A high-current proton accelerator system for boron neutron capture therapy system

Abstract

The present application discloses a high-current proton accelerator system for a boron neutron capture therapy system, and relates to the technical field of electrical digital data processing. The system comprises an imaging unit, a processing unit and a control unit. The imaging unit generates three-dimensional image data of a target tissue by a multimodal imaging method. The processing unit identifies and divides the core area, edge area and transition area of ​​a tumor based on edge detection and density gradient algorithm, and extracts depth information, density information and morphological feature information. The control unit generates energy step information based on the extracted features using a neural network model, and controls the energy output of a proton accelerator through a radio frequency system, so as to realize personalized treatment of different tumor areas. The system further optimizes the energy distribution by dynamically adjusting the fusion weights of the features of each area, and has the multimodal image processing and dynamic calculation capabilities for complex tumor morphology and feature information, thereby improving the effectiveness and safety of treatment.

Description

Technical Field

[0001] The present application relates to the technical field of electrical digital data processing, and in particular to a high-current proton accelerator system for a boron neutron capture therapy system. Background Art

[0002] High-current proton accelerator systems are widely used in the fields of data processing and image analysis, especially in medical imaging. However, current high-current proton accelerator systems still have deficiencies in processing complex medical image data and optimizing system parameters, especially in image preprocessing, feature extraction, and multimodal data fusion. This makes it difficult for the system to effectively deal with complex tumor morphology and affects energy allocation. Existing systems rely on static models for image processing, making it difficult to optimize energy allocation strategies in real time, especially when dealing with different tumor areas, and lack the ability to adaptively adjust.

[0003] For example, the patent with publication number CN117875063A discloses a method for obtaining biological dose for boron neutron capture therapy, which generates a dynamic biological dose model by obtaining a three-dimensional boron-10 concentration distribution model of an organism and combining it with the physical dose model of neutrons and photons. The biological dose of the organism is obtained based on the verified dynamic biological dose model. Through this method, the accuracy of the dynamic biological dose model can be ensured, thereby ensuring the accuracy of the biological dose obtained based on the model.

[0004] Although this method can adjust the treatment plan in real time, it mainly relies on the calculation of the biological dose model and relies less on image data processing and real-time optimization. Its disadvantage is that it lacks multimodal image processing and dynamic calculation capabilities for complex tumor morphology and characteristic information, and cannot effectively respond to the personalized energy requirements of different tumor areas. Summary of the invention

[0005] In view of the shortcomings of the prior art, the present application discloses a high-current proton accelerator system for a boron neutron capture therapy system, comprising:

[0006] An imaging unit, which generates three-dimensional image data of a target tissue based on a multimodal imaging method;

[0007] A processing unit is used to determine, based on the edge detection method, a first lesion area in the target tissue for the three-dimensional image data of the target tissue; and to determine, based on a density gradient algorithm, the area range and area information of a core area, an edge area, and a transition area in the first lesion area; wherein the area information includes: depth information, density information, and morphological feature information;

[0008] A control unit determines an energy step switching point based on the area range; inputs the area information into a neural network model to generate energy step information; and controls the energy output of a proton accelerator in the first lesion area based on the energy step information and the energy step switching point.

[0009] As an optional implementation, the multimodal imaging method includes: magnetic resonance imaging, positron emission tomography and computer tomography; the imaging unit is also used for:

[0010] Acquiring treatment information of the target tissue; wherein the treatment information includes: treatment requirement information, equipment condition information and user status information;

[0011] Based on the treatment information of the target tissue, three-dimensional image data of the target tissue is generated using at least one of the multimodal imaging methods.

[0012] As an optional implementation manner, the determining of the region range and region information of the core region, the edge region, and the transition region in the first lesion region includes:

[0013] Calculating the density gradient of each voxel in the target tissue to generate a density gradient field; wherein the density gradient is the density change rate between adjacent voxels;

[0014] In the density gradient field, determining a region where the density gradient is higher than a first gradient threshold as a candidate region;

[0015] In the candidate area, the area where the density gradient value is higher than the second gradient threshold and the density change rate is continuously distributed within the first area is taken as the core area;

[0016] In the candidate area, a region having a target gradient change feature is determined as an edge region;

[0017] The non-core area and non-edge area in the candidate area is taken as the transition area.

[0018] As an optional implementation manner, the determining the energy step switching point includes: determining the regional boundaries of the core area, the edge area, and the transition area based on the regional ranges of the core area, the edge area, and the transition area;

[0019] The region boundary between the core region and the edge region is set as a first switching point;

[0020] The boundary between the edge area and the transition area is set as a second switching point.

[0021] As an optional implementation manner, the morphological feature information of the core area includes: a first morphological feature; the morphological feature information of the edge area includes: a second morphological feature; the morphological feature information of the transition area includes: a third morphological feature;

[0022] The first morphological characteristics include: density concentration, volume uniformity and morphological symmetry;

[0023] The second morphological features include: edge irregularity, multi-scale shape spectrum, and edge thickness variation;

[0024] The third morphological feature includes: density gradient smoothness, gradient direction consistency and morphological transition characteristics.

[0025] As an optional implementation, the energy step information includes: edge area energy level, core area energy peak value and transition area energy attenuation curve; the neural network model includes a first input channel, a second input channel and a third input channel;

[0026] Wherein, the first input channel is used to input the first morphological feature; the second input channel is used to input the second morphological feature; the third input channel is used to input the third morphological feature;

[0027] The output nodes of the neural network model include: edge area energy level, core area energy peak value and transition area energy attenuation curve.

[0028] As an optional implementation, the neural network model further includes: a cross-fusion layer; the cross-fusion layer is used to fuse the first morphological feature, the second morphological feature and the third morphological feature.

