CN110415785A - The method and system of artificial intelligence guidance radiotherapy planning - Google Patents
The method and system of artificial intelligence guidance radiotherapy planning Download PDFInfo
- Publication number
- CN110415785A CN110415785A CN201910820650.3A CN201910820650A CN110415785A CN 110415785 A CN110415785 A CN 110415785A CN 201910820650 A CN201910820650 A CN 201910820650A CN 110415785 A CN110415785 A CN 110415785A
- Authority
- CN
- China
- Prior art keywords
- radiotherapy planning
- radiotherapy
- automatically
- plan
- planning
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0012—Biomedical image inspection
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H15/00—ICT specially adapted for medical reports, e.g. generation or transmission thereof
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H20/00—ICT specially adapted for therapies or health-improving plans, e.g. for handling prescriptions, for steering therapy or for monitoring patient compliance
- G16H20/40—ICT specially adapted for therapies or health-improving plans, e.g. for handling prescriptions, for steering therapy or for monitoring patient compliance relating to mechanical, radiation or invasive therapies, e.g. surgery, laser therapy, dialysis or acupuncture
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H30/00—ICT specially adapted for the handling or processing of medical images
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30096—Tumor; Lesion
Abstract
The invention discloses the method and system of artificial intelligence guidance radiotherapy planning, including obtain CT image, are delineated automatically to the ROI region of CT image；By CT image and automatically the ROI region delineated is input in prediction model, and automatic Prediction goes out dosage distribution or DVH；The distribution of the dosage of prediction or DVH as the distribution of the dosage of reference or are referred into DVH, processing is optimized using the reverse optimization algorithm based on dosage distribution or DVH guidance, generates executable radiotherapy planning；Executable radiotherapy planning includes positive radiotherapy planning, stereotactic radiotherapy plan and intensity modulated radiation therapy plan, and wherein intensity modulated radiation therapy plan includes dynamic intensity-modulated radiation therapy plan, static intensity modulating radiotherapy planning, the strong radiotherapy planning of adjustment with volume and rotation intensity modulated radiation therapy plan.The present invention has preferable accuracy, stability and normalization, so as to improve medical software and hardware resources utilization rate, solves the problems, such as the high-level radiotherapy planning of the more difficult formulation of infirmary.
Description
Technical field
The present invention relates to field of medical technology, and in particular to the method and system of artificial intelligence guidance radiotherapy planning.
Background technique
Found according to investigation, the automatic method for the treatment of plan there are several types of: be based on KBP (Knowledge-based
Planning, Heuristics production plan) method, be based on PB-AIO (Protocol-based Automatic
Iterative Optimisation, agreement/template automatic Iterative optimization) method, the method (Multi- based on MCO
Criteria Optimisation, multiple criteria optimization) and the automatic radiotherapy planning based on artificial intelligence method.
Method based on KBP includes the following two kinds method: first method is the mode (atlas- based on atlas
Based approach), the mode based on atlas is usually to be made into a database to the patient's plan for having done treatment,
As soon as often carrying out a new patient, similar case is searched from database, thus using the parameter planned accordingly as new patient
The initial parameter of plan；Second method is the mode (model-based approach) based on model, and is based on DVH (agent
Amount volume histogram) guidance method belong to the method based on model；This method is usually, first using treated to
A large amount of cases and profile diagram therein establish the mapping relations of anatomy and geometrical characteristic, to establish a set of DVH model, often
It newly arrives a case, possible DVH is predicted according to geometrical characteristic, the generation of plan is then guided with the DVH.This method is
It is used by the TPS system Eclipse of Varian Associates, Inc. (US) 611 Hansen Way, Palo Alto, California 94303, U.S.A., referred to as RapidPlan.
It is based on the advantages of KBP method: can be improved the efficiency for the treatment of plan production, studies have shown that the method based on KBP
VMAT planned time can be made hour to be reduced to 10 to 15 minutes from 1 to 1.5；The otherness for reducing plan production is (same
Physics teacher makes the plan in different time or different physics teachers makes the plan to the same patient)；The shortcomings that based on KBP method
It is: needs that model is carefully adjusted and optimized, otherwise tumour conformal degree, target coverage is all not so good as original manual plan；
The plan of prediction is only clinically-acceptable, not necessarily optimal.Based on DVH guidance method the shortcomings that be, can only be pre-
The DVH in ROI (area-of-interest) region is measured, and exceeds the tissue of ROI, can not be predicted.In addition DVH cannot provide any sky
Between information.The method or voxel-based 3-dimensional dose prediction technique of these disadvantages are overcome at present, rather than predict DVH.Mesh
Preceding majority researcher develops the 3-dimensional dose prediction technique based on AI, however most of researcher does not consider difference
The situation of beam setting, they generally use same beam setting, cause to collect enough trained cases for same disease
Need to train different models as a problem, and for different beam settings, the application range of institute's training pattern by
To great limitation.However there is researcher to propose the solution that a kind of pair of AI model increases the channel beam.
Method based on PB-AIO has following specific method: (1) method of adjust automatically optimization aim and bound term, user
As long as the primary constraint of input and target item；(2) mouse is automatically moved by script realization to adjust the DVH of specific ROI；(3) it adopts
The mankind are imitated with AI does the process that ginseng is adjusted in optimization；(4) AutoPlanning in the TPS Pinnacle of PHILIPS Co. from
The gradually tune that the treatment model agreement that dynamic schedule module is inputted according to user carries out five layers of circulation is joined.
The advantages of automatic optimization method based on PB-AIO, is: generating the efficiency and quality and hand of IMRT or VMAT plan
Dynamic plan quite or more preferably, has researcher to compare the appraisal result planned automatically and manually using a set of plan points-scoring system
The appraisal result of plan, the average planned automatically is 62.3, and planning average manually is 59.1, plans points-scoring system
Total score be 100 points；Planned time can halve, and test VMAT, and the time became half an hour from 64 minutes.It is based on
The shortcomings that automatic optimization method of PB-AIO is: the parameter of input template directly determines the quality of plan, if template parameter
It is arranged not good enough, then what the plan automatically generated was made not as good as experienced physics teacher by optimization manually, therefore should
The use of method is limited to the experience of physics teacher.
Method based on MCO: MCO also cries multiple-objection optimization.The core of this method be obtain Pareto optimal solution, that is,
Any one of one plan objective function cannot all be further continued for improving, unless at least reducing some objective function.Specifically
It is divided into posteriority method and transcendental method again.
Transcendental method can only generate a good plan, and current software only has Erasmus-iCycle, can support
A kind of a kind of Cyberknife (ejected wave knife, treatment mode) and IMPT (the reverse intensity modulated therapy of proton beam, treatment mode).Elekta
Company prepares a kind of automatic planning model of one photon beam (treatment mode) of publication into TPS, that is to say, that the software is also
In development phase.The advantages of transcendental method: obtaining original intent inventory from history plan and constantly automatically updates inventory list,
Adjustment parameter is participated in without artificial completely；Automatically the plan quality planned is equal to or is higher than plan manually；Optimization process is saturating
It is bright intuitive；Method based on wish list is very flexible, and the experience of history plan not only can be used, and can also will treat personnel
Experience or other experiences be added wishlist；After changing treatment agreement regulation, re-starts automatic plan and do not need hand
Dynamic adjustment parameter；The same wish list can be used for different treatment modes, such as VMAT, Cyberknife, proton etc.；
It is only a text-only file beneficial to wish list, different medical institutions can very easily be compared to each other plan quality；
Generative capacity with multi-purpose project；The dosage in the region OAR is accomplished low as far as possible automatically；It can be carried out beam orientation optimization；It can be with
Unbiased comparison is carried out to different treatment modes using identical wish list.The disadvantage is that this method is still in automation
Scope, the method for not using AI, resulting radiotherapy planning have mechanicalness；There is no Program Assessment and 3-dimensional dose to verify, nothing
Method ensures to plan outstanding property and reliability.
