Guide wire posture prediction method based on long short-term memory network

Abstract

The invention belongs to the technical field of interventional surgical robots, and discloses a guide wire posture prediction method based on a long short-term memory network, which comprises the following steps: inputting a plurality of sample images, extracting morphological characteristics of each sample image, measuring extracted morphological characteristic parameters, manually inputting guide wire parameters corresponding to the morphological characteristic parameters, and predicting the posture of the guide wire according to the morphological characteristic parameters. Combining the morphological characteristic parameters and the corresponding guide wire parameters as sample data; dividing all the sample data into a training set and a verification set, classifying all the sample data in the training set, and finally performing standardization processing on all the sample data in the training set and the verification set; building a guide wire posture prediction model based on the sample data; training a guide wire posture prediction model through the standardized sample data; inputting new sample data to predict the posture of the guide wire; the problem that in the prior art, the calculation operation time of guide wire posture prediction is long, and the requirement in an operation cannot be met is solved, and the guide wire posture prediction efficiency is effectively improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of interventional surgical robots, and in particular relates to a guidewire posture prediction method based on a long short-term memory network. Background Art

[0002] With the continuous development of science and technology, robotics technology is affecting human life in many ways. In addition to revolutionizing the manufacturing industry, robots have also been integrated into operating rooms around the world, improving many new minimally invasive surgical procedures. Minimally invasive surgery has great advantages over traditional open surgery. It can reduce patient discomfort, hospital stay and surgical costs. In interventional surgery, the use of robotic technology makes the surgery more precise and controllable, and can establish a pathway through the robot to allow the interventional guidewire to reach a more distant lesion at one time without the need to frequently change the guidewire, thereby improving surgical efficiency.

[0003] During the operation, the guidewire interacts with the human body. Therefore, in order to complete the precise and smooth control of the robot, it is necessary to be able to predict the position and shape information of the guidewire, and the construction and training of the guidewire posture prediction model is an indispensable part. At present, there are some simulation methods for guidewires in the existing technology, which mainly use the finite element method and the connecting beam unit method for simulation calculation. However, since the finite element method is a general numerical method framework, combined with the specific connecting beam unit method, it is usually necessary to discretize and solve the target structure, which will generate a large number of nodes, units and equation groups. Especially in the guidewire simulation that requires higher precision, it will significantly increase the computational complexity, resulting in more computing resources and time consumption. Therefore, there is a problem in the existing technology that the calculation running time of the guidewire posture prediction is long and cannot meet the needs during surgery. Summary of the invention

[0004] In view of the shortcomings of the prior art, the purpose of the present invention is to provide a guidewire posture prediction method based on a long short-term memory network, which solves the problem in the prior art that the calculation running time of guidewire posture prediction is long and cannot meet the needs during surgery.

[0005] The purpose of the present invention can be achieved through the following technical solutions:

[0006] The guidewire posture prediction method based on long short-term memory network includes the following steps:

[0007] Input multiple sample images, extract morphological features of each sample image, measure the extracted morphological feature parameters, manually input the guide wire parameters corresponding to the morphological feature parameters, and combine the morphological feature parameters and the corresponding guide wire parameters as sample data;

[0008] All sample data are divided into training set and validation set, all sample data in the training set are classified, and finally all sample data in the training set and validation set are standardized;

[0009] Build a guidewire posture prediction model based on sample data;

[0010] The guidewire posture prediction model is trained using the standardized sample data;

[0011] Input new sample data for guidewire posture prediction.

[0012] The specific steps for classifying all sample data in the training set are as follows:

[0013] Count the time steps of all sample data in the training set;

[0014] Set the time step threshold;

[0015] Sample data whose time step is less than the threshold is short sequence data, and sample data whose time step is greater than or equal to the threshold is long sequence data.

[0016] The specific steps for standardizing sample data are as follows:

[0017] Determine the step length of the longest feature among all features corresponding to each single sample data;

[0018] The step lengths of all features in each sample data are filled to the same length as the step length of its longest feature to complete the sample data standardization process.

[0019] The specific steps to build a guidewire posture prediction model are as follows:

[0020] The guidewire posture prediction model includes two fully connected layers, an LSTM layer, and an output layer. The sample data is processed in sequence by two fully connected layers, an LSTM layer, and an output layer.

[0021] The sample data is dimensionally upgraded and features are extracted through two fully connected layers;

[0022] Add a Dropout layer between the two fully connected layers and the LSTM layer, and set the Dropout parameters;

[0023] The LSTM layer is used to model the correlation of sample data in time and sequence dimensions, extracting and learning the dynamic change characteristics and patterns therein;

[0024] The output layer predicts a coordinate sequence representing the guidewire posture.

