Method and device for automatic radiotherapy plan generation based on artificial intelligence

Abstract

The present invention discloses a method and device for automatically generating radiotherapy plans based on artificial intelligence. The method comprises the following steps: S1: analyzing the similarity between the current case and each case in the case library, and extracting the most similar N cases; S2: combining the clinical indicators of different tumor hospitals, forming different evaluation criteria, and extracting the evaluation criteria as the target function for optimization; S3: using the target function established in S2, using distributed iterative optimization, and finally automatically generating a treatment plan. The present invention eliminates the very cumbersome manual design process, and the plan designer does not need to manually modify and optimize the plan multiple times, and does not need to manually compare the optimization results each time, which significantly improves the efficiency of plan making.

Description

Technical Field

[0001] The present invention belongs to the technical field of radiotherapy, and in particular relates to a method and device for automatically generating a radiotherapy plan based on artificial intelligence. Background Art

[0002] Conventional radiotherapy plan design is a time-consuming and labor-intensive process, which requires physicists to continuously adjust the field parameters and optimization parameters during plan optimization to find the optimal plan. In addition, it is closely related to the experience differences between plan designers, the time invested in plan design, and the clinical indicators of medical institutions. Therefore, automatic planning has developed rapidly in recent years. It can improve the efficiency of plan design while ensuring the quality of the plan, free up manpower, and automatically complete the complex and tedious process of plan making through machines.

[0003] Currently, there are some methods dedicated to the automation of radiotherapy plan design, such as the Auto-Planning function in Pinnacle TPS. The automatic plan completed by Auto Planning in the Radiotherapy Planning System (TPS) needs to be evaluated. If the clinical indicators are not met, the system background will continuously and automatically create auxiliary structures and automatically adjust the optimization conditions to optimize and iterate, and finally complete the plan that meets the clinical indicators. This method can reduce manual intervention, but the time for a single plan design is long, and it cannot accurately adjust a certain organ that does not meet the standard. Often such automatic adjustments will affect the overall situation, thereby making the effect of a certain plan worse. Summary of the invention

[0004] In order to solve the above technical problems, the present invention proposes a method and device for automatically generating a radiotherapy plan based on artificial intelligence.

[0005] In order to achieve the above object, the technical solution of the present invention is as follows:

[0006] In one aspect, the present invention discloses a method for automatically generating a radiotherapy plan based on artificial intelligence, comprising the following steps:

[0007] S1: Analyze the similarity between the current case and the cases in the case database, and extract the most similar N cases;

[0008] S2: Combine the clinical indicators of different cancer hospitals to form different evaluation criteria, and extract the evaluation criteria as the objective function for optimization;

[0009] S3: Through the objective function established in S2, distributed iterative optimization is used to automatically generate a treatment plan.

[0010] Based on the above technical solution, the following improvements can be made:

[0011] As a preferred solution, S1 specifically includes the following steps:

[0012] S1.1: Extract the OVH information of each case from the case database;

[0013] S1.2: Calculate the OVH information of the current case;

[0014] S1.3: Reduce the dimension of the OVH information of the current case;

[0015] S1.4: Calculate the cosine of the angle between the anatomical feature vector of the current case and each case in the case database as the similarity;

[0016] S1.5: Find the N cases with the greatest similarity to the current case.

[0017] As a preferred solution, S3 specifically includes the following steps:

[0018] S3.1: Set prescription;

[0019] S3.2: Solve OVH information based on the delineation information, find the most similar cases, and obtain the most suitable optimization conditions;

[0020] S3.3: Automatically add radiation fields and optimization parameters;

[0021] S3.4: Calculate the deposition matrix;

[0022] S3.5: Transform optimization parameters into optimization objective function;

[0023] S3.6: CG optimized photon flux;

[0024] S3.7: Calculate dose and DVH;

[0025] S3.8: Calculate the plan score of the plan and / or the organ score of the organ according to the indicators of the evaluation criteria, and determine whether the plan score meets the average score of the plan of this type or whether the organ score meets the evaluation criteria;

[0026] If satisfied, the plan is final;

[0027] If not, an optimization adjustment operation is performed to recalculate the dose and DVH until the conditions are met.