Posteriority method can generate a large amount of plan, very strong to the dependence of computing resource, at present Raysearch company
Raystation and the Eclipse of Varian Associates, Inc. (US) 611 Hansen Way, Palo Alto, California 94303, U.S.A. use this mode.The advantages of posteriority method: it can generate relative to hand
Dynamic plan is more excellent or the plan that is equal to, small to the experience dependence of operator, reduces 10 minutes than manual planned time and arrives
It differs within 45 minutes.There is researcher that can complete each case, and mass ratio in 1 hour with the automatic plan of RayStation
It is manually good.The shortcomings that posteriority method is: obtained plan is the Pareto optimal solution in flux range, without directly considering
The optimization of machine parameter, final plan needs are converted to the plan that can treat, and conversion process middle dosage characteristic can be sent out
It is raw to change, especially there is the case of low-density tissue to will appear apparent dose difference afterwards before conversion on target area, at this time
With regard to needing manually to participate in careful adjusting parameter.
Method based on artificial intelligence: medical treatment of connecting with the heart proposes the artificial intelligence based on cloud in disclosed patent and puts
Planning system is treated, using Optimum distribution formula algorithmization cloud radiotherapy planning；In the scheme, patient is hopeful the mould from image room
It is come in the time of therapeutic room under quasi- localization machine, the radiotherapy automated planning system that AI energizes has completed the putting of the patient
Plan is treated, and completes the automatic Program Assessment of cloud, and is controlled by cloud radiotherapy quality, is formed by Program Assessment report, quality is protected
Card report and radiotherapy planning can be transmitted to doctor workstation by cloud, can be used to treat after finally auditing for doctor.
The advantages of method based on artificial intelligence: it is one really based on the radiotherapy planning of artificial intelligence, greatly reduces
The artificial dependence of radiotherapy planning；And proposing artificial intelligence needs Program Assessment and quality assurance to carry out assisting to ensure manually
The reliability of intelligent predicting.The shortcomings that method based on artificial intelligence: the radiotherapy planning based on cloud needs to consider the guarantor of image
There is public cloud and challenge in close property, unless being applied to the private clound of hospital internal, and want to the speed and stability of network
Ask relatively high.
For planning automatically, other than itself generates treatment plan, there can also be following application: will be intended to be automatically
The QA and checking tool of plan；It is (a kind of to treat mould for the unbiased comparison of different treatment technologies, such as IMRT, VMAT, SBRT
Formula) etc.；For Treatment decsion, such as treatment means selection and progress personalized treatment, it can select which patient is suitble to matter
Son treatment, and which patient is enough good as long as photon therapy, reduces the unreasonable scheduling of medical resource；For adaptive radiotherapy
Deng.
The dosage for being currently based on AI technology plans to be more heavily weighted toward progress dose prediction first automatically, then uses specifically certainly
Dynamic optimization adjusts the mode of ginseng to carry out.And AI training process itself is a the processes of parameter optimization, theoretically from dosage to
The generation of plan should also be and can train.So the generation from patient image to treatment plan can should directly be trained
It completes.Reducing the uncertainty that unnecessary intermediate steps introduce is beneficial for the generation for the treatment of plan.And it directly instructs
The time for practicing the plan of generation goes out the time more much shorter planned compared to Automatic Optimal after dose prediction.
Summary of the invention
Aiming at the shortcomings existing in the above problems, the present invention provides the method for artificial intelligence guidance radiotherapy planning and is
System.
An object of the present disclosure is to provide a kind of method of artificial intelligence guidance radiotherapy planning, comprising:
Obtain CT image；
The ROI region of the CT image is delineated automatically；By the CT image and automatically the ROI region input delineated
Into prediction model, automatic Prediction goes out dosage distribution or DVH；
The distribution of the dosage of prediction or DVH are distributed to as reference dose or are referred to DVH, using based on dosage distribution or DVH
The reverse optimization algorithm of guidance optimizes processing, generates executable radiotherapy planning；The executable radiotherapy planning includes forward direction
Radiotherapy planning, stereotactic radiotherapy plan and intensity modulated radiation therapy plan, wherein the intensity modulated radiation therapy plan includes dynamic intensity-modulated radiation therapy
Plan, static intensity modulating radiotherapy planning, the strong radiotherapy planning of adjustment with volume and rotation intensity modulated radiation therapy plan.
As a further improvement of the present invention, the acquisition CT image, comprising:
CT image is obtained by CT radiotherapy simulative generator or MR radiotherapy simulative generator, image archiving is established and communication system carries out shadow
As management, is stored and transmitted with Dicom；
The ROI region to the CT image is delineated automatically, comprising:
The automatic identification of normal organ with delineate automatically: delineate each organ of Whole Body automatically based on machine learning；
The automatic identification of tumor locus with delineate: delineated if systemic organs can be carried out, tumour is using reversely delineating；It will
Jeopardize after the completion of organ delineates, remainder is tumor locus；It is (swollen using the disease PTV (plan field) and GTV of machine learning
Tumor target area) the relationship extended out, remainder is delineated automatically.
As a further improvement of the present invention, the construction method of the prediction model are as follows:
Case is collected and automatic screening；
Construction considers the automatic dosage prediction model of beam (beam) angle；
Artificial intelligence (AI) training of case；
Obtain prediction model.
As a further improvement of the present invention, described to be distributed or join using the distribution of the dosage of prediction or DVH as reference dose
DVH is examined, processing is optimized using the reverse optimization algorithm based on dosage distribution or DVH guidance, generates executable radiotherapy planning；
Include:
Based on flux pattern optimization algorithm (Fluence Map Optimization, abbreviation FMO), optimization flux weight
Figure；Then, executable dynamic intensity-modulated radiation therapy plan is automatically generated using blade sequence algorithm in conjunction with the machine information of accelerator；
Or,
Based on direct Ziye optimization method (Direct Aperture Optimization, abbreviation DAO), automatically generating can
Execute static intensity modulating radiotherapy planning；Or,
Based on genetic algorithm, perhaps col-generating arithmetic automatically generates the strong radiotherapy planning of adjustment with volume or rotation intensity modulated radiation therapy meter
It draws；Or,
Positive radiotherapy planning；Or,
Stereotactic radiotherapy plan.
As a further improvement of the present invention, further includes:
Through unified prescription standard in conjunction with artificial intelligence, scores executable radiotherapy planning generated, obtain
To the scoring total score of Program Assessment；
By Monte Carlo 3-dimensional dose verification technique, 2D or 3D is carried out to executable radiotherapy planning generated
Gamma analysis, obtains the percent of pass situation of Gamma analysis；
The percent of pass situation of scoring total score and Gamma analysis based on executable radiotherapy planning, Program Assessment is given birth to automatically
It is reported at radiotherapy planning；
Doctor audits radiotherapy planning report.
The system that the present invention second is designed to provide a kind of artificial intelligence guidance radiotherapy planning, comprising:
Patient's CT module, for obtaining CT image；
Automatically module is delineated, is delineated automatically for the ROI region to the CT image；
Automatic dosage prediction module, the ROI region for delineating by the CT image and automatically are input in prediction model,
Automatic Prediction goes out dosage distribution or DVH；
Automatic Inverse Planning module, dosage distribution or DVH for that will predict are distributed as reference dose or refer to DVH,
Processing is optimized using the reverse optimization algorithm based on dosage distribution or DVH guidance, generates executable radiotherapy planning；It is described can
Executing radiotherapy planning includes positive radiotherapy planning, stereotactic radiotherapy plan and intensity modulated radiation therapy plan, wherein the intensity modulated radiation therapy
Plan includes dynamic intensity-modulated radiation therapy plan, static intensity modulating radiotherapy planning, the strong radiotherapy planning of adjustment with volume and rotation intensity modulated radiation therapy plan.
