielectric guidewire resistance according to claim 3, characterized in that: The conversion method for the main end feedback force includes: Establish a dynamic model of guidewire movement during intervention in a vascular model. ,in, Main feedback force, A spring dynamics model representing the contact between the guidewire and the blood vessel. This represents a viscous resistance dynamic model of the guidewire in contact with blood vessels. A dynamic model representing the frictional forces in contact between the guidewire and the blood vessel wall; ,in, The elastic modulus in contact with blood vessels. , They are respectively The three components on the x, y, and z axes; ,in, This represents the damping coefficient of the vascular environment. The force feedback control lever's movement speed; ,in, The coefficient of friction for blood vessel contact. , To extract The symbolic function.
5. The method for restoring the tactile sensation of dielectric guidewire resistance according to claim 4, characterized in that: The method for constructing the master-end feedback force calculation model includes: The main feedback force, calculated by converting the resistance value output from the end resistance measurement model with the speed at which the operator moves the handle in the corresponding enhanced dataset, is added to the enhanced dataset to form a new dataset. On the new dataset, a convolutional neural network (CNN) was trained to construct a master-end feedback force calculation model that takes blood flow signals and displacement values as inputs and master-end feedback force as output. ,in, The main feedback force output by the main feedback force calculation model. This is a blood flow signal. This is the displacement value; The training loss of the master feedback force measurement model .
6. A dielectric guidewire resistance tactile feedback restoration system, characterized in that, The system, applied to the method for restoring tactile feedback of dielectric guidewire resistance according to any one of claims 1-5, comprises: The main operation module includes a force feedback joystick and a tactile feedback interface. The force feedback joystick is used to receive the action signals of the doctor's control over the guidewire, and at the same time provides the main feedback force for the doctor to perceive. The tactile feedback interface is used to transmit the main feedback force to the force feedback joystick. The controller module includes an input shaper and a PID controller, which are used to shape the motion signal and transmit it to the slave actuator. The slave execution module includes a guidewire driver and an information sensor. The guidewire driver is used to drive the guidewire to perform interventional movements according to the action signal, and the information sensor is used to collect blood flow signals and displacement values at the tip of the guidewire. The numerical conversion module has a built-in master feedback force calculation model, which is used to calculate the master feedback force based on the blood flow signal and displacement signal at the tip of the guidewire and transmit it to the tactile feedback interface.
7. The dielectric guidewire resistance tactile feedback restoration system according to claim 6, characterized in that: The main feedback force calculation model in the numerical conversion module ,in, The main feedback force output by the main feedback force calculation model. This is a blood flow signal. This represents the displacement value.
8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, implement the method as described in any one of claims 1-5.