[0025] The specific steps for training the guidewire posture prediction model are as follows:

[0026] Set different training batches for long sequence data and short sequence data respectively;

[0027] Set the hyperparameters of the guidewire posture prediction model;

[0028] Set the optimization target, use the mean square error of the coordinate sequence as the loss function, and use the Adam optimization algorithm to iteratively optimize the guidewire posture prediction model;

[0029] Add a learning rate regulator to gradually adjust the learning rate;

[0030] Set the number of iterations, and calculate the loss function of the training set and the validation set in each iteration;

[0031] Add an early stopping mechanism and set the early stopping parameters. When the loss function no longer decreases and the number of iterations exceeds the early stopping parameters, stop training in advance to obtain the optimal model.

[0032] The optimization goal is to minimize the loss function.

[0033] The specific steps for inputting new sample data for guidewire posture prediction are as follows:

[0034] New sample data is input into the trained guidewire posture prediction model to obtain a coordinate sequence representing the guidewire posture, and the coordinate sequence is drawn on the image, and the generated curve is the predicted guidewire posture.

[0035] Beneficial effects of the present invention:

[0036] The present invention can reduce the amount of calculation in the guidewire posture prediction process through the guidewire posture prediction model, and can effectively reduce the prediction time through continuous training of the guidewire posture prediction model, thereby improving the efficiency of guidewire posture prediction during surgery. The guidewire posture can be obtained in advance, which can not only be used for preoperative path planning, but also for speed adjustment during surgery to improve wire feeding stability and accuracy. At the same time, the contact force between the guidewire and the channel can be estimated through the guidewire posture to improve surgical safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0038] Figure 1 It is a flow chart of the guidewire posture prediction method based on long short-term memory network of the present invention;

[0039] Figure 2 It is a sample data processing flow chart of the guidewire posture prediction model of the present invention. DETAILED DESCRIPTION

[0040] 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 of the present invention. 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.

[0041] like Figure 1 to Figure 2 As shown, the guidewire posture prediction method based on long short-term memory network includes the following steps:

[0042] Input multiple sample images, extract morphological features of each sample image, measure the extracted morphological feature parameters, manually input the guide wire parameters corresponding to the morphological feature parameters, and combine the morphological feature parameters and the corresponding guide wire parameters as sample data;

[0043] All sample data are divided into training set and validation set, all sample data in the training set are classified, and finally all sample data in the training set and validation set are standardized;

[0044] Build a guidewire posture prediction model based on sample data;

[0045] The guidewire posture prediction model is trained using the standardized sample data;

[0046] Input new sample data for guidewire posture prediction.

[0047] The specific steps for classifying all sample data in the training set are as follows:

[0048] Count the time steps of all sample data in the training set;

[0049] Set the time step threshold;

[0050] Sample data whose time step is less than the threshold is short sequence data, and sample data whose time step is greater than or equal to the threshold is long sequence data.

[0051] The specific steps for standardizing sample data are as follows:

[0052] Determine the step length of the longest feature among all features corresponding to each single sample data;

[0053] The step lengths of all features in each sample data are filled to the same length as the step length of its longest feature to complete the sample data standardization process.

[0054] The specific steps to build a guidewire posture prediction model are as follows:

[0055] The guidewire posture prediction model includes two fully connected layers, an LSTM layer, and an output layer. The sample data is processed in sequence by two fully connected layers, an LSTM layer, and an output layer.

[0056] The sample data is dimensionally upgraded and features are extracted through two fully connected layers.

[0057] Add a Dropout layer between the two fully connected layers and the LSTM layer, and set the Dropout parameters to prevent overfitting;

[0058] The LSTM layer is used to model the correlation of sample data in time and sequence dimensions, extracting and learning the dynamic change characteristics and patterns therein;

[0059] The output layer predicts a coordinate sequence representing the guidewire posture.

[0060] The sample data processing flow of the prediction model is as follows:

[0061] After the sample data is input, the input features are mapped to a higher dimension through the fully connected layer 1;

[0062] The fully connected layer 2 further maps the input features to the dimensions required by the LSTM layer and extracts nonlinear features to provide LSTM with more expressive input.

[0063] The Dropout layer randomly discards the output of some neurons to prevent overfitting and obtain the feature sequence required by the LSTM layer;

[0064] The nonlinear features of the input are extracted through the LSTM layer, the temporal dependency of the input sequence is captured, and the feature representation of each time step, i.e., the hidden state sequence, is generated as the input of the output layer;

[0065] The hidden state of LSTM is mapped to the final prediction value, i.e. the coordinate sequence of the guide wire, through the output layer.