As a further improvement of the present invention, the patient CT module, is used for:
CT image is obtained by CT radiotherapy simulative generator or MR radiotherapy simulative generator, image archiving is established and communication system carries out shadow
As management, is stored and transmitted with Dicom；
It is described to delineate module automatically, it is used for:
The automatic identification of normal organ with delineate automatically: delineate each organ of Whole Body automatically based on machine learning；
The automatic identification of tumor locus with delineate: delineated if systemic organs can be carried out, tumour is using reversely delineating；It will
Jeopardize after the completion of organ delineates, remainder is tumor locus；It is (swollen using the disease PTV (plan field) and GTV of machine learning
Tumor target area) the relationship extended out, remainder is delineated automatically.
As a further improvement of the present invention, in the automatic dosage prediction module, the building side of the prediction model
Method are as follows:
Case is collected and automatic screening；
Construction considers the automatic dosage prediction model of beam (beam) angle；
Artificial intelligence (AI) training of case；
Obtain prediction model.
As a further improvement of the present invention, the automatic Inverse Planning module, is used for:
Based on flux pattern optimization algorithm (Fluence Map Optimization, abbreviation FMO), optimization flux weight
Figure；Then, executable dynamic intensity-modulated radiation therapy plan is automatically generated using blade sequence algorithm in conjunction with the machine information of accelerator；
Or,
Based on direct Ziye optimization method (Direct Aperture Optimization, abbreviation DAO), automatically generating can
Execute static intensity modulating radiotherapy planning；Or,
Based on genetic algorithm, perhaps col-generating arithmetic automatically generates the strong radiotherapy planning of adjustment with volume or rotation intensity modulated radiation therapy meter
It draws；Or,
Positive radiotherapy planning；Or,
Stereotactic radiotherapy plan.
As a further improvement of the present invention, further includes:
Automatic Program Assessment module, for by unified prescription standard in conjunction with AI, to executable radiotherapy generated
Plan is scored；
Automatic 3D QA module, for passing through Monte Carlo 3-dimensional dose verification technique, to executable radiotherapy meter generated
It draws and carries out 2D 3D Gamma analysis, obtain the percent of pass situation of Gamma analysis；
Report generation module, for based on executable radiotherapy planning, Program Assessment scoring total score and Gamma analysis
Percent of pass situation automatically generates radiotherapy planning report；
Doctor's audit report module, for being audited to radiotherapy planning report.
Compared with prior art, the invention has the benefit that
Manpower is saved in less manual intervention during the present invention makes a plan；The plan quality of generation is high, meets industry
Standard or national standard；Plan is quickly generated, patient is benefited；With preferable accuracy, stability and normalization, so as to
It is enough to improve medical software and hardware resources utilization rate, solve the problems, such as the high-level radiotherapy planning of the more difficult formulation of infirmary.
Detailed description of the invention
Fig. 1 is the flow chart for the method that artificial intelligence disclosed in an embodiment of the present invention guides radiotherapy planning；
Fig. 2 is the frame diagram for the system that artificial intelligence disclosed in an embodiment of the present invention guides radiotherapy planning；
Fig. 3 is the mapping dictionary list of standard name and alias disclosed in an embodiment of the present invention；
Fig. 4, which is that an embodiment of the present invention is disclosed, assesses software prototype figure；
Fig. 5 is CGAN schematic diagram disclosed in an embodiment of the present invention；
Fig. 6 is data prediction flow chart disclosed in an embodiment of the present invention；
Fig. 7 is beam data statistics figure disclosed in an embodiment of the present invention；
Fig. 8 is beam data screening figure disclosed in an embodiment of the present invention；
Fig. 9 is the disclosed effect picture predicted using GAN network of an embodiment of the present invention；
Figure 10 is DVH comparison result figure disclosed in an embodiment of the present invention；
Figure 11, which is that an embodiment of the present invention is disclosed, simulates H/N tumors cross section.
In figure:
1, patient CT module；2, module is delineated automatically；3, automatic dosage prediction module；4, automatic Inverse Planning module；5,
Automatic Program Assessment module；6, automatic 3D QA module；7, report generation module；8, doctor's audit report module.
Specific embodiment
In order to make the object, technical scheme and advantages of the embodiment of the invention clearer, below in conjunction with the embodiment of the present invention
In attached drawing, technical scheme in the embodiment of the invention is clearly and completely described, it is clear that described embodiment is
A part of the embodiments of the present invention, instead of all the embodiments.Based on the embodiments of the present invention, ordinary skill people
Member's every other embodiment obtained without making creative work, shall fall within the protection scope of the present invention.
The present invention provides method, system, equipment and the storage medium of artificial intelligence guidance radiotherapy planning, by using depth
Learning art, and based on automatic program evaluation and review technique and Monte Carlo 3-dimensional dose verification technique, hundreds and thousands of examples are obtained automatically
High quality radiotherapy planning data be training sample, carry out machine learning, finally automatically derive the radiotherapy planning of high quality, and
The outstanding property of automatic Program Assessment and Monte Carlo 3-dimensional dose verification technique confirmation machine learning achievement and reliable is utilized again
Property.Wherein:
AI technology has been energized this big technological innovation in field three, to substantially increase the effect of the production radiotherapy planning in this field
The curative effect of rate and treatment patient.
Automatic program evaluation and review technique: it through unified prescription standard in conjunction with AI, gives each radiotherapy planning and objectively comments
Point, the otherness of the Prescription evaluation of doctor is greatly reduced, to obtain more objective appraisal, it is ensured that the training of machine learning
Collection and learning outcome are high quality in terms of meeting prescription requirements, improve the outstanding property of training set and learning outcome, thus
Improve curative effect.
Monte Carlo 3-dimensional dose verification technique: Monte Carlo technique is the goldstandard of the Rapid Dose Calculation of radiotherapy, but existing
Radiotherapy planning used mostly based on non-homogeneous modified analytical algorithm (such as pencil beam, convolution superposition algorithm), these algorithms
At some in the works there may be biggish error, show that the Gamma percent of pass in 3-dimensional dose verifying is poor.If drawn
Enter the verifying of Monte Carlo 3-dimensional dose, improved by parallel algorithms such as GPU, can be used for clinical 3-dimensional dose verifying, will cover
The analysis of the Rapid Dose Calculation Comparative result of special Caro Rapid Dose Calculation result and existing radiotherapy planning obtains Gamma percent of pass, chooses high
The radiotherapy planning of Gamma percent of pass is used as training sample, it is ensured that it is high-quality that the sample of machine learning, which meets dosage accuracy aspect,
Amount, improve the reliability of training set and learning outcome.
Automatic plan forecast technology: AI establishes deep learning neural network model, passes through in conjunction with radiotherapy planning data set
After deep learning method is trained outstanding radiotherapy planning data set, the automatic planning model of high quality can produce, utilize
The medical image that the model can be inputted according to user carries out radiotherapy planning prediction, and passes through automatic Program Assessment and Monte Carlo three
Tie up the outstanding property and reliability of dosage verifying technology confirmation machine learning achievement.Above-mentioned whole process is completed under unmanned intervene,
When only taking up computer machine, substantially increase production radiotherapy planning efficiency, thus greatly reduce patient etc. it is to be treated when
Between, curative effect is also improved indirectly.