[0028] As a preferred solution, the optimization and adjustment operation in S3.8 specifically includes the following steps:

[0029] (a) Identify the organ that needs adjustment;

[0030] (b) Determine the appropriate radiation field;

[0031] (c) Adjust the intensity map of the current field;

[0032] (d) Leaf segmentation.

[0033] As a preferred solution, in S3.8, the plan score of the plan is calculated by combining the scores of the dose constraint, DVH constraint, HI, CI and each item in the evaluation criteria.

[0034] In addition, on the other hand, the present invention also discloses an artificial intelligence-based radiotherapy automatic plan generation device, comprising:

[0035] Case similarity analysis module: The case similarity analysis module is used to analyze the similarity between the current case and each case in the case database, and extract the most similar N cases;

[0036] The objective function extraction module is used to combine the clinical indicators of different cancer hospitals to form different evaluation criteria and extract the evaluation criteria as the objective function for optimization;

[0037] The treatment plan automatic generation module is used to use the objective function established by the objective function extraction module to perform iterative optimization using distributed methods, and finally automatically generate a treatment plan.

[0038] As a preferred solution, the case similarity analysis module includes:

[0039] An OVH information extraction unit, the OVH information extraction unit is used to extract the OVH information of each case from the case database;

[0040] An OVH information calculation unit, the OVH information calculation unit is used to calculate the OVH information of the current case;

[0041] A dimension reduction unit, which is used to reduce the dimension of the OVH information of the current case;

[0042] A similarity calculation unit, the similarity calculation unit is used to calculate the cosine of the angle between the anatomical feature vector of the current case and each case in the case database as the similarity;

[0043] The similarity search unit is used to find the N cases with the greatest similarity to the current case.

[0044] As a preferred solution, the treatment plan automatic generation module is specifically operated through the following steps:

[0045] T1: setting prescription;

[0046] T2: Solve OVH information based on the delineation information, find the most similar cases, and obtain the most suitable optimization conditions;

[0047] T3: Automatically add radiation fields and optimization parameters;

[0048] T4: Calculate the deposition matrix;

[0049] T5: The optimization parameters are converted into the optimization objective function;

[0050] T6: CG optimized photon flux;

[0051] T7: Calculate dose and DVH;

[0052] T8: Calculate the plan score of the plan and / or the organ score of the organ according to the indicators of the evaluation criteria, and determine whether the plan score meets the average score of the plan of this type or whether the organ score meets the evaluation criteria;

[0053] If satisfied, the plan is final;

[0054] If not, an optimization adjustment operation is performed to recalculate the dose and DVH until the conditions are met.

[0055] As a preferred solution, the optimization and adjustment operation in S3.8 specifically includes the following steps:

[0056] (A) Identify the organ that needs adjustment;

[0057] (B) Determine the appropriate radiation field;

[0058] (C) Adjust the intensity map of the current field;

[0059] (D) Leaf segmentation.

[0060] As a preferred solution, in T8, the plan score of the plan is calculated by combining the scores of the dose constraint, DVH constraint, HI, CI and each item in the evaluation criteria.

[0061] The present invention provides an artificial intelligence-based radiotherapy automatic plan generation method and device, which has the following beneficial effects:

[0062] Compared with the Auto-Planning function in the Pinnacle system, the optimization engine of the present invention can use artificial intelligence to screen out the most similar cases from the big data case library and obtain the most suitable optimization conditions. At the same time, an accurate objective function is established in combination with the clinical indicators of major cancer hospitals. When the clinical indicators are not met, the organs that do not meet the indicators can be found, and then the corresponding radiation fields can be accurately located. By adjusting the photon flux map of the radiation field, the dose distribution of the organ can be directly improved to meet the clinical indicators. This process has little effect on the dose distribution of other organs, and the overall effect of the plan has been significantly improved.