[0029] As an optional implementation manner, the feature fusion mode of the cross fusion layer is weighted fusion; the mathematical expression of the weighted fusion is:

[0030] ;

[0031] in, is the fusion feature vector, The first morphological feature is The second morphological feature is The third morphological feature is , , are the fusion weights of the core area, the transition area and the edge area respectively.

[0032] As an optional implementation, the cross-fusion layer further includes: a feature driving module; the feature driving module is used to adjust the fusion weights of the core area, the transition area and the edge area; the adjustment formula is as follows:

[0033] ;

[0034] in, , , are the adjustment coefficients of the core area, the transition area and the edge area respectively, is the adjustment function of the core area based on relative density, is the adjustment function of the edge region based on morphological complexity, It is the adjustment function of the transition zone based on the depth change.

[0035] Compared with the prior art, the beneficial effects of the present invention are: by combining multimodal imaging, edge detection and density gradient algorithms, the core area, edge area and transition area of ​​complex tumors can be accurately identified, the image processing capability of the system is optimized, and the energy allocation strategy for each area is dynamically generated through a neural network model. The system integrates a cross-fusion layer and a feature-driven module, which can dynamically fuse and adjust the weights of multimodal image data, thereby enhancing the adaptability to the characteristics of different regions and significantly improving the accuracy of image analysis. Through the optimized image processing process, the system shows higher efficiency and flexibility when processing large-scale medical data, provides more accurate input information for subsequent energy allocation and processing, and further improves the robustness of the overall system. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 A schematic diagram of the structure of a high-current proton accelerator system for a boron neutron capture therapy system provided in this application;

[0037] Figure 2 This is an example diagram of the first lesion area provided in this application;

[0038] Figure 3 A flow chart of the method for determining the area range of the core area, edge area and transition area in the first lesion area provided in the present application. DETAILED DESCRIPTION

[0039] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0040] See also Figure 1 , Figure 1A schematic diagram of the structure of a high-current proton accelerator system for a boron neutron capture therapy system provided in this application, the system comprising:

[0041] An imaging unit 10 generates three-dimensional image data of a target tissue based on a multimodal imaging method;

[0042] The processing unit 20 is used to determine, based on the edge detection method, a first lesion area in the target tissue for the three-dimensional image data of the target tissue; and to determine, based on the density gradient algorithm, the area range and area information of the core area, the edge area and the transition area in the first lesion area; wherein the area information includes: depth information, density information and morphological feature information;

[0043] The control unit 30 determines an energy step switching point based on the area range; inputs the area information into a neural network model to generate energy step information; and controls the energy output of the proton accelerator in the first lesion area based on the energy step information and the energy step switching point.

[0044] This application uses multimodal imaging to generate three-dimensional image data of target tissues to capture detailed structural information of the tumor and its surrounding tissues. Through edge detection and density gradient algorithms, the system can identify and distinguish the core area, edge area and transition area of ​​the tumor, and extract depth information, density information and morphological feature information. This information is the basis for subsequent neutron beam energy distribution calculations.

[0045] For imaging unit 10:

[0046] In a specific implementation, the imaging unit 10 generates three-dimensional image data of the target tissue using a multimodal imaging method. Image data of different modalities are registered and fused to generate a comprehensive three-dimensional model. The registration process aligns images of different modalities through an algorithm to ensure their spatial consistency. The fused three-dimensional model not only contains the anatomical structure of the tumor, but also its metabolic activity and density information, providing comprehensive and accurate basic data for subsequent stratified analysis and treatment planning. Through this multimodal imaging process, the system can fully capture and characterize the complex characteristics of the lesion area, thereby providing reliable data support for personalized treatment.

[0047] As an optional implementation, multimodal imaging methods include: Magnetic Resonance Imaging (MRI), Positron Emission Tomography (PET) and Computed Tomography (CT), etc., which respectively capture different biological and structural information of the target tissue.

[0048] Multimodal imaging combines the advantages of different imaging technologies and can provide comprehensive and detailed target tissue information. However, in actual treatment, the required imaging information may vary due to different clinical needs, patient status or equipment availability. Therefore, in the present application, the imaging unit 10 is also used to: obtain the treatment information of the target tissue; based on the treatment information of the target tissue, use at least one of the multimodal imaging methods to generate three-dimensional image data of the target tissue.

[0049] In specific implementations, each imaging technology has different advantages. For example, MRI is good at soft tissue imaging, PET provides metabolic information, and CT can quickly generate density and structural information. By combining these imaging technologies, comprehensive three-dimensional image data of target tissues can be generated.

[0050] Based on the treatment information of the target tissue, the imaging unit 10 can selectively use one or more of the above imaging modalities to meet the specific treatment requirements. For example, when detailed metabolic activity information is required, PET imaging can be used alone or in priority; when high-resolution anatomical structures are required, MRI can be the first choice.

[0051] In addition, when equipment resources are limited or the patient's condition does not allow for long-term imaging, a single imaging modality can be selectively used to complete the generation of three-dimensional image data.

[0052] In a specific implementation, the imaging unit 10 determines the current treatment requirements and equipment conditions by acquiring treatment information of the target tissue, so as to make the best imaging modality selection. The treatment information includes the location, size, metabolic activity level of the tumor, and the overall health status of the patient.

[0053] The treatment information also includes: treatment demand information, equipment condition information and user status information;

[0054] In this way, the system can work effectively in various clinical scenarios. According to different treatment needs, the imaging unit 10 can dynamically adjust the imaging technology used to ensure that the generated three-dimensional image data is both comprehensive and meets the current treatment planning requirements. This method of selectively using imaging modalities not only improves treatment efficiency, but also can provide high-quality imaging data even when resources are limited or the patient is in poor condition, ensuring the smooth progress of the treatment process.

[0055] Through the selective combination of these three imaging techniques, the anatomical information, metabolic information, and density information of the target tissue can be fully acquired, providing a data basis for subsequent treatment planning.

[0056] It is understandable that the registration and fusion of multimodal image data is not the main content of the present application. For its specific method, reference may be made to the prior art and will not be repeated here.