Dosage based on AI plan to aim to solve the problem that automatically present physical teacher on existing treatment planning systems (TPS) into
The problem of needing continuously to manually adjust the parameters of constraint and target item in the calculating process of row reverse optimization.In order to
Meet clinical, laws and regulations requirement and the enforceability in view of plan, physics teacher need constantly to modify these parameters to improve PTV
The dosage of (target area) is distributed coverage rate, while the dosage for avoiding OAR (jeopardizing organ) from receiving is excessively high.This process it is very time-consuming and
Very strong to the dependence of physics teacher's experience, adjusting target component every time can all have an impact to result next time, lead to last life
At the difference of plan may be very big.Due to process time-consuming, physics teacher is often high-quality to generate without time enough and energy
Amount, so that the efficiency that plan generates is very low.And plan is automatically generated by AI, it is done in whole process almost without artificial
Pre- process, the efficiency for generating plan is just very high, and can reduce the introducing of human error, avoids the people with different experiences
Member's problem big in the plan difference that the same case is made into.
For hospital, present physics Shi Yitian does 1 to 3 plans, it is also possible to can just do within several days a plan.Using
Intelligence plan can do more than ten to tens with one day, and hospital can thus be allowed to guarantee that patient obtains within the shorter time
Preferably while treatment, higher income is obtained.Because if the plan of patient delays the long period and works it out, cancer cell is such as
How what transfer and the state of an illness, which can deteriorate, is all difficult to expect, control of the timeliness of traditional radiotherapy planning to patient's state of an illness
Guarantee with survival rate just seems very unfavorable.
Dosage based on AI is planned " to adequately protect to health tissues and right in data preprocessing phase in order to reach automatically
Tumor region is sufficiently destroyed " Management Treatment Plan target, it will usually have hospital internal formulate unified standard, the world RTOG mark
Quasi- or relevant other industry standard is standardized.According to these standards, the case screen selecting formwork of specific disease is formulated.It opens
It sends out the treatment plan assessment software based on correlate template and unified screening is carried out to history case, the treatment meter after capable of guaranteeing screening
Drawing all is the high quality plan for meeting relevant criterion.For the sake of prudent, need again using objective Program Assessment standard pair
The plan of AI prediction carries out automatic assessment screening and the verifying of automatic Monte Carlo dose.
The present invention is described in further detail with reference to the accompanying drawing:
As shown in Figure 1, the present invention provides a kind of method of artificial intelligence guidance radiotherapy planning, comprising:
S1, CT image is obtained；Wherein:
Disease is obtained by CT radiotherapy simulative generator or MR radiotherapy simulative generator (have MR image be converted to CT image function)
People's CT image is established PACS (image archiving and communication system) Lai Shixian patient image management, is stored and passed with Dicom
It is defeated.
S2, the ROI region of CT image is delineated automatically；Wherein:
Automatic Target delineations based on deep learning (Deep Learning), it can be achieved that:
1, it the automatic identification of normal (jeopardizing) organ and delineates automatically: machine learning can be based on and delineate Whole Body automatically
Each organ；
2, the automatic identification of tumor locus with delineate: delineated if systemic organs can be carried out, tumour using reversely delineating,
It will jeopardize after the completion of organ delineates, remainder is tumor locus, using the pass of the disease PTV and GTV of machine learning extended out
System, delineates remaining position automatically.
S3, the ROI region delineated by CT image and automatically are input in prediction model, automatic Prediction go out dosage distribution or
DVH；Wherein:
1, construction considers the automatic dosage prediction model of beam angle
If can consider the difference that dosage caused by beam angle is different is distributed, trained and prediction accuracy can be improved；
Solve the problems, such as case source deficiency；The algorithm is also the key problem for solving automatic dosage prediction universality.Possible solution
Scheme is that prediction and training all consider beam angle.Specifically there are several types of schemes.
Scheme one:
Using beam angle as a priori conditions when training and prediction, case is labeled, using with item
The machine learning model of part probability is trained and predicts that user inputs a customized angle beam when being predicted
Degree combination (being multiple beam angles, referred to herein as a beam angle combinations because a plan has multiple beam) is i.e.
It can.If input of the user without beam angle, system default can provide one group of recommendation beam angle automatically.The recommendation
It is one group of best beam angle being found using the clustering method in machine learning according to same disease case.It can also be according to
According to the anatomical features of history case and image, it is established that prediction model, according to the CT image of input, automatic Prediction is suitable out
Beam angle combinations.The model that can be referred to, which is had ready conditions, generates confrontation network C GAN, and what is had been carried out at present is original life
At confrontation network model.CGAN schematic diagram is as shown in Figure 5.
Scheme two:
Data prediction is carried out to the CT (and RS) of input before training and prediction.Here there are two types of may for bracket representative
Resolving ideas, include RS processing or processing not comprising RS.By taking the processing comprising RS as an example, to CT according to beam angle
(and the MU weight of existing case in the works, equally for comprising MU weight) around some axis, (recommendation makes to CT data
The axis rotated with gantry) it rotated, cut, position, it then synthesizes, synthesis is to be obtained using MU as weight using gravity model appoach
The CT data of synthesis, if not considering MU, being all set to weight is 1.Finally reconstruct virtual CT.It is then to seat to RS
Mark is rotated, and final coordinate is generated also according to gravity model appoach to the coordinate of rotation, weight refers to the synthesis mode of CT.Process
Figure is as shown in fig. 6, left figure represents the process of data preprocessing for not considering beam angle；There are three be sliced to be with every set CT for right figure
Example illustrates the process rebuild.
Scheme three:
For the case plan of different beam angles, is screened out using pretreated mode and deviate main beam combination
Plan is given it up, and is filtered out and is screened according to the formula proposed as follows to the history case plan being collected into.
It is screened according to the standard of 1 times of standard deviation, the case exceeded is not used in training.It is remaining to may act as instructing
Practice collection.This mode, which reduces beam angle difference to a certain extent and is distributed on predicted dose, to be influenced, but is subtracted to a certain extent
Lack and can be used for trained case.It is exemplified below with 20 sets of plan cases.All beam are extracted from 20 sets of cases first
Angle.
Wherein every a line represents the angle array of a plan.Array length is determined according to beam number is most, Mei Geji
The beam number drawn is recorded in advance with a variable.Beam data are counted first, obtain figure as shown in Figure 7
Picture.
According to the distribution situation of beam number, learn that the more for the treatment of is 8 beam.Frequency is 14.This 14
It is intended to be the plan screened, using screening formula.Obtain result as shown in Figure 8；That is, three horizontal lines divide from top to bottom
Average value 1 times of standard deviation bigger than normal, average value and average value 1 times of standard deviation less than normal are not represented.Here it is considered that between 1 times of standard deviation model
It is the close case of beam angle in enclosing, can then excludes 2 cases, remaining 12 sets of plan cases can be used for instructing
Practice.
2, the solution that prescribed dose difference influences trained and prediction accuracy
Other than Beam angle difference causes case limited source and prediction accuracy uncontrollable, prescribed dose difference
It may cause these situations.It is possible that solution be that DVH is normalized according to prescribed dose.It can thus gather around
There is more extensive case source.
3, automatic dosage is predicted
Automatic dosage prediction carries out case collection by assessing automatically, (falls ill early advanced stage, prescription for disease classification
Dosage, cancer site etc.) it delineates then by CT using unified standard template progress case screening and dosage is trained together,
Prediction model is generated, to carry out dose prediction, the input of prediction is CT and delineating (RS) automatically based on CT, the output of prediction
As a result the DVH generated in turn for the corresponding slice dosage of each CT sectioning image or thus.
Prediction effect is shown
An effect picture for using GAN network to be predicted neck case below, as shown in Figure 9.
It can be seen that the prediction result in the part PTV is marginally better than baseline results.Certainly it also needs largely to be tested.