[0063] The method of directly optimizing the photon flux map greatly reduces the number of iterations of the present invention, relies on the calculation of the existing distributed system, and rapidly improves the speed of iteration, thereby realizing the automatic preparation of the plan accurately and quickly. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0065] Figure 1 The present invention provides a flowchart of a method for automatically generating a radiotherapy plan.

[0066] Figure 2 This is a flow chart of S1 provided in an embodiment of the present invention.

[0067] Figure 3 A clinical indicator chart of a well-known cancer hospital provided in an embodiment of the present invention.

[0068] Figure 4 A template diagram of the evaluation criteria provided in an embodiment of the present invention.

[0069] Figure 5 This is a flowchart of S3 provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0070] The preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0071] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0072] Using ordinal numbers “first,” “second,” “third,” etc. to describe common objects merely indicates that different instances of similar objects are involved and is not intended to imply that the objects so described must have a given order in time, space, order, or in any other manner.

[0073] In addition, the expression of “comprising” an element is an “open” expression, which simply means that a corresponding component exists, and should not be interpreted as excluding additional components.

[0074] In order to achieve the purpose of the present invention, in some embodiments of a method and device for automatically generating a radiotherapy plan based on artificial intelligence, as Figure 1 As shown, the method for automatically generating a radiotherapy plan includes the following steps:

[0075] S1: Analyze the similarity between the current case and the cases in the case database, and extract the three most similar cases;

[0076] S2: Combine the clinical indicators of different cancer hospitals to form different evaluation criteria, and extract the evaluation criteria as the objective function for optimization;

[0077] S3: Through the objective function established in S2, distributed iterative optimization is used to automatically generate a treatment plan.

[0078] Clinical indicators of a well-known cancer hospital are as follows: Figure 2 ,According to these clinical indicators, the evaluation criteria for the plan are extracted, and the template is as follows Figure 3 shown.

[0079] In order to further optimize the implementation effect of the present invention, in some other embodiments, the remaining characteristic technologies are the same, except that, Figure 4 As shown, S1 specifically includes the following steps:

[0080] S1.1: Extract the Overlap Volume Histogram (OVH) information of each case from the case database;

[0081] S1.2: Calculate the OVH information of the current case;

[0082] S1.3: Perform PCA dimensionality reduction on the OVH information of the current case;

[0083] S1.4: Calculate the cosine of the angle between the anatomical feature vector of the current case and each case in the case database as the similarity;

[0084] S1.5: Find the three cases that are most similar to the current case.

[0085] Further, based on the above embodiments, Figure 5 As shown, S3 specifically includes the following steps:

[0086] S3.1: Set prescription;

[0087] S3.2: Solve OVH information based on the delineation information, find the most similar cases, and obtain the most suitable optimization conditions;

[0088] S3.3: Automatically add radiation fields and optimization parameters;

[0089] S3.4: Calculate the deposition matrix and calculate the dose contribution of each pencil beam to each voxel point, denoted as α jm ;

[0090] α jm That is, the dose contribution of the mth beam of unit intensity to the jth voxel point. jm The matrix formed is called the energy characteristic matrix (or deposition matrix);

[0091] S3.5: Transform optimization parameters into optimization objective function;

[0092]

[0093]

[0094]

[0095]

[0096] in, is the objective function related to all OARs, Then it is the objective function related to the target area;

[0097] Noar is the number of OARs, NT i is the number of sampling points in the i-th OAR;

[0098] NT PTV is the number of sampling points in the target area;

[0099] When the point dose in an organ violates the constraint δ j =1, otherwise δ j =0.ω j is the weight of the jth point;

[0100] d j is the calculated dose at the jth point, p j is the prescribed dose at the jth point;

[0101] α and β are normalization factors that control the importance of target volume and OAR;

[0102] Nray is the total number of beams (ray, also called pencil beam);

