WO2024098659A1 - 肿瘤放疗计划设计方法、装置、电子设备及存储介质 - Google Patents

肿瘤放疗计划设计方法、装置、电子设备及存储介质 Download PDF

Info

Publication number
WO2024098659A1
WO2024098659A1 PCT/CN2023/087510 CN2023087510W WO2024098659A1 WO 2024098659 A1 WO2024098659 A1 WO 2024098659A1 CN 2023087510 W CN2023087510 W CN 2023087510W WO 2024098659 A1 WO2024098659 A1 WO 2024098659A1
Authority
WO
WIPO (PCT)
Prior art keywords
parameter vector
current
optimization parameter
cost function
function value
Prior art date
Application number
PCT/CN2023/087510
Other languages
English (en)
French (fr)
Inventor
杨晓喻
赵于前
杨振
魏瑞
曹瑛
李书舟
邵其刚
唐杜
彭昭
Original Assignee
中南大学湘雅医院
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by 中南大学湘雅医院 filed Critical 中南大学湘雅医院
Priority to US18/354,621 priority Critical patent/US20240157171A1/en
Publication of WO2024098659A1 publication Critical patent/WO2024098659A1/zh

Links

Classifications

    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61N—ELECTROTHERAPY; MAGNETOTHERAPY; RADIATION THERAPY; ULTRASOUND THERAPY
    • A61N5/00—Radiation therapy
    • A61N5/10—X-ray therapy; Gamma-ray therapy; Particle-irradiation therapy
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G06N20/20—Ensemble learning
    • 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
    • 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
    • G06N3/0464—Convolutional networks [CNN, ConvNet]
    • 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
    • G06N3/048—Activation functions
    • 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
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N5/00—Computing arrangements using knowledge-based models
    • G06N5/01—Dynamic search techniques; Heuristics; Dynamic trees; Branch-and-bound
    • 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
    • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/20—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for computer-aided diagnosis, e.g. based on medical expert systems
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
    • Y02T10/00—Road transport of goods or passengers
    • Y02T10/10—Internal combustion engine [ICE] based vehicles
    • Y02T10/40—Engine management systems