S4, the distribution of the dosage of prediction or DVH as reference dose or are referred into DVH, drawn using based on dosage distribution or DVH
The reverse optimization algorithm led optimizes processing, generates executable radiotherapy planning；Executable radiotherapy planning includes positive radiotherapy meter
Draw, stereotactic radiotherapy plan and intensity modulated radiation therapy plan, wherein the intensity modulated radiation therapy plan include dynamic intensity-modulated radiation therapy plan, it is quiet
The plan of state intensity modulated radiation therapy, the strong radiotherapy planning of adjustment with volume and rotation intensity modulated radiation therapy plan；Wherein:
Based on flux pattern optimization algorithm (Fluence Map Optimization, abbreviation FMO), optimization flux weight
Figure；Then, executable dynamic intensity-modulated radiation therapy plan is automatically generated using blade sequence algorithm in conjunction with the machine information of accelerator；
Based on direct Ziye optimization method (Direct Aperture Optimization, abbreviation DAO), automatically generating can
Execute static intensity modulating radiotherapy planning；
Based on genetic algorithm, perhaps col-generating arithmetic automatically generates the strong radiotherapy planning of adjustment with volume or rotation intensity modulated radiation therapy meter
It draws；
Or positive radiotherapy planning；
Or stereotactic radiotherapy plan.
It is specific:
Automatic Inverse Planning is exactly to be distributed the distribution of the dosage of above-mentioned prediction or DVH as reference dose or with reference to DVH, adopt
With based on dosage distribution or DVH guidance reverse optimization algorithm, in conjunction with the voxel agent of the calculating by specific dose calculation engine
Unit is measured, optimization flux weight map is automatically derived most in conjunction with the machine information of particular-accelerator using blade sequence algorithm
Whole treatment plan；Wherein optimization algorithm can use linear model, can also be realized using nonlinear model.
Automatic Program Assessment is called in generated plan automatically --- including every score and total score, it is supplied to doctor crowd
Standard uses；
By taking the linear model of Automatic Optimal model as an example:
Planning optimization engine constitutes optimization of IMRT inverse planning problem using a series of linear objective functions:
For each OAR setting dosage maximum value objective function, dosage average value objective function, equivalent uniform target letter
Number；
Maximum value objective function:
Average value objective function:
Equivalent uniform objective function:
For each PTV setting dosage maximum value objective function, deviate prescribed dose degree objective function；
Maximum value objective function:
Deviate prescribed dose degree target:
Ensure to plan enforceability, sets Fluence Map smoothness constraint objective function；
Smooth objective function:
Entire optimization of IMRT inverse planning problem can linearly be expressed as following formula:
……
……
……
Wherein γiβiκiβtφt It is exactly the weight parameter of each objective function, that is, physics teacher needs again and again
The parameter of adjustment, lucky optimization of IMRT inverse planning model can according to the Dose distribution data optimization of AI dose prediction model prediction this
A little parameters, that is, AI model learn from history case to how planned design.Above-mentioned optimization of IMRT inverse planning problem table
Reach the form of matrix:
Subject to Ax >=b,
x≥0.
α is weight parameter vector
C is objective function expression matrix
X is decision variable
G is smooth objective function expression matrix
A is constraint condition factor matrix
B is restrained boundary
Here is the dual problem of former Reverse Problem:
Subject to C ' α+g >=A ' p,
p≥0.
P is the dual variable of former problem constraint condition
So former problem can be become the form of an absolute duality gap of optimization to optimization α:
Subject to C ' α+g >=A ' p,
α >=0, p >=0.
It is the value of each objective function, these values can learn to obtain by AI dose prediction engine；
It is a constant for a clinical program, optimization problem is not contributed, can be cast out.
Last problem becomes:
Subject to C ' α+g >=A ' p,
α > 0, p >=0.
Obviously last Optimized model is the linear programming problem of a standard, the value of model energy optimization, in value generation
Enter to inside the optimization of IMRT inverse planning model of most original:
Subject to Ax >=b,
x≥0.
As soon as α it is known that model above is the linear programming problem of a standard again, also can optimization x (Fluence map),
Ultimately produce the plan of high quality.
Automatic plan prototype test effect
For the neck case in national standard YY0889 standard, test of heuristics has been carried out.Basis without manual intervention
On can Automatic Optimal go out to meet the plan of laws and regulations requirement.Used time amounts at 5 minutes or so.What user uniquely needed to input is place
The 3-dimensional dose data of square dose requirements and prediction.
Final DVH comparison result is as shown in Figure 10.Solid line and dotted line respectively represent the DVH of artificial optimization and automatic reverse
The DVH of optimization, the prediction DVH that wherein Automatic Optimal uses.
The constraint of example in national standard and target call are as shown in figure 11.
Plan automatically as seen from Figure 11 the DVH result provided than the change manually done the part of PTV slightly almost,
But also it is satisfactory plan.And plan only to need general 5 minutes or so automatically, and manually plan at least to want one small
When.
S5, it is combined by unified prescription standard and artificial intelligence (Artificial Intelligence, abbreviation AI),
It scores executable radiotherapy planning generated, obtains the scoring total score of Program Assessment；Wherein:
1, for the standardization of specific disease name
Structure alias is delineated as Uniform Name in order to guarantee to automatically extract when training preprocessed data, needs to formulate
The mapping dictionary list of standard name and alias, as shown in Figure 3；
2, automatic Program Assessment
Automatically the premise planned is that have outstanding plan database.Therefore, it is necessary to the outstanding plans of an automatic screening to calculate
Method and tool.The tool can both give plan Select to use, carry out scoring use, such as Fig. 4 after can also generating to automatic plan
It is shown；
Assessment software is positioned at multi-functional information and data-management application, is used to help doctor and physics teacher to radiotherapy meter
Drawing is improved and is promoted normalization, meets international standard for what is intelligently planned to filter out according to history case here
Or the case of specification.The input of application program is DICOM data derived from treatment planning systems (TPS), including RS, RP, RD,
Tetra- part CT.Output is appraisal result.
It is above-mentioned to be selected for our independently developed assessment software prototypes for carrying out scoring screening to existing plan case
The higher and lower case of score is respectively used to machine learning training and prediction test out.
Scoring template needed for assessment software is all that each hospital's uniform rules oneself is formulated, even for different diseases
More specifically classification, can formulate different templates.It can be with reference to the inside of international standard and hospital oneself mark when formulation
It is quasi-.
S6, pass through Monte Carlo 3-dimensional dose verification technique, 2D or 3D is carried out to executable radiotherapy planning generated
Gamma analysis, obtains the percent of pass situation of Gamma analysis；Wherein:
Generated plan is analyzed using based on the 3D Gamma for covering card QA automatically；Provide the percent of pass feelings of Gamma analysis
Condition, for being made reference to doctor's finally approval plan.It is promoted for accurate radiotherapy.
It is twentieth century four using Monte Carlo method (Monte Carlo method), also referred to as statistical simulation methods
Ten mid-nineties 90s are due to the development of science and technology and the invention of electronic computer, and the one kind being suggested is with Probability Statistics Theory
The very important numerical computation method of one kind of guidance.Refer to using random number (or more common pseudo random number) and solves very much
The method of computational problem.Corresponding with it is deterministic algorithm.Monte Carlo method is in financial engineering, macroeconomics,
Computational physics (such as PARTICLE TRANSPORT FROM calculating, quantum calculation of thermodynamics, aerodynamics calculate) field is widely used.It is radiating
Field, which is used widely, originates from the Manhattan project of Americanized atom bomb in World War II.
