X-ray high-precision control method and system based on BP neural network

Abstract

The present invention discloses a high-precision X-ray control method and system based on BP neural network, including the following steps: creating a BP neural network deep learning model; generating a number of X-ray control system response delay sample data by a CT ray source subsystem; training the BP neural network deep learning model by using the sample data; deploying the trained BP neural network deep learning model to the CT ray source subsystem; when the X-ray dose is ready to change, pre-inputting the voltage and current parameters of the current stage and the voltage and current parameters of the next stage into the BP neural network deep learning model, and inferring the response delay time by the BP neural network deep learning model; according to the response delay time output by the BP neural network deep learning model, the voltage and current of the next stage are input to the CT ray source subsystem in advance of the response delay time. The technical problem of system response delay in the X-ray source subsystem of the CT system in the prior art when controlling X-rays is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of CT system control, and in particular to a high-precision X-ray control method and system based on a BP neural network. Background Art

[0002] At present, the X-ray source is the core component of the CT system, and the X-rays generated by the X-ray source are used to obtain images. Although the CT system has many advantages, the radiation exposure it generates to the patient will have irreversible effects, so the dose control of the X-ray source becomes particularly important.

[0003] In the prior art, in order to reduce the radiation X-ray dose as much as possible, the current CT system supports the function of dynamically changing the X-ray dose according to the different positions and angles of the patient during the scanning process. However, since the X-ray source subsystem of the CT system has a system response delay problem when controlling the X-ray, how to ensure the high real-time performance and high accuracy of the X-ray in the process of reducing the dose is the key to ensuring the quality of reconstructing the patient's image. Summary of the invention

[0004] To this end, the present invention provides a high-precision X-ray control method and system based on a BP neural network, which aims to effectively ensure image quality while significantly reducing radiation dose, so as to solve the technical problem of system response delay in the X-ray source subsystem of the CT system in the prior art when controlling X-rays.

[0005] In order to achieve the above object, the present invention provides the following technical solutions:

[0006] A high-precision X-ray control method based on BP neural network includes the following steps:

[0007] Create a BP neural network deep learning model;

[0008] Based on the created BP neural network deep learning model, the CT ray source subsystem generates several X-ray control system response delay sample data under different parameters and conditions;

[0009] A BP neural network deep learning model is trained using several X-ray control system response delay sample data to fit the delay time values ​​between different input parameters and the delay time values ​​between the input parameters after the change to the corresponding output specified X-ray dose;

[0010] Deploy the trained BP neural network deep learning model to the CT ray source subsystem;

[0011] When the X-ray dose is about to change, the voltage and current parameters of the current stage and the voltage and current parameters of the next stage are pre-input into the BP neural network deep learning model, and the BP neural network deep learning model infers and outputs the response delay time of the CT ray source subsystem;

[0012] According to the response delay time output by the BP neural network deep learning model, the voltage and current of the next stage are input into the CT ray source subsystem in advance of the response delay time.

[0013] On the basis of the above technical solution, the present invention is further described as follows:

[0014] As a further solution of the present invention, the creation of the BP neural network deep learning model specifically includes:

[0015] Creating a BP neural network deep learning model, wherein the training process of the BP neural network deep learning model specifically includes a working signal forward transmission process and an error signal back propagation process;

[0016] The BP neural network consists of an input layer, several hidden layers and an output layer;

[0017] The construction process of the BP neural network includes:

[0018] Construct the number of input layer nodes of BP neural network;

[0019] Construct the number of output layer nodes of BP neural network;

[0020] Construct the number of hidden layer nodes (hyperparameter) of BP neural network;

[0021] The number of input layer nodes for constructing the BP neural network specifically includes:

[0022] The input layer of the BP neural network consists of 4 input nodes, including the voltage (kV1) and current value (mA1) of the previous stage and the voltage value (kV2) and current value (mA2) of the next stage;

[0023] The number of output layer nodes of the BP neural network is constructed as follows:

[0024] The output layer of the BP neural network consists of 1 output node with a delay time, and the delay time t is in microseconds / us;

[0025] The number of hidden layer nodes (hyperparameters) of constructing the BP neural network specifically includes:

[0026] The number of nodes in the hidden layer of the BP neural network is determined by the following strategy calculation formula (1):

[0027]

[0028] Among them, h is the number of hidden layer nodes, m is the number of input layer nodes, n is the number of output layer nodes, and a is a tuning constant between 1 and 10;

[0029] According to the number of input layer nodes and the number of output layer nodes of the BP neural network constructed above, it can be known that the number of input layer nodes m of the BP neural network is equal to 4, and the number of output layer nodes n is equal to 1;

[0030] According to the principle that the number of hidden layer nodes h is not greater than twice the number of input layer nodes, the adjustment constant is set to 6. The number of hidden layer nodes is calculated by the hidden layer node number strategy formula (1) to be h≈8. Further bias compensation is given and a bias b is added. Finally, the number of neuron nodes in the hidden layer is set to 9. The number of neuron nodes in each layer is the same, that is, 9 neurons.