[0066] The specific steps for training the guidewire posture prediction model are as follows:

[0067] Set different training batches for long sequence data and short sequence data respectively;

[0068] Set the hyperparameters of the guidewire posture prediction model;

[0069] Set the optimization target, use the mean square error of the coordinate sequence as the loss function, and use the Adam optimization algorithm to iteratively optimize the guidewire posture prediction model;

[0070] Add a learning rate regulator to gradually adjust the learning rate;

[0071] Set the number of iterations, and calculate the loss function of the training set and the validation set in each iteration;

[0072] Add an early stopping mechanism and set the early stopping parameters. When the loss function no longer decreases and the number of iterations exceeds the early stopping parameters, stop training in advance to obtain the optimal model and prevent overfitting.

[0073] The optimization goal is to minimize the loss function.

[0074] The specific steps for inputting new sample data for guidewire posture prediction are as follows:

[0075] New sample data is input into the trained guidewire posture prediction model to obtain a coordinate sequence representing the guidewire posture, and the coordinate sequence is drawn on the image, and the generated curve is the predicted guidewire posture.

[0076] The present invention can reduce the amount of calculation in the guidewire posture prediction process through the guidewire posture prediction model, and can effectively reduce the prediction time through continuous training of the guidewire posture prediction model, thereby improving the efficiency of guidewire posture prediction during surgery. The guidewire posture can be obtained in advance, which can not only be used for preoperative path planning, but also for speed adjustment during surgery to improve wire feeding stability and accuracy. At the same time, the contact force between the guidewire and the channel can be estimated through the guidewire posture to improve surgical safety.

[0077] In the description of this specification, the description with reference to the terms "one embodiment", "example", "specific example", etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0078] The above shows and describes the basic principles, main features and advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited by the above embodiments, and the above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, and these changes and improvements all fall within the scope of the present invention to be protected.

Claims

1. A guidewire posture prediction method based on long short-term memory network, characterized in that: The following steps are involved: Input multiple sample images, extract morphological features of each sample image, measure the extracted morphological feature parameters, manually input the guide wire parameters corresponding to the morphological feature parameters, and combine the morphological feature parameters and the corresponding guide wire parameters as sample data; All sample data are divided into training set and validation set, all sample data in the training set are classified, and finally all sample data in the training set and validation set are standardized; Build a guidewire posture prediction model based on sample data; The guidewire posture prediction model is trained using the standardized sample data; Input new sample data for guidewire posture prediction.

2. The guidewire posture prediction method based on long short-term memory network according to claim 1 is characterized in that: The specific steps for classifying all sample data in the training set are as follows: Count the time steps of all sample data in the training set; Set the time step threshold; Sample data whose time step is less than the threshold is short sequence data, and sample data whose time step is greater than or equal to the threshold is long sequence data.

3. The guidewire posture prediction method based on long short-term memory network according to claim 2 is characterized in that: The specific steps for standardizing sample data are as follows: Determine the step length of the longest feature among all features corresponding to each single sample data; The step lengths of all features in each sample data are filled to the same length as the step length of its longest feature to complete the sample data standardization process.

4. The guidewire posture prediction method based on long short-term memory network according to claim 3 is characterized in that: The specific steps to build a guidewire posture prediction model are as follows: The guidewire posture prediction model includes two fully connected layers, an LSTM layer, and an output layer. The sample data is processed in sequence by two fully connected layers, an LSTM layer, and an output layer. The sample data is dimensionally upgraded and features are extracted through two fully connected layers. Add a Dropout layer between the two fully connected layers and the LSTM layer, and set the Dropout parameters; The LSTM layer is used to model the correlation of sample data in time and sequence dimensions, extracting and learning the dynamic change characteristics and patterns therein; The output layer predicts a coordinate sequence representing the guidewire posture.

5. The guidewire posture prediction method based on long short-term memory network according to claim 4 is characterized in that: The specific steps for training the guidewire posture prediction model are as follows: Set different training batches for long sequence data and short sequence data respectively; Set the hyperparameters of the guidewire posture prediction model; Set the optimization target, use the mean square error of the coordinate sequence as the loss function, and use the Adam optimization algorithm to iteratively optimize the guidewire posture prediction model; Add a learning rate regulator to gradually adjust the learning rate; Set the number of iterations, and calculate the loss function of the training set and the validation set in each iteration; Add an early stopping mechanism and set the early stopping parameters. When the loss function no longer decreases and the number of iterations exceeds the early stopping parameters, stop training in advance to obtain the optimal model.

6. The guidewire posture prediction method based on long short-term memory network according to claim 5 is characterized in that: The optimization goal is to minimize the loss function.

7. The guidewire posture prediction method based on long short-term memory network according to claim 6 is characterized in that: The specific steps for inputting new sample data for guidewire posture prediction are as follows: New sample data is input into the trained guidewire posture prediction model to obtain a coordinate sequence representing the guidewire posture, and the coordinate sequence is drawn on the image, and the generated curve is the predicted guidewire posture.

Images