Monte Carlo (MC) analogy method rely on its PARTICLE TRANSPORT FROM process and Geometric Modeling accuracy, medical physics with
Radiation dosimetry field plays an important role.Rapid Dose Calculation important role in the formulation of radiotherapy planning, MC simulation with
It is referred to as the goldstandard of Rapid Dose Calculation by means of its accuracy.Common MC software for calculation mainly include EGS4/5, EGSnrc, MCNP,
PENELOPE, GEANT4 etc., but the speed of MC simulation and time-consuming problem become the principal element for limiting its clinical application, such as exist
, be in the case where meeting 2.5% uncertainty degree during actual clinical target dose calculates, the conventional MC dosage based on CPU, which is simulated, to be calculated
6h need to be spent.Therefore using photon, the electronics, matter of the method difference speeding-up simulation different-energy range of multi-core CPU parallel computation
Son transports, and becomes the important research direction that the simulation of MC dosage calculates.It has developed simultaneously some by simplifying physics mistake
The MC program that Cheng Jinhang operation accelerates, such as VMC++, MCDOSE/MCSIM, dose plan method (Dose Planning
Method, DPM) although etc. above method speed that MC is calculated constantly promoted, required be similar to reality apart from clinical
When MC calculate there are also very big development space.Graphics processor (GPU) due to its numerous arithmetic element is parallel, high memory bandwidth,
The advantages that cost for supporting floating number algorithm, unit to calculate is low, routine interface is open has more wide in MC analogue technique field
Wealthy application prospect.For this purpose, medical treatment of connecting with the heart develops the quick MC program based on GPU, clinical disease can be realized in several minutes
The 3-dimensional dose of example calculates.Compared to the Dose calculation algorithm of analytic expression, Monte-carlo Simulation Method is in uneven density region
It is high to calculate accuracy, without additional experiential modification, is suitble to carry out the QA verifying of 3-dimensional dose in clinic.
The percent of pass situation of S7, the scoring total score based on executable radiotherapy planning, Program Assessment and Gamma analysis are automatic
Generate radiotherapy planning report；
S8, doctor audit radiotherapy planning report；Wherein:
Audit report has abstract and details content.Abstract describes the percent of pass situation of the Gamma analysis of 3D QA, meter
The scoring total score of assessment is drawn, and is ready to carry out the abstract of plan；Being discussed in detail for each content in detail, facilitates doctor to this
Automatic plan carefully audit.If approved, it is published on medical accelerator and is ready to carry out；If examination & approval do not pass through,
Manpower intervention is allowed to carry out manual modification plan.
Further, the method that automatic dosage is predicted in S3 of the present invention are as follows:
A) case collection and automatic screening；
B) construction considers the automatic dosage prediction model of beam (beam) angle；
C) artificial intelligence (AI) training of case；
D) the automatic dosage prediction of case.
Above-mentioned automatic dosage prediction steps are alternatively predicted at: automatic dosage volume histogram (DVH)
Method 1: the DVH prediction based on machine learning:
A) case is collected and is screened；
B) construction considers the automatic dosage prediction model of beam (beam) angle；
C) artificial intelligence (AI) training of case；
D) the DVH prediction of case.
Method 2: the DVH prediction based on statistical method:
A) case is collected and is screened；
B) the DVH statistics of case；
C) the DVH prediction of case.
Method 3: the DVH prediction based on template
A) the goal constraint item initial using preset template generation；
B) algorithm adds bound term automatically, adjusts weight；
C) algorithm adds accessory organ automatically；
The DVH of case is predicted.
As shown in Fig. 2, the present invention provides a kind of system of artificial intelligence guidance radiotherapy planning, comprising:
Patient CT module 1, for realizing above-mentioned S1；
Automatically module 2 is delineated, for realizing above-mentioned S2；
Automatic dosage prediction module 3, for realizing above-mentioned S3；
Automatic Inverse Planning module 4, for realizing above-mentioned S4；
Automatic Program Assessment module 5, for realizing above-mentioned S5；
Automatic 3D QA module 6, for realizing above-mentioned S6；
Report generation module 7, for realizing above-mentioned S7；
Doctor's audit report module 8, for realizing above-mentioned S8.
The present invention also provides a kind of calculating equipment, comprising:
One or more processors；
Memory；And
One or more programs, wherein one or more programs are stored in memory and are configured as by one or more
Processor executes, one or more programs include above-mentioned artificial intelligence guidance radiotherapy planning method instruction.
The present invention also provides a kind of computer readable storage medium for storing one or more programs, one or more journeys
Sequence includes instruction, and instruction is suitable for being loaded by memory and being executed the method for above-mentioned artificial intelligence guidance radiotherapy planning.
Advantages of the present invention are as follows:
Manpower is saved in less manual intervention during the present invention makes a plan；The plan quality of generation is high, meets industry
Standard or national standard；Plan is quickly generated, patient is benefited；With preferable accuracy, stability and normalization, so as to
It is enough to improve medical software and hardware resources utilization rate, solve the problems, such as the high-level radiotherapy planning of the more difficult formulation of infirmary.
These are only the preferred embodiment of the present invention, is not intended to restrict the invention, for those skilled in the art
For member, the invention may be variously modified and varied.All within the spirits and principles of the present invention, it is made it is any modification,
Equivalent replacement, improvement etc., should all be included in the protection scope of the present invention.
Claims (10)
1. a kind of method of artificial intelligence guidance radiotherapy planning characterized by comprising
Obtain CT image；
The ROI region of the CT image is delineated automatically；
By the CT image and automatically the ROI region delineated is input in prediction model, and automatic Prediction goes out dosage distribution or DVH；
The distribution of the dosage of prediction or DVH are distributed to as reference dose or are referred to DVH, is guided using based on dosage distribution or DVH
Reverse optimization algorithm optimize processing, generate executable radiotherapy planning；The executable radiotherapy planning includes positive radiotherapy
Plan, stereotactic radiotherapy plan and intensity modulated radiation therapy plan, wherein the intensity modulated radiation therapy plan include dynamic intensity-modulated radiation therapy plan,
The strong radiotherapy planning of static intensity modulating radiotherapy planning, adjustment with volume and rotation intensity modulated radiation therapy plan.
2. the method as described in claim 1, which is characterized in that the acquisition CT image, comprising:
CT image is obtained by CT radiotherapy simulative generator or MR radiotherapy simulative generator, image archiving is established and communication system carries out camera tube
Reason, is stored and is transmitted with Dicom；
The ROI region to the CT image is delineated automatically, comprising:
The automatic identification of normal organ with delineate automatically: delineate each organ of Whole Body automatically based on machine learning；
The automatic identification of tumor locus with delineate: delineated if systemic organs can be carried out, tumour is using reversely delineating；It will jeopardize
After the completion of organ is delineated, remainder is tumor locus；Using the relationship of the disease PTV and GTV of machine learning extended out, to surplus
Remaining part point is delineated automatically.
3. the method as described in claim 1, which is characterized in that the construction method of the prediction model are as follows:
Case is collected and automatic screening；
Construction considers the automatic dosage prediction model of beam angle；
The artificial intelligence training of case；
Obtain prediction model.
4. the method as described in claim 1, which is characterized in that described using the distribution of the dosage of prediction or DVH as reference dose
Distribution refers to DVH, optimizes processing using the reverse optimization algorithm based on dosage distribution or DVH guidance, generates executable
Radiotherapy planning；Include:
Based on flux pattern optimization algorithm, optimization flux weight map；Then, in conjunction with the machine information of accelerator, using blade sequence
Column algorithm automatically generates executable dynamic intensity-modulated radiation therapy plan；Or,
Based on direct Ziye optimization method, executable static intensity modulating radiotherapy planning is automatically generated；Or,
Based on genetic algorithm, perhaps col-generating arithmetic automatically generates the strong radiotherapy planning of adjustment with volume or rotation intensity modulated radiation therapy plan；
Or,
Positive radiotherapy planning；Or,
Stereotactic radiotherapy plan.
5. the method as described in claim 1, which is characterized in that further include:
Through unified prescription standard in conjunction with artificial intelligence, scores executable radiotherapy planning generated, counted
Draw the scoring total score of assessment；
By Monte Carlo 3-dimensional dose verification technique, 2D 3D Gamma points are carried out to executable radiotherapy planning generated
Analysis obtains the percent of pass situation of Gamma analysis；
The percent of pass situation of scoring total score and Gamma analysis based on executable radiotherapy planning, Program Assessment, which automatically generates, puts
Treat planning report；
Doctor audits radiotherapy planning report.