Definitions

  • the present invention belongs to the field of radiotherapy technology, and specifically relates to a method, device, electronic equipment and storage medium for designing a tumor radiotherapy plan.
  • IMRT Intensity modulated radiation therapy
  • Plan design is a key link in IMRT, including two steps: plan optimization and plan verification.
  • plan optimization stage clinical staff need to conduct trial and error based on their own experience, and continuously adjust the plan optimization parameters until the dosimetric indicators of the target area and organs at risk meet clinical requirements and reach the optimal state considered by the designer.
  • plan verification stage clinical staff use dose measurement equipment to verify the execution accuracy of the plan, in which the gamma passing rate (GPR) is used to quantify the execution accuracy. If the execution accuracy does not meet the standard, it is necessary to start redesigning the plan until both the dosimetric indicators and execution accuracy meet the standards before patient treatment can be carried out.
  • GPR gamma passing rate
  • KBP knowledge-based planning
  • PBP protocol-based planning
  • MCOP multicriteria optimization-based planning
  • the optimization process tightens the optimization parameters of the organs at risk one by one according to the preset script until the hard constraints are about to be violated and the dose of the organs at risk cannot be further reduced, then the update is stopped and the final optimization parameters are determined; the MCOP method automatically generates a plan database based on the initial dosimetric optimization conditions, which contains multiple Pareto optimal plans. Then the plan designer selects the optimal plan that meets clinical requirements based on clinical experience.
  • the MCOP method requires the plan designer to weigh and select the final plan in the Pareto optimal plan set, and the plan quality still depends on the subjective experience of the selector.
  • the present invention provides a tumor radiotherapy plan design method, device, electronic device and storage medium to solve the problem that the existing tumor radiotherapy plan design method cannot simultaneously optimize the radiotherapy plan dosimetry quality and execution accuracy, and is too dependent on manual plan design experience.
  • an embodiment of the present invention provides a tumor radiotherapy plan design method, including: obtaining an initial optimization parameter vector set as the current optimization parameter vector set, and calculating the current cost function value; randomly correcting the current optimization parameter vector set to generate a preset number of sets of alternative optimization parameter vector sets; performing planning parameter optimization on the preset number of sets of alternative optimization parameter vector sets in parallel, and respectively calculating the total cost function value of each set of the alternative optimization parameter vector sets; determining the current optimal and suboptimal alternative optimization parameter vector sets according to the total cost function value, sampling and updating the current optimization parameter vector set and the current cost function value according to the total cost function value of the current alternative optimization parameter vector set, and performing repeated iterations until the convergence condition is met; outputting the final optimal optimization parameter vector set after the iteration is completed, determining the plan, and calculating and outputting the planned leaf movement and dose distribution of the multi-leaf collimator.
  • the initial optimization parameter vector set as the current optimization parameter vector set and calculating the current cost function value, it includes: collecting planning parameters for radiotherapy of tumors in multiple locations and corresponding gamma pass rate data to establish a training data set; training a gamma pass rate prediction model based on the training data set to obtain the trained gamma pass rate prediction model.
  • the parameter optimization algorithm is used to perform planning parameter optimization on the preset number of sets of alternative optimization parameter vector sets in parallel, and the total cost function value of each set of the alternative optimization parameter vector sets is calculated respectively, including: performing planning parameter optimization on the preset number of sets of alternative optimization parameter vector sets in parallel, obtaining current planning parameters and current dosimetric indicators of each set of the alternative optimization parameter vector sets; and calculating the total cost function value of each set of the alternative optimization parameter vector sets by applying a multifunctional optimization total cost function according to the current planning parameters, the current dosimetric indicators and the target dosimetric indicators.
  • the planning parameter optimization is performed on the preset number of optimization parameter vector sets to obtain the current planning parameters and current dosimetric indicators of each set of the alternative optimization parameter vector sets, including: for any set of the alternative optimization parameter vector sets, determining the field angle or range based on the CT image and structure; optimizing the planning parameters according to the field angle by applying an optimization algorithm based on gradient information to obtain the current planning parameters and current dosimetric indicators, wherein the current dosimetric indicators include Target homogeneity value, target conformality value, volume dose and average dose of organs at risk.
  • the total cost function value of each set of the alternative optimization parameter vector sets is calculated based on the current planning parameters, the current dosimetric indicators and the target dosimetric indicators using a multifunctional optimization total cost function, including: for any set of the alternative optimization parameter vector sets, a trained gamma pass rate prediction model is used for prediction according to the current planning parameters to obtain a predicted gamma pass rate, and a plan execution accuracy cost function is used to calculate a plan execution accuracy cost function value based on the predicted gamma pass rate; a plan dosimetric quality cost function is used to calculate a plan dosimetric quality cost function value based on the current dosimetric indicators and the target dosimetric indicators; a total cost function value is calculated based on the plan execution accuracy cost function value and the plan dosimetric quality cost function value using a multifunctional optimization total cost function.
  • the current optimization parameter vector set and the current cost function value are sampled and updated, and repeated iterations are performed until the convergence condition is met, including: sampling decision whether to adopt the current optimal or suboptimal alternative optimization parameter vector set; if the current optimal or suboptimal alternative optimization parameter vector set is adopted, the current optimal or suboptimal alternative optimization parameter vector set and the corresponding total cost function value are used as the current optimization parameter vector set and the current cost function value respectively; judging whether the convergence condition is met, wherein the convergence condition is reaching a preset maximum number of iterations or the longest iteration time, or the cost function value decreases and converges; if the convergence condition is not met, returning to the step of randomly modifying the current optimization parameter vector set to generate a preset number of sets of alternative optimization parameter vector sets.
  • the sampling decision whether to adopt the current optimal or suboptimal alternative optimization parameter vector set includes: calculating the acceptance probability of the current optimal or suboptimal alternative optimization parameter vector set based on the total cost function value of the current optimal or suboptimal alternative optimization parameter vector set and the current cost function value; randomly generating a random number between 0 and 1 based on a uniform distribution; if the random number is less than the calculated acceptance probability, determining to adopt the current optimal or suboptimal alternative optimization parameter vector set; otherwise, not adopting the current optimal or suboptimal alternative optimization parameter vector set.
  • an embodiment of the present invention also proposes a tumor radiotherapy plan design device, including: an initialization unit, used to obtain an initial optimization parameter vector set as the current optimization parameter vector set, and calculate the current cost function value; an alternative set acquisition unit, used to generate a preset number of sets of alternative optimization parameter vector sets based on the randomly corrected current optimization parameter vector set; a total cost calculation unit, used to perform plan parameter optimization on the preset number of sets of alternative optimization parameter vector sets in parallel, and calculate the total cost function value of each set of the alternative optimization parameter vector sets respectively; an update iteration unit, used to determine the current optimal or suboptimal alternative optimization parameter vector set according to the total cost function value, sample and update the current optimization parameter vector set and the current cost function value according to the total cost function value of the current alternative optimization parameter vector set, and repeat iteration until the convergence condition is met; a plan output unit, used to output the final optimal optimization parameter vector set after the iteration is completed, determine the plan parameters, calculate and output the planned leaf movement and dose distribution of the multi-