[0103] α jm is the dose contribution of the mth beam of unit intensity to the jth voxel point;

[0104] The whole α jmThe matrix formed is called the energy characteristic matrix (or deposition matrix);

[0105] is the intensity of the mth beam;

[0106] S3.6: Conjugate Gradient (CG) method to optimize photon flux;

[0107] S3.7: Calculate dose and dose volume histogram (DVH);

[0108] S3.8: Calculate the plan score of the plan and / or the organ score of the organ according to the indicators of the evaluation criteria, and determine whether the plan score meets the average score of the plan of this type or whether the organ score meets the evaluation criteria;

[0109] If satisfied, the plan is final;

[0110] If not, an optimization adjustment operation is performed to recalculate the dose and DVH until the conditions are met.

[0111] Further, based on the above embodiment, the optimization and adjustment operation in S3.8 specifically includes the following steps:

[0112] (a) Identify the organ that needs adjustment;

[0113] (b) Determine the appropriate radiation field;

[0114] (c) Adjust the intensity map of the current field;

[0115] (d) Leaf segmentation.

[0116] Furthermore, in S3.8, the plan score of the plan is calculated by combining the dose constraint, DVH constraint, HI (dose uniformity), CI (conformity index) and the scores of each item in the evaluation criteria.

[0117] First, confirm whether the unsatisfied score item is among the user-configured ignorable items. If it is not ignorable, perform the following operations to optimize and adjust.

[0118] The following is a detailed description of each step of the optimization and adjustment operation.

[0119] (a) Determine the organ that needs to be adjusted and find the corresponding organ according to the items that do not meet the standards in the evaluation criteria. The types of items corresponding to the organs can be divided into high and low doses of the target area, CI / HI of the target area, dose constraints and DVH constraints of OARs, and dose constraints and DVH constraints of auxiliary organs. These types are explained in detail below.

[0120] 1) Target area. The planning target volume (PTV) may have high dose and low dose. At this time, the system will save the part surrounded by the high dose area dose line (generally 110% of the prescribed dose) as an organ. Through Boolean operations with the target area, the high dose within the target area, the high dose outside the target area and the low dose within the target area can be subdivided and saved as corresponding organs respectively.

[0121] 2) Conformity Index (CI): It is the ratio of the treatment volume to the PTV. Usually, the minimum dose of the target area represented by the 90% isodose line is used as the lower limit of the treatment area. The more the shape and size of the treatment area conform to the planned target area, the better the plan. Ideally, 1 is the best, so the closer the CI value is to 1, the higher the score will be. In the case of a poor CI score, the system will save the part surrounded by the high-dose area dose line as an organ. Through Boolean operations with the target area, it can generate outlines of high doses within the target area, high doses outside the target area, and insufficient doses within the target area, and finally save them as the corresponding organs.

[0122] 3) Dose uniformity (HI): It is the uniformity of dose in the target area. Ideally, we hope that the DVH curve of PTV can drop directly from D98 to D2, and the plan will be better. The program background judges the score based on the value of (D2-D98) / D50 to get HI. In summary, ideally, when D2=D98, HI is the best, so the closer the HI value is to 0, the higher the score will be. When the HI score is poor, the system will save the high-dose part in the target area as an organ and adjust it to remove the high-dose.

[0123] 4) OAR: Organs at Risk (OAR) generally have dose constraints or DVH constraints that do not meet the standards. When the dose constraint of an OAR exceeds the limit, the isodose curve of the current dose will be saved as an organ, and a Boolean operation of the organ with the current OAR will be performed to obtain a new organ as the organ to be adjusted. When the DVH constraint of an OAR does not meet the standards, for example, the V30 for the small intestine in the evaluation standard is <30%, but the DVH result is V35 <30%, then the background will save the 3000 and 3500 dose lines as organs, perform a Boolean operation of the organ with the current OAR to generate the organ to be adjusted, and perform an operation to reduce the intensity value in the intensity map to meet the DVH standards.

