Mobile medical radiation protection shielding optimization method and system

Abstract

The invention relates to the technical field of medical systems, in particular to a mobile medical radiation protection shielding optimization method and system, and the method comprises the steps: obtaining the dose data of a historical radiation field, and constructing a learning data set containing a plurality of shielding layouts; acquiring height, weight, BMI and lesion position data of the detected person; based on the learning data set, training a deep learning model for predicting X-ray dose field distribution; according to the deep learning model, the optimal shielding body position is extracted, and the shielding body layout is determined; x-ray simulation calculation is carried out based on the data of the detected person; constructing a shielding position discretization strategy according to an X-ray simulation calculation result; constructing a reinforcement learning optimization protection scheme based on a shielding position discretization strategy; and outputting an optimized shielding strategy and a dose field monitoring result, the deep learning model can learn a complex radiation distribution mode from massive historical data, and the reinforcement learning algorithm can continuously optimize a decision strategy in a dynamic environment.

Description

Technical Field

[0001] The present invention relates to the technical field of medical systems, and more specifically, to a mobile medical radiation protection shielding optimization method and system thereof. Background Art

[0002] With the continuous advancement of medical technology, X-rays are increasingly used in diagnosis and treatment. However, the potential harm of X-ray radiation to human health is also increasingly attracting attention. In order to protect the safety of medical staff and patients, effective radiation protection has become an important issue for medical institutions. Traditional radiation protection methods mainly rely on fixed shielding facilities and protective clothing, but these methods often fail to meet the requirements of flexibility and accuracy in the application scenarios of mobile X-ray equipment.

[0003] In recent years, with the development of artificial intelligence technology, some researchers have begun to try to apply machine learning algorithms to the field of radiation protection. For example, some studies have proposed using neural networks to predict the scattering distribution of X-rays to assist in shielding design. Other scholars have explored radiation dose assessment methods based on Monte Carlo simulations in an attempt to improve the accuracy of dose calculations. These methods have improved the protection effect to a certain extent, but there are still some significant limitations.

[0004] First, most of the existing intelligent protection methods are designed for static scenes and are difficult to adapt to the frequently changing working environment of mobile X-ray equipment. Secondly, these methods often treat dose prediction and shielding optimization as independent steps, lacking a unified optimization framework, resulting in less than ideal overall protection effects. Furthermore, existing methods lack accuracy when dealing with complex scattered radiation, especially when the human body model and equipment geometry are complex. In addition, most methods ignore the real-time requirements and are unable to adjust the protection strategy in a timely manner according to dynamic changes during the inspection process.

[0005] More importantly, existing technologies generally lack consideration for individualized protection. Different patients have different body shapes, examination sites, and health conditions, and a unified protection plan is difficult to meet individual needs. At the same time, existing methods also need to be improved in terms of user-friendliness. Medical staff often need complex operations to achieve protection optimization, which may cause unnecessary delays in emergency medical situations.

[0006] In view of the above problems, there is an urgent need for a medical radiation protection shielding optimization method that can comprehensively consider mobility, real-time, accuracy and personalized needs. This method should be able to adaptively optimize the shielding strategy in real time according to different examination scenarios and patient characteristics, while ensuring the simplicity and intuitiveness of operation. Summary of the invention

[0007] The present invention aims at the deficiencies of the existing technology and proposes an innovative mobile medical radiation protection shielding optimization method and system. The method realizes the automation of the whole process from data collection, model training to real-time optimization through the organic combination of intelligent algorithms. It can not only accurately predict the radiation distribution in complex environments, but also dynamically adjust the shielding strategy according to real-time data, truly realizing personalized and precise radiation protection.

[0008] The present invention provides a mobile medical radiation protection shielding optimization method, comprising:

[0009] The acquisition steps include:

[0010] Obtain historical radiation field dose data and construct a learning dataset containing various shielding layouts;

[0011] Obtain the height, weight, BMI and lesion location data of the person being tested;

[0012] Processing steps include:

[0013] Based on the learning data set, a deep learning model is trained to predict the X-ray dose field distribution;

[0014] According to the deep learning model, extracting the optimal shielding body position and determining the shielding body layout;

[0015] Performing X-ray simulation calculation based on the detected person data;

[0016] constructing a shielding position discretization strategy according to the X-ray simulation calculation results;

[0017] Based on the shielding position discretization strategy, a reinforcement learning optimization protection scheme is constructed;

[0018] Output steps include:

[0019] Output optimized shielding strategy and dose field monitoring results.