[0057] For example, the patient's treatment information can be obtained through the system's electronic medical record (EMR) system, including the basic characteristics, location, size, and physiological condition of the tumor. For example, the system can extract the patient's brain tumor data, including that it is located in the left frontal lobe of the brain, is 3 cm in size, has an irregular shape, and is close to key neural pathways.

[0058] At the same time, the imaging unit 10 analyzes this information to determine the most suitable imaging technique, which can be achieved by deploying a lightweight neural network in the imaging unit 10. For example, if the tumor is close to the brain stem or other important structures, the system can choose MRI with high-resolution imaging capabilities.

[0059] After selecting the imaging mode, the imaging unit 10 can generate three-dimensional image data of the target tissue. For example, for a malignant tumor located on the left side of the brain, the imaging unit 10 can first use MRI to generate detailed soft tissue images, then use PET to generate metabolic activity images of the tumor, and fuse the MRI and PET images.

[0060] In a specific implementation, since the multimodal imaging data comes from different imaging technologies, their coordinate systems and resolutions may be different, the imaging unit 10 can spatially align images of different modalities through an image registration algorithm. Common registration methods include feature point-based registration and intensity-based registration, so that images of different modalities are aligned in the same coordinate system.

[0061] For processing unit 20:

[0062] The processing unit 20 receives and analyzes the three-dimensional image data from the imaging unit 10. First, the first lesion area in the target tissue is determined based on the edge detection method, and the part that needs to be treated is identified. Subsequently, the first lesion area is further analyzed using the density gradient algorithm to identify and divide the core area, edge area and transition area. The depth information, density information and morphological feature information of each area will be extracted and used for subsequent treatment planning.

[0063] In a specific implementation, an automatic segmentation algorithm, such as a region growing method or an edge detection method, may be used to separate the lesion region (the first lesion region) from surrounding normal tissues.

[0064] During the segmentation process, the segmentation results in the two-dimensional image are converted into voxels, each of which represents a small cube in the three-dimensional space.

[0065] The voxel data of all two-dimensional images are superimposed layer by layer to form a complete three-dimensional image. The position and density value of each voxel is determined by the pixel information in the original image. In this way, the system can accurately reconstruct the morphology of the target tissue in three-dimensional space.

[0066] Among them, voxel is the basic unit in three-dimensional space, used to represent and process three-dimensional data. It is similar to a pixel in a two-dimensional image, but the difference is that a voxel has length, width and height, representing a small cube in three-dimensional space. Each voxel usually contains specific information about its spatial position, such as density, color or intensity.

[0067] See also Figure 2 and Figure 3 , Figure 2 This is an example diagram of the first lesion area provided in this application. Figure 3 A flow chart of the method for determining the area range of the core area, edge area and transition area in the first lesion area provided in the present application.

[0068] The method comprises steps S101 to S105, wherein:

[0069] S101: Calculate the density gradient of each voxel in the target tissue to generate a density gradient field; wherein the density gradient is the density change rate between adjacent voxels;

[0070] S102: In the density gradient field, determine an area where the density gradient is higher than a first gradient threshold as a candidate area;

[0071] S103: In the candidate area, a region whose density gradient value is higher than the second gradient threshold and whose density change rate is continuously distributed within the first region is taken as a core region;

[0072] S104: within the candidate area, determining an area having a target gradient change feature as an edge area;

[0073] S105: taking the area in the candidate area that is neither the core area nor the edge area as a transition area.

[0074] In order to effectively plan treatment for the lesion area, it is necessary to understand the different regional characteristics of the target tissue in detail, such as density, depth and morphological characteristics. The density gradient algorithm can accurately divide the core area, edge area and transition area by calculating the density change rate of voxels in the target tissue.

[0075] In a specific implementation, in the generated density gradient field, the processing unit 20 identifies regions with density gradient values ​​higher than a first gradient threshold and marks these regions as candidate regions, wherein the first gradient threshold can be determined by density gradient distribution statistics of the target tissue, and the selected region has significant density variation characteristics.

[0076] In the candidate area, further screen the areas with density gradient values ​​higher than the second gradient threshold, and analyze the density change rate distribution of these areas. If the density gradients of these areas are continuously distributed within a range, they are marked as core areas. The core area corresponds to the center or main part of the lesion, and its density gradient is significantly higher than that of the surrounding tissue.

[0077] The system analyzes regions with specific gradient change characteristics within the candidate region, which correspond to the edge regions of the lesion. For example, the edge region can be determined by identifying regions where the density gradient gradually decreases.

[0078] The non-core and non-edge areas in the candidate area are marked as transition areas. The transition area is usually located at the periphery of the lesion, with a small density change, but still has certain gradient characteristics.

[0079] In a specific implementation, the first gradient threshold is used to preliminarily screen out areas with significant density changes in the target tissue. It is a key criterion for identifying candidate areas. By setting this threshold, the system can filter out areas with insignificant density changes to ensure that the density changes in the candidate areas are significant enough to warrant further analysis. The threshold can be determined by statistical analysis. For example, first analyze the overall density distribution in the target tissue and calculate the density change rate distribution curve of all voxels. Then, select a suitable threshold (for example, the upper 25% quantile of the density change rate) as the first gradient threshold. This threshold may be adjusted according to different tissue types, lesion characteristics, and treatment needs to optimize the recognition accuracy of candidate areas.

[0080] The second gradient threshold is used to further refine the core area of ​​the candidate area. This threshold is higher than the first gradient threshold and is mainly used to identify areas with more dramatic density changes, which usually correspond to the central part or active area of ​​the tumor.

[0081] The setting of the second gradient threshold is similar to the first gradient threshold, and a higher quantile in the first gradient threshold can be selected. For example, in the candidate area, the system may select the upper 10% quantile of the density change rate distribution as the second gradient threshold. Through this screening, the system can more accurately determine the location of the core area.

