Radiotherapy lattice parameter optimization method and system based on space segmentation technology

Abstract

The invention discloses a radiotherapy lattice parameter optimization method based on a space segmentation technology. The method comprises the following steps: obtaining radiotherapy image data of a patient; calculating geometric features of the tumor according to the radiotherapy image data and preset geometric constraint conditions; and inputting the tumor geometric features and the initial lattice parameters into a lattice parameter optimization model constructed based on a PPO algorithm for identification to obtain optimal lattice parameters, wherein the lattice parameter optimization model comprises a strategy network, a value network, a parameter and optimization module and a reward calculation module. The method can output the optimal lattice parameter according to the radiotherapy image data of the patient, provides a reliable basis for space segmentation radiotherapy, does not depend on artificial experience, and can effectively improve the efficiency.

Description

Technical Field

[0001] The present invention relates to the technical field of lattice parameter optimization for spatially fractionated radiotherapy, and specifically relates to a method and system for optimizing radiotherapy lattice parameters based on spatial segmentation technology. Background Art

[0002] Spatially Fractionated Radiotherapy (SFRT) is a new radiotherapy method that improves the treatment effect by generating an alternating high-low dose distribution pattern inside the tumor. SFRT discretely distributes multiple high-dose lattice target areas in the tumor to form a dose "peak-valley" distribution, that is, a high-dose peak is formed in the area where the lattice is located, while the tumor area not covered by the lattice forms a low-dose valley area. Research shows that SFRT can increase the dose peak inside the tumor without significantly increasing the damage to normal tissues, can activate the immune response inside the tumor, and achieve a higher local tumor control rate. Especially for tumors that are difficult to treat effectively with traditional radiotherapy (such as large-volume and highly tolerant tumors), SFRT shows good efficacy.

[0003] However, a key challenge in the practical application of SFRT is how to reasonably set and optimize the parameters of the lattice, such as the lattice spacing, radius, and Peak-to-Valley Dose Ratio (PVDR). These parameters have a direct impact on the treatment method selection of doctors. However, due to the different sizes and shapes of tumors in each patient, manual or static algorithm-based parameter setting is often difficult to meet personalized needs. If the lattice distribution is too dense, it may lead to a decrease in the peak-to-valley dose ratio and increase the damage to normal tissues; if the lattice distribution is too sparse, it may not effectively cover the entire tumor area. Therefore, in order to achieve the best dose distribution, a lattice optimization method that can adaptively adjust according to the specific tumor morphology of the patient is needed.

[0004] In the prior art, SFRT mainly relies on operators to manually arrange the position and size of the lattice. Usually, a radiotherapy planner places the lattice one by one through a two-dimensional image display or a three-dimensional model to achieve the required dose peak-valley distribution. However, this method highly depends on the experience and judgment of the operator, with low efficiency and large errors. When the tumor volume is large or the shape is complex, manually arranging the lattice is not only time-consuming but also difficult to achieve the ideal uniformity, resulting in the dose distribution being difficult to meet the requirements of the best peak-to-valley ratio. In addition, due to the relatively fixed position and size of the manually arranged lattice, it is difficult for the operator to flexibly adjust the lattice parameters during the radiotherapy planning process, thereby reducing the treatment effect of SFRT.

[0005] To address the deficiencies of manual arrangement, some studies have proposed lattice arrangement methods based on geometric algorithms, such as the closest packing method. One study proposed a solution using the closest packing algorithm. This method determines the lattice spacing and initial position based on the tumor volume and shape, and arranges multiple spherical lattices using the principle of closest packing to cover the interior of the tumor target area as evenly as possible. This method achieves spatial segmentation of the dose through a uniform geometric distribution and is applicable to tumors with regular shapes and medium sizes. However, since the parameter settings in geometric algorithms are usually fixed, it is difficult to flexibly handle complex tumor morphologies.

[0006] Other studies have proposed an optimized lattice arrangement scheme based on multivariate Gaussian distribution to achieve dose control within the tumor area through iterative optimization of fixed parameters. This method sets fixed lattice parameters (such as spacing and diameter) through multimodal imaging and arranges lattices within the tumor area, which is applicable to optimization requirements within a certain range, but has limited adaptability under different tumor morphologies.

[0007] With the advancement of artificial intelligence technology, deep learning has been gradually applied to the prediction and optimization of radiotherapy dose distribution. For example, some literature has attempted to use convolutional neural networks (CNNs) and generative adversarial networks (GANs) to predict radiotherapy dose distribution and achieved remarkable results in actual cases. However, most of these studies focus on traditional radiotherapy with uniform dose distribution and are mostly applied to simple two-dimensional or three-dimensional models, lacking exploration of the application of non-uniform dose distribution in SFRT technology. In addition, deep learning models usually rely on a large amount of labeled data for training. In the field of SFRT, clinical data on non-uniform dose distribution is relatively scarce, making it difficult to obtain sufficient training data. Therefore, at present, the dose prediction model based on deep learning is not fully applicable to the lattice parameter optimization of SFRT.

[0008] In summary, the existing SFRT lattice arrangement schemes in the prior art mainly have the following problems: low efficiency of manual arrangement, difficulty in meeting personalized needs due to fixed parameter settings based on geometric algorithms, and the radiotherapy optimization schemes based on deep learning and reinforcement learning have not been fully applied to the dose distribution optimization of SFRT. Summary of the Invention

[0009] To solve the deficiencies in the prior art, the present invention provides a radiotherapy lattice parameter optimization method and system based on space segmentation technology, which can output optimal lattice parameters according to the radiotherapy image data of patients, provide a reliable basis for space segmentation radiotherapy, and do not rely on artificial experience, effectively improving efficiency.