[0020] Preferably, the deep learning model adopts a self-encoding stacked denoising auto-encoding network or a neural network, and the self-encoding stacked denoising auto-encoding network or the neural network has at least one denoising auto-encoding layer.

[0021] Preferably, the method for obtaining historical radiation field dose data is:

[0022] A Monte Carlo algorithm is used to simulate mobile X-ray equipment, phantom positions, shielding strategies, and radiation dose fields. Simulation software is used to conduct experiments to obtain radiation field dose information for a variety of shielding layouts and form the learning data set.

[0023] Preferably, the learning data set includes:

[0024] Body models, mobile X-ray equipment, shielding arrangements, radiation field and dose field images;

[0025] The radiation field represents the position and angle of the X-ray emission point and the vector of the radiation intensity parameters; the dose field image is a collection of dose values ​​of multiple voxels, represented as a three-dimensional matrix of the dose field.

[0026] Preferably, the dose field image is calculated according to the following formula:

[0027] dI=R·S·I,

[0028] Among them, R is the ray path matrix, S is the tissue sensitivity matrix, and I is the ray intensity matrix.

[0029] Preferably, the step of constructing a reinforcement learning optimization protection scheme includes:

[0030] Using reinforcement learning algorithms, we calculate the reward function of the optimized shielding strategy.

[0031] The reward function of the optimized shielding strategy is obtained by the following method:

[0032] A reinforcement learning model is established. The reinforcement learning model is set with an expected target benefit. The parameters of the model are obtained according to the optimized shielding strategy. The training process of the model is as follows:

[0033] Define the current dose field as S t , define the coordinates of the shielding body as a t , define the action of reinforcement learning as S t+1 , define the reward obtained after the shield moves as r t ;

[0034] Define the target return as R t , the target benefit is:

[0035] R t =β1R1+(1-β1)R2+β n (1-β n-1 )…+β N-1 (1-β N-2 ) / (1-β N-1 ),

[0036] Among them, β is the iteration reward coefficient.

[0037] As an advantage, it also includes the steps of:

[0038] According to the optimized shielding strategy, the actual mobile X-ray equipment is controlled to perform X-ray operations and output dose field monitoring.

[0039] Preferably, the method of dose field monitoring is:

[0040] The dose field monitoring adopts a plurality of distributedly arranged LED lights with adjustable intensity, and the positions of the LED lights are arranged around the body model;

[0041] Adjust the brightness of the LED lamp according to the location, radiation parameters, and shielding scheme of the mobile X-ray equipment.

[0042] Preferably, the step of X-ray simulation calculation comprises:

[0043] Predict single scan dose distribution based on two-dimensional Gaussian model;

[0044] Perform X-ray CT reconstruction;

[0045] Based on the Gaussian approximate model, the Gaussian approximate model is modified by weighting the prior probability density reconstructed by X-ray CT, and the optimal Gaussian model is solved by the maximum likelihood method;

[0046] Based on the optical propagation model, a dose-shielding position data set is constructed.

[0047] A mobile medical radiation protection shielding optimization system, comprising:

[0048] Main control computer, X-ray generator, flat panel detector, movable shield, dose detection detector, image generation unit, database and computing unit;

[0049] The X-ray generator is connected to a flat panel detector, the flat panel detector is built into a detector housing, and the detector housing is supported by a shielding body;

[0050] The dose detection detector is installed below the detector housing;

[0051] The main control machine is signal-connected to the X-ray generator, the flat panel detector, the movable shield, the dose detection detector, the image generation unit, the database and the calculation unit;

[0052] The computing unit has a built-in dose prediction network and a dose-shielding position data set;

[0053] The dose prediction network includes a dose distribution model, a Monte Carlo simulation, a deep reinforcement learning module, a Monte Carlo simulation and a dose simulation module connected in sequence;

[0054] The deep reinforcement learning module includes a data set, an actor-critic network and an optimization module;

[0055] The input of the data set is the energy of X-rays, the distance between the tube and the focus, the focus of the ray, the irradiation range and angle of the ray, the plain scan CT image and the position of the movable shielding body, and the output is the single scan dose and the multiple scan dose distribution;

[0056] The actor-critic network is used to perform X-ray CT reconstruction, predict dose distribution based on a Gaussian model, update the new position of the movable shield to calculate the single scan dose prediction, perform X-ray CT reconstruction and Monte Carlo simulation, and calculate rewards;

[0057] The optimization module optimizes the position of the movable shielding body through an actor-critic network to achieve the purpose of optimizing the shielding position.