  • an embodiment of the present invention further proposes an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the aforementioned method when executing the program.
  • an embodiment of the present invention further proposes a computer storage medium, in which at least one executable instruction is stored, and the executable instruction enables a processor to execute the aforementioned method.
  • an embodiment of the present invention provides a method, device, electronic device and storage medium for designing a tumor radiotherapy plan, the method comprising: obtaining an initial optimization parameter vector set as a current optimization parameter vector set, and calculating a current cost function value; generating a preset number of sets of alternative optimization parameter vector sets according to the randomly corrected current optimization parameter vector set; performing plan parameter optimization on the preset number of sets of the alternative optimization parameter vector sets, and respectively calculating the total cost function value of each set of the alternative optimization parameter vector sets; determining the current optimal or suboptimal alternative optimization parameter vector set according to the total cost function value, sampling and updating the current optimization parameter vector set and the current cost function value according to the total cost function value of the current alternative optimization parameter vector set, and performing repeated iterations until the convergence condition is met; outputting the final optimal optimization parameter vector set after the iteration is completed, determining the plan parameters, calculating and outputting the planned leaf movement and dose distribution of the multi-leaf collimator,
  • FIG1 is a schematic flow chart of a method for designing a tumor radiotherapy plan in an embodiment of the present invention
  • FIG2 is a schematic diagram of the detailed process of the tumor radiotherapy plan design method in an embodiment of the present invention.
  • FIG. 3 is a schematic diagram of the structure of a GPR prediction model in a tumor radiotherapy plan design method according to an embodiment of the present invention
  • FIG4 is a schematic diagram of the structure of a tumor radiotherapy plan design device in an embodiment of the present invention.
  • FIG. 5 is a schematic diagram of an electronic device in an embodiment of the present invention.
  • An embodiment of the present invention provides a method for designing a tumor radiotherapy plan. As shown in FIG1 , the method for designing a tumor radiotherapy plan includes:
  • Step S11 Obtain an initial optimization parameter vector set as the current optimization parameter vector set, and calculate the current cost function value.
  • the initial optimization parameters are vector-encoded to obtain an initial optimization parameter vector set, the initial optimization parameter vector set is used as the current optimization parameter vector set, and the current cost function value is calculated based on the current optimization parameter vector set.
  • the initial optimization parameters include dose parameters, volume parameters and weight parameters.
  • the initial optimization parameter vector set includes a dose optimization parameter vector, a volume optimization parameter vector and a weight optimization parameter vector.
  • the dose optimization parameter vector only contains the dose optimization parameters of all serial organs
  • the volume optimization parameter vector only contains the volume optimization parameters of all parallel organs
  • the weight optimization parameter vector contains the weight optimization parameters of all target areas and organs at risk.
  • step S11 it is necessary to train the gamma passing rate (GPR) prediction model to obtain the trained GPR prediction model.
  • GPR gamma passing rate
  • the planning parameters and corresponding gamma passing rate data of tumor radiotherapy at multiple sites are collected to establish a training data set; the gamma passing rate prediction model is trained based on the training data set to obtain the trained gamma passing rate prediction model.
  • the planning parameter may be a field flux or an aperture shape.
  • the training data set is divided into a training set, a validation set, and a test set, and the training set and the validation set are used to train the gamma passing rate prediction model, and the test set is used to test the prediction accuracy of the gamma passing rate prediction model.
  • the input of the GPR prediction model is the planning parameter (such as a field flux map or an aperture shape), and the output is the GPR.
  • the GPR prediction model may be a classical machine learning model, such as Poisson regression, AdaBoost, and random forest models, or a deep learning model, such as a convolutional neural network, a residual neural network, a transformer, and the like.
  • the optimization algorithm based on the gradient information performs a single plan parameter optimization on the current optimization parameter vector set to obtain the current plan parameters and current dosimetric indicators corresponding to the current optimization parameter vector set; the trained GPR prediction model is used to predict the GPR to obtain the predicted GPR; the value of the plan execution accuracy cost function is calculated according to the predicted GPR; the value of the plan quality cost function is calculated according to the current dosimetric indicators and the target dosimetric indicators, and then the value of the multifunctional optimization total cost function is calculated to obtain the total cost function value corresponding to the current optimization parameter vector set as the current cost function value.
  • the optimization algorithm based on the gradient information can be a gradient descent algorithm, interior point optimization or Newton method, etc., which is not limited here.
  • CT images and structures are obtained, and then the radiation field angle or range is obtained, and then the planning parameters are optimized based on the optimization algorithm of the gradient information, and the initial average dose of the organ at risk is calculated.
  • the plan execution accuracy cost function satisfies the following relationship:
  • GPR + and GPR- are the upper and lower limits of the preset GPR, with preferred values of 1.00 and 0.85 respectively.
  • Nf is the number of planned fields
  • GPRi is the GPR of the i-th field predicted by the trained artificial intelligence model.
  • the planned dosimetry quality cost function satisfies the following relationship:
  • ND is the number of dosimetric indicators to be evaluated
  • tj is the weight coefficient parameter of the jth dosimetric indicator Dj
  • Dj includes the target area uniformity index DHI , the target area conformality index DCI and the dose index of each organ at risk D OAR .
  • HI is the target area uniformity value
  • HI + and HI - are the preset upper and lower limits of uniformity
  • HI + and HI - are 8.0 and 3.0 respectively
  • CI is the target area conformality value
  • CI + and CI - are the preset upper and lower limits of conformality
  • CI + and CI - are 0.95 and 0.65 respectively
  • the target dosimetry indicators include but are not limited to: the upper and lower limits of the uniformity of the target uniformity value HI, the upper and lower limits of the conformality of the target conformality value CI, the upper and lower limits of the average dose of the organs at risk, etc.
  • the upper and lower limits of the average dose of the organs at risk are 1.1 times and 0.9 times the initial average dose, respectively.
  • the upper and lower limits of the average dose of the organs at risk are 1.1 times and 0.9 times the initial average dose, respectively.
  • D 5 and D 95 are the maximum doses achieved by 5% and 95% of the target volume, respectively, and D p is the corresponding target prescription dose;
  • TV 95 is the volume within the target that reaches 95% of the prescription dose, TV is the target volume, and
  • V 95 is the volume that reaches 95% of the prescription dose, including Inside and outside the target area.
  • the multifunctional optimization total cost function is obtained by combining the plan execution accuracy cost function and the plan quality cost function using the weighted summation method.
  • ⁇ is the relative weight factor of the plan execution accuracy cost function and the plan quality cost function, preferably 10.0-30.0.
  • Step S12 Randomly modify the current optimization parameter vector set to generate a preset number of candidate optimization parameter vector sets.
  • step S12 in each iteration, the current optimization parameter vector set is randomly modified to generate a preset number N sets of candidate optimization parameter vector sets.