[0124] 5) Auxiliary organs: Auxiliary organs are generally presented in the evaluation criteria in the following two ways: one is the expansion of the target area, and this type of organ focuses on the drop gradient of the target area dose; the other is the expansion of OAR, and this type of organ is used to detect whether the dose constraint of the OAR expansion part meets the standard.

[0125] When the dose of the auxiliary organs generated by the expansion of the target area does not meet the standard, the dose line will be saved as the organ and a Boolean operation will be performed with the auxiliary organ to obtain the underdose area, and the underdose area will be saved as the organ; when the dose constraint of the auxiliary organ of the OAR expansion exceeds the limit, for example, the dose of Cord+5mm (spinal cord expansion 5mm) is less than 5000, the area with a dose greater than 5000 in the Cord+5mm organ will be saved as a new organ as the organ to be adjusted, and the intensity value in the intensity map will be reduced.

[0126] (b) Determine the appropriate radiation field. Based on the outline of the organ and the current dose of the organ, the background can solve the dose contribution value of each radiation field to the current organ, from which the radiation field with the largest or smallest dose contribution can be found. For PTV underdose, the radiation field with the smallest contribution value will be selected, and for high dose, the radiation field with the largest contribution value will be selected; for OAR, the radiation field with the largest contribution value will be selected for adjustment.

[0127] (c) Adjust the intensity map of the current radiation field. After the radiation field is determined, the corresponding serial number of the current organ in the intensity map can be obtained according to the projection of the organ. The background automatically adjusts the intensity value of the current serial number (such as: increase or decrease by 10%).

[0128] (d) Leaf segmentation.

[0129] The automatic radiotherapy plan generation method of the present invention analyzes the most similar cases based on artificial intelligence. First, the OVH of each critical organ in the case library is obtained, and then PCA dimensionality reduction processing is performed to find the most similar case in the case library in combination with the OVH of the new case OAR. The present invention combines the clinical indicators of major cancer hospitals to compile a set of plan evaluation standards that meet clinical requirements. The present invention performs iterative optimization by automatically adjusting the photon flux map in the background, so as to accurately optimize the organs that do not meet the standards, thereby improving the overall quality of the plan.

[0130] This method uses artificial intelligence to analyze the most similar cases, dynamically adjusts the optimization objective function according to the different clinical indicators of different hospitals, and then relies on the TPS optimization engine to quickly and automatically make plans, and continuously iterates based on the optimization results to approximate the evaluation criteria, thereby completing the production of automatic plans and eliminating the very tedious manual design process. Plan designers do not need to manually modify and optimize the plans multiple times, and do not need to manually compare the optimization results each time, which significantly improves the efficiency of plan making.

[0131] In addition, an embodiment of the present invention further discloses an artificial intelligence-based radiotherapy automatic plan generation device, comprising:

[0132] Case similarity analysis module: The case similarity analysis module is used to analyze the similarity between the current case and each case in the case database, and extract the three most similar cases;

[0133] The objective function extraction module is used to combine the clinical indicators of different cancer hospitals to form different evaluation criteria and extract the evaluation criteria as the objective function for optimization;

[0134] The treatment plan automatic generation module is used to use the objective function established by the objective function extraction module to perform iterative optimization using distributed methods, and finally automatically generate a treatment plan.

[0135] In order to further optimize the implementation effect of the present invention, in some other embodiments, the remaining feature technologies are the same, except that the case similarity analysis module includes:

[0136] An OVH information extraction unit, the OVH information extraction unit is used to extract the OVH information of each case from the case database;

[0137] An OVH information calculation unit, the OVH information calculation unit is used to calculate the OVH information of the current case;

[0138] A dimension reduction unit is used to perform PCA dimension reduction on the OVH information of the current case;

[0139] A similarity calculation unit, the similarity calculation unit is used to calculate the cosine of the angle between the anatomical feature vector of the current case and each case in the case database as the similarity;

[0140] Similarity finding unit,The similarity finding unit is used to find the three cases with the greatest similarity to the current case.