[0058] The beneficial effects of the present invention are mainly reflected in the following aspects:

[0059] First, it cleverly combines deep learning and reinforcement learning algorithms to overcome the limitations of a single algorithm. The deep learning model can learn complex radiation distribution patterns from massive historical data, while the reinforcement learning algorithm can continuously optimize decision-making strategies in a dynamic environment. This combination not only improves the accuracy of predictions, but also achieves adaptive optimization of protection strategies.

[0060] Secondly, the dose field calculation formula proposed in this invention takes into account multiple factors such as the ray path, tissue sensitivity and ray intensity, greatly improving the accuracy of dose assessment. This multi-factor comprehensive method can better simulate the complex propagation process of X-rays in the human body, providing a more reliable basis for subsequent shielding optimization.

[0061] Furthermore, the method of the present invention also has a major breakthrough in real-time and visualization. Through the innovative LED visualization system, medical staff can intuitively observe the real-time dose distribution, which greatly improves the convenience and safety of operation. At the same time, the modular design and high degree of automation of the system greatly reduce the difficulty of operation, so that even in emergency situations, the optimal protection can be quickly achieved.

[0062] In addition, the method of the present invention pays special attention to personalized needs. By considering the individual characteristics of the patient, such as body shape and examination site, the system can customize the most suitable protection plan for each patient. This not only improves the protection effect, but also optimizes the patient experience.

[0063] Finally, the method of the present invention has good scalability and adaptability. It is not only applicable to current X-ray equipment, but can also be easily extended to other types of medical radiation protection fields. This versatility and foresight make the present invention have broad application prospects.

[0064] In general, the mobile medical radiation protection shielding optimization method and system proposed in the present invention comprehensively solves various challenges faced by existing technologies by innovatively integrating multiple advanced technologies. It not only significantly improves the protection effect and operation efficiency, but also brings considerable economic and social benefits to medical institutions. This innovative solution is expected to promote technological innovation in the field of medical radiation protection and provide stronger protection for the health and safety of patients and medical staff. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 It is a system logic block diagram of the present invention.

[0066] Figure 2 is a flow chart of the method of the present invention.

[0067] Figure 3 It is a logic block diagram of the computing unit of the present invention.

[0068] Figure 4 It is a logic block diagram of the dose field monitoring of the present invention. DETAILED DESCRIPTION

[0069] Please refer to Figure 1-4 The present invention provides a mobile medical radiation protection shielding optimization method and system. The method realizes intelligent optimization of medical radiation protection shielding by combining deep learning and reinforcement learning algorithms, significantly improving the protection effect and operation efficiency.

[0070] The mobile medical radiation protection shielding optimization method of the present invention comprises an acquisition step, a processing step and an output step.

[0071] In the acquisition step, historical radiation field dose data is first acquired to construct a learning data set containing multiple shielding layouts. The historical radiation field dose data here can be derived from actual measurement data or obtained through simulation. Preferably, the present invention uses a Monte Carlo algorithm for simulation to obtain a large amount of high-quality training data. Specifically, different models of mobile X-ray equipment, patient models of different body shapes, and various possible shielding strategies can be simulated to obtain comprehensive radiation field dose distribution data.

[0072] In addition, this method also obtains the height, weight, BMI and lesion location data of the person being tested. These individualized data are crucial for subsequent precise protection optimization. For example, for patients with a higher BMI, an increased X-ray dose may be required to obtain a clear image, so shielding measures also need to be strengthened accordingly.

[0073] In the processing step, firstly, based on the learning data set, a deep learning model is trained to predict the X-ray dose field distribution. The present invention preferably uses a self-encoding stacked denoising autoencoding network or a neural network as a deep learning model. This network structure has powerful feature extraction and denoising capabilities and can effectively process complex radiation field data.

[0074] Next, based on the trained deep learning model, the optimal shielding position is extracted and the shielding layout is determined. This step makes full use of the deep learning model's accurate prediction of the radiation field distribution, so that the best shielding strategy can be found.