[0010] To achieve the above objective, the specific solution adopted by the present invention is: A radiotherapy lattice parameter optimization method based on space segmentation technology, comprising the following steps:

[0011] Obtain the radiotherapy image data of the patient;

[0012] Calculate the tumor geometric features according to the radiotherapy image data and the preset geometric constraint conditions;

[0013] Input the tumor geometric features and the initial lattice parameters into the lattice parameter optimization model constructed based on the PPO (Proximal Policy Optimization) algorithm. The lattice parameter optimization model includes a policy network, a parameter optimization module, a reward calculation module, and a value network;

[0014] Use the policy network to iteratively generate the lattice optimization strategy according to the tumor geometric features and the initial lattice parameters;

[0015] Use the parameter optimization module to generate the optimized lattice parameters based on the lattice optimization strategy;

[0016] Use the reward calculation module to calculate the reward value for the optimized lattice parameters, and when the reward value meets the preset conditions, output the optimized lattice parameters as the optimal lattice parameters;

[0017] When the reward value does not meet the preset conditions, use the value network to evaluate the value of the policy network to obtain the state value, and the state value is used to guide the policy network to iteratively update the lattice optimization strategy.

[0018] Preferably, the tumor geometric features include the tumor volume, the shape factor, and the boundary features, where the shape factor is used to describe the tumor as a regular geometric body according to the radiotherapy image data.

[0019] Preferably, after calculating the tumor geometric features, determine the state space and action space of the lattice parameter optimization model based on the tumor geometric features.

[0020] Preferably, the state space of the lattice parameter optimization model is represented as:

[0021] S = {Volume, Shape Factor, Boundary Features, Current Parameters};

[0022] where Volume is the tumor volume, Shape Factor is the shape factor, Boundary Features is the boundary feature, and Current Parameters is the current lattice parameter state;

[0023] The action space of the lattice parameter optimization model is represented as:

[0024] A = [Spacing, Radius, Edge distance];

[0025] Among them, Spacing is the distance between lattice points; Radius is the action radius of lattice points; Edge distance is the distance between the lattice and the edge of the tumor target area.

[0026] Preferably, the specific method for reward calculation includes:

[0027] R = ω 1 ·R PVDR +ω 2 ·R ADR +ω 3 ·R Coverage -ω 4 ·R Penalty ;

[0028] Among them, R is the reward value, R PVDR is the dose peak-to-valley ratio in radiotherapy, R ADR is the absorbed dose ratio, ω 1 is the weight of R PVDR is the weight of ω 2 is the weight of R ADR and there is ω 1 +ω 2 = 1, R Coverage is the coverage rate, R Penalty is the penalty term.

[0029] Preferably, the calculation method of the dose peak-to-valley ratio R PVDR in radiotherapy is:

[0030]

[0031] Among them, D peak is the average high dose of the lattice target area, D valley is the average low dose of the area outside the lattice, and D peak and D valley are both determined by optimizing the lattice parameters;

[0032] The calculation method of the absorbed dose ratio R ADR is:

[0033]

[0034] Among them, D ablative is the total dose in the ablation area, D total is the total dose in the tumor area, and D ablative and D total are both determined by optimizing the lattice parameters.

[0035] Preferably, the calculation method of the coverage rate R Coverage is:

[0036]

[0037] Among them, Coverage Area is the tumor volume covered by the lattice layout and is determined by optimizing the lattice parameters. Tumor Area is the total tumor volume and is determined by the tumor geometric features.

[0038] Preferably, the penalty term R Penalty is calculated as follows:

[0039] R Penalty = R Penalty-Density + R Penalty-Overlap ;

[0040] Among them, R Penalty-Density is the density penalty factor, and R Penalty-Overlap is the overlap penalty factor.

[0041] A radiotherapy lattice parameter optimization system based on space segmentation technology includes:

[0042] An input module for obtaining the radiotherapy image data of a patient and preprocessing the radiotherapy image data;

[0043] A tumor recognition module for calculating the tumor geometric features based on the radiotherapy image data and preset geometric constraint conditions;

[0044] A lattice parameter optimization module for identifying the optimal lattice parameters by using a lattice parameter optimization model for the tumor geometric features.

[0045] Preferably, the system further includes a lattice arrangement and DICOM update module for generating lattice arrangement data according to the optimal lattice parameters and generating a data transfer file.

[0046] The radiotherapy lattice parameter optimization method of the present invention can generate optimal lattice parameters with better dose peak-to-valley ratio PVDR and absorbed dose ratio ADR, providing more targeted and effective parameter guidance for doctors to perform radiotherapy on patients. Especially in the treatment of large-volume or complex-shaped tumors, it can provide favorable assistance for the radiotherapy process. During the process of optimizing the lattice parameters, the present invention can adaptively adjust the layout parameters of the lattice according to the specific tumor characteristics of the patient, effectively meet the dose distribution requirements of irregular and complex-shaped tumors, overcome the limitations of fixed parameter settings in existing methods, and meet the specific requirements of different patients for dose distribution during treatment. Based on the deep reinforcement learning technology, the present invention can automatically generate optimal lattice parameters according to the radiotherapy image data of the patient, thereby reducing the dependence on the experience of operators and reducing the time and labor costs in the manual arrangement process, which is beneficial to improving the overall efficiency of the treatment process. Description of the Drawings

[0047] To more clearly illustrate the technical solutions in the embodiments of the present invention or in the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0048] Figure 1 is a flowchart of the parameter optimization method of the present invention;

[0049] Figure 2 is a schematic structural diagram of the lattice parameter optimization model;

[0050] Figure 3 is a schematic diagram of the change process of the lattice parameters;

[0051] Figure 4 is a block diagram of the structure of the lattice parameter optimization system of the present invention. Specific embodiments

[0052] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0053] Embodiment 1.