  • the random modification method can be based on Gaussian distribution or other random modification methods, such as uniform distribution, etc., which are not limited here.
  • N is between 5 and 20.
  • Different optimization parameter vectors are modified according to the following relationship:
  • Optimize the parameter vector for the current dose, volume or weight is a random number vector generated based on a standard normal distribution, with the same number of elements as Consistent, ⁇ is the correction rate.
  • ⁇ is the correction rate.
  • ⁇ is about 0.05, and different dose optimization parameter vectors can take the same ⁇ value or different ⁇ values.
  • ⁇ is about 0.20, and different weight optimization parameter vectors can take the same ⁇ value or different ⁇ values.
  • Step S13 performing planning parameter optimization on the preset number of sets of candidate optimization parameter vector sets in parallel, and calculating the total cost function value of each set of the candidate optimization parameter vector sets respectively.
  • step S13 referring to FIG. 2, firstly, the preset number of sets of the alternative optimization parameter vector sets are optimized in parallel for planning parameters, and the current planning parameters and current dosimetric indicators of each set of the alternative optimization parameter vector sets are obtained.
  • the field angle or range is determined based on the CT image and structure; the planning parameters are optimized according to the field angle using an optimization algorithm based on gradient information, and the current planning parameters and current dosimetric indicators are obtained, wherein the current dosimetric indicators include the target area uniformity value, the target area conformality value, and the volume dose and average dose of the organ at risk.
  • the total cost function value of each set of the alternative optimization parameter vector set is calculated by applying the multifunctional optimization total cost function according to the current planning parameters, the current dosimetric indicators and the target dosimetric indicators.
  • the trained gamma pass rate prediction model is used for prediction according to the current planning parameters to obtain the predicted gamma pass rate
  • the plan execution accuracy cost function is used to calculate the plan execution accuracy cost function value according to the predicted gamma pass rate
  • the plan dosimetric quality cost function is used to calculate the plan dosimetric quality cost function value according to the current dosimetric indicators and the target dosimetric indicators
  • the plan execution accuracy cost function value and the plan dosimetric quality cost function value are calculated according to the plan execution accuracy cost function value and the plan dosimetric quality cost function value.
  • the quantitative quality cost function value is calculated by applying the multifunctional optimization total cost function.
  • the calculation method of the total cost function value is the same as the calculation method of the total cost function value in step S11, which will
  • Step S14 Determine the current optimal and suboptimal candidate optimization parameter vector sets according to the total cost function value, sample and update the current optimization parameter vector set and the current cost function value according to the total cost function value of the current candidate optimization parameter vector set, and repeat iterations until convergence conditions are met.
  • an alternative optimization parameter vector set with the lowest total cost function value is selected from a preset number N sets of alternative optimization parameter vector sets and determined as the current optimal alternative optimization parameter vector set and the current suboptimal alternative optimization parameter vector set.
  • the sampling decision is whether to adopt the current optimal or suboptimal alternative optimization parameter vector set. If the current optimal or suboptimal alternative optimization parameter vector set is adopted, the current optimal or suboptimal alternative optimization parameter vector set and the corresponding total cost function value are used as the current optimization parameter vector set and the current cost function value, respectively. Then, it is determined whether the convergence condition is met, wherein the convergence condition is reaching the preset maximum number of iterations or the longest iteration time, or the cost function value decreases and converges. If the current optimal or suboptimal alternative optimization parameter vector set is not adopted, it is directly determined whether the convergence condition is met. If the convergence condition is not met, return to step S12 for the next iteration. If the convergence condition is met, the iteration ends.
  • step S14 when sampling and deciding whether to adopt the current optimal or suboptimal candidate optimization parameter vector set, the acceptance probability of the current optimal or suboptimal candidate optimization parameter vector set is first calculated according to the total cost function value of the current optimal or suboptimal candidate optimization parameter vector set and the current cost function value by applying the following relationship:
  • ⁇ is the initial temperature parameter, preferably 0.01-0.10
  • ⁇ is the attenuation factor parameter, preferably 0.01-0.10
  • k is the number of optimization iterations.
  • a random number between 0 and 1 is randomly generated based on uniform distribution; if the random number is less than the calculated acceptance probability, the current optimal or suboptimal candidate optimization parameter vector set is adopted; otherwise, the current optimal or suboptimal candidate optimization parameter vector set is not adopted.
  • the process of manual trial and error in the prior art is: optimize the plan parameters according to the set optimization parameters, calculate the dosimetric indicators of this optimization, and judge whether the dosimetric indicators of all target areas or organs at risk meet the clinical requirements and whether there is room for improvement. If the plan does not meet the clinical requirements or there is room for improvement, the designer needs to manually modify the corresponding plan optimization parameters based on experience and repeat the plan parameter optimization until the plan meets the clinical requirements and reaches the optimal state considered by the designer. At this point, the trial and error is completed and the plan can be output. The trial and error process takes up most of the time for clinical plan optimization and is heavily dependent on Designer experience.
  • the embodiment of the present invention simulates manual trial and error of plans based on meta-heuristic algorithms, automatically adjusts the optimization parameters of the plans, and realizes the automatic design of radiotherapy plans.
  • the evolutionary strategy and simulated annealing fusion algorithm are used to simulate manual trial and error of plans, which is less dependent on manual plan design experience, has a small subjective error, a high degree of automation, and does not require the establishment of a plan database or optimization template for a specific medical institution, a specific disease, or a specific designer, thereby saving manpower and time costs, being easy to promote, and reducing the waiting time for patients to receive treatment.
  • Step S15 Output the final optimal optimization parameter vector set after the iteration is completed, determine the planning parameters, and calculate and output the planned leaf movement and dose distribution of the multi-leaf collimator.
  • the final optimal optimization parameter vector set after the iteration is completed is output, the planning parameters are determined, and the planned leaf movement and dose distribution of the multi-leaf collimator are calculated and output, so as to obtain the final tumor radiotherapy plan.
  • the final optimal optimization parameter vector set after the iteration is completed refers to the optimization parameter vector set with the smallest total cost function in all iteration steps.
  • the embodiment of the present invention also compares the automatic planning method of the present invention with the manual planning method for brain tumors and lung cancer to verify the tumor radiotherapy planning design method of the embodiment of the present invention: including the radiotherapy planning design of 7 brain tumor patients and 7 lung cancer patients, collecting 1541 pairs of field flux and corresponding GPR of multi-site plans to form a data set.
  • the data set is randomly divided into a training set, a validation set, and a test set.
  • the training set and the validation set are used to train the GRP prediction model, and the prediction accuracy of the model is tested using the test set, with a mean absolute error of 0.015.
  • the GPR prediction model in the tumor radiotherapy plan design method of the embodiment of the present invention adopts a deep learning model, and the main body of the model is a residual convolutional network.
  • the specific model structure is shown in Figure 3, where MC Dropout represents Monte Carlo sampling (abbreviated as Monte Carlo sampling), and the Rectified Linear Unit (ReLU) is a neural activation function.