[0141] Further, based on the above embodiment, the treatment plan automatic generation module is specifically operated through the following steps:

[0142] T1: setting prescription;

[0143] T2: Solve OVH information based on the delineation information, find the most similar cases, and obtain the most suitable optimization conditions;

[0144] T3: Automatically add radiation fields and optimization parameters;

[0145] T4: Calculate the deposition matrix;

[0146] T5: The optimization parameters are converted into the optimization objective function;

[0147] T6: CG optimized photon flux;

[0148] T7: Calculate dose and DVH;

[0149] T8: Calculate the plan score of the plan and / or the organ score of the organ according to the indicators of the evaluation criteria, and determine whether the plan score meets the average score of the plan of this type or whether the organ score meets the evaluation criteria;

[0150] If satisfied, the plan is final;

[0151] If not, an optimization adjustment operation is performed to recalculate the dose and DVH until the conditions are met.

[0152] Further, based on the above embodiment, the optimization and adjustment operation in S3.8 specifically includes the following steps:

[0153] (A) Identify the organ that needs adjustment;

[0154] (B) Determine the appropriate radiation field;

[0155] (C) Adjust the intensity map of the current field;

[0156] (D) Leaf segmentation.

[0157] Further, based on the above embodiment, in T8, the plan score of the plan is calculated by combining the dose constraint, DVH constraint, HI, CI and the scores of each item in the evaluation criteria.

[0158] The formulas involved in the specific embodiment of the automatic radiotherapy plan generation device based on artificial intelligence are similar to the above-mentioned automatic radiotherapy plan generation method, and will not be repeated here.

[0159] It is worth noting that the above embodiment adopts the PCA dimensionality reduction method, but in other embodiments, other dimensionality reduction methods may be adopted.

[0160] The Auto-Planning function in the Pinnacle system performs optimization iterations by continuously and automatically creating auxiliary structures and automatically adjusting optimization conditions. Unlike the present invention, which accurately adjusts the photon flux map of a certain organ that does not meet the standard, the Auto-Planning method often affects the overall situation during the automatic adjustment process, thereby making the planning effect worse, failing to achieve accurate optimization, and greatly increasing the time required for planning.

[0161] Compared with the Auto-Planning function in the Pinnacle system, the optimization engine of the present invention can use artificial intelligence to screen out the most similar cases from the big data case library and obtain the most suitable optimization conditions. At the same time, an accurate objective function is established in combination with the clinical indicators of major cancer hospitals. When the clinical indicators are not met, the organs that do not meet the indicators can be found, and then the corresponding radiation fields can be accurately located. By adjusting the photon flux map of the radiation field, the dose distribution of the organ can be directly improved to meet the clinical indicators. This process has little effect on the dose distribution of other organs, and the overall effect of the plan has been significantly improved.

[0162] The method of directly optimizing the photon flux map greatly reduces the number of iterations of the present invention, relies on the calculation of the existing distributed system, and rapidly improves the speed of iteration, thereby realizing the automatic preparation of the plan accurately and quickly.

[0163] It should be understood that the various techniques described herein can be implemented in combination with hardware or software, or a combination thereof. Thus, the method and apparatus of the present invention, or certain aspects or portions of the method and apparatus of the present invention, can take the form of program code (i.e., instructions) embedded in a tangible medium, such as a floppy disk, a CD-ROM, a hard disk drive, or any other machine-readable storage medium, wherein when the program is loaded into a machine such as a computer and executed by the machine, the machine becomes a device for practicing the present invention.