[0075] Then, based on the acquired data of the detected person, an X-ray simulation calculation is performed. Here, a Gaussian model is used to approximate the dose distribution of a single scan. Preferably, the present invention uses the following formula to calculate the dose distribution of a single scan:

[0076]

[0077] Where D(x, y) is the dose rate at the spatial position (x, y), A is the normalization coefficient, and σ is the standard deviation of the Gaussian distribution.

[0078] Based on the X-ray simulation calculation results, this method further constructs a shield position discretization strategy. This step discretizes the continuous shield position space to facilitate subsequent reinforcement learning optimization. Preferably, the shield position space can be divided into an nxn grid, where n can be selected according to the specific application scenario, and is usually between 10 and 20.

[0079] Finally, based on the shield position discretization strategy, a reinforcement learning optimization protection scheme is constructed. The present invention adopts the deep Q network (DQN) algorithm for reinforcement learning. Specifically, the state space S is defined as the current dose field distribution and shield position, the action space A is the possible shield movement direction, and the reward function R is the weighted sum of the dose reduction and the operation complexity. The goal of DQN is to maximize the cumulative reward:

[0080] Q * (s, a) = E s′ [r+γmax a′ Q * (s′, a′)|s, a],

[0081] Among them, Q * (s, a) is the optimal action value function, r is the discount factor, which usually takes a value between 0.95 and 0.99.

[0082] In the output step, the method outputs the optimized shielding strategy and the dose field monitoring results. The optimized shielding strategy includes the optimal position and shape of the shielding body, while the dose field monitoring results intuitively display the optimized radiation distribution.

[0083] The deep learning model used in the present invention is a self-encoding stacked denoising auto-encoding network or neural network, and at least one denoising auto-encoding layer. This network structure has a multi-level feature extraction capability and can effectively process complex radiation field data. At the same time, the introduction of the denoising auto-encoding layer greatly improves the robustness of the model, enabling it to cope with data noise that may exist in practical applications.

[0084] Preferably, the self-encoding stacked denoising auto-encoding network of the present invention comprises 3 to 5 encoding layers and corresponding decoding layers. The number of neurons in each encoding layer can be reduced successively, for example, a structure of [1024, 512, 256, 128, 64] can be adopted. The decoding layer accordingly adopts a structure of [64, 128, 256, 512, 1024]. Between each encoding and decoding layer, a batch normalization layer and a ReLU activation function can be added to accelerate the training process and improve the nonlinear expression ability of the model.

[0085] The method of obtaining historical radiation field dose data of the present invention is to use Monte Carlo algorithm for simulation. This method can generate a large amount of high-quality training data without the need for actual radiation measurement experiments, which saves time and cost and avoids unnecessary radiation exposure.

[0086] In the specific implementation, mature Monte Carlo simulation software packages such as GEANT4 or MCNP can be used. During the simulation process, the following key factors need to be considered:

[0087] 1. Parameters of mobile X-ray equipment, such as X-ray energy spectrum, focal spot size, filter type, etc.

[0088] 2. The geometric shape and material composition of the phantom. A standard human body model such as the ICRP standard human body model can be used.

[0089] 3. The material properties and geometric shape of the shielding body. Commonly used shielding materials include lead, tungsten alloy, etc.

[0090] 4. Spatial distribution characteristics of the radiation field, including the effects of scattering and secondary radiation.

[0091] By systematically varying these parameters, a comprehensive dataset containing a variety of shielding layouts can be generated. For example, 10 different X-ray equipment configurations, 20 different phantoms, and 50 different shielding strategies can be simulated, resulting in 10,000 different radiation field dose distribution data. Such a large-scale dataset can effectively support subsequent deep learning model training.

[0092] The method of the present invention realizes intelligent optimization of medical radiation protection shielding by combining deep learning and reinforcement learning algorithms. Compared with traditional methods, the present invention has the following significant advantages:

[0093] 1. Accuracy: The radiation field distribution is accurately predicted through deep learning models, thereby achieving more accurate shielding optimization.

[0094] 2. Adaptability: The reinforcement learning algorithm can automatically adjust the shielding strategy according to different examination scenarios and patient characteristics.

[0095] 3. Efficiency: Intelligent algorithms greatly improve the speed of shielding optimization, making real-time optimization possible.

[0096] 4. Safety: Through better shielding strategies, the risk of radiation exposure to medical staff and patients can be significantly reduced.