[0054] As Figures 1 to 4 shown, the radiotherapy lattice parameter optimization method based on the space segmentation technology includes S1 to S7.

[0055] S1. Obtain the radiotherapy image data of the patient. The radiotherapy image data mainly includes common medical images such as CT images and MRI images. Since radiotherapy is carried out after the patient is diagnosed with a tumor, and when the doctor diagnoses the tumor, the position of the tumor in the medical image will be determined. Therefore, when obtaining the radiotherapy image data of the patient, the radiotherapy image data can directly include the ROI information, that is, the region of interest, marked by the doctor during the diagnosis process, so as to facilitate subsequent determination of the position and size and other characteristics of the tumor based on the radiotherapy image data.

[0056] S2. Calculate the geometric features of the tumor based on the radiotherapy image data and the preset geometric constraint conditions. Among them, the geometric constraint conditions are used to simplify the irregular tumor into a regular geometric body, so as to facilitate subsequent lattice division. In an embodiment of the present invention, the tumor geometric features include tumor volume, shape factor, and boundary features, where the shape factor is used to describe the tumor as a regular geometric body according to the radiotherapy image data. More specifically, the shape factor describes the irregular tumor as a regular sphere, and then calculates the volume of the tumor. Where r is the radius of the tumor, which can be calculated by the method of graphic fitting according to the ROI information, and the boundary condition is set as the boundary length, that is, the perimeter of a cross-section passing through the center of the sphere.

[0057] On the other hand, after calculating the tumor geometric features, determine the state space and action space of the lattice parameter optimization model based on the tumor geometric features. The state space of the lattice parameter optimization model is expressed as: S = {Volume, Shape Factor, Boundary Features, Current Parameters}; Where, Volume is the tumor volume, Shape Factor is the shape factor, Boundary Features is the boundary feature, and Current Parameters is the current lattice parameter state.

[0058] The action space of the lattice parameter optimization model is expressed as: A = [Spacing, Radius, Edge distance]; Where, Spacing is the distance between lattice points; Radius is the action radius of lattice points, which is used to control the coverage range of the radiotherapy area; Edge distance is the distance between the lattice and the edge of the tumor target area.

[0059] The state space and the action space cooperate to perform spatial constraints on the lattice parameter optimization model, ensuring that the lattice parameter optimization model can output the optimal lattice parameters that match the tumor geometric features better, that is, match the patient's tumor state better, and provide more accurate and sufficient data references for doctors to perform radiotherapy on patients.

[0060] S3. Input the tumor geometric features and the initial lattice parameters into the lattice parameter optimization model constructed based on the PPO (Proximal Policy Optimization) algorithm for identification. The lattice parameter optimization model includes a policy network, a parameter optimization module, a reward calculation module, and a value network.

[0061] S4. Use the policy network to iteratively generate a lattice optimization strategy based on the tumor geometric features and the initial lattice parameters. As Figure 1 and 2 shown, the policy network is used to generate a lattice optimization strategy for optimizing the lattice parameters and the corresponding optimized lattice parameters. The policy network uses a multi-layer perceptron structure and consists of four main parts, including an input layer, two hidden layers, and an output layer. The input layer is used to receive the tumor geometric features and the initially input lattice parameters. Among them, the tumor geometric features include volume, shape, boundary, etc. These inputs provide the policy network with complete background information for subsequent decision-making. The hidden layer structure includes two fully connected layers with the number of nodes being 256 and 128 respectively, and the ReLU activation function is used to enhance the network's non-linear modeling ability. This design can capture the complex relationships in the lattice layout and efficiently process the optimization problem. The output layer is used to output the lattice optimization strategy outward, specifically including the lattice spacing, lattice radius, and lattice layout. These parameters are directly used to guide the parameter optimization module in the process of generating the optimized lattice parameters. To fully improve the performance of the policy network, each layer undergoes a non-linear transformation through the ReLU activation function to enhance the feature extraction ability of the policy network; in the output layer, the tanh function is used to map the result to a reasonable range. The specific formula is a = tanh(W·ReLU(H(s)) + b), where s is the state vector, including the tumor geometric features and the current parameter state, that is, the lattice optimization strategy, a is the output parameter adjustment suggestion, and W and b are the network weights and biases respectively. In actual applications, the policy network adopts a multi-scale optimization strategy. First, it determines the lattice distribution at a coarse-grained level, and then performs local fine optimization. At the same time, an experience replay mechanism is introduced to store historical optimization data for batch training to improve the learning effect. In addition, randomness can be introduced when the policy network performs parameter search to avoid falling into local optimal solutions. The policy network and the value network cooperate with each other through backpropagation, thereby continuously adjusting the lattice optimization strategy generated by the policy network. To avoid the situation where the policy network and the value network have long-term ineffective iterations, that is, the lattice optimization strategy has no obvious improvement during multiple iterations, the present invention also adopts an early stopping mechanism. When there is no obvious difference in the lattice optimization strategy during consecutive multiple iterations, the iteration is stopped, thereby avoiding wasting computing resources while ensuring the optimization effect.

[0062] S5. Use the parameter optimization module to generate optimized lattice parameters based on the lattice optimization strategy. As Figure 1As shown, the parameter optimization module is used to generate optimized lattice parameters according to the lattice optimization strategy. Specifically, the method for generating optimized lattice parameters includes three parts, namely, lattice spacing calculation, lattice radius optimization, and lattice layout adjustment, corresponding to the lattice optimization strategy. More specifically, each iteration of the policy network can generate a lattice optimization strategy. Correspondingly, each iteration of the policy network, the parameter optimization module generates an optimized lattice parameter. When generating the first optimized lattice parameter, the basis is the initial lattice parameter input manually, and the first optimized lattice parameter is generated by optimizing the initial lattice parameter through the lattice optimization strategy. In the subsequent process, the basis is the optimized lattice parameter generated in the previous time, and then the optimized lattice parameter is optimized and updated using the new lattice optimization strategy to realize the continuous iterative update of the optimized lattice parameter.