[0031] As a further solution of the present invention, the construction process of the BP neural network also includes:

[0032] Construct the number of hidden layers (hyperparameters) of the BP neural network, including:

[0033] Based on the fact that the input parameters of the CT ray source subsystem and the delay time of the output X-rays are nonlinearly mapped, and its goal is to control the X-ray dose generated by the CT ray source subsystem with high precision, according to the BP neural network deep learning model, the number of hidden layers is determined to be 5 layers, combining the characteristics of the CT ray source subsystem and the input and output parameters of the BP neural network, while taking into account the network performance and training complexity.

[0034] As a further solution of the present invention, the construction process of the BP neural network also includes:

[0035] Construct the activation function of the BP neural network, including:

[0036] The ELU activation function has a zero-mean output distribution and can speed up training. The ELU activation function is unilaterally saturated and has better convergence characteristics.

[0037] Accurately controlling the X-ray dose according to the CT ray source subsystem is a typical regression task. To ensure the training speed and better convergence, the BP neural network uses the ELU activation function.

[0038] The ELU activation function curve expression is:

[0039]

[0040] Among them, α is a predefined hyperparameter used to control the saturation of the output when x is negative;

[0041] The predefined hyperparameter α is equal to 1.

[0042] As a further solution of the present invention, the construction process of the BP neural network also includes:

[0043] Construct the weight hyperparameters and bias hyperparameters of the BP neural network, including:

[0044] The BP neural network is based on the set ELU activation function, and the He initialization method is used for weight initialization. The expression of the He initialization function is:

[0045]

[0046] Among them, w is the weight hyperparameter, The mean is 0 and the variance is Normal distribution, h is the number of hidden layer nodes;

[0047] The bias in the BP neural network is set to 1, that is, the bias hyperparameter b=1, which is used to adjust the threshold of the activation function of each node in the BP neural network, help the neural network to better fit the training data, improve the ability to handle complex problems and the nonlinear performance of the model, and prevent the model from overfitting.

[0048] As a further solution of the present invention, the construction process of the BP neural network also includes:

[0049] Construct the signal forward propagation of BP neural network;

[0050] Construct error back propagation of BP neural network;

[0051] The signal forward propagation of the BP neural network is constructed as follows:

[0052] Signal forward propagation refers to the process in a neural network where the input layer is transmitted to the output layer after being processed by the weighted sum of neurons in each layer and the activation function, that is, the forward transmission process of the working signal mentioned above;

[0053] The basic formula design for signal forward propagation includes linear functions and activation functions;

[0054] Among them, the linear function formula is:

[0055] y=wx + b (4)

[0056] In the formula, y is the input value of the activation function, w is the weight hyperparameter, and b is the bias hyperparameter; the activation function formula is:

[0057] z=σ(y) (5)

[0058] In the formula, z is the neuron input value of the next layer, and σ is the activation function;

[0059] The formula from the input layer to the first hidden layer is:

[0060]

[0061] Z1=σ(Y1) (7)

[0062] The formula for hidden layer (k-1) and hidden layer (k) is:

[0063]

[0064] Z k =σ(Y k ) (9)

[0065] The formulas for the hidden layer and the output layer are:

[0066]

[0067] The error back propagation of constructing the BP neural network specifically includes:

[0068] Error back propagation, that is, the process of back propagation of error signals, is from the output layer through the hidden layer to the input layer. In the back propagation process, the weight parameters and bias parameters in each layer of the neural network are updated in turn using the gradient descent method according to the propagated error signal.

[0069] Among them, the update formula of the weight parameter is:

[0070]

[0071] In the formula, w ′ is the updated weight, w is the weight before update, η is the learning rate, and L is the loss function;

[0072] The update formula of the bias parameter is:

[0073]

[0074] Where b ′ is the bias after update, b is the bias before update, η is the learning rate, and L is the loss function;

[0075] The step size of parameter update is controlled by setting the learning rate, and the initial value of the learning rate η is set to 0.01.

[0076] As a further solution of the present invention, the construction process of the BP neural network also includes:

[0077] Construct the loss function of BP neural network, including:

[0078] The loss function is used to measure the gap between the output of the neural network model and the actual result. The effect of the loss function is directly related to the degree of final convergence of the model and the degree of loss reduction.