[0082] The first area range is the continuity criterion for defining the core area. Even if the density gradient value is higher than the second gradient threshold, it is still necessary to determine whether these high gradient areas are distributed continuously in space. If there are significant density faults between some high gradient areas, these areas are excluded from the core area.

[0083] The first region range can be defined based on spatial distance. A distance threshold can be set, for example, if the distance between adjacent high gradient regions exceeds 2 mm, they are not considered continuous. The range is set based on the specific morphological characteristics of the tumor and treatment requirements.

[0084] The target gradient change feature is used to identify edge areas. Edge areas are usually characterized by a gradually decreasing density gradient. This change feature can help the system select areas that meet specific gradient patterns from candidate areas.

[0085] The target gradient change characteristics can be determined by analyzing the typical density change patterns of the tumor edge. For example, a large amount of tumor data can be learned through machine learning algorithms to extract common edge density change patterns, and these patterns can be used as features to identify edge areas. These features include the speed at which the density gradient gradually decreases, the shape of the changing curve, etc.

[0086] For example, if the target tissue is a malignant tumor in the brain, the tumor appears as an irregularly shaped area with significant internal density changes in the three-dimensional image generated by the imaging unit 10. In order to perform precise treatment, the system needs to divide the tumor area in detail to determine the area range and information of the core area, edge area and transition area.

[0087] The multimodal imaging method of MRI and CT is used to obtain three-dimensional image data of the tumor. MRI images provide soft tissue contrast information of the tumor, while CT images provide density distribution information.

[0088] In the processing unit 20, the density of each voxel is calculated to generate a density gradient field. For example, in a CT image, if the density of a voxel is 120 Hounsfield units (HU), the density of an adjacent voxel is 100 HU, and the distance between the two is 1 mm, then the density gradient is 20 HU / mm.

[0089] By analyzing the density gradient field, the system identifies areas with a density gradient higher than 50 HU / mm and marks them as candidate areas. These areas show significant density changes and usually correspond to active parts of the tumor.

[0090] Among the candidate areas, we further screened out areas with a density gradient higher than 80 HU / mm, and determined whether the density change rate of these areas was continuously distributed within a certain range. For example, if the density change of a certain part of the tumor was between 80-100 HU / mm and was evenly distributed, it would be marked as the core area. The core area represents the central area of ​​the tumor, where metabolic activity is higher and the response to treatment is most significant.

[0091] Continue to analyze the area where the density gradient gradually decreases in the candidate area and determine it as the edge area. For example, the area with a density gradient between 50-80 HU / mm and a relatively continuous distribution is marked as the edge area. This part of the area is usually the extension of the tumor and has a certain transition with the surrounding healthy tissue.

[0092] The remaining non-core and non-edge areas in the candidate region, which usually have a smaller density gradient but still have some difference from healthy tissue, are marked as transition areas. The transition area is usually located at the periphery of the tumor and represents the boundary area of ​​tumor extension.

[0093] In this way, the system can effectively divide the core area, edge area and transition area of ​​the tumor area (first lesion area) and provide accurate regional information. This provides accurate data for subsequent treatment planning, allowing treatment to cover different parts of the tumor more targetedly, improve the overall treatment effect, and reduce the impact on surrounding healthy tissues.

[0094] With respect to the processing unit 20, in the present application, the processing unit 20 is used to determine the regional information of each region in the first lesion region, and the regional information includes: depth information, density information and morphological feature information.

[0095] In a specific implementation, the processing unit 20 obtains the depth information of each voxel by calculating the distance from each voxel to the surface or a reference point (such as the epidermis or an important anatomical structure). This depth information is used to analyze the internal structure of the lesion area, such as identifying the center and edge positions of the tumor.

[0096] The acquisition of density information has been completed in the process of dividing the core area, edge area and transition area above, and will not be repeated here. Density information can help the system identify high-density core areas and areas with gradually decreasing density.

[0097] As an optional implementation, determining the energy step switching point includes: determining the regional boundary of the core area, the edge area and the transition area based on the regional range of the core area, the edge area and the transition area; setting the regional boundary between the core area and the edge area as the first switching point; setting the regional boundary between the edge area and the transition area as the second switching point.

[0098] In a specific implementation, the control unit 30 determines the boundary between the core area and the edge area based on the area range of each area obtained by the processing unit 20, and sets this position as the first switching point. At this switching point, the proton beam can be controlled to switch from the energy level of the edge area to the energy peak of the core area to ensure that the core area can receive enough energy to maximize the tumor killing effect.

[0099] Next, the control unit 30 identifies the boundary between the edge zone and the transition zone and sets it as a second switching point. At this switching point, the energy of the proton beam can gradually transition from the energy level of the edge zone to an energy attenuation curve suitable for the transition zone, ensuring that the energy is smoothly distributed in the transition zone, thereby effectively avoiding over-irradiation of healthy tissues.

[0100] In a specific implementation, the control unit 30 can directly use the outer boundary of the core area as the boundary position between the core area and the edge area. Specifically, in three-dimensional space, this boundary point is the surface formed by the outermost voxels of the core area. The outer boundary of the edge area is used as the boundary position between the edge area and the transition area. The boundary information of the edge area can also be directly extracted from the three-dimensional image data and marked as the switching point for transitioning to the next area.

[0101] As an optional implementation, the middle point between the boundaries of two adjacent areas (such as the core area and the edge area) is taken as the intersection point, especially when the area boundaries are complex or irregular.

[0102] This method ensures the precise switching of energy between different areas, optimizes the energy distribution of the proton beam, improves the accuracy of treatment, and reduces damage to normal tissue. This segmented control strategy not only simplifies the decision-making process of energy allocation, but also enables the system to flexibly respond to different tumor morphologies and patient needs, achieving more personalized treatment plans.