[0063] S6. Use the reward calculation module to calculate the reward for the optimized lattice parameter to obtain a reward value, and when the reward value meets the preset conditions, output the optimized lattice parameter as the optimal lattice parameter. The reward calculation module is used to perform real-time evaluation and feedback on the optimized lattice parameter through physical model calculation and index evaluation. Specifically, the specific method of reward calculation includes: R = ω 1 ·R PVDR +ω 2 ·R ADR +ω 3 ·R Coverage -ω 4 ·R Penalty ; where R is the reward value, R PVDR is the dose peak-to-valley ratio in radiotherapy, R ADR is the absorbed dose ratio, ω 1 is the weight of R PVDR ω 2 is the weight of R ADR and ω 1 +ω 2 = 1, R Coverage is the coverage rate, R Penalty is the penalty term. More specifically, the dose peak-to-valley ratio and the absorbed dose ratio are conventional parameters in this field. The dose peak-to-valley ratio is a core index for evaluating the treatment effect, reflecting the contrast between high and low dose regions. The absorbed dose ratio is used to evaluate the effective treatment range and calculate the proportion of the high dose region in the total treatment volume.

[0064] The calculation methods of each parameter are as follows.

[0065] The calculation method of the dose peak-to-valley ratio R PVDR in radiotherapy is: where D peakis the average high dose of the lattice target area, D valley is the average low dose of the area outside the lattice, and D peak and D valley are both determined by optimizing the lattice parameters.

[0066] The calculation method of the absorbed dose ratio R ADR is as follows: where, D ablative is the total dose within the ablation area, D total is the total dose of the tumor area, and D ablative and D total are both determined by optimizing the lattice parameters.

[0067] The calculation method of the coverage rate R Coverage is as follows: where, Coverage Area is the tumor volume covered by the lattice layout and is determined by optimizing the lattice parameters, Tumor Area is the total tumor volume and is determined by the geometric characteristics of the tumor.

[0068] The penalty term mainly targets unreasonable lattice distribution characteristics, including two aspects: density penalty and overlap penalty. The density penalty constrains the lattice density to be within a reasonable range in the form of a quadratic function. The overlap penalty adopts a soft constraint method, allowing a small amount of overlap but imposing corresponding penalties, and this design provides a larger optimization space. The calculation method of the penalty term R Penalty is as follows: R Penalty = R Penalty-Density + R Penalty-Overlap ; where, R Penalty-Density is the density penalty factor, used to prevent the lattice layout from being too dense or too sparse, R Penalty-Overlap is the overlap penalty factor, used to prevent uneven dose distribution caused by excessive overlap between lattices.

[0069] Furthermore, where, is the average covered volume of each lattice, Optimal Density is the preset optimal lattice density. R Penalty-Overlap = ∈·Overlap Area, where, ∈ is the constraint factor, Overlap Area is the area of the overlap region.

[0070] S7. When the reward value does not meet the preset conditions, the value network is used to evaluate the policy network to obtain the state value, and the state value is used to guide the policy network to iteratively update the lattice optimization strategy. As Figure 1 and2 As shown, the value network is used to evaluate the optimized state of the current lattice, that is, to evaluate the output result of the policy network. The input layer of the value network is the same as that of the policy network, ensuring that the value evaluation is based on the same information features, thereby ensuring the accuracy of the evaluated state value. The hidden layer is a fully connected layer containing 128 nodes, and the ReLU activation function is also used. The output layer has only one node, which is used to generate the current state value. The state value reflects the quality of the optimized lattice parameters and can provide feedback on the optimization direction for the policy network. The core optimization objective of the value network can be expressed as where r t (θ) represents the probability ratio of the old and new policies, and ∈ is the clipping parameter used to limit the policy update amplitude. In this embodiment, ∈ is set to 0.2.

[0071] In the present invention, the reward calculation module takes the dose peak-to-valley ratio, absorption dose ratio, and tumor coverage rate as the main optimization objectives, and combines the penalty terms of lattice density and overlap to achieve comprehensive control of the dose distribution. In specific implementation, the system adopts an adaptive weight adjustment mechanism. According to the change trend of each index during the optimization process, the weight coefficient is dynamically adjusted. When a certain index deviates significantly from the expectation, its weight is increased accordingly; when each index tends to be balanced, the weight is kept stable. This adaptive mechanism ensures the stability and reliability of the optimization process.

[0072] The optimization process of the present invention adopts an iterative design. Each round of optimization includes four main steps: state observation, action selection, execution, and evaluation. The optimization controller coordinates each link and determines the iteration process according to the convergence condition. The core iteration formula is: where θ represents the optimization parameter, α is the learning rate, and L(θ) is the loss function. An adaptive step size mechanism is introduced during the optimization process, which can dynamically adjust the learning rate according to the optimization effect. When the improvement of the optimization effect is not obvious for several consecutive rounds, the early stopping mechanism is triggered to avoid excessive calculation.

[0073] Finally, after obtaining the optimal lattice parameters, the dose during the radiotherapy process can be calculated as a reference for the real radiotherapy process. The calculation process takes into account the beam physical characteristics and tissue inhomogeneity, and the core calculation model is: where K is the dose deposition kernel, ρ is the tissue density distribution, and Φ is the incident photon flux. This method can accurately reflect the influence of the lattice layout on the dose distribution.

[0074] In addition, a lookup table can be established based on the dose distribution data of common configurations in the actual radiotherapy process, which can improve the calculation efficiency while ensuring the calculation accuracy, and can also provide a reliable basis for the training of the lattice optimization model.