[0079] This embodiment uses the mean square error function as the loss function;

[0080] If the output value is set to 1, the mean square error formula is:

[0081] L=(t ′ -t us ) 2 (13)

[0082] Where L is the loss function, t ′ is the predicted value, t us is the output value.

[0083] A high-precision X-ray control system for executing the high-precision X-ray control method based on a BP neural network, the high-precision X-ray control system comprising:

[0084] Training model creation module, used to create BP neural network deep learning model;

[0085] A sample acquisition module is used to generate a number of X-ray control system response delay sample data under different parameters and conditions by the CT ray source subsystem based on the created BP neural network deep learning model;

[0086] A sample training module is used to train a BP neural network deep learning model using a number of X-ray control system response delay sample data, and fit the delay time value between the different input parameters and the delay time value between the input parameters after the change to the corresponding output specified X-ray dose;

[0087] A training model deployment module is used to deploy the trained BP neural network deep learning model to the CT ray source subsystem;

[0088] The parameter input module is used to input the voltage and current parameters of the current stage and the voltage and current parameters of the next stage into the BP neural network deep learning model when the X-ray dose is ready to change, and the BP neural network deep learning model infers and outputs the response delay time of the CT ray source subsystem;

[0089] The delayed advance input module is used to advance the response delay time according to the response delay time output by the BP neural network deep learning model to input the voltage and current of the next stage into the CT ray source subsystem.

[0090] An electronic device comprises at least one processor and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can perform the steps of the method.

[0091] A storage medium stores a computer program, which implements the steps of the method when executed by a processor.

[0092] The present invention has the following beneficial effects:

[0093] The method and system set up three parts: model training, model reasoning, and X-ray control. The feedback delay sample data of the X-ray control system under different conditions are collected and sent to the BP neural network model for training. The trained BP neural network model is then used to infer the response delay duration of the subsequent process in real time based on the current position and parameters. Then, according to the inferred response delay duration, the control parameters are sent to the X-ray control system in advance. When the specified position and angle are reached, the X-ray control system has accurately adjusted the X-ray to reach the expected dose. This can effectively achieve high-power CT systems, and can simultaneously ensure imaging quality as well as high real-time and high-accuracy control of the time and position of the specified X-ray dose, thereby achieving better images with smaller doses and improving the practicality of the overall functional architecture. BRIEF DESCRIPTION OF THE DRAWINGS

[0094] In order to more clearly illustrate the implementation mode of the present invention or the technical solution in the prior art, the drawings required for the implementation mode or the description of the prior art will be briefly introduced below. The structures, proportions, sizes, etc. illustrated in this specification are only used to match the contents disclosed in the specification for people familiar with this technology to understand and read. Any structural modification, change in proportional relationship or adjustment of size should still fall within the scope of the technical contents disclosed in the present invention without affecting the effects and purposes that can be achieved by the present invention.

[0095] Figure 1 The present invention provides an overall flow chart of a high-precision X-ray control method based on a BP neural network.

[0096] Figure 2 A schematic diagram of the overall process principle of the X-ray high-precision control method based on BP neural network provided in an embodiment of the present invention.

[0097] Figure 3 A schematic diagram of the functional architecture principle corresponding to the BP neural network-based X-ray high-precision control method provided in an embodiment of the present invention.

[0098] Figure 4A schematic diagram of the BP neural network structure constructed in the BP neural network-based X-ray high-precision control method provided in an embodiment of the present invention.

[0099] Figure 5 A schematic diagram of an ELU activation function curve for constructing a BP neural network in the BP neural network-based X-ray high-precision control method provided in an embodiment of the present invention.

[0100] Figure 6 A schematic diagram of the functional architecture of a high-precision X-ray control system based on a BP neural network provided in an embodiment of the present invention.

[0101] Figure 7 A schematic diagram of the physical structure of an electronic device provided by an embodiment of the present invention.

[0102] In the accompanying drawings, the components represented by the reference numerals are listed as follows:

[0103] Training model creation module 10; sample collection module 20; sample training module 30; training model deployment module 40; parameter input module 50; delay advance input module 60;

[0104] Electronic device 90 : processor 901 , memory 902 , internal bus 903 . DETAILED DESCRIPTION

[0105] The following is a description of the implementation of the present invention by specific embodiments. People familiar with the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. Obviously, the described embodiments are 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.

[0106] The terms such as "upper", "lower", "left", "right", and "middle" used in this specification are only for the convenience of description and are not intended to limit the scope of the invention. Changes or adjustments to their relative relationships should be regarded as within the scope of the invention without substantially changing the technical content.