[0164] The above embodiments are only for illustrating the technical concept and features of the present invention, and their purpose is to enable ordinary technicians in the field to understand the content of the present invention and implement it. They cannot be used to limit the protection scope of the present invention. All equivalent changes or modifications made according to the spirit of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for automatically generating radiotherapy plans based on artificial intelligence, characterized in that: The following steps are involved: S1: Analyze the similarity between the current case and the cases in the case database, and extract the most similar N cases; S2: Combine the clinical indicators of different cancer hospitals to form different evaluation criteria, and extract the evaluation criteria as the objective function for optimization; S3: Through the objective function established in S2, distributed iterative optimization is used to automatically generate a treatment plan. S3 specifically includes the following steps: S3.1: Set prescription; S3.2: Solve OVH information based on the delineation information, find the most similar cases, and obtain the most suitable optimization conditions; S3.3: Automatically add radiation fields and optimization parameters; S3.4: Calculate the deposition matrix; S3.5: Transform optimization parameters into optimization objective function; S3.6: CG optimized photon flux; S3.7: Calculate dose and DVH; S3.8: Calculate the plan score of the plan and / or the organ score of the organ according to the indicators of the evaluation criteria, and determine whether the plan score meets the average score of the plan of this type or whether the organ score meets the evaluation criteria; If satisfied, the plan is final; If not, optimization and adjustment operations are performed to recalculate the dose and DVH until the conditions are met; The optimization and adjustment operation in S3.8 specifically includes the following steps: Identify the organs that need adjustment; Determine the appropriate field of view; Adjust the intensity map of the current field; Leaf segmentation; In S3.8, the plan score of the plan is calculated by combining the scores of dose constraints, DVH constraints, HI, CI and each item in the evaluation criteria.

2. The method for automatically generating a radiotherapy plan according to claim 1, characterized in that: S1 specifically includes the following steps: S1.1: Extract the OVH information of each case from the case database; S1.2: Calculate the OVH information of the current case; S1.3: Reduce the dimension of the OVH information of the current case; S1.4: Calculate the cosine of the angle between the anatomical feature vector of the current case and each case in the case database as the similarity; S1.5: Find the N cases with the greatest similarity to the current case.

3. An artificial intelligence-based radiotherapy automatic plan generation device, characterized in that: include: A case similarity analysis module, which is used to analyze the similarity between the current case and each case in the case library, and extract the most similar N cases; An objective function extraction module, wherein the objective function extraction module is used to combine clinical indicators of different tumor hospitals to form different evaluation criteria, and extract the evaluation criteria as an optimized objective function; A treatment plan automatic generation module, which is used to perform iterative optimization using distributed methods based on the objective function established by the objective function extraction module, and finally automatically generate a treatment plan; The treatment plan automatic generation module is specifically operated by the following steps: T1: setting prescription; T2: Solve OVH information based on the delineation information, find the most similar cases, and obtain the most suitable optimization conditions; T3: Automatically add radiation fields and optimization parameters; T4: Calculate the deposition matrix; T5: The optimization parameters are converted into the optimization objective function; T6: CG optimized photon flux; T7: Calculate dose and DVH; T8: Calculate the plan score of the plan and / or the organ score of the organ according to the indicators of the evaluation criteria, and determine whether the plan score meets the average score of the plan of this type or whether the organ score meets the evaluation criteria; If satisfied, the plan is final; If not, optimization and adjustment operations are performed to recalculate the dose and DVH until the conditions are met; The optimization and adjustment operation in T8 specifically includes the following steps: Identify the organs that need adjustment; Determine the appropriate field of view; Adjust the intensity map of the current field; Leaf segmentation; In T8, the plan score of the plan is calculated by combining the scores of dose constraints, DVH constraints, HI, CI and each item in the evaluation criteria.

4. The radiotherapy automatic plan generation device according to claim 3, characterized in that: The case similarity analysis module includes: An OVH information extraction unit, configured to extract OVH information of each case from the case database; An OVH information calculation unit, the OVH information calculation unit is used to calculate the OVH information of the current case; A dimension reduction unit, wherein the dimension reduction unit is used to reduce the dimension of the OVH information of the current case; A similarity calculation unit, wherein the similarity calculation unit is used to calculate the cosine of the angle between the anatomical feature vector of the current case and each case in the case database as the similarity; The similarity finding unit is used to find N cases with the greatest similarity to the current case.

Images