6. a kind of system of artificial intelligence guidance radiotherapy planning characterized by comprising
Patient's CT module, for obtaining CT image；
Automatically module is delineated, is delineated automatically for the ROI region to the CT image；
Automatic dosage prediction module, the ROI region for delineating by the CT image and automatically are input in prediction model, automatically
Predict dosage distribution or DVH；
Automatic Inverse Planning module, dosage distribution or DVH for that will predict are distributed or are referred to as reference dose DVH, use
Reverse optimization algorithm based on dosage distribution or DVH guidance optimizes processing, generates executable radiotherapy planning；It is described executable
Radiotherapy planning includes positive radiotherapy planning, stereotactic radiotherapy plan and intensity modulated radiation therapy plan, wherein the intensity modulated radiation therapy plan
Including dynamic intensity-modulated radiation therapy plan, static intensity modulating radiotherapy planning, the strong radiotherapy planning of adjustment with volume and rotation intensity modulated radiation therapy plan.
7. system as claimed in claim 6, which is characterized in that the patient CT module is used for:
CT image is obtained by CT radiotherapy simulative generator or MR radiotherapy simulative generator, image archiving is established and communication system carries out camera tube
Reason, is stored and is transmitted with Dicom；
It is described to delineate module automatically, it is used for:
The automatic identification of normal organ with delineate automatically: delineate each organ of Whole Body automatically based on machine learning；
The automatic identification of tumor locus with delineate: delineated if systemic organs can be carried out, tumour is using reversely delineating；It will jeopardize
After the completion of organ is delineated, remainder is tumor locus；Using the relationship of the disease PTV and GTV of machine learning extended out, to surplus
Remaining part point is delineated automatically.
8. system as claimed in claim 6, which is characterized in that in the automatic dosage prediction module, the prediction model
Construction method are as follows:
Case is collected and automatic screening；
Construction considers the automatic dosage prediction model of beam angle；
The artificial intelligence training of case；
Obtain prediction model.
9. system as claimed in claim 6, which is characterized in that the automatic Inverse Planning module is used for:
Based on flux pattern optimization algorithm, optimization flux weight map；Then, in conjunction with the machine information of accelerator, using blade sequence
Column algorithm automatically generates executable dynamic intensity-modulated radiation therapy plan；Or,
Based on direct Ziye optimization method, executable static intensity modulating radiotherapy planning is automatically generated；Or,
Based on genetic algorithm, perhaps col-generating arithmetic automatically generates the strong radiotherapy planning of adjustment with volume or rotation intensity modulated radiation therapy plan；
Or,
Positive radiotherapy planning；Or,
Stereotactic radiotherapy plan.
10. system as claimed in claim 6, which is characterized in that further include:
Automatic Program Assessment module, for by unified prescription standard in conjunction with AI, to executable radiotherapy planning generated
It scores；
Automatic 3D QA module, for by Monte Carlo 3-dimensional dose verification technique, to executable radiotherapy planning generated into
Row 2D or 3D Gamma analysis, obtain the percent of pass situation of Gamma analysis；
Report generation module, for based on executable radiotherapy planning, Program Assessment scoring total score and Gamma analysis pass through
Rate situation automatically generates radiotherapy planning report；
Doctor's audit report module, for being audited to radiotherapy planning report.
Priority Applications (3)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910820650.3A CN110415785A (en) | 2019-08-29 | 2019-08-29 | The method and system of artificial intelligence guidance radiotherapy planning |
| US16/977,095 US11964170B2 (en) | 2019-08-29 | 2020-05-22 | Standardized artificial intelligence automatic radiation therapy planning method and system |
| PCT/CN2020/091843 WO2021036366A1 (en) | 2019-08-29 | 2020-05-22 | Standardized artificial intelligence automatic radiotherapy planning method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910820650.3A CN110415785A (en) | 2019-08-29 | 2019-08-29 | The method and system of artificial intelligence guidance radiotherapy planning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110415785A true CN110415785A (en) | 2019-11-05 |
Family
ID=68369527
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910820650.3A Pending CN110415785A (en) | 2019-08-29 | 2019-08-29 | The method and system of artificial intelligence guidance radiotherapy planning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110415785A (en) |
Cited By (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111028914A (en) * | 2019-12-04 | 2020-04-17 | 北京连心医疗科技有限公司 | Artificial intelligence guided dose prediction method and system |
| CN111790065A (en) * | 2020-08-11 | 2020-10-20 | 北京易康医疗科技有限公司 | Multi-beam combined radiotherapy method for tumor |
| WO2021036366A1 (en) * | 2019-08-29 | 2021-03-04 | 北京连心医疗科技有限公司 | Standardized artificial intelligence automatic radiotherapy planning method and system |
| CN112447278A (en) * | 2020-11-17 | 2021-03-05 | 北京全域医疗技术集团有限公司 | Patient plan data verification method and device and computer equipment |
| CN112597702A (en) * | 2020-12-21 | 2021-04-02 | 电子科技大学 | Pneumatic modeling generation type confrontation network model training method based on radial basis function |
| CN112618967A (en) * | 2020-12-17 | 2021-04-09 | 程明霞 | Device for adjusting radiotherapy dose distribution and generating corresponding radiotherapy plan |
| CN112972912A (en) * | 2021-04-22 | 2021-06-18 | 北京大学第三医院（北京大学第三临床医学院） | Non-coplanar radiotherapy beam incident path optimization method |
| CN113241151A (en) * | 2021-02-25 | 2021-08-10 | 中科超精(南京)科技有限公司 | Intelligent radiotherapy cloud system and implementation method |
| CN113517072A (en) * | 2021-09-13 | 2021-10-19 | 四川大学 | Prediction method of VMAT radiotherapy plan based on deep neural network |
| CN113516233A (en) * | 2021-09-13 | 2021-10-19 | 四川大学 | Neural network prediction method for VMAT radiotherapy plan |
| CN114864070A (en) * | 2022-04-28 | 2022-08-05 | 福建自贸试验区厦门片区Manteia数据科技有限公司 | Radiotherapy information management system |
| CN115829972A (en) * | 2022-12-02 | 2023-03-21 | 中山大学肿瘤防治中心（中山大学附属肿瘤医院、中山大学肿瘤研究所） | Radiotherapy plan three-dimensional dose distribution and flux synchronous prediction method and device |
| CN116153465A (en) * | 2023-04-24 | 2023-05-23 | 福建自贸试验区厦门片区Manteia数据科技有限公司 | Radiotherapy plan generation device based on reference DVH, electronic equipment and storage medium |
| CN115829972B (en) * | 2022-12-02 | 2024-04-19 | 中山大学肿瘤防治中心（中山大学附属肿瘤医院、中山大学肿瘤研究所） | Synchronous prediction method and device for three-dimensional dose distribution and flux of radiotherapy plan |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105930636A (en) * | 2016-03-29 | 2016-09-07 | 中北大学 | Radiotherapy scheme optimization system capable of automatically determining weight of target function |
| CN107403201A (en) * | 2017-08-11 | 2017-11-28 | 强深智能医疗科技（昆山）有限公司 | Tumour radiotherapy target area and jeopardize that organ is intelligent, automation delineation method |