[0107] like Figures 1 to 5As shown, an embodiment of the present invention provides a high-precision X-ray control method based on a BP neural network, and its basic principles are divided into three parts: model training, model reasoning, and X-ray control. That is, first, feedback delay sample data of the X-ray control system under different conditions are collected, and the collected feedback delay sample data are sent to the BP neural network model for training, and then the trained BP neural network model is used to infer the response delay duration of the subsequent process in real time based on the current position and parameters, and then the control parameters are sent to the X-ray control system in advance based on the inferred response delay duration. When the specified position and angle are reached, the X-ray control system has accurately adjusted the X-ray to reach the expected dose, which is intended to simultaneously ensure the imaging quality and high real-time and high-accuracy control of the time and position of the X-ray specified dose for high-power CT systems, thereby achieving better images with smaller doses.

[0108] The above-mentioned X-ray high-precision control method based on BP neural network specifically includes the following steps:

[0109] S100: Create a BP neural network deep learning model;

[0110] The specific process is: creating a BP neural network deep learning model, wherein the training process of the BP neural network deep learning model specifically includes a working signal forward transmission process and an error signal back propagation process;

[0111] The BP neural network consists of an input layer, several hidden layers and an output layer. Figure 4 For a schematic diagram of the BP neural network structure constructed in the X-ray high-precision control method based on the BP neural network provided in this embodiment, please refer to Figure 4 , the more specific construction process of the BP neural network is:

[0112] S101: the number of input layer nodes for constructing BP neural network;

[0113] The input layer of the BP neural network consists of 4 input nodes, including the voltage (kV1) and current value (mA1) of the previous stage and the voltage value (kV2) and current value (mA2) of the next stage;

[0114] S102: Construct the number of output layer nodes of BP neural network;

[0115] The output layer of the BP neural network consists of 1 output node with a delay time, and the delay time t is in microseconds / us;

[0116] S103: construct the number of hidden layer nodes (hyperparameter) of the BP neural network;

[0117] The number of nodes in the hidden layer of the BP neural network is determined by the following strategy calculation formula (1):

[0118]

[0119] Among them, h is the number of hidden layer nodes, m is the number of input layer nodes, n is the number of output layer nodes, and a is a tuning constant between 1 and 10;

[0120] According to the number of input layer nodes and the number of output layer nodes of the BP neural network constructed above, it can be known that the number of input layer nodes m of the BP neural network is equal to 4, and the number of output layer nodes n is equal to 1;

[0121] According to the principle that the number of hidden layer nodes h is not greater than twice the number of input layer nodes, the adjustment constant is set to 6. The number of hidden layer nodes is calculated by the hidden layer node number strategy formula (1) to be h≈8. Further bias compensation is given and a bias b is added. Finally, the number of neuron nodes in the hidden layer is set to 9. The number of neuron nodes in each layer is the same, that is, 9 neurons.

[0122] S104: construct the number of hidden layers (hyperparameter) of the BP neural network;

[0123] The number of hidden layers of BP neural network is an important factor affecting the performance of neural network. Several hidden layers are used to fit nonlinear functions. In theory, the more hidden layers there are, the stronger the network's nonlinear mapping ability to input data is. However, in practice, it also increases the complexity and training difficulty of the network, and may lead to overfitting, that is, the model performs well on the training data, but poorly on new data. It may also make the BP neural network more likely to fall into local minima during training, affecting the generalization ability of the model. Similarly, fewer hidden layers may lead to underfitting, limiting the model's ability to capture complex patterns in the data, making the model perform poorly when facing new data or complex tasks.

[0124] In this embodiment, the input parameters of the CT ray source subsystem and the delay time of the output X-ray are nonlinearly mapped, and the goal is to control the X-ray dose generated by the CT ray source subsystem with high precision. Therefore, based on the BP neural network deep learning model, the characteristics of the CT ray source subsystem and the input and output parameters of the BP neural network are comprehensively considered, while taking into account the network performance and training complexity, the number of hidden layers is determined to be a 5-layer structure;

[0125] S105: constructing an activation function of a BP neural network;

[0126] At present, commonly used activation functions include Sigmod (S-shaped function), Softmax (flexible maximum transfer function), tanh function (hyperbolic tangent function), ReLU (rectified linear unit function), LeakyReLU (leaky linear unit function), ELU (exponential linear unit function) and other functions; among them, the Sigmod activation function maps real numbers to the interval (0, 1), which is often used to solve binary classification problems; the Softmax activation function normalizes real numbers to the interval (0, 1) and makes the sum 1, which is often used to solve multi-classification problems; the Tanh activation function maps real numbers to the interval (-1, 1), which is often used to solve multi-classification or regression problems; the ReLU activation function outputs the positive number itself for positive numbers and outputs 0 for negative numbers, but the disadvantage is that negative values ​​may cause the problem of neuron "death"; the LeakyReLU function solves the problem of neuron "death" of the ReLU function, but there is a problem of poor convergence; the ELU activation function has a zero-mean output distribution, which can speed up the training speed, and the ELU activation function is unilaterally saturated, which can converge better;