[0075] The radiotherapy lattice parameter optimization method of the present invention can generate optimal lattice parameters with a better peak-to-valley dose ratio (PVDR) and absorbed dose ratio (ADR), providing more targeted and effective parameter guidance for doctors to perform radiotherapy on patients. Especially in the treatment of large-volume or complex-shaped tumors, it can provide favorable assistance for the radiotherapy process.

[0076] The present invention can take multi-modal image data including CT images and MRI images as input, and generate tumor geometric features through RTSTRUCT files in DICOM format. It has a wide range of applications, and this data-driven precise layout technology can adapt to tumors of different sizes, shapes, and positions, ensuring that the lattice layout strictly covers the tumor area, with higher accuracy and practicality.

[0077] In the process of optimizing the lattice parameters, the present invention can adaptively adjust the layout parameters of the lattice according to the specific tumor characteristics of the patient, effectively meeting the dose distribution requirements of irregular and complex-shaped tumors, overcoming the limitations of fixed parameter settings in the existing methods, and satisfying the specific dose distribution requirements of different patients during treatment. Based on the deep reinforcement learning technology, the present invention can automatically generate optimal lattice parameters according to the radiotherapy image data of the patient, thereby reducing the dependence on the experience of the operator, and reducing the time and labor costs in the manual layout process, which is beneficial to improving the overall efficiency of the treatment process.

[0078] The present invention can provide support for the wide application of the SFRT technology in clinics, promote its development in the field of radiation oncology, and improve the success rate of clinical treatment and the quality of life of patients.

[0079] In summary, by introducing the deep reinforcement learning technology, the present invention aims to solve the limitations existing in the prior art, improve the accuracy and personalization of spatial fractionated radiotherapy, and ultimately achieve a more efficient and safer radiotherapy effect.

[0080] Embodiment 2.

[0081] This embodiment is further optimized on the basis of Embodiment 1. Specifically, the following steps are further included in Embodiment 2.

[0082] After outputting the optimal lattice parameters, empirical data is generated based on the current tumor geometric features, initial lattice parameters, and optimal lattice parameters, and added to the empirical pool. The empirical data contains complete information on the optimization process, including the parameter change path, phased evaluation results, and final effect data. The empirical pool is managed using a hierarchical indexing structure to support fast retrieval based on tumor features. When processing a new case, the system selects similar cases from the empirical pool through feature matching, extracts their optimization experiences to guide the current parameter optimization process, significantly shortening the optimization time and improving the optimization effect.

[0083] Furthermore, the method for generating experience includes: First, establish a first mapping relationship between tumor geometric features and optimal lattice parameters; then, establish an extended relationship between tumor-derived features and optimal lattice parameters, where tumor-derived features include centroid position, principal axis direction, spatial anisotropy, tissue density distribution characteristics, spatial distribution of surrounding organs, and surrounding organ risk data, etc.; after that, establish a second mapping relationship between initial lattice parameters and optimal lattice parameters; subsequently, combine the tumor geometric features, initial lattice parameters, and optimal lattice parameters into experience based on the first mapping relationship, extended relationship, and second mapping relationship, and add it to the empirical pool; then, cluster the experience based on the type of tumor, the staging result of the tumor, and the geometric features of the tumor, so as to establish a connection between the experience and a pre-set empirical index table; finally, optimize the way the value network evaluates the state value of the policy network based on the empirical pool. It should be noted that data such as tumor-derived features, the type of tumor, and the staging result of the tumor can all be obtained from the patient's diagnosis results, which will not be elaborated here.

[0084] When optimizing the radiotherapy lattice parameters for other patients in the future, first, the experience closest to the actual situation of the patient can be selected from the empirical pool based on the empirical index table, and then the first mapping relationship in this experience can be used as a guiding parameter and input into the policy network, thus accelerating the process of the policy network generating lattice optimization strategies; on the other hand, correct the initial lattice parameters based on the second mapping relationship in this experience. Specifically, a feasible region range can be generated based on the second mapping relationship and the optimal lattice parameters, and when the initial lattice parameters exceed the feasible region range, correct them, thus avoiding the initial lattice parameters from deviating too much from the actual situation and being able to further accelerate the overall process of lattice parameter optimization.

[0085] Embodiment III.

[0086] This embodiment is further improved based on Embodiment II, specifically as follows.

[0087] Although the radiotherapy lattice parameters can be optimized more quickly based on Embodiment 2, each optimization still requires multiple rounds of iteration. In practical applications, a new case often requires 100 to 200 iterations, and the overall process still consumes a large amount of time. To fully improve efficiency and shorten the optimization time, in Embodiment 3, the present invention constructs a neural network model based on the transfer learning method, which can also quickly optimize the radiotherapy lattice parameters for new cases.

[0088] First, a feature dimensionality reduction and normalization method based on an autoencoder is used to convert experience into the form of feature vectors, ensuring the expression efficiency and computational efficiency of the feature vectors. The autoencoder compresses high-dimensional features into a low-dimensional latent space through a multi-layer neural network structure during the encoding stage, and then reconstructs the original features during the decoding stage. This process not only achieves dimensionality reduction of the features but also retains the key correlations between the features. The normalization process ensures the scale consistency of features in different dimensions, effectively avoiding the impact of dimensional differences on the subsequent optimization process. This multi-level and multi-dimensional feature representation method provides a reliable data basis for subsequent parameter optimization, effectively supporting knowledge transfer and rapid optimization.

[0089] Second, after each radiotherapy lattice parameter optimization process is completed, that is, after the optimal lattice parameters are generated, in addition to generating experience, all current optimized lattice parameters, corresponding reward values, and state evaluation results are recorded. Special attention is paid to the impact of parameter changes on the dose distribution, including the change trends of key indicators such as the peak-to-valley dose ratio (PVDR) and target coverage. In addition, the decision-making basis of the policy network and the state value evaluated by the value network are also recorded.