[0097] In summary, the mobile medical radiation protection shielding optimization method provided by the present invention provides an efficient, accurate and safe solution for the field of medical radiation protection by innovatively combining multiple advanced algorithms. The method of the present invention further refines the composition of the learning data set. As described in claim 4, the learning data set includes a body model, a mobile X-ray device, a shielding body layout scheme, and radiation field and dose field images. This comprehensive data set design ensures that the deep learning model can fully learn the impact of various factors on radiation protection.

[0098] In a preferred embodiment of the present invention, the human body model uses a standard human body model, such as the ICRP reference human model. This model contains detailed information about the human anatomical structure and tissue composition, and can accurately simulate the propagation and absorption process of X-rays in the human body. Preferably, a variety of models of different body shapes can be set to cover patients of different ages, genders and body shapes.

[0099] The model of the mobile X-ray device contains information such as the device geometry, X-ray tube parameters, and filter configuration. This method takes into account different models and configurations of X-ray devices to improve the versatility of the model. For example, various device configurations with tube voltages ranging from 60 kV to 150 kV and tube currents ranging from 0.5 mA to 5 mA can be simulated.

[0100] The shielding arrangement is the core optimization object of the present invention. In the data set, various possible shielding shapes, sizes, materials and position information are included. Preferably, the present method considers a multi-layer shielding structure, such as a combination of a lead layer and a glass layer, to achieve a better protection effect.

[0101] The radiation field is represented by an innovative vectorization method. The radiation field is represented as a vector of the position and angle of the X-ray emission point and the radiation intensity parameters. Specifically, it can be represented by the following vectors:

[0102]

[0103] Among them, (x, y, z) is the spatial coordinates of the point where the X-ray is emitted, θ and φ are the zenith angle and azimuth angle of the ray respectively, and I is the radiation intensity parameter. This representation method is both concise and comprehensive, and provides a good input format for subsequent deep learning models.

[0104] The dose field image is represented by a three-dimensional matrix, and each element corresponds to a dose value of a voxel. Preferably, the method adopts a voxel size of 0.5 cm×0.5 cm×0.5 cm, which can provide sufficiently detailed dose distribution information without causing excessive computational burden.

[0105] The present invention proposes an innovative dose field calculation formula:

[0106] dI=R·S·I,

[0107] This formula takes into account a variety of influencing factors and can more accurately simulate the propagation and dose deposition process of X-rays in the human body. In this formula, R is the ray path matrix, S is the tissue sensitivity matrix, and I is the ray intensity matrix.

[0108] The ray path matrix R describes the propagation path of the X-ray in the human body. Preferably, the method uses a ray tracing algorithm to calculate the R matrix. For each voxel, the path length of the X-ray through the voxel is calculated, and possible scattering effects are taken into account.

[0109] The tissue sensitivity matrix S reflects the sensitivity of different human tissues to radiation. In one embodiment of the present invention, the value of the S matrix is ​​set according to the tissue weight factor published by ICRP. For example, the gonads, red bone marrow and colon have higher sensitivity and can be assigned a larger weight value; while the skin and bone surface have lower sensitivity and are assigned a smaller weight value.

[0110] The ray intensity matrix I describes the energy distribution of the incident X-rays. Preferably, the I matrix is ​​represented in the form of an energy spectrum to more accurately reflect the energy distribution characteristics of the X-rays. For example, for a 120 kV X-ray tube, the energy spectrum can be divided into 20 energy segments, and the width of each energy segment is 6 keV.

[0111] Through this sophisticated dose field calculation method, the present invention can more accurately evaluate the protective effects of different shielding strategies and provide a reliable basis for subsequent optimization processes.

[0112] The present invention adopts a reinforcement learning algorithm to optimize the shielding strategy and proposes a new reward function calculation method. This method guides the reinforcement learning algorithm to gradually optimize the shielding strategy by setting the expected target return.

[0113] In a preferred embodiment of the present invention, the reinforcement learning model adopts a deep Q network (DQN) structure. The state space is defined as the current dose field distribution S t and shielding body coordinates a t , the action space is defined as the possible movement directions S of the shield t +1. After each move, the reward r is calculated based on the new dose distribution t .

[0114] Target return R t The calculation uses an innovative recursive formula:

[0115] R t =β1R1+(1-β1)R2+β n (1-β n-1 )...+β N-1 (1-β N-2 ) / (1-β N-1 ),

[0116] Among them, β is the iterative reward coefficient, which usually takes a value between 0.9 and 0.99. This calculation method takes into account both the short-term benefit R1 and the long-term benefit R2, and can achieve a good balance between the immediate protection effect and the long-term optimization goal. Preferably, the short-term benefit R1 can be defined as the percentage of dose reduction after a single move, and the long-term benefit R2 can be defined as the reciprocal of the minimum number of moves required to reach the target protection level. Such a definition encourages the algorithm to find a significant movement strategy and avoids too frequent movement operations.