[0127] Since the precise control of X-ray dose by the CT ray source subsystem in this embodiment is a typical regression task, in order to ensure the training speed and better convergence, the BP neural network adopts the ELU activation function;

[0128] Figure 5 For a schematic diagram of the ELU activation function curve of the BP neural network constructed in the X-ray high-precision control method based on the BP neural network provided in this embodiment, please refer to Figure 5 , the ELU activation function curve expression is:

[0129]

[0130] Among them, α is a predefined hyperparameter used to control the saturation of the output when x is negative;

[0131] In this embodiment, the predefined hyperparameter α is equal to 1;

[0132] S106: construct weight hyperparameters and bias hyperparameters of the BP neural network;

[0133] In BP neural network, weight initialization is crucial to the training effect. Improper selection of weights may cause the network to fail to train correctly or overfit the data. Common methods of weight initialization include uniform distribution, Gaussian distribution, Xavier initialization, He initialization, etc.

[0134] The BP neural network in this embodiment is based on the set ELU activation function, and the He initialization method (also known as Kaiming initialization) is specifically used for weight initialization. The expression of the He initialization function is:

[0135]

[0136] Among them, w is the weight hyperparameter, The mean is 0 and the variance is Normal distribution, h is the number of hidden layer nodes;

[0137] The bias in the BP neural network is used to adjust the threshold of the activation function of each node in the neural network, so as to help the neural network better fit the training data; and the bias can ensure better fitting and improve the ability to handle complex problems. More importantly, it can improve the nonlinear performance of the model and prevent the model from overfitting.

[0138] In this embodiment, the bias is set to 1, that is, the bias hyperparameter b=1;

[0139] S107: Construct the signal forward propagation of BP neural network (Forward Propagation);

[0140] Signal forward propagation refers to the process in a neural network where the input layer is transmitted to the output layer after being processed by the weighted sum of neurons in each layer and the activation function, that is, the forward transmission process of the working signal mentioned above;

[0141] The basic formula design for signal forward propagation includes linear functions and activation functions;

[0142] Among them, the linear function formula is:

[0143] y=wx + b (4)

[0144] In the formula, y is the input value of the activation function, w is the weight hyperparameter, and b is the bias hyperparameter; the activation function formula is:

[0145] z=σ(y) (5)

[0146] In the formula, z is the neuron input value of the next layer, and σ is the activation function;

[0147] The formula from the input layer to the first hidden layer is:

[0148]

[0149] Z1=σ(Y1) (7)

[0150] The formula for hidden layer (k-1) and hidden layer (k) is:

[0151]

[0152] Z k =σ(Y k ) (9)

[0153] The formulas for the hidden layer and the output layer are:

[0154]

[0155] S108: Constructing error back propagation of BP neural network;

[0156] Error back propagation, that is, the process of back propagation of error signals, is from the output layer through the hidden layer to the input layer. In the back propagation process, the weight parameters and bias parameters in each layer of the neural network are updated in turn using the gradient descent method according to the propagated error signal.

[0157] Among them, the update formula of the weight parameter is:

[0158]

[0159] In the formula, w ′ is the updated weight, w is the weight before update, η is the learning rate, and L is the loss function;

[0160] The update formula of the bias parameter is:

[0161]

[0162] Where b ′ is the bias after update, b is the bias before update, η is the learning rate, and L is the loss function;

[0163] By setting the learning rate control parameter update step, this embodiment sets the initial value of the learning rate to 0.01;

[0164] S109: Construct the loss function of BP neural network;

[0165] The loss function is used to measure the gap between the output of the neural network model and the actual result. The effect of the loss function is directly related to the degree of final convergence of the model and the degree of loss reduction.