[0090] Third, based on the above data, establish a typical case library. For different types of tumor characteristics and treatment requirements, screen and store representative typical cases. Each typical case contains a complete optimization history, including all process data from the initial lattice parameter setting to the output of the optimal lattice parameter. The typical case library is classified and managed according to the morphological characteristics, location features, and optimization difficulty of tumors. In the typical case library, to improve the data access efficiency and storage space utilization rate, compression storage technology can be used to reduce the data occupied space while maintaining a fast retrieval speed. For the intermediate states during the optimization process, adopt a key-frame storage strategy, only retain the state points with significant changes, and restore the complete optimization trajectory through interpolation methods. In addition, a data indexing mechanism also needs to be established. To be able to efficiently retrieve in the typical case library, use Locality-Sensitive Hashing (LSH) technology to construct an efficient case retrieval mechanism. The LSH algorithm maps similar cases to the same or adjacent hash buckets by designing a family of feature-sensitive hash functions. This hash mapping maintains the local similarity in the feature space, enabling quick location of typical cases similar to the features of new cases. To improve the accuracy and robustness of the retrieval, adopt a multi-table hashing strategy, and perform parallel retrieval through multiple independent hash tables, effectively reducing the risk of missed detection caused by hash conflicts.

[0091] Fourth, adopt a similarity calculation method based on multi-feature weighting to facilitate quick querying of the typical case library. Specifically, when calculating the similarity, comprehensively consider multiple dimensions such as tumor geometric features, centroid position, and dose requirements, and accurately evaluate the similarity between the new case and the cases in the typical case library through measurement methods such as weighted Euclidean distance or cosine similarity, so as to quickly query similar cases. To further improve the retrieval efficiency, establish a three-dimensional index structure in the typical case library. At the top layer, conduct coarse-grained classification according to the basic type and approximate location of the tumor; the middle layer is subdivided based on more detailed features, such as the main axis direction, spatial anisotropy, and tissue density distribution characteristics, etc.; the bottom layer contains specific cases and their complete feature information. This hierarchical structure enables the adoption of a top-down retrieval strategy, quickly narrowing the search range and improving the retrieval efficiency. At the same time, support the dynamic update of the index structure, which can automatically adjust and optimize the index structure as new cases are added, maintaining the stability of the retrieval performance.

[0092] Fifth, a prediction value of the initial lattice parameters is generated by the parameter interpolation method of the nearest neighbor cases. After retrieving similar cases, instead of simply directly adopting the initial lattice parameters or the optimal lattice parameters of a single most similar case, multiple cases with higher similarity are selected for parameter interpolation. The distance-weighted strategy is adopted in the interpolation process, that is, the higher the similarity of the case, the greater the weight in parameter prediction. At the same time, the time decay factor is considered to make the parameters of the newer cases have higher reference value, so as to ensure that the prediction results can reflect the latest optimization experience. Further, considering that even for similar tumors, there may still be differences in local geometric features, a parameter adjustment mechanism based on local features is designed. By analyzing the differences between the new case and the reference case in the local area, including features such as curvature change and boundary complexity, the initial parameters obtained by interpolation are adjusted specifically. This adjustment not only considers the differences in geometric features, but also takes into account the position constraints of surrounding organs and the dose protection requirements, ensuring that the adjusted parameters are more in line with the actual treatment needs.

[0093] Based on the above multiple mechanisms, in order to improve the stability and reliability of parameter prediction, an ensemble learning method is used to integrate the results of multiple prediction models. The specific implementation includes: training multiple base predictors based on different feature subsets, and each predictor focuses on a specific feature combination, such as a tumor morphology feature predictor, a position feature predictor, and a boundary feature predictor, etc. Through the Bagging strategy of random forest and the Boosting strategy of gradient boosting tree, the outputs of these predictors are weighted and combined to generate the final parameter prediction result. A prediction uncertainty evaluation mechanism is also introduced. When the variance of the prediction result exceeds the preset threshold, the system automatically increases the exploration range and reduces the local search intensity to ensure the reliability of the optimization result.

[0094] More specifically, in the pre-training stage of the policy network, the historical optimization data is fully utilized to construct the initial policy model. By learning the optimization trajectories of a large number of successful cases, the policy network can master the basic laws of parameter adjustment. The hierarchical training strategy is adopted in the pre-training process. First, the general parameter adjustment mode is learned, and then specialized training is carried out for specific types of tumors to form a hierarchical optimization policy library.

[0095] In order to achieve knowledge transfer between different tumor features, a feature adaptation layer is designed. This layer matches the feature distributions of typical cases and new cases through feature mapping and distribution alignment techniques, reducing the impact of inter-domain differences on policy transfer. At the same time, a progressive fine-tuning mechanism is adopted to gradually adapt to the features of new cases while maintaining the basic performance of the model, ensuring the stability and effectiveness of the optimization strategy.

[0096] In terms of value assessment migration, a benchmark model for value assessment is constructed by analyzing the correlation between parameter adjustment and treatment effect in typical cases. Not only specific assessment values are migrated, but also assessment criteria and weight assignment schemes are included to ensure the comparability and consistency of assessment results. The reliability of value assessment is guaranteed by a confidence assessment mechanism. The uncertainty of each assessment result is calculated, and the confidence level is determined based on factors such as feature similarity and sample size. For assessment results with low confidence, the proportion of exploratory search is increased to avoid over-reliance on uncertain experience. The migration weight is dynamically adjusted according to the confidence of assessment and the actual optimization effect to achieve adaptive control of experience migration.