  • the input of the GPR prediction model is the planned field flux map, which is automatically zero-filled and center-cropped to standardize the model input, and then after 9 residual convolution module encodings, 2 fully connected encodings and Sigmoid function activation, the predicted GPR is output.
  • GPR gamma passing rate
  • Table 1 is a statistical comparison of the planning quality of the automatic plan and the manual plan using the tumor radiotherapy planning design method of the embodiment of the present invention.
  • the target area uniformity of the automatic plan of the embodiment of the present invention is significantly better than that of the manual plan, and the target area conformality is close to that of the manual plan.
  • the dosimetric indicators of the brainstem, lens, optic nerve, optic chiasm, pituitary gland, spinal cord, lung, heart and whole body are all better than those of the manual plan.
  • Table 2 is a statistical comparison of the execution accuracy measurement results of the automatic plan and the manual plan using the method of the present invention.
  • the GPR of the automatic plan based on multifunctional optimization of the present invention is slightly higher than that of the manual plan, and the number of fields that failed the measurement (measured GPR ⁇ 90% is defined as failed) is significantly reduced.
  • the tumor radiotherapy plan design method of the embodiment of the present invention has multifunctional synchronous optimization capabilities, incorporates the GPR prediction model into the plan optimization framework, and combines the meta-inspiration algorithm and the field flux optimization algorithm to achieve synchronous optimization of the radiotherapy plan quality and plan execution accuracy, taking into account both quality and execution accuracy, reducing the proportion of failed plans, and is expected to improve patient efficacy; using evolutionary strategies and simulated annealing fusion algorithms to simulate the manual trial and error process of plan optimization, it relies less on manual plan design experience, has a small subjective error, and a high degree of automation. There is no need to establish a plan database or optimization template for a specific medical institution, a specific disease, or a specific designer, which saves manpower and time costs, is easy to promote, and reduces patients' treatment waiting time.
  • the tumor radiotherapy plan design method of the embodiment of the present invention obtains an initial optimization parameter vector set as the current optimization parameter vector set and calculates the current cost function value; randomly corrects the current optimization parameter vector set to generate a preset number of sets of alternative optimization parameter vector sets; performs planning parameter optimization on the preset number of sets of alternative optimization parameter vector sets in parallel, and calculates the total cost function value of each set of the alternative optimization parameter vector sets respectively; determines the current optimal and suboptimal alternative optimization parameter vector sets according to the total cost function value, samples and updates the current optimization parameter vector set and the current cost function value according to the total cost function value of the current optimal alternative optimization parameter vector set, and repeats iterations until the convergence condition is met; outputs the final optimal optimization parameter vector set after the iteration is completed, determines the planning parameters, calculates and outputs the planned leaf movement and dose distribution of the multi-leaf collimator, can synchronously optimize the dosimetric quality and execution accuracy of the radiotherapy plan, realizes the automation of plan design, relies less on manual experience, and can
  • an embodiment of the present invention also provides a tumor radiotherapy plan design device.
  • the tumor radiotherapy plan design device includes: an initialization unit, an alternative set acquisition unit, a total cost calculation unit, an update iteration unit, and a plan output unit.
  • an initialization unit used to obtain an initial optimization parameter vector set as a current optimization parameter vector set, and calculate a current cost function value
  • An alternative set acquisition unit used for randomly modifying the current optimization parameter vector set to generate a preset number of sets of alternative optimization parameter vector sets
  • a total cost calculation unit used for performing planning parameter optimization on the preset number of sets of the candidate optimization parameter vector sets in parallel, and calculating the total cost function value of each set of the candidate optimization parameter vector sets respectively;
  • An updating and iterating unit configured to determine a current optimal and suboptimal candidate optimization parameter vector set according to the total cost function value, and to update the current optimization parameter vector set and the current cost function value by set sampling according to the total cost function value of the current optimal candidate optimization parameter vector and perform repeated iterations until a convergence condition is met;
  • the plan output unit is used to output the final optimal optimization parameter vector set after the iteration is completed, determine the plan parameters, calculate and output the planned leaf movement and dose distribution of the multi-leaf collimator.
  • each module can be implemented in the same or multiple software and/or hardware.
  • the device of the above embodiment is applied to the corresponding method in the above embodiment, and has the beneficial effects of the corresponding method embodiment, which will not be described in detail here.
  • an embodiment of the present invention further provides an electronic device, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the method described in any one of the above embodiments is implemented.
  • An embodiment of the present invention provides a non-volatile computer storage medium, wherein the computer storage medium stores at least one executable instruction, and the computer executable instruction can execute the method described in any of the above embodiments.
  • FIG5 shows a more specific schematic diagram of the hardware structure of an electronic device provided in this embodiment, and the device may include: a processor 501, a memory 502, an input/output interface 503, a communication interface 504, and a bus 505.
  • the processor 501, the memory 502, the input/output interface 503, and the communication interface 504 are connected to each other in communication within the device through the bus 505.
  • the processor 501 may be a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits. Implementation is used to execute relevant programs to implement the technical solution provided by the method embodiment of the present invention.
  • CPU Central Processing Unit
  • ASIC application-specific integrated circuit
  • the memory 502 can be implemented in the form of ROM (Read Only Memory), RAM (Random Access Memory), static storage device, dynamic storage device, etc.
  • the memory 502 can store operating systems and other application programs.
  • the relevant program code is stored in the memory 502 and called and executed by the processor 501.
  • the input/output interface 503 is used to connect the input/output module to realize information input and output.
  • the input/output module can be configured in the device as a component (not shown in the figure), or it can be externally connected to the device to provide corresponding functions.
  • the input device may include a keyboard, a mouse, a touch screen, a microphone, various sensors, etc.
  • the output device may include a display, a speaker, a vibrator, an indicator light, etc.
  • the communication interface 504 is used to connect a communication module (not shown) to realize communication interaction between the device and other devices.
  • the communication module can realize communication through wired mode (such as USB, network cable, etc.) or wireless mode (such as mobile network, WIFI, Bluetooth, etc.).
  • the bus 505 includes a path for transmitting information between various components of the device (eg, the processor 501 , the memory 502 , the input/output interface 503 , and the communication interface 504 ).
  • the device may also include other components necessary for normal operation.
  • the above device may also only include the components necessary for implementing the embodiments of the present invention, and does not necessarily include all the components shown in the figure.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Physics & Mathematics (AREA)
  • Biomedical Technology (AREA)
  • General Health & Medical Sciences (AREA)
  • Software Systems (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • General Physics & Mathematics (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • Computational Linguistics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Medical Informatics (AREA)
  • Molecular Biology (AREA)
  • Public Health (AREA)
  • Biophysics (AREA)
  • Pathology (AREA)
  • Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
  • Epidemiology (AREA)
  • Primary Health Care (AREA)
  • Veterinary Medicine (AREA)
  • Animal Behavior & Ethology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Radiology & Medical Imaging (AREA)
  • Surgery (AREA)
  • Urology & Nephrology (AREA)
  • Databases & Information Systems (AREA)
  • Radiation-Therapy Devices (AREA)