[0166] This embodiment uses the mean square error function as the loss function;

[0167] If the output value is set to 1, the mean square error formula is:

[0168] L=(t ′ -t us ) 2 (13)

[0169] Where L is the loss function, t ′ is the predicted value, t us is the output value;

[0170] S200: Based on the created BP neural network deep learning model, the CT ray source subsystem generates several X-ray control system response delay sample data under different parameters and conditions;

[0171] S300: using a plurality of X-ray control system response delay sample data to train a BP neural network deep learning model, fitting the delay time values ​​of changes between different input parameters and the delay time values ​​between the input parameters after the change and the corresponding output specified X-ray dose;

[0172] S400: deploying the trained BP neural network deep learning model to the CT ray source subsystem;

[0173] S500: When the X-ray dose is ready to change, the voltage and current parameters of the current stage and the voltage and current parameters of the next stage are pre-input into the BP neural network deep learning model, and the BP neural network deep learning model infers and outputs the response delay time of the CT ray source subsystem;

[0174] S600: according to the response delay time output by the BP neural network deep learning model, the voltage and current of the next stage are input into the CT ray source subsystem in advance of the response delay time;

[0175] Finally, the CT ray source subsystem controls the change of the X-ray dose, specifically, the X-ray dose is changed by controlling the change of the filament voltage (kilovolt / kV) and the tube current (milliampere / mA);

[0176] The CT ray source subsystem periodically outputs the current tube current value (milliampere / mA) during the process of generating the ray source, and the cycle frequency is milliseconds, preferably 5-10ms.

[0177] Through the above control process, the CT ray source subsystem can accurately achieve the (theoretical) expected X-ray dose at a specified position, specified angle, and specified time point.

[0178] The embodiment of the present invention also provides a high-precision X-ray control system based on BP neural network. Figure 6 The functional architecture diagram of the high-precision X-ray control system based on BP neural network provided in the embodiment of the present invention is as follows: Figure 6 As shown, the X-ray high-precision control system comprises:

[0179] The training model creation module 10 is used to create a BP neural network deep learning model;

[0180] The sample acquisition module 20 is used to generate a plurality of X-ray control system response delay sample data under different parameters and conditions by the CT ray source subsystem based on the created BP neural network deep learning model;

[0181] The sample training module 30 is used to train the BP neural network deep learning model using a number of X-ray control system response delay sample data, and fit the delay time value between the different input parameters and the delay time value between the input parameters after the change to the corresponding output specified X-ray dose;

[0182] A training model deployment module 40 is used to deploy the trained BP neural network deep learning model to the CT ray source subsystem;

[0183] The parameter input module 50 is used to input the voltage and current parameters of the current stage and the voltage and current parameters of the next stage into the BP neural network deep learning model when the X-ray dose is ready to change, and the BP neural network deep learning model infers and outputs the response delay time of the CT ray source subsystem;

[0184] The delayed advance input module 60 is used to advance the response delay time according to the response delay time output by the BP neural network deep learning model to input the voltage and current of the next stage into the CT ray source subsystem.

[0185] Figure 7 FIG. 1 is a schematic diagram of the physical structure of an electronic device according to an embodiment of the present invention. Figure 7 As shown, the electronic device 90 includes: a processor 901 (processor), a memory 902 (memory) and an internal bus 903;

[0186] The processor 901 and the memory 902 communicate with each other via the internal bus 903;

[0187] The processor 901 is used to call the program instructions in the memory 902 to execute the methods provided by the above-mentioned method embodiments, for example, including: creating a BP neural network deep learning model; based on the created BP neural network deep learning model, generating a plurality of X-ray control system response delay sample data under different parameters and conditions by the CT ray source subsystem; using a plurality of X-ray control system response delay sample data to train the BP neural network deep learning model, fitting the delay time value of the change between different input parameters and the delay time value between the input parameters after the change to the corresponding output specified X-ray dose; deploying the trained BP neural network deep learning model to the CT ray source subsystem; when the X-ray dose is ready to change, pre-inputting the voltage and current parameters of the current stage and the voltage and current parameters of the next stage into the BP neural network deep learning model, and the BP neural network deep learning model infers and outputs the response delay time of the CT ray source subsystem; according to the response delay time output by the BP neural network deep learning model, inputting the voltage and current of the next stage into the CT ray source subsystem in advance of the response delay time.

[0188] The present embodiment provides a non-transitory computer-readable storage medium, which stores computer instructions. The computer instructions enable a computer to execute the methods provided by the above-mentioned method embodiments, for example, including: creating a BP neural network deep learning model; based on the created BP neural network deep learning model, generating a plurality of X-ray control system response delay sample data under different parameters and conditions by a CT ray source subsystem; using a plurality of X-ray control system response delay sample data to train the BP neural network deep learning model, fitting the delay time value of the change between different input parameters and the delay time value between the input parameters after the change to the corresponding output specified X-ray dose; deploying the trained BP neural network deep learning model to the CT ray source subsystem; when the X-ray dose is ready to change, pre-inputting the voltage and current parameters of the current stage and the voltage and current parameters of the next stage into the BP neural network deep learning model, and inferring and outputting the response delay duration of the CT ray source subsystem by the BP neural network deep learning model; and inputting the voltage and current of the next stage into the CT ray source subsystem in advance of the response delay duration according to the response delay duration output by the BP neural network deep learning model.