[0103] As an optional implementation manner, the morphological feature information of the core area includes: a first morphological feature; the morphological feature information of the edge area includes: a second morphological feature; the morphological feature information of the transition area includes: a third morphological feature;

[0104] The first morphological characteristics include: density concentration, volume uniformity and morphological symmetry;

[0105] The second morphological features include: edge irregularity, multi-scale shape spectrum, and edge thickness variation;

[0106] The third morphological feature includes: density gradient smoothness, gradient direction consistency and morphological transition characteristics.

[0107] In the specific implementation, the morphological feature information of the core area is defined as the first morphological feature, which specifically includes the following three aspects: density concentration, volume uniformity and morphological symmetry. Density concentration refers to the distribution of tissue density in the core area, reflecting the concentration of tumor cells in the core area. Volume uniformity measures whether the core area is evenly distributed in space. Higher volume uniformity means the balance of the tumor structure. Morphological symmetry is used to evaluate the geometric symmetry of the core area, which can reflect the biological behavior of the tumor in some cases. For example, higher symmetry corresponds to a more stable tumor morphology.

[0108] Secondly, the morphological feature information of the edge area is defined as the second morphological feature, including: edge irregularity, multi-scale shape spectrum and edge thickness variation. Edge irregularity describes the morphological complexity and irregularity of the tumor edge, which is usually related to the aggressiveness of the tumor. The multi-scale shape spectrum describes the geometric characteristics of the edge by analyzing the edge shape characteristics at different scales. This multi-scale analysis can reveal the changes of the edge at different observation scales. Edge thickness variation is used to capture the uneven thickness of the edge area, which can reflect the degree of tumor expansion in different directions and the sharpness of the edge.

[0109] The morphological feature information of the transition zone is defined as the third morphological feature, which specifically includes: density gradient smoothness, gradient direction consistency and morphological transition characteristics. Density gradient smoothness is used to describe the smoothness of tissue density changes in the transition zone. A transition zone with higher smoothness indicates that the density changes are more uniform, which is conducive to the smooth distribution of energy. Gradient direction consistency refers to the degree of consistency of the density gradient direction in the transition zone. A higher consistency means that the energy can be smoothly transitioned in the transition zone, reducing damage to healthy tissues. The morphological transition feature is used to describe the morphological change characteristics from the core area to the edge area, or from the edge area to the transition zone. This feature is conducive to adjusting the energy allocation strategy.

[0110] In a specific implementation, each feature can be determined as follows: by analyzing the gray value distribution of the CT or MRI image in the core area, the average density and standard deviation of each voxel in the core area are calculated. The density concentration is determined by these statistical values, reflecting the concentration of tumor cells in the core area.

[0111] The uniformity of voxel distribution in the core area is calculated. For example, the core area is divided into several sub-areas, and the number of voxels in each sub-area is compared. If the number of voxels in each sub-area is close, the volume uniformity is high.

[0112] The geometric center of the core is calculated and its shape symmetry is analyzed. Symmetry indices such as the asymmetry index are used to quantify the morphological symmetry of the core, which is determined by comparing the tumor's morphology with its mirror image.

[0113] Through edge detection algorithms (such as Canny algorithm or Sobel operator), the fractal dimension or irregularity index is calculated. These indices quantify the complexity and irregularity of the edge.

[0114] The morphology of the edge area is analyzed at different scales by spectrum analysis. The Fourier transform results of the edge area are decomposed at multiple scales to generate a multi-scale shape spectrum, thereby capturing the morphological characteristics of the edge area at different scales.

[0115] By analyzing the thickness distribution of the edge area, the thickness variation in each direction is calculated. The thickness in each direction of the edge area can be determined by radial distance measurement or morphological analysis, and its variation range and distribution unevenness can be calculated.

[0116] The smoothness of density changes is evaluated by calculating the gradient of voxel density in the transition region. The continuity and smoothness of density gradients can be analyzed using Laplacian or convolution-based gradient calculation methods.

[0117] By analyzing the directional distribution of density gradients, the directional consistency of density gradients in the transition region can be evaluated. The gradient direction of each voxel can be calculated, and the distribution of these directions can be statistically analyzed, and the consistency of the direction can be quantified using standard deviation or entropy.

[0118] The morphological transition characteristics are determined by analyzing the morphological changes between the core area, edge area, and transition area. For example, morphological analysis tools (such as morphological closing and opening operations) are used to evaluate the morphological transition between regions and quantify the smoothness and continuity of morphological changes.

[0119] For depth information, the tumor surface can be defined in the three-dimensional image data, and then the shortest distance from each voxel to the tumor surface can be calculated through the distance transformation algorithm to obtain the depth information of each area.

[0120] In this way, by defining and distinguishing these morphological features, the system can better analyze and treat different areas of the tumor and provide more precise treatment plans, thereby improving the effectiveness and safety of BNCT treatment.

[0121] As an optional implementation, the neural network model includes a first input channel, a second input channel and a third input channel;

[0122] Wherein, the first input channel is used to input the first morphological feature; the second input channel is used to input the second morphological feature; the third input channel is used to input the third morphological feature;

[0123] The output nodes of the neural network model include: edge area energy level, core area energy peak value and transition area energy attenuation curve.

[0124] In the specific implementation, the neural network model in this application is designed to include multiple input channels and corresponding output nodes to ensure accurate processing of the characteristics of each tumor region and optimization of the energy distribution strategy. The neural network model includes three input channels, which are used to process the morphological feature information of the core area, edge area and transition area, respectively, and multiple output nodes, which are used to generate energy control parameters for each area.

[0125] The first input channel is specifically used to input the morphological feature information of the core area, namely the first morphological feature. The extracted core area features, including density concentration, volume uniformity, and morphological symmetry, are passed to this channel as input data. The neural network receives and processes these feature information through the first input channel to extract deep features related to the energy demand of the core area and use them for subsequent energy peak calculations.