Abstract

本发明提供一种肿瘤放疗计划设计方法、装置、电子设备及存储介质，方法包括：获取作为当前优化参数向量集的初始优化参数向量集，计算当前成本函数值；随机修正当前优化参数向量集，产生预设数量套备选优化参数向量集；对预设数量套备选优化参数向量集进行计划参数优化，计算总成本函数值；根据当前备选优化参数向量集的总成本函数值，抽样更新当前优化参数向量集和当前成本函数值，进行重复迭代，直至满足收敛条件；输出迭代后的最终最优的优化参数向量集，确定计划参数，计算并输出多叶准直器的计划叶片走位和剂量分布。本发明能够同步优化放疗计划剂量学质量和执行准确度，实现计划设计自动化，较少依赖人工经验，能够节省临床人力和时间成本。

Description

肿瘤放疗计划设计方法、装置、电子设备及存储介质 技术领域
本发明属于放疗技术领域，具体涉及到一种肿瘤放疗计划设计方法、装置、电子设备及存储介质。
背景技术
调强放射治疗(放疗)是目前的主流放疗技术。计划设计是调强放疗的关键环节，包括计划优化和计划验证两个步骤。在计划优化阶段，临床工作人员需要根据自己的经验进行试错，不断调整计划优化参数，直至靶区和危及器官的剂量学指标满足临床要求、达到设计者认为的最优状态。在计划验证阶段，临床工作人员使用剂量测量设备验证计划的执行准确度，其中使用伽马通过率(Gamma passing rate，GPR)量化执行准确度。若执行准确度不达标，则需开始重新进行计划设计，直至剂量学指标和执行准确度均达标后，才能进行患者治疗。其中，优化参数的人工试错和验证不通过计划的再设计是造成临床工作负荷重的两个主要因素。实现高质量和高执行准确度的自动计划设计是提高临床效率、降低患者治疗等待时间的关键，有望改善患者疗效，具有重要的临床意义。
目前已有的自动计划设计方法有三种：基于知识的自动计划(Knowledge based planning，KBP)、基于模板的自动计划(Protocol based planning，PBP)和基于多目标优化的自动计划(Multicriteria optimization based planning，MCOP)。KBP方法收集以往的临床治疗计划建立特定肿瘤部位的数据库，训练人工智能模型学习先验知识，预测新病人临床可实现的剂量学指标，进而指导计划设计的优化参数设置；PBP方法首先根据临床经验建立特定肿瘤部位的剂量学优化参数模板，在满足硬性约束条件下，优化过程根据预设的脚本逐个收紧危及器官的优化参数，直至将要违背硬性约束条件、无法进一步降低危机器官剂量，停止更新，确定最终的优化参数；MCOP方法根据初始剂量学优化条件自动生成一个计划数据库，包含多个帕累托最优计划，之后计划设计者再根据临床经验从中挑选出符合临床要求的最优计划。
但是，现有的自动计划方法仅能面向计划剂量学质量(即靶区和危及器官的剂量学指标)进行优化，不涉及计划执行准确度的优化。计划的执行准确度需要使用辐射探测器测量后才能确定，优化阶段难以量化执行准确度，自然无法优化执行准确度。即使在计划优化阶段能够预知执行准确度，由于难以解析地描述执行准确度成本函数与计划参数的梯度关系，常规的通量优化算法无法实现计划质量和执行准确度同步优化。若执行准确度不达标，则需重新进行计划设计，浪费时间和人力，延误患者治疗。另外，现有的自动计划方法过于依赖人工计划设计经验，存在主观误差，方法扩展性差。例如，KBP和PBP方法需要根据临床经验建立对应的计划数据库或优化参数模板，这些数据库或优化参数模板的质量决定了最终自动计 划设计的质量，而它们往往是通过回顾性收集临床计划获得，若收集的计划质量欠佳，自动计划质量亦会欠佳，不可避免地存在主观误差。同时，考虑到不同肿瘤部位的目标剂量学指标不同，不同医疗机构具有不同的治疗规范，不同计划设计者有不同的临床经验和计划设计习惯，为了保证自动计划的质量，需要针对不同肿瘤部位、医疗机构或计划设计者建立不同的数据库或模板，过于繁琐，难以扩展。MCOP方法则需要计划设计者在帕累托最优计划集中权衡选择最终计划，计划质量仍依赖于选择者的主观经验。
发明内容
本发明提供一种肿瘤放疗计划设计方法、装置、电子设备及存储介质，以解决现有肿瘤放疗计划设计方法不能同步优化放疗计划剂量学质量和执行准确度，且过于依赖人工计划设计经验的问题。
基于上述目的，本发明实施例提供了一种肿瘤放疗计划设计方法，包括：获取作为当前优化参数向量集的初始优化参数向量集，并计算当前成本函数值；随机修正所述当前优化参数向量集，产生预设数量套备选优化参数向量集；并行对所述预设数量套所述备选优化参数向量集进行计划参数优化，并分别计算各套所述备选优化参数向量集的总成本函数值；根据所述总成本函数值确定当前最优及次优备选优化参数向量集，根据所述当前备选优化参数向量集的总成本函数值，抽样更新所述当前优化参数向量集和所述当前成本函数值，并进行重复迭代，直至满足收敛条件；输出迭代完成后的最终最优的优化参数向量集，确定计划，计算并输出多叶准直器的计划叶片走位和剂量分布。
可选的，所述获取作为当前优化参数向量集的初始优化参数向量集，并计算当前成本函数值之前，包括：收集多个部位肿瘤放疗的计划参数和对应的伽马通过率数据，建立训练数据集；基于所述训练数据集对伽马通过率预测模型进行训练，得到训练后的所述伽马通过率预测模型。
可选的，所述基于参数优化算法并行对所述预设数量套所述备选优化参数向量集进行计划参数优化，并分别计算各套所述备选优化参数向量集的总成本函数值，包括：对所述预设数量套所述备选优化参数向量集并行进行计划参数优化，获取各套所述备选优化参数向量集的当前计划参数和当前剂量学指标；根据所述当前计划参数、所述当前剂量学指标以及目标剂量学指标应用多功能优化总成本函数计算各套所述备选优化参数向量集的总成本函数值。
可选的，所述对所述预设数量套优化参数向量集进行计划参数优化，获取各套所述备选优化参数向量集的当前计划参数和当前剂量学指标，包括：针对任一套所述备选优化参数向量集，基于CT影像及结构确定射野角度或范围；根据所述射野角度应用基于梯度信息的优化算法优化计划参数，获取当前计划参数和当前剂量学指标，其中所述当前剂量学指标包括 靶区均匀性数值、靶区适形性数值以及危及器官的体积剂量及平均剂量。
可选的，所述根据所述当前计划参数、所述当前剂量学指标以及目标剂量学指标应用多功能优化总成本函数计算各套所述备选优化参数向量集的总成本函数值，包括：针对任一套所述备选优化参数向量集，根据所述当前计划参数应用训练后的伽马通过率预测模型进行预测，获取预测伽马通过率，并根据所述预测伽马通过率应用计划执行准确度成本函数计算计划执行准确度成本函数值；根据所述当前剂量学指标和所述目标剂量学指标应用计划剂量学质量成本函数计算计划剂量学质量成本函数值；根据所述计划执行准确度成本函数值和所述计划剂量学质量成本函数值应用多功能优化总成本函数计算总成本函数值。