[0117] The method of the present invention also includes controlling the actual mobile X-ray device to perform X-ray operation according to the optimized shielding strategy and outputting the dose field monitoring result. This step applies the optimization result to the actual operation, realizing the combination of theory and practice.

[0118] In one embodiment of the present invention, the optimized shielding strategy is transmitted to the control system of the mobile X-ray device via a digital signal. The control system automatically adjusts the position and angle of the shielding body according to the received strategy. Preferably, a stepper motor is used to drive the shielding body to move so as to achieve precise position control. The accuracy of each adjustment can reach 0.1 mm, ensuring the accurate execution of the optimization strategy.

[0119] The output of the dose field monitoring results adopts an innovative visualization method. The present invention preferably uses a color heat map to intuitively display the dose distribution, while combining it with an isodose line representation method. The color of the heat map changes from blue to red, corresponding to the change of dose from low to high. The isodose lines clearly mark the boundaries of specific dose levels. This combined representation method is both intuitive and accurate, which helps medical staff quickly evaluate the protection effect and make necessary adjustments.

[0120] Through the above detailed design and implementation, the mobile medical radiation protection shielding optimization method of the present invention realizes the full process intelligence from data collection, model training to practical application. This method not only improves the protection effect, but also greatly improves the operating efficiency, bringing revolutionary progress to the field of medical radiation protection. The method of the present invention further innovatively proposes a real-time dose field monitoring scheme. As described in claim 8, the scheme uses multiple distributed adjustable intensity LED lights to visualize the dose distribution. This method intuitively displays the spatial distribution of the radiation field, which helps medical staff better understand and control radiation risks.

[0121] In a preferred embodiment of the present invention, an LED array is arranged around the phantom. Specifically, a 5×5 LED matrix can be arranged in each of the four directions of the front, back, left, and right of the phantom, with a total of 100 LEDs. The brightness of each LED can be adjusted in 256 levels between 0 and 255 to accurately reflect the dose intensity at different locations.

[0122] The brightness of the LED lamp is adjusted based on the position, radiation parameters and shielding scheme of the mobile X-ray device. This method uses an innovative mapping algorithm to convert the calculated dose value into the corresponding LED brightness. Preferably, a logarithmic mapping relationship is used:

[0123]

[0124] Among them, B is the LED brightness value (0-255), D is the dose value at that position, D0 is the reference dose value (usually taken as the background radiation dose), and k is the adjustment coefficient. This logarithmic mapping relationship can provide good visual resolution within a larger dose range.

[0125] The method of the present invention also takes into account the human eye's perception characteristics of brightness changes. According to the Weber-Fechner law, the brightness perceived by the human eye is proportional to the logarithm of the physical brightness. Therefore, the above logarithmic mapping algorithm is not only mathematically reasonable, but also conforms to the characteristics of human visual perception and can provide a more intuitive dose distribution display.

[0126] The X-ray simulation calculation steps of the present invention include a series of innovative algorithms and models. The combination of these algorithms and models realizes efficient and accurate dose distribution prediction, providing a reliable basis for subsequent shielding optimization.

[0127] First, this method predicts the single scan dose distribution based on a two-dimensional Gaussian model. In practical applications, the dose distribution of an X-ray beam usually presents characteristics similar to a Gaussian distribution. Preferably, the following two-dimensional Gaussian model is used:

[0128]

[0129] Where D(x, y) is the dose rate at position (x, y), A is the normalization coefficient, (x0, y0) is the beam center position, σ x and σ y are the standard deviations in the x and y directions, respectively. This model is both simple and effective, and can quickly give a preliminary estimate of the dose distribution.

[0130] Next, the method of the present invention performs X-ray CT reconstruction. Preferably, the reconstruction is performed using a filtered back projection (FBP) algorithm. The FBP algorithm has high computational efficiency and can quickly provide acceptable reconstruction results. In practical applications, a Ram-Lak filter or a Shepp-Logan filter can be used to strike a balance between image quality and computational speed according to specific requirements.