[0126] The second input channel is used to input the morphological feature information of the edge area, namely the second morphological feature. The features extracted from the edge area, including edge irregularity, multi-scale shape spectrum, and edge thickness change, are passed to this channel. The neural network processes these complex edge features through the second input channel, identifies the important factors affecting the energy level of the edge area, and provides data support for energy regulation.

[0127] The third input channel is used to input the morphological feature information of the transition zone, namely the third morphological feature. The density gradient smoothness, gradient direction consistency and morphological transition characteristics of the transition zone are passed to this channel as input data. Through the third input channel, the neural network can understand and process the characteristic information of the transition zone, and on this basis generate parameters for the energy attenuation curve to ensure the rationality and continuity of the energy distribution in the transition zone.

[0128] After the neural network model processes the features of each input channel, it generates energy control parameters for different regions through the output nodes. In the specific implementation, the output nodes of the model include the following three parts:

[0129] Marginal zone energy level: The neural network generates marginal zone energy level parameters based on the morphological feature information input from the marginal zone. This parameter is used to control the energy distribution of the proton beam in the marginal zone to ensure that the complex morphology of the marginal zone is fully treated without excessively damaging healthy tissue.

[0130] Core area energy peak: The neural network generates the core area energy peak based on the morphological feature information input by the core area. This parameter is used to set the highest energy of the core area to ensure that the proton beam can achieve the best treatment effect in the core area and maximize tumor killing.

[0131] Energy attenuation curve of transition zone: The neural network generates an energy attenuation curve of transition zone based on the morphological feature information input in the transition zone. This output is used to control the gradual weakening of the energy of the proton beam in the transition zone, ensuring a smooth transition of energy from the core zone to the edge zone and then to the transition zone, avoiding treatment faults or over-irradiation of healthy tissues.

[0132] In the specific implementation, the control unit 30 first extracts the morphological feature information of each area, and inputs these features into the corresponding input channels of the neural network model. The model receives and processes these input features, and finally generates corresponding energy control parameters at the output node through layer-by-layer neural network calculations. These parameters will be directly applied to the energy control module of the proton accelerator to achieve accurate energy distribution to meet the treatment needs of different areas.

[0133] In this way, through the design of multiple input channels, the neural network model can fully consider the characteristic differences of various tumor regions and generate targeted energy control parameters, significantly improving the accuracy and personalization of treatment.

[0134] Exemplarily, the neural network model may be a convolutional neural network (CNN), a multilayer perceptron (MLP), a long short-term memory network (LSTM), a hybrid neural network (Hybrid Networks) or other models.

[0135] For example, for the training of the neural network model, three-dimensional medical imaging data (such as CT, MRI) can be collected from tumor treatment cases, and the morphological feature information of the core area, edge area and transition area can be extracted based on these imaging data. The data of each case includes the first morphological feature of the core area, the second morphological feature of the edge area and the third morphological feature of the transition area. In addition to the morphological features, the ideal energy level, energy peak and energy attenuation curve corresponding to each area need to be collected as label data.

[0136] At the beginning of training, the weight parameters of the neural network model are randomly initialized. The input of the model includes the first input channel, the second input channel and the third input channel, each receiving the morphological features of different regions.

[0137] The input data is calculated through each layer of the network to generate predicted values ​​of the energy level of the edge area, the energy peak of the core area, and the energy decay curve of the transition area. The system calculates the error between these outputs and the true label, and the error can be measured using the mean square error (MSE) or other suitable loss function.

[0138] In addition, the back propagation algorithm can also be used to calculate the gradient of the error relative to the model parameters. Use optimization algorithms (such as Adam, SGD) to update the weights of the network to minimize the loss function. This process will be repeated in multiple training iterations, gradually adjusting the weight parameters of the model so that the energy parameters output by the model are closer to the true value.

[0139] As an optional implementation, the neural network model further includes: a cross-fusion layer; the cross-fusion layer is used to fuse the first morphological feature, the second morphological feature and the third morphological feature.

[0140] In the specific implementation, the neural network model further includes a cross-fusion layer for fusing morphological feature information from different regions. Specifically, the cross-fusion layer integrates the first morphological feature of the core area, the second morphological feature of the edge area, and the third morphological feature of the transition area, thereby generating a more comprehensive and integrated feature expression in the subsequent layers of the neural network.

[0141] In the process of tumor treatment, the features of the core area, edge area and transition area do not exist independently, and there may be complex dependencies or interactions between them. By fusing these features, the cross-fusion layer can generate more representative comprehensive features, providing more accurate input for subsequent energy regulation strategies.

[0142] The cross-fusion layer can initially combine features from different regions through linear combination or nonlinear transformation. The combined features can be processed by weighted summation, concatenation, or through a small neural network (such as a multilayer perceptron).

[0143] The cross-fusion layer can obtain the importance of each region feature and dynamically adjust the weights of these features. The adjustment of feature weights can be achieved through the feature-driven module, which analyzes the relative importance of each region feature and weights it during the fusion process.

[0144] Finally, the cross-fusion layer outputs a comprehensive feature vector, which contains both the independent features of each region and the interactive information between regions. This comprehensive feature will be passed to the subsequent layers of the neural network to generate specific energy control parameters.

[0145] By introducing the cross-fusion layer, the neural network model can more effectively integrate and utilize the feature information of different regions, thereby improving the overall model's expressiveness and prediction accuracy. The cross-fusion layer not only enables the model to capture the global information of features in different regions, but also dynamically adjusts the weights of different features to ensure that the energy control parameters output by the model are more in line with actual treatment needs.

[0146] In addition, the cross-fusion layer reduces the independence assumption between features in each region through feature cross-fertilization and comprehensive processing, making the model more robust and able to adapt to more complex and variable tumor morphologies.

[0147] It should be noted that in order to perform the fusion operation, we must first ensure that the dimensions and formats of the input features are consistent, which can be achieved by aligning the features through linear transformation, convolutional layer, or fully connected layer.