可选的，所述根据所述当前备选优化参数向量集的总成本函数值，抽样更新所述当前优化参数向量集和所述当前成本函数值，并进行重复迭代，直至满足收敛条件，包括：抽样决策是否采纳所述当前最优或次优备选优化参数向量集；如果采纳所述当前最优或次优备选优化参数向量集，则将所述当前最优或次优备选优化参数向量集和对应的总成本函数值分别作为当前优化参数向量集和当前成本函数值；判断是否满足收敛条件，其中所述收敛条件为达到预设的最大迭代次数或最长迭代时间、或者成本函数值降幅收敛；如果不满足收敛条件，则返回所述随机修正当前优化参数向量集，产生预设数量套备选优化参数向量集的步骤。
可选的，所述抽样决策是否采纳所述当前最优或次优备选优化参数向量集，包括：根据所述当前最优或次优备选优化参数向量集的所述总成本函数值以及所述当前成本函数值计算当前最优或次优备选优化参数向量集的接受概率；基于均匀分布随机产生0至1之间的随机数；如果所述随机数小于计算的所述接受概率，则确定采纳所述当前最优或次优备选优化参数向量集；否则不采纳所述当前最优或次优备选优化参数向量集。
基于同一发明构思，本发明实施例还提出了一种肿瘤放疗计划设计装置，包括：初始化单元，用于获取作为当前优化参数向量集的初始优化参数向量集，并计算当前成本函数值；备选集获取单元，用于根据所述随机修正当前优化参数向量集，产生预设数量套备选优化参数向量集；总成本计算单元，用于并行对所述预设数量套所述备选优化参数向量集进行计划参数优化，并分别计算各套所述备选优化参数向量集的总成本函数值；更新迭代单元，用于根据所述总成本函数值确定当前最优或次优备选优化参数向量集，根据所述当前备选优化参数向量集的总成本函数值，抽样更新所述当前优化参数向量集和所述当前成本函数值并进行重复迭代，直至满足收敛条件；计划输出单元，用于输出迭代完成后的最终最优的优化参数向量集，确定计划参数，计算并输出多叶准直器的计划叶片走位和剂量分布。
基于同一发明构思，本发明实施例还提出了一种电子设备，包括存储器、处理器及存储在存储器上并可在处理器上运行的计算机程序，所述处理器执行所述程序时实现前述的方法。
基于同一发明构思，本发明实施例还提出了一种计算机存储介质，存储介质中存储有至少一项可执行指令，所述可执行指令使处理器执行前述的方法。
本发明的有益效果是：从上面所述可以看出，本发明实施例提供的一种肿瘤放疗计划设计方法、装置、电子设备及存储介质，方法包括：获取作为当前优化参数向量集的初始优化参数向量集，并计算当前成本函数值；根据所述随机修正当前优化参数向量集，产生预设数量套备选优化参数向量集；对所述预设数量套所述备选优化参数向量集进行计划参数优化，并分别计算各套所述备选优化参数向量集的总成本函数值；根据所述总成本函数值确定当前最优或次优备选优化参数向量集，根据所述当前备选优化参数向量集的总成本函数值，抽样更新所述当前优化参数向量集和所述当前成本函数值，并进行重复迭代，直至满足收敛条件；输出迭代完成后的最终最优的优化参数向量集，确定计划参数，计算并输出多叶准直器的计划叶片走位和剂量分布，能够同步优化放疗计划质量和执行准确度，实现自动计划设计，较少依赖人工经验，能够节省临床人力和时间成本。
附图说明
为了更清楚地说明本发明实施例或现有技术中的技术方案，下面将对实施例或现有技术描述中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图仅仅是本发明实施例，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图获得其他的附图。
图1为本发明实施例中的肿瘤放疗计划设计方法的流程示意图；
图2为本发明实施例中的肿瘤放疗计划设计方法的详细过程示意图；
图3为本发明实施例中的肿瘤放疗计划设计方法中的GPR预测模型的结构示意图；
图4为本发明实施例中的肿瘤放疗计划设计装置的结构示意图；
图5为本发明实施例中电子设备示意图。
具体实施方式
为使本公开的目的、技术方案和优点更加清楚明白，以下结合具体实施例，并参照附图，对本公开进一步详细说明。
需要说明的是，除非另外定义，本发明实施例使用的技术术语或者科学术语应当为本公开所属领域内具有一般技能的人士所理解的通常意义。本发明实施例中使用的“第一”、“第二”以及类似的词语并不表示任何顺序、数量或者重要性，而只是用来区分不同的组成部分。“包括”或者“包含”等类似的词语意指出现该词前面的元件或者物件涵盖出现在该词后面列举的元件或者物件及其等同，而不排除其他元件或者物件。“连接”或者“相连”等类似的词语并非 限定于物理的或者机械的连接，而是可以包括电性的连接，不管是直接的还是间接的。“上”、“下”、“左”、“右”等仅用于表示相对位置关系，当被描述对象的绝对位置改变后，则该相对位置关系也可能相应地改变。
本发明实施例提供了一种肿瘤放疗计划设计方法，如附图1所示，肿瘤放疗计划设计方法包括：
步骤S11：获取作为当前优化参数向量集的初始优化参数向量集，并计算当前成本函数值。
在本发明实施例中，对初始的优化参数进行向量编码，获得初始优化参数向量集，以所述初始优化参数向量集作为当前优化参数向量集，并根据所述当前优化参数向量集计算当前成本函数值。初始的优化参数包括剂量参数、体积参数和权重参数。初始优化参数向量集包括剂量优化参数向量、体积优化参数向量和权重优化参数向量。剂量优化参数向量仅包含所有串行器官的剂量优化参数，体积优化参数向量仅包含所有并行器官的体积优化参数，权重优化参数向量包含所有靶区和危及器官的权重优化参数。
在本发明实施例中，在步骤S11之前，需要对伽马通过率(Gamma passing rate，GPR)预测模型进行训练，获取训练后的GPR预测模型。具体地，参见图2，收集多个部位肿瘤放疗的计划参数和对应的伽马通过率数据，建立训练数据集；基于所述训练数据集对伽马通过率预测模型进行训练，得到训练后的所述伽马通过率预测模型。其中计划参数可以为射野通量或孔径形状。在本发明实施例中，将训练数据集分为训练集、验证集和测试集，使用训练集和验证集训练伽马通过率预测模型，使用测试集测试伽马通过率预测模型的预测准确度。GPR预测模型的输入为计划参数(如射野通量图或孔径形状)，输出为GPR。GPR预测模型可以为经典的机器学习模型，如泊松回归、AdaBoost和随机森林等模型，也可以为深度学习模型，如卷积神经网络、残差神经网络、transformer等模型。
根据当前优化参数向量集计算当前成本函数值时，首先基于梯度信息的优化算法对当前优化参数向量集进行单次计划参数优化，获取与当前优化参数向量集对应的当前计划参数和当前剂量学指标；应用训练后的GPR预测模型对GPR进行预测，获取预测GPR；根据预测GPR计算计划执行准确度成本函数的值；根据当前剂量学指标和目标剂量学指标计算计划质量成本函数的值，进而计算多功能优化总成本函数的值，得到与当前优化参数向量集对应的总成本函数值，作为当前成本函数值。其中，基于梯度信息的优化算法可以为梯度下降算法、内点优化或牛顿法等，在此并不作限制。
在本发明实施例中，继续参见图2，获取CT影像及结构，进而获取射野角度或范围，再基于梯度信息的优化算法优化计划参数，并计算危及器官的初始平均剂量以获取目标 剂量学指标。计划执行准确度成本函数满足以下关系式：