[0131] Based on the Gaussian approximation model and CT reconstruction results, this method further modifies the Gaussian model. This step adopts the idea of ​​Bayesian inference, takes the CT reconstruction results as prior information, and updates the initial Gaussian model. Specifically, the maximum a posteriori probability (MAP) estimation method can be used:

[0132] The maximum a posteriori probability (MAP) estimation method can be used:

[0133]

[0134] Among them, θ is the parameter set of the Gaussian model, and D is the CT reconstruction data. Through this method, a more accurate dose distribution model can be obtained.

[0135] Finally, the method of the present invention constructs a dose-shielding position data set based on the modified model. This data set is the basis for subsequent shielding optimization. Preferably, a grid sampling method can be used to uniformly sample the possible position space of the shielding body and calculate the dose distribution corresponding to each position. The sampling density can be adjusted according to the computing resources and accuracy requirements, for example, a 10cm×10cm×10cm grid division can be used.

[0136] The present invention also proposes an innovative mobile medical radiation protection shielding optimization system, which integrates a variety of advanced hardware and software modules and realizes the automation of the entire process from data collection to shielding optimization.

[0137] The core components of the system include the main control machine 1, X-ray generator 2, flat panel detector 3, movable shield 4, dose detection detector 5, image generation unit 6, database 7 and computing unit 8. These components form a highly integrated intelligent system through carefully designed connection and control methods.

[0138] The X-ray generator 2 is connected to the flat panel detector 3, and together they constitute the basic unit of X-ray imaging. The flat panel detector 3 is built into the detector housing and supported by the shielding body 4. This design not only protects the detector, but also provides preliminary radiation shielding. The dose detection detector 5 is installed under the detector housing to monitor the radiation dose in real time.

[0139] The main control machine 1 is the control center of the entire system, and it maintains signal connection with all other components. The main control machine 1 is responsible for coordinating the work of each module to ensure the synchronization and consistency of the system operation. For example, it can control the position adjustment of the movable shielding body 4 according to the output of the optimization algorithm.

[0140] Computational unit 8 is the "brain" of the system, with a built-in dose prediction network and dose-shielding position data set. The dose prediction network is a complex neural network structure, including a dose distribution model, Monte Carlo simulation, deep reinforcement learning module, Monte Carlo simulation and dose simulation module connected in sequence. This multi-level network structure can accurately simulate the complex radiation propagation process and provide a reliable prediction basis for shielding optimization.

[0141] The deep reinforcement learning module is the core part of the computational unit 8, which includes the data set, the actor-critic network, and the optimization module. The input of the data set includes the energy of the X-ray, the distance from the tube to the focal point, the focus of the ray, the range and angle of the ray irradiation, the plain scan CT image, and the position of the movable shield. These comprehensive input data ensure that the model can fully consider various influencing factors.

[0142] The actor-critic network is responsible for X-ray CT reconstruction, dose distribution prediction and shield position update. This network uses the latest deep reinforcement learning technology to find the optimal shielding strategy in a complex state space. The optimization module continuously adjusts the position of the movable shield based on the output of the actor-critic network until the best protection effect is achieved.

[0143] Through this carefully designed system architecture, the present invention realizes the intelligentization and automation of medical radiation protection. The system can adaptively adjust the shielding strategy according to the real-time collected data, greatly improving the protection effect and operation efficiency. At the same time, the modular design of the system also facilitates future upgrades and expansions.

[0144] In summary, the mobile medical radiation protection shielding optimization method and system proposed in this invention provide a comprehensive, efficient and intelligent solution for the field of medical radiation protection by innovatively combining multiple advanced algorithms and hardware designs. This method and system can not only significantly improve the protection effect and reduce the radiation risk of medical staff and patients, but also greatly improve the operating efficiency and bring significant economic and social benefits to medical institutions.

[0145] It should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modification, replacement, and improvement made within the principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A mobile medical radiation protection shielding optimization method, characterized in that: include: The acquisition steps include: Obtain historical radiation field dose data and construct a learning dataset containing various shielding layouts; Obtain the height, weight, BMI and lesion location data of the person being tested; Processing steps include: Based on the learning data set, training a deep learning model for predicting X-ray dose field distribution; According to the deep learning model, extracting the optimal shielding body position and determining the shielding body layout; Performing X-ray simulation calculation based on the detected person data; constructing a shielding position discretization strategy according to the X-ray simulation calculation results; Based on the shielding position discretization strategy, a reinforcement learning optimization protection scheme is constructed; Output steps include: Output optimized shielding strategy and dose field monitoring results.