[0097] The search space guiding mechanism significantly improves the parameter optimization efficiency by using historical optimization experience. By analyzing the distribution law of parameters in typical cases, the approximate range of effective parameters is determined, thus narrowing the search space. This reduction is not a simple range limitation, but a dynamic adjustment based on the characteristics of the current case to ensure the rationality of the search space. An adaptive parameter sampling strategy is adopted, and the sampling process is guided by the probability density of parameter distribution in historical experience. In terms of the balance between exploration and exploitation, an improved UCB (Upper Confidence Bound) algorithm is used to dynamically adjust the ratio of exploring new parameter combinations and exploiting known excellent parameters to avoid falling into local optimal solutions.

[0098] The convergence acceleration mechanism improves the optimization efficiency through various strategies. By using historical optimization trajectory data, the optimal direction of parameter adjustment is predicted to reduce ineffective parameter exploration. During the optimization process, the learning rate is dynamically adjusted according to the parameter convergence situation. A larger learning rate is adopted at the initial stage of optimization to quickly approach the optimal solution region, and the learning rate is reduced when approaching convergence for fine-tuning. To avoid over-iteration, an intelligent early stopping mechanism is designed. By monitoring the change trend of optimization metrics, when the improvement amplitude of continuous multiple rounds of optimization is lower than the preset threshold, the optimization process will be automatically terminated. At the same time, the optimal parameter combination during the optimization process is recorded, and when the performance drops, it can be rolled back to the previous optimal state to ensure the reliability of the optimization result.

[0099] As Figure 4 shown, the present invention further provides a radiotherapy lattice parameter optimization system based on space segmentation technology, including an input module, a tumor recognition module, a lattice parameter optimization module, a lattice layout and DICOM update module, and an output module.

[0100] An input module for acquiring radiotherapy image data of a patient and preprocessing the radiotherapy image data. Specifically, the input module includes a file loading unit and a parameter input unit. The file loading unit is used to read an RTSTRUCT file in DICOM format, and the parameter input unit is used to obtain initial parameters provided by the user, such as lattice spacing and radius. The lattice parameter optimization model can be continuously optimized based on these input initial parameters to obtain the optimal lattice parameters.

[0101] A tumor recognition module for calculating tumor geometric features based on the radiotherapy image data and preset geometric constraint conditions. Specifically, the tumor recognition module includes a DICOM parsing unit and a GTV contour extraction unit. The DICOM parsing unit is used to parse the read RTSTRUCT file in DICOM format, and the GTV contour extraction unit is used to obtain the tumor geometric features by contour extraction according to the parsing result of the DICOM parsing unit.

[0102] A lattice parameter optimization module for identifying the optimal lattice parameters by using the lattice parameter optimization model for the tumor geometric features. Specifically, the lattice parameter optimization module includes an agent initialization unit, a lattice arrangement unit, a reward calculation unit, and a policy update unit. The agent initialization unit is used to initialize the PPO algorithm. The lattice arrangement unit is used to arrange the lattice according to the initial parameters and continuously optimize the lattice arrangement method by using the PPO algorithm to obtain the original lattice parameters. The reward calculation unit is used to determine whether the original lattice parameters meet the PVDR and ADR conditions to determine the optimal lattice parameters. The policy update unit is used to update and optimize the policy network and the value network.

[0103] A lattice arrangement and DICOM update module for generating lattice arrangement data according to the optimal lattice parameters and generating a data transfer file. Specifically, the lattice arrangement and DICOM update module includes a lattice arrangement unit and a DICOM update unit. The lattice arrangement unit is used to arrange the lattice according to the optimal lattice parameters to generate optimal lattice point distribution data. The DICOM update unit is used to write the optimal lattice point distribution data into the RTSTRUCT file in DICOM format.

[0104] An output module including a file saving unit and a result evaluation unit. The file saving unit is used to save the generated RTSTRUCT file to a storage medium, and the result evaluation unit is used to evaluate the optimal lattice point distribution data and output an evaluation report.

[0105] When this system is used to implement the above-mentioned Embodiment 2, it further includes a feature correlation analysis unit, a case retrieval unit, a policy optimization unit, and a parameter estimation unit.

[0106] The feature correlation analysis unit is used to establish the mapping relationship between tumor features and optimization parameters. The feature correlation analysis unit constructs a complete feature vector by extracting basic feature information such as the volume, shape, and boundary of the tumor, and combining environmental features such as the distribution of surrounding organs. The corresponding relationship between this feature information and the optimal lattice parameters is systematically stored and managed, providing a data basis for subsequent optimization processes.

[0107] The case retrieval unit is used to identify cases with similar features from the experience pool. By establishing a multi-level index structure, the system can efficiently calculate the feature similarity degree between cases and achieve fast retrieval and matching. This similarity-based retrieval mechanism can effectively narrow the parameter search space and provide valuable reference information for the optimization process.

[0108] The strategy optimization unit is used to dynamically adjust the optimization strategy according to historical optimization experience. This unit optimizes the training process of the strategy network by analyzing historical data and adjusts the reference standard for value evaluation. This strategy update mechanism can enable the optimization process to better adapt to different types of tumor features and improve the optimization efficiency.

[0109] The parameter estimation unit is responsible for generating the initial optimization parameters. This unit predicts the feasible range of parameters by analyzing historical data, evaluates the rationality of parameter combinations, and can dynamically adjust the parameter generation strategy. This parameter estimation mechanism can significantly improve the quality of the initial parameters and accelerate the optimization convergence speed.

[0110] Those skilled in the art can clearly understand that for the convenience and conciseness of description, the specific working processes of the systems and devices described above can refer to the corresponding processes in the foregoing method embodiments, which will not be elaborated herein. In several embodiments provided in this disclosure, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of the modules is only a logical function division, and there may be other division methods in actual implementation. For another example, multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections to each other can be through some communication interfaces. The indirect couplings or communication connections of the devices or modules can be in electrical, mechanical, or other forms.