[0148] As an optional implementation manner, the feature fusion mode of the cross fusion layer is weighted fusion; the mathematical expression of the weighted fusion is:

[0149] ;

[0150] in, is the fusion feature vector, The first morphological feature is The second morphological feature is The third morphological feature is , , , are the fusion weights of the core area, the transition area and the edge area respectively.

[0151] In the specific implementation, the features of different regions are weighted and summed according to certain weights. The weights can be fixed or dynamically generated by the feature-driven module.

[0152] As an optional implementation, the cross fusion layer further includes: a feature driving module; the feature driving module is used to adjust the fusion weight of each region; the adjustment formula is as follows:

[0153] ;

[0154] in, , , are the adjustment coefficients of the core area, the transition area and the edge area respectively, is the adjustment function of the core area based on relative density, is the adjustment function of the edge region based on morphological complexity, It is the adjustment function of the transition zone based on the depth change.

[0155] In practice, the fixed-weight fusion method has an obvious limitation: it cannot dynamically respond to changes in regional features under different circumstances. For example, if the core area of ​​a tumor has a significant change in density, or the morphological complexity of the edge area suddenly increases, the fixed weight coefficient may not be adjusted in a timely and flexible manner, resulting in inaccurate fusion results. In order to overcome the limitations of fixed weights, a feature-driven module is introduced to dynamically adjust the weights by analyzing the features of each area.

[0156] The adjustment function is used to calculate the dynamic adjustment value of each regional feature, and the weight is changed based on specific biological characteristics or complexity; the adjustment coefficient is used to amplify or reduce the impact of the adjustment function and determines the size of the adjustment range.

[0157] For example, you can control , controls the degree of influence of the edge area on the final fusion feature when the feature changes. The specific adjustment coefficients of each region can be determined based on the treatment goal and the location of each region from the important organ. For example, if the transition area is very close to an important organ but far from the core area, then a larger adjustment coefficient can be set for the core area based on the treatment goal.

[0158] For example, if If the setting is larger, then even a small Changes can also significantly affect the weight of the core area. This is particularly effective in precision treatment of the core area.

[0159] In a specific implementation, the relative density of the core area refers to the relative value of the density of the core area compared with the density of its surrounding area (usually the edge area). It is used to dynamically adjust the weight of the core area according to the density difference between the core area and the edge area. This function can reflect the importance of the core area in a specific situation. For example, when the density of the core area is significantly higher than that of the edge area, it may mean that the core area is the most active or important part of the tumor, and thus requires a higher weight. For example, when the density of the core area is significantly higher than that of the edge area, it indicates that the importance of the core area in the current situation has increased.

[0160] The morphological complexity of the margin zone generally reflects the irregularity and diversity of the tumor margin, and these features may be associated with the invasiveness and spread of the tumor. It is used to dynamically adjust the weight of the edge region according to its morphological complexity. The morphological complexity can be quantified by fractal dimension, shape spectrum analysis or other irregularity indicators. When the morphological complexity of the edge region increases, The value of becomes larger, indicating that the edge area needs a higher weight. Exemplarily, nonlinear characteristics can be introduced into the function, for example, by setting a morphological complexity threshold. When the current morphological complexity exceeds this threshold, the change of weight adjustment will be more significant.

[0161] The depth change of the transition zone usually refers to the depth gradient when the tumor transitions from the edge zone to the normal tissue. It is used to dynamically adjust the weight of the transition zone according to the depth change. The depth change can be measured by calculating the distance gradient from the voxel in the transition zone to the tumor surface. When the depth gradient of the transition zone changes dramatically (i.e., the standard deviation is large), The value of increases, indicating that more attention needs to be paid to the features in the transition region. If the depth gradient changes gently (i.e., the standard deviation is small), the weight adjustment is small, indicating that the features in the transition region may be relatively less important in the fusion process.

[0162] In this specific embodiment, the system accurately controls the energy output of the proton accelerator in the first lesion area based on the energy step information generated by the neural network model and the predetermined energy step switching points to achieve personalized treatment of different areas of the tumor.

[0163] The neural network model generates energy control parameters for different regions through its output nodes. These output nodes include:

[0164] During the treatment process, the control unit 30 controls the energy output of the proton accelerator based on the energy step information output by the neural network model and the determined switching point:

[0165] In the core area, the proton accelerator output energy reaches the peak value of the core area to achieve a strong attack on the core part of the tumor. The energy setting at this time is determined by the peak energy of the core area, ensuring that the proton beam can provide sufficient energy when penetrating into the core area.

[0166] In the margin zone, the energy output of the proton accelerator is adjusted to the margin zone energy level, which is lower than that in the core zone, but sufficient to destroy tumor cells in the margin zone and avoid damage to surrounding healthy tissue through fine control.

[0167] In the transition zone, the proton accelerator gradually reduces the output energy according to the energy attenuation curve, so that the energy is smoothly distributed in the transition zone, ensuring that the proton beam can naturally decay to a level safe for normal tissue when penetrating the transition zone.

[0168] In the process of executing energy output control, the system uses a real-time monitoring feedback mechanism to adjust the energy output of the proton accelerator at any time. According to the patient's real-time physiological data and tumor response, the energy ladder information and switching point position are dynamically adjusted to further optimize the treatment effect and ensure that the output energy of the proton beam is always at the optimal level at different treatment stages.

[0169] In this way, based on the energy ladder information and switching point control strategy generated by the neural network, the proton accelerator can achieve accurate energy distribution to different tumor areas during the treatment process, maximizing the effectiveness of treatment while reducing damage to normal tissues. This precise energy control method is particularly suitable for the treatment of complex tumors and provides technical support for personalized radiotherapy.

[0170] In specific implementations, the energy output of the proton accelerator depends on the control of the radio frequency system, which accelerates and adjusts the energy of the proton beam through the radio frequency cavity. During the acceleration of the proton beam, the radio frequency system controls the acceleration gradient of the proton beam by adjusting the electric field strength and phase of the radio frequency cavity, thereby achieving control of the proton energy.