2. The method according to claim 1, characterized in that The deep learning model adopts a self-encoding stacked denoising auto-encoding network or a neural network, and the self-encoding stacked denoising auto-encoding network or the neural network has at least one denoising auto-encoding layer.

3. The method according to claim 1, characterized in that The method for obtaining historical radiation field dose data is: A Monte Carlo algorithm is used to simulate mobile X-ray equipment, phantom positions, shielding strategies, and radiation dose fields. Simulation software is used to conduct experiments to obtain radiation field dose information for a variety of shielding layouts and form the learning data set.

4. The method according to claim 1, characterized in that: The learning data set includes: Body models, mobile X-ray equipment, shielding arrangements, radiation field and dose field images; The radiation field represents the position and angle of the X-ray emission point and the vector of the radiation intensity parameters; the dose field image is a collection of dose values ​​of multiple voxels, represented as a three-dimensional matrix of the dose field.

5. The method according to claim 4, characterized in that The dose field image is calculated according to the following formula: dI=R·S·I, Among them, R is the ray path matrix, S is the tissue sensitivity matrix, and I is the ray intensity matrix.

6. The method according to claim 1, characterized in that The steps of constructing the reinforcement learning optimization protection solution include: Using reinforcement learning algorithms, we calculate the reward function of the optimized shielding strategy. The reward function of the optimized shielding strategy is obtained by the following method: A reinforcement learning model is established. The reinforcement learning model is set with an expected target benefit. The parameters of the model are obtained according to the optimized shielding strategy. The training process of the model is as follows: Define the current dose field as S t , define the coordinates of the shielding body as a t , define the action of reinforcement learning as S t+1 , define the reward obtained after the shield moves as r t ; Define the target return as R t , the target benefit is: R t =β1R1+(1-β1)R2+β n (1-b) n-1 )…+b N-1 (1-b) N-2 ) / (1-b N-1 ), Among them, β is the iteration reward coefficient.

7. The method according to claim 1, characterized in that Also includes the steps: According to the optimized shielding strategy, the actual mobile X-ray equipment is controlled to perform X-ray operations and output dose field monitoring.

8. The method according to claim 7, characterized in that The method of dose field monitoring is: The dose field monitoring adopts a plurality of distributedly arranged LED lights with adjustable intensity, and the positions of the LED lights are arranged around the body model; Adjust the brightness of the LED lamp according to the location, radiation parameters, and shielding scheme of the mobile X-ray equipment.

9. The method according to claim 1, characterized in that: The steps of the X-ray simulation calculation include: Predict single scan dose distribution based on two-dimensional Gaussian model; Perform X-ray CT reconstruction; Based on the Gaussian approximate model, the Gaussian approximate model is modified by weighting the prior probability density reconstructed by X-ray CT, and the optimal Gaussian model is solved by the maximum likelihood method; Based on the optical propagation model, a dose-shielding position data set is constructed.

10. A mobile medical radiation protection shielding optimization system, characterized in that: include: Main control computer, X-ray generator, flat panel detector, movable shield, dose detection detector, image generation unit, database and computing unit; The X-ray generator is connected to a flat panel detector, the flat panel detector is built into a detector housing, and the detector housing is supported by a shielding body; The dose detection detector is installed below the detector housing; The main control machine is signal-connected to the X-ray generator, the flat panel detector, the movable shield, the dose detection detector, the image generation unit, the database and the calculation unit; The computing unit has a built-in dose prediction network and a dose-shielding position data set; The dose prediction network includes a dose distribution model, a Monte Carlo simulation, a deep reinforcement learning module, a Monte Carlo simulation and a dose simulation module connected in sequence; The deep reinforcement learning module includes a data set, an actor-critic network and an optimization module; The input of the data set is the energy of X-rays, the distance between the tube and the focus, the focus of the ray, the irradiation range and angle of the ray, the plain scan CT image and the position of the movable shielding body, and the output is the single scan dose and the multiple scan dose distribution; The actor-critic network is used to perform X-ray CT reconstruction, predict dose distribution based on a Gaussian model, update the new position of the movable shield to calculate the single scan dose prediction, perform X-ray CT reconstruction and Monte Carlo simulation, and calculate rewards; The optimization module optimizes the position of the movable shielding body through an actor-critic network to achieve the purpose of optimizing the shielding position.

Images