[0189] A person skilled in the art can understand that all or part of the steps of implementing the above method embodiment can be completed by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps of the above method embodiment; and the aforementioned storage medium includes: ROM, RAM, magnetic disk or optical disk, etc., various storage media that can store program codes.

[0190] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, i.e., they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Those of ordinary skill in the art may understand and implement it without creative effort.

[0191] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods of each embodiment or some parts of the embodiment.

[0192] Although the present invention has been described in detail above by general description and specific embodiments, it is obvious to those skilled in the art that some modifications or improvements can be made to the present invention. Therefore, these modifications or improvements made without departing from the spirit of the present invention all belong to the scope of protection claimed by the present invention.

Claims

1. A high-precision X-ray control method based on BP neural network, characterized in that: The steps include: Create a BP neural network deep learning model; Based on the created BP neural network deep learning model, the CT ray source subsystem generates several X-ray control system response delay sample data under different parameters and conditions; A BP neural network deep learning model is trained using several X-ray control system response delay sample data to fit the delay time values ​​between different input parameters and the delay time values ​​between the input parameters after the change to the corresponding output specified X-ray dose; Deploy the trained BP neural network deep learning model to the CT ray source subsystem; When the X-ray dose is about to change, the voltage and current parameters of the current stage and the voltage and current parameters of the next stage are pre-input into the BP neural network deep learning model, and the BP neural network deep learning model infers and outputs the response delay time of the CT ray source subsystem; According to the response delay time output by the BP neural network deep learning model, the voltage and current of the next stage are input into the CT ray source subsystem in advance of the response delay time.

2. The X-ray high-precision control method based on BP neural network according to claim 1 is characterized in that: The creation of the BP neural network deep learning model specifically includes: Creating a BP neural network deep learning model, wherein the training process of the BP neural network deep learning model specifically includes a working signal forward transmission process and an error signal back propagation process; The BP neural network consists of an input layer, several hidden layers and an output layer; The construction process of the BP neural network includes: Construct the number of input layer nodes of BP neural network; Construct the number of output layer nodes of BP neural network; Construct the hyperparameter of the number of hidden layer nodes of the BP neural network; The number of input layer nodes for constructing the BP neural network specifically includes: The input layer of the BP neural network consists of 4 input nodes, including the voltage value kV1 and current value mA1 of the previous stage and the voltage value kV2 and current value mA2 of the next stage. The number of output layer nodes of the BP neural network is constructed as follows: The output layer of the BP neural network consists of 1 output node with a delay time, and the delay time t is in microseconds / us; The hyperparameters of the number of hidden layer nodes for constructing the BP neural network specifically include: The number of nodes in the hidden layer of the BP neural network is determined by the following strategy calculation formula (1): Among them, h is the number of hidden layer nodes, m is the number of input layer nodes, n is the number of output layer nodes, and a is a tuning constant between 1 and 10; According to the number of input layer nodes and the number of output layer nodes of the BP neural network constructed above, it can be known that the number of input layer nodes m of the BP neural network is equal to 4, and the number of output layer nodes n is equal to 1; According to the principle that the number of hidden layer nodes h is not greater than twice the number of input layer nodes, the adjustment constant is set to 6. The number of hidden layer nodes is calculated by the hidden layer node number strategy formula (1) to be h≈8. Further bias compensation is given and a bias b is added. Finally, the number of neuron nodes in the hidden layer is set to 9. The number of neuron nodes in each layer is the same, that is, 9 neurons.

3. The X-ray high-precision control method based on BP neural network according to claim 2 is characterized in that: The construction process of the BP neural network also includes: Construct the hidden layer number hyperparameters of the BP neural network, including: Based on the fact that the input parameters of the CT ray source subsystem and the delay time of the output X-rays are nonlinearly mapped, and its goal is to control the X-ray dose generated by the CT ray source subsystem with high precision, according to the BP neural network deep learning model, the number of hidden layers is determined to be 5 layers, combining the characteristics of the CT ray source subsystem and the input and output parameters of the BP neural network, while taking into account the network performance and training complexity.