[0111] The modules described as separate components may or may not be physically separated. The components shown as modules may or may not be physical modules, that is, they can be located in one place, or distributed to multiple network modules. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0112] The present invention also provides a computer device, including a memory and a processor.

[0113] The memory is used for storing a computer program.

[0114] The processor is configured to read and execute the computer program to implement the above-mentioned radiotherapy lattice parameter optimization method based on the space segmentation technology.

[0115] Finally, the present invention provides a storage medium storing a computer program, which when executed implements the above-mentioned radiotherapy lattice parameter optimization method based on the space segmentation technology.

[0116] Based on this storage medium, the technical solution of the present disclosure, in essence or the part that contributes to the prior art or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present disclosure. The aforementioned storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM), random access memories (RAM), magnetic disks, or optical discs that can store program codes.

[0117] The various embodiments in this specification are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. For the same or similar parts among the various embodiments, reference can be made to each other.

[0118] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present invention. Various modifications to these embodiments will be obvious to those skilled in the art. The general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the broadest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for optimizing radiotherapy lattice parameters based on space segmentation technology, characterized in that: The steps include: Obtain radiotherapy imaging data of patients; Calculate tumor geometry based on radiotherapy imaging data and pre-set geometric constraints; The tumor geometry and initial lattice parameters are input into a lattice parameter optimization model based on the Proximal Policy Optimization (PPO) algorithm. The lattice parameter optimization model includes a policy network, a parameter optimization module, a reward calculation module, and a value network. The strategy network is used to iteratively generate a lattice optimization strategy based on the tumor geometry characteristics and initial lattice parameters; Generate optimized lattice parameters based on lattice optimization strategy using parameter optimization module; The reward calculation module is used to calculate the reward for the optimized lattice parameters to obtain a reward value, and when the reward value meets the preset conditions, the optimized lattice parameters are output as the optimal lattice parameters; When the reward value does not meet the preset conditions, the value network is used to evaluate the value of the policy network to obtain the state value, which is used to guide the policy network to iteratively update the lattice optimization strategy.

2. The method for optimizing radiotherapy lattice parameters based on space segmentation technology according to claim 1, characterized in that: Tumor geometric features include tumor volume, shape factor, and boundary features, where the shape factor is used to describe the tumor as a regular geometric body based on radiotherapy imaging data.

3. The method for optimizing radiotherapy lattice parameters based on space segmentation technology according to claim 1, characterized in that: After calculating the geometric features of the tumor, the state space and action space of the lattice parameter optimization model are determined based on the geometric features of the tumor.

4. The method for optimizing radiotherapy lattice parameters based on space segmentation technology according to claim 3, characterized in that: The state space representation of the lattice parameter optimization model is: S={Volume,Shape Factor,Boundary Features,Current Parameters}; Among them, Volume is the tumor volume, Shape Factor is the shape factor, Boundary Features is the boundary features, and Current Parameters is the current lattice parameter status; The action space of the lattice parameter optimization model is expressed as: A=[Spacing, Radius, Edge distance]; Among them, Spacing is the distance between lattice points; Radius is the effective radius of the lattice point, and Edge distance is the distance between the lattice and the edge of the tumor target area.

5. The method for optimizing radiotherapy lattice parameters based on space segmentation technology according to claim 1, characterized in that: The specific method of reward calculation includes: R=ω1·R PVDR +ω2·R ADR +ω3·R Coverage -ω4·R Penalty ; Among them, R is the total reward value, R PVDR is the peak-to-valley ratio of dose in radiotherapy, R ADR is the absorbed dose ratio, ω1 is R PVDR The weight of R ADR weight, and ω1+ω2=1, R Coverage is the coverage rate, R Penalty For penalty items.

6. The method for optimizing radiotherapy lattice parameters based on space segmentation technology according to claim 5, characterized in that: Peak-to-valley ratio of dose in radiotherapy PVDR The calculation method is: Among them, D peak is the average high dose of the lattice target area, D valley is the average low dose in the region outside the lattice, and D peak and D valley All are determined by optimizing the lattice parameters; Absorbed dose ratio R ADR The calculation method is: Among them, D ablative is the total dose in the ablation area, D total is the total dose to the tumor area, and D ablative and D total All are determined by optimizing the lattice parameters.

7. The method for optimizing radiotherapy lattice parameters based on space segmentation technology according to claim 5, characterized in that: Coverage R Coverage The calculation method is: Among them, Coverage Area is the covered tumor volume of the lattice layout and is determined by the optimized lattice parameters, and TumorArea is the total tumor volume and is determined by the tumor geometric characteristics.

8. The method for optimizing radiotherapy lattice parameters based on space segmentation technology according to claim 5, characterized in that: Penalty term R Penalty The calculation method is: R Penalty =R Penalty-Density +R Penalty-Overlap ; Among them, R Penalty-Density is the density penalty factor, R Penalty-Overlap is the overlap penalty factor.

9. A radiotherapy lattice parameter optimization system based on space segmentation technology, characterized in that: include: An input module, used for acquiring radiotherapy image data of a patient and preprocessing the radiotherapy image data; A tumor recognition module, which is used to calculate the geometric features of the tumor based on the radiotherapy image data and the preset geometric constraints; The lattice parameter optimization module is used to identify the geometric features of the tumor using the lattice parameter optimization model to obtain the optimal lattice parameters.

10. The radiotherapy lattice parameter optimization system based on space segmentation technology according to claim 9, characterized in that: The system further comprises a lattice arrangement and DICOM update module for generating lattice arrangement data according to optimal lattice parameters and generating a data transfer file.

Images