[0171] When the proton beam enters the core area, the RF system can provide a higher acceleration gradient by increasing the electric field strength of the RF cavity to ensure that the proton beam reaches the required energy peak. The RF system will adjust the RF power in real time based on the core area energy peak parameters output by the neural network model to ensure that the proton beam has enough energy to effectively destroy tumor cells when penetrating the core area.

[0172] When the proton beam enters the edge zone, the radio frequency system can adjust the energy of the proton beam to the set level of the edge zone by controlling the radio frequency phase and electric field strength.

[0173] In the transition zone, the energy of the proton beam needs to decay gradually to achieve a smooth energy transition. The radio frequency system can achieve a gentle decay of the proton beam energy by gradually reducing the electric field strength of the radio frequency cavity and adjusting the radio frequency frequency. This decay process must match the energy decay curve output by the neural network model to ensure that the energy distribution meets the treatment requirements of the transition zone.

[0174] It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0175] It should be understood that determining B based on A does not mean determining B only based on A. B can also be determined based on A and / or other information.

[0176] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed in the present invention can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0177] In the description of this specification, the description with reference to the terms "one embodiment", "example", "specific example", etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0178] The preferred embodiments of the present invention disclosed above are only used to help explain the present application. The preferred embodiments do not describe all the details in detail, nor do they limit the present application to specific implementation methods. Obviously, many modifications and changes can be made according to the content of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the present application, so that those skilled in the art can understand and use the present application well. The present application is limited only by the claims and their full scope and equivalents.

Claims

1. A high-current proton accelerator system for a boron neutron capture therapy system, characterized in that: include: An imaging unit, which generates three-dimensional image data of a target tissue based on a multimodal imaging method; A processing unit, configured to determine a first lesion area in the target tissue based on an edge detection method with respect to the three-dimensional image data of the target tissue; Based on a density gradient algorithm, determining the region range and region information of a core region, a marginal region, and a transition region in the first lesion region; wherein the region information includes: depth information, density information, and morphological feature information; A control unit, based on the region range, determines an energy step switching point; inputs the region information into a neural network model to generate energy step information; and controls the energy output of a proton accelerator in the first lesion region based on the energy step information and the energy step switching point; The morphological feature information of the core area includes: a first morphological feature; the morphological feature information of the edge area includes: a second morphological feature; the morphological feature information of the transition area includes: a third morphological feature; The first morphological characteristics include: density concentration, volume uniformity and morphological symmetry; The second morphological features include: edge irregularity, multi-scale shape spectrum, and edge thickness variation; The third morphological features include: density gradient smoothness, gradient direction consistency and morphological transition characteristics; Determining the energy step switching point includes: determining the regional boundaries of the core area, the edge area, and the transition area based on the regional ranges of the core area, the edge area, and the transition area; The region boundary between the core region and the edge region is set as a first switching point; the first switching point is used to control the energy of the proton beam to switch from the energy level of the edge region to the energy peak of the core region; The boundary between the edge region and the transition region is set as a second switching point; the second switching point is used to control the energy of the proton beam to gradually transition from the energy level of the edge region to an energy attenuation curve suitable for the transition region; The energy step information includes: edge area energy level, core area energy peak value and transition area energy attenuation curve; the neural network model includes a first input channel, a second input channel and a third input channel; Wherein, the first input channel is used to input the first morphological feature; the second input channel is used to input the second morphological feature; the third input channel is used to input the third morphological feature; The output nodes of the neural network model include: edge area energy level, core area energy peak value and transition area energy attenuation curve.

2. The high-current proton accelerator system for a boron neutron capture therapy system according to claim 1, characterized in that: The multimodal imaging method includes: magnetic resonance imaging, positron emission tomography and computer tomography; the imaging unit is also used for: Acquiring treatment information of the target tissue; wherein the treatment information includes: treatment requirement information, equipment condition information and user status information; Based on the treatment information of the target tissue, three-dimensional image data of the target tissue is generated using at least one of the multimodal imaging methods.

3. The high-current proton accelerator system for a boron neutron capture therapy system according to claim 2, characterized in that: The determining of the region range and region information of the core region, the edge region and the transition region in the first lesion region comprises: Calculating the density gradient of each voxel in the target tissue to generate a density gradient field; wherein the density gradient is the density change rate between adjacent voxels; In the density gradient field, determining a region where the density gradient is higher than a first gradient threshold as a candidate region; In the candidate area, the area where the density gradient value is higher than the second gradient threshold and the density change rate is continuously distributed within the first area is taken as the core area; In the candidate area, a region having a target gradient change feature is determined as an edge region; The non-core area and non-edge area in the candidate area is taken as a transition area.

4. The high-current proton accelerator system for a boron neutron capture therapy system according to claim 3, characterized in that: The neural network model also includes: a cross fusion layer; the cross fusion layer is used to fuse the first morphological feature, the second morphological feature and the third morphological feature.

5. The high-current proton accelerator system for a boron neutron capture therapy system according to claim 4, characterized in that: The feature fusion mode of the cross fusion layer is weighted fusion; the mathematical expression of the weighted fusion is: ； in, is the fusion feature vector, The first morphological feature is The second morphological feature is The third morphological feature is , , are the fusion weights of the core area, the transition area and the edge area respectively.

6. The high-current proton accelerator system for a boron neutron capture therapy system according to claim 5, characterized in that: The cross-fusion layer further includes: a feature driving module; the feature driving module is used to adjust the fusion weights of the core area, the transition area and the edge area; the adjustment formula is as follows: ； in, , , are the adjustment coefficients of the core area, the transition area and the edge area respectively, is the adjustment function of the core area based on relative density, is the adjustment function of the edge region based on morphological complexity, is the adjustment function of the transition zone based on the depth change.

Images