4. The X-ray high-precision control method based on BP neural network according to claim 3 is characterized in that: The construction process of the BP neural network also includes: Construct the activation function of the BP neural network, including: The ELU activation function has a zero-mean output distribution and can speed up training. The ELU activation function is unilaterally saturated and has better convergence characteristics. Accurately controlling the X-ray dose according to the CT ray source subsystem is a typical regression task. To ensure the training speed and better convergence, the BP neural network uses the ELU activation function. The ELU activation function curve expression is: Among them, α is a predefined hyperparameter used to control the saturation of the output when x is negative; The predefined hyperparameter α is equal to 1.

5. The X-ray high-precision control method based on BP neural network according to claim 4 is characterized in that: The construction process of the BP neural network also includes: Construct the weight hyperparameters and bias hyperparameters of the BP neural network, including: The BP neural network is based on the set ELU activation function, and the He initialization method is used for weight initialization. The expression of the He initialization function is: Among them, w is the weight hyperparameter, The mean is 0 and the variance is Normal distribution, h is the number of hidden layer nodes; The bias in the BP neural network is set to 1, that is, the bias hyperparameter b=1, which is used to adjust the threshold of the activation function of each node in the BP neural network, help the neural network to better fit the training data, improve the ability to handle complex problems and the nonlinear performance of the model, and prevent the model from overfitting.

6. The X-ray high-precision control method based on BP neural network according to claim 5 is characterized in that: The construction process of the BP neural network also includes: Construct the signal forward propagation of BP neural network; Construct error back propagation of BP neural network; The signal forward propagation of the BP neural network is constructed as follows: Signal forward propagation refers to the process in a neural network where the input layer is transmitted to the output layer after being processed by the weighted sum of neurons in each layer and the activation function, that is, the forward transmission process of the working signal mentioned above; The basic formula design for signal forward propagation includes linear functions and activation functions; Among them, the linear function formula is: y=wx + b (4) In the formula, y is the input value of the activation function, w is the weight hyperparameter, and b is the bias hyperparameter; The activation function formula is: z=σ(y) (5) In the formula, z is the neuron input value of the next layer, and σ is the activation function; The formula from the input layer to the first hidden layer is: Z1=σ(Y1) (7) The formula for hidden layer k-1 and hidden layer k is: Z k =σ(Y k ) (9) The formulas for the hidden layer and the output layer are: The error back propagation of constructing the BP neural network specifically includes: Error back propagation, that is, the process of back propagation of error signals, is from the output layer through the hidden layer to the input layer. In the back propagation process, the weight parameters and bias parameters in each layer of the neural network are updated in turn using the gradient descent method according to the propagated error signal. Among them, the update formula of the weight parameter is: In the formula, w ′ is the updated weight, w is the weight before update, η is the learning rate, and L is the loss function; The update formula of the bias parameter is: Where b ′ is the bias after update, b is the bias before update, η is the learning rate, and L is the loss function; The step size of parameter update is controlled by setting the learning rate, and the initial value of the learning rate η is set to 0.

01.

7. The X-ray high-precision control method based on BP neural network according to claim 6 is characterized in that: The construction process of the BP neural network also includes: Construct the loss function of BP neural network, including: The loss function is used to measure the gap between the output of the neural network model and the actual result. The effect of the loss function is directly related to the degree of final convergence of the model and the degree of loss reduction. Use mean square error function as loss function; If the output value is set to 1, the mean square error formula is: L=(t ′ -t us ) 2 (13) Where L is the loss function, t ′ is the predicted value, t us is the output value.

8. An X-ray high-precision control system for executing the X-ray high-precision control method based on BP neural network as claimed in claim 7, characterized in that: The X-ray high-precision control system comprises: Training model creation module, used to create BP neural network deep learning model; A sample acquisition module is used to generate a number of X-ray control system response delay sample data under different parameters and conditions by the CT ray source subsystem based on the created BP neural network deep learning model; A sample training module is used to train a BP neural network deep learning model using a number of X-ray control system response delay sample data, and fit the delay time value between the different input parameters and the delay time value between the input parameters after the change to the corresponding output specified X-ray dose; A training model deployment module is used to deploy the trained BP neural network deep learning model to the CT ray source subsystem; The parameter input module is used to input the voltage and current parameters of the current stage and the voltage and current parameters of the next stage into the BP neural network deep learning model when the X-ray dose is ready to change, and the BP neural network deep learning model infers and outputs the response delay time of the CT ray source subsystem; The delayed advance input module is used to advance the response delay time according to the response delay time output by the BP neural network deep learning model to input the voltage and current of the next stage into the CT ray source subsystem.

9. An electronic device comprising at least one processor and a memory communicatively connected to the at least one processor, characterized in that: The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the steps of the method according to any one of claims 1 to 7.

10. A storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method described in any one of claims 1 to 7 are implemented.

Images