| CN108717866A (en) * | 2018-04-03 | 2018-10-30 | 陈辛元 | A kind of method, apparatus, equipment and the storage medium of the distribution of prediction radiotherapy planning dosage |
| CN109801696A (en) * | 2017-11-17 | 2019-05-24 | 北京连心医疗科技有限公司 | A kind of cloud radiotherapy planning method, equipment, storage medium and the system of artificial intelligence |
-
2019
- 2019-08-29 CN CN201910820650.3A patent/CN110415785A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105930636A (en) * | 2016-03-29 | 2016-09-07 | 中北大学 | Radiotherapy scheme optimization system capable of automatically determining weight of target function |
| CN107403201A (en) * | 2017-08-11 | 2017-11-28 | 强深智能医疗科技（昆山）有限公司 | Tumour radiotherapy target area and jeopardize that organ is intelligent, automation delineation method |
| CN109801696A (en) * | 2017-11-17 | 2019-05-24 | 北京连心医疗科技有限公司 | A kind of cloud radiotherapy planning method, equipment, storage medium and the system of artificial intelligence |
| CN108717866A (en) * | 2018-04-03 | 2018-10-30 | 陈辛元 | A kind of method, apparatus, equipment and the storage medium of the distribution of prediction radiotherapy planning dosage |
Cited By (23)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20230128148A1 (en) * | 2019-08-29 | 2023-04-27 | Beijing Linking Medical Technology Co., Ltd | Standardized Artificial Intelligence Automatic Radiation Therapy Planning Method and System |
| WO2021036366A1 (en) * | 2019-08-29 | 2021-03-04 | 北京连心医疗科技有限公司 | Standardized artificial intelligence automatic radiotherapy planning method and system |
| US11964170B2 (en) | 2019-08-29 | 2024-04-23 | Beijing Linking Medical Technology Co., Ltd. | Standardized artificial intelligence automatic radiation therapy planning method and system |
| CN111028914A (en) * | 2019-12-04 | 2020-04-17 | 北京连心医疗科技有限公司 | Artificial intelligence guided dose prediction method and system |
| CN111028914B (en) * | 2019-12-04 | 2024-03-22 | 北京连心医疗科技有限公司 | Artificial intelligence guided dose prediction method and system |
| CN111790065A (en) * | 2020-08-11 | 2020-10-20 | 北京易康医疗科技有限公司 | Multi-beam combined radiotherapy method for tumor |
| CN112447278A (en) * | 2020-11-17 | 2021-03-05 | 北京全域医疗技术集团有限公司 | Patient plan data verification method and device and computer equipment |
| CN112447278B (en) * | 2020-11-17 | 2024-04-26 | 北京全域医疗技术集团有限公司 | Verification method and device for patient plan data and computer equipment |
| CN112618967A (en) * | 2020-12-17 | 2021-04-09 | 程明霞 | Device for adjusting radiotherapy dose distribution and generating corresponding radiotherapy plan |
| CN112618967B (en) * | 2020-12-17 | 2024-03-15 | 程明霞 | Device for adjusting radiotherapy dosage distribution and generating corresponding radiotherapy plan |
| CN112597702A (en) * | 2020-12-21 | 2021-04-02 | 电子科技大学 | Pneumatic modeling generation type confrontation network model training method based on radial basis function |
| CN113241151B (en) * | 2021-02-25 | 2024-04-02 | 中科超精(南京)科技有限公司 | Intelligent radiotherapy cloud system and implementation method |
| CN113241151A (en) * | 2021-02-25 | 2021-08-10 | 中科超精(南京)科技有限公司 | Intelligent radiotherapy cloud system and implementation method |
| CN112972912B (en) * | 2021-04-22 | 2021-08-17 | 北京大学第三医院（北京大学第三临床医学院） | Non-coplanar radiotherapy beam incident path optimization method |
| CN112972912A (en) * | 2021-04-22 | 2021-06-18 | 北京大学第三医院（北京大学第三临床医学院） | Non-coplanar radiotherapy beam incident path optimization method |
| CN113517072A (en) * | 2021-09-13 | 2021-10-19 | 四川大学 | Prediction method of VMAT radiotherapy plan based on deep neural network |
| CN113517072B (en) * | 2021-09-13 | 2022-01-28 | 四川大学 | VMAT radiotherapy plan prediction device based on deep neural network |
| CN113516233B (en) * | 2021-09-13 | 2022-01-28 | 四川大学 | Neural network prediction device for VMAT radiotherapy plan |
| CN113516233A (en) * | 2021-09-13 | 2021-10-19 | 四川大学 | Neural network prediction method for VMAT radiotherapy plan |
| CN114864070A (en) * | 2022-04-28 | 2022-08-05 | 福建自贸试验区厦门片区Manteia数据科技有限公司 | Radiotherapy information management system |
| CN115829972A (en) * | 2022-12-02 | 2023-03-21 | 中山大学肿瘤防治中心（中山大学附属肿瘤医院、中山大学肿瘤研究所） | Radiotherapy plan three-dimensional dose distribution and flux synchronous prediction method and device |
| CN115829972B (en) * | 2022-12-02 | 2024-04-19 | 中山大学肿瘤防治中心（中山大学附属肿瘤医院、中山大学肿瘤研究所） | Synchronous prediction method and device for three-dimensional dose distribution and flux of radiotherapy plan |
| CN116153465A (en) * | 2023-04-24 | 2023-05-23 | 福建自贸试验区厦门片区Manteia数据科技有限公司 | Radiotherapy plan generation device based on reference DVH, electronic equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110415785A (en) | The method and system of artificial intelligence guidance radiotherapy planning | |
| Van Dyk | THE MODERN TECHNOLOGY OF RADIATION ONCOLOGY: A Compendium for Medical Physicists and Radiation Oncologists. Volume 4 | |
| US11278737B2 (en) | Systems and methods for automatic treatment planning and optimization | |
| Bodensteiner | RayStation: External beam treatment planning system | |
| CN103282967B (en) | For radiotherapeutic automatization treatment plan | |
| US20190046813A1 (en) | Systems and Methods for Radiation Treatment Planning | |
| CN110124214B (en) | Predictive dose distribution guided intensity modulated radiotherapy plan optimization systems, media and devices | |
| CN114681813B (en) | Automatic radiation therapy planning system, automatic radiation therapy planning method, and storage medium | |
| EP3742702B1 (en) | Standardized cloud radiotherapy planning method and storage medium | |
| US10076673B2 (en) | Interactive dose gradient based optimization technique to control IMRT delivery complexity | |
| CN110085298A (en) | Intensity modulated radiation therapy plan 3-dimensional dose distribution forecasting method based on depth e-learning | |
| CN101820827A (en) | The method of the fractionation of radiation dosage of accommodation radiotherapy dosage | |
| CN110327554B (en) | Intensity modulated radiotherapy plan optimization method based on predicted dose distribution guidance and application | |
| CN102184330A (en) | Image feature and smart regressive model-based method for optimizing intensity modulated radiotherapy plan | |
| Amit et al. | Automatic learning‐based beam angle selection for thoracic IMRT | |
| CN110021399A (en) | A kind of automatic design method of radiotherapy treatment planning | |
| Leal et al. | MLC leaf width impact on the clinical dose distribution: a Monte Carlo approach | |
| CN113178242B (en) | Automatic plan optimization system based on coupled generation countermeasure network | |
| Alhamada et al. | Monte Carlo as quality control tool of stereotactic body radiation therapy treatment plans | |
| US20220241614A1 (en) | Fluence map prediction and treatment plan generation for automatic radiation treatment planning | |
| Wang et al. | Artificial intelligence in adaptive radiation therapy | |
| CN112635024A (en) | Automatic planning and designing system for radiotherapy and construction method thereof | |
| Purdy et al. | Three-dimensional treatment planning and conformal therapy | |
| Teng et al. | Beam-wise dose composition learning for head and neck cancer dose prediction in radiotherapy | |
| US11964170B2 (en) | Standardized artificial intelligence automatic radiation therapy planning method and system |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20191105 |
|
| RJ01 | Rejection of invention patent application after publication |