其中，为计划执行准确度成本函数，为计划射野预测GPR的均值，使用公式(2)计算GPR+和GPR-分别为预设GPR的上下限，优选值分别为1.00和0.85，Nf为计划射野的数目，GPRi为所训练人工智能模型预测的第i个射野的GPR。
计划剂量学质量成本函数满足以下关系式：
其中，为计划质量成本函数，ND为待评估剂量学指标的个数，tj为第j个剂量学指标Dj的权重系数参数，Dj包括靶区的均匀性指标DHI、靶区的适形性指标DCI和各个危及器官剂量指标DOAR。


HI为靶区均匀性数值，HI+和HI-分别为预设的均匀性上下限，HI+和HI-分别取8.0和3.0；CI为靶区适形性数值，CI+和CI-分别为预设的适形性上下限，CI+和CI-分别取0.95和0.65；为某个危及器官的平均剂量，和分别为对应危及器官平均剂量的预设上下限。目标剂量学指标包括但不限于：靶区均匀性数值HI的均匀性上下限、靶区适形性数值CI的适形性上下限、危及器官平均剂量的上下限等，优选的，危及器官平均剂量的上下限分别取1.1倍和0.9倍的初始平均剂量可选的，

D5和D95分别为5％和95％的靶区体积所达到的最大剂量，Dp为对应靶区处方剂量；TV95为靶区内达到95％处方剂量的体积，TV为靶区体积，V95为达到95％处方剂量的体积，包括 靶区内外。
使用加权求和的方法结合计划执行准确度成本函数和计划质量成本函数，得到多功能优化总成本函数
其中，λ为计划执行准确度成本函数和计划质量成本函数的相对权重因子，优选10.0-30.0。
步骤S12：随机修正所述当前优化参数向量集，产生预设数量套备选优化参数向量集。
在步骤S12中，如图2所示，在每次迭代中，随机修正当前优化参数向量集，生成预设数量N套备选优化参数向量集。其中随机修正方法可以是基于高斯分布进行随机修正，也可以是其他随机修正方法，如均匀分布等，在此不作限制。优选的，N取5～20之间。根据以下关系式修正不同的优化参数向量：
其中，为生成的剂量、体积或权重备选优化参数向量，为当前剂量、体积或权重优化参数向量，为基于标准正态分布生成的随机数向量，元素个数与一致，α为修正率。优选的，对于剂量优化参数向量和体积优化参数向量的修正，α取0.05左右，不同的剂量优化参数向量可以取相同的α值，也可以取不同α值。对于权重优化参数向量的修正，α取0.20左右，不同的权重优化参数向量可以取相同的α值，也可以取不同α值。
步骤S13：并行对所述预设数量套所述备选优化参数向量集进行计划参数优化，并分别计算各套所述备选优化参数向量集的总成本函数值。
在步骤S13中，继续参见图2，首先对所述预设数量套所述备选优化参数向量集并行进行计划参数优化，获取各套所述备选优化参数向量集的当前计划参数和当前剂量学指标。可选的，针对任一套所述备选优化参数向量集，基于CT影像及结构确定射野角度或范围；根据所述射野角度应用基于梯度信息的优化算法优化计划参数，获取当前计划参数和当前剂量学指标，其中所述当前剂量学指标包括靶区均匀性数值、靶区适形性数值以及危及器官的体积剂量及平均剂量。
然后根据所述当前计划参数、所述当前剂量学指标以及目标剂量学指标应用多功能优化总成本函数计算各套所述备选优化参数向量集的总成本函数值。可选的，针对任一套所述备选优化参数向量集，根据所述当前计划参数应用训练后的伽马通过率预测模型进行预测，获取预测伽马通过率，并根据所述预测伽马通过率应用计划执行准确度成本函数计算计划执行准确度成本函数值；根据所述当前剂量学指标和所述目标剂量学指标应用计划剂量学质量成本函数计算计划剂量学质量成本函数值；根据所述计划执行准确度成本函数值和所述计划剂 量学质量成本函数值应用多功能优化总成本函数计算总成本函数值。总成本函数值的计算方法与步骤S11中的总成本函数值的计算方法相同，在此不再赘述。
步骤S14：根据所述总成本函数值确定当前最优及次优备选优化参数向量集，根据所述当前备选优化参数向量集的总成本函数值，抽样更新所述当前优化参数向量集和所述当前成本函数值，并进行重复迭代，直至满足收敛条件。
在本发明实施例中，从预设数量N套备选优化参数向量集中选择总成本函数值最低的备选优化参数向量集，确定为当前最优备选优化参数向量集和当前次优备选优化参数向量集。
继续参见图2，抽样更新当前优化参数向量集和当前成本函数值时，抽样决策是否采纳所述当前最优或次优备选优化参数向量集。如果采纳所述当前最优或次优备选优化参数向量集，则将所述当前最优或次优备选优化参数向量集和对应的总成本函数值分别作为当前优化参数向量集和当前成本函数值。然后判断是否满足收敛条件，其中所述收敛条件为达到预设的最大迭代次数或最长迭代时间、或者成本函数值降幅收敛等。如果不采纳当前最优或次优备选优化参数向量集，则直接判断是否满足收敛条件。如果不满足收敛条件，则返回步骤S12，以进行下一次的迭代。如果满足收敛条件，则迭代结束。
在步骤S14中，抽样决策是否采纳所述当前最优或次优备选优化参数向量集时，首先根据所述当前最优或次优备选优化参数向量集的所述总成本函数值以及所述当前成本函数值应用以下关系式计算所述当前最优或次优备选优化参数向量集的接受概率
其中，和分别为当前最优或次优备选优化参数向量集对应的总成本函数值和当前成本函数值，β为初始温度参数，优选为0.01-0.10，γ为衰减因子参数，优选为0.01-0.10，k为优化迭代次数。
然后基于均匀分布随机产生0至1之间的随机数；如果所述随机数小于计算的所述接受概率，则确定采纳所述当前最优或次优备选优化参数向量集；否则不采纳所述当前最优或次优备选优化参数向量集。
现有技术中的人工试错的过程为：根据所设置的优化参数进行计划参数优化，计算本次优化的剂量学指标，判断所有靶区或危及器官的剂量学指标是否达到临床要求、是否还有改进空间，若计划未达到临床要求或尚有改进空间，设计者需根据经验人工修改对应的计划优化参数，重复进行计划参数优化，直至计划达到临床要求，并达到设计者认为的最优状态，至此试错完成，可以输出计划。试错过程占据了临床计划优化的大部分时间，且严重依赖于 设计者的经验。
本发明实施例基于元启发算法模拟人工计划试错，自动调整计划的优化参数，实现放疗计划的自动设计。具体如步骤S12-步骤S14，应用进化策略和模拟退火融合算法模拟人工计划试错，较少依赖人工计划设计经验，主观误差小，自动化程度高，无需建立特定医疗机构、特定病种、特定设计者的计划数据库或优化模板，节省人力和时间成本，易于推广，减少患者的治疗等待时间。
步骤S15：输出迭代完成后的最终最优的优化参数向量集，确定计划参数，计算并输出多叶准直器的计划叶片走位和剂量分布。
在本发明实施例中，如果满足收敛条件，则迭代结束，输出迭代完成后的最终最优的优化参数向量集，确定计划参数，计算并输出多叶准直器的计划叶片走位和剂量分布，即得到最终的肿瘤放疗计划。其中迭代完成后的最终最优的优化参数向量集是指所有迭代步骤内总成本函数最小的优化参数向量集。
本发明实施例还进行脑瘤和肺癌的本发明自动计划方法与人工计划方法对比以验证本发明实施例的肿瘤放疗计划设计方法：包含7例脑瘤患者和7例肺癌患者的放疗计划设计，收集了多部位计划的1541对射野通量和对应的GPR，组成数据集。将数据集随机分为训练集、验证集和测试集，使用训练集和验证集训练GRP预测模型，使用测试集测试模型的预测准确度，平均绝对误差为0.015。
本发明实施例的肿瘤放疗计划设计方法中GPR预测模型采用深度学习模型，模型主体为残差卷积网络，具体模型结构参见图3，其中，MC Dropout表示蒙特卡罗抽样(简称蒙卡抽样)，修正线性单元(Rectified Linear Unit，ReLU)是一种神经激活函数。GPR预测模型的输入为计划射野通量图，经过自动补零和中心裁剪标准化模型输入，再经过9个残差卷积模块编码、2个全连接编码和Sigmoid函数激活后，输出预测的GPR。
为了保证放疗的准确度和疗效，在对患者实施治疗之前，需要使用射线探测器阵列对患者的放疗计划进行实验测量，并分析测量结果的伽马通过率(Gama Passing Rate,GPR)。GPR由放疗计划计算剂量和实测剂量的差异决定，数值在0％-100％之间，越接近100％表示计划通过率越高、准确度越好。美国医学物理学家协会(American Association of Physicists in Medicine，AAPM)出版的TG 218号报告推荐使用90％GPR(计算参数3％,2mm)作为计划通过与否的干预标准，GPR≥90％表示计划通过预警，GPR＜90％表示不通过。
表1自动计划与人工计划的计划质量统计对比表

表1为使用本发明实施例的肿瘤放疗计划设计方法的自动计划与人工计划的计划质量统计对比。本发明实施例自动计划的靶区均匀性明显优于人工计划，靶区适形性接近人工计划，脑干、晶体、视神经、视交叉、垂体、脊髓、肺、心脏和全身的剂量学指标均优于人工计划。
表2.自动计划与人工计划的执行准确度对比
表2为使用本发明方法的自动计划与人工计划的执行准确度测量结果统计对比。本发明基于多功能优化自动计划的GPR略高于人工计划，测量未通过(测量GPR<90％定义为未通过)射野数量显著降低。
本发明实施例的肿瘤放疗计划设计方法具备多功能同步优化能力，将GPR预测模型纳入计划优化框架，结合元启发算法和射野通量优化算法能够实现放疗计划质量和计划执行准确度同步优化，兼顾质量和执行准确度，降低不通过计划比例，有望改善患者疗效；使用进化策略和模拟退火融合算法模拟计划优化的人工试错过程，较少依赖人工计划设计经验，主观误差小，自动化程度高，无需建立特定医疗机构、特定病种、特定设计者的计划数据库或优化模板，节省人力和时间成本，易于推广，减少患者的治疗等待时间。
本发明实施例的肿瘤放疗计划设计方法通过获取作为当前优化参数向量集的初始优化参数向量集，并计算当前成本函数值；随机修正所述当前优化参数向量集，产生预设数量套备选优化参数向量集；并行对所述预设数量套所述备选优化参数向量集进行计划参数优化，并分别计算各套所述备选优化参数向量集的总成本函数值；根据所述总成本函数值确定当前最优及次优备选优化参数向量集，根据所述当前最优备选优化参数向量集的总成本函数值，抽样更新所述当前优化参数向量集和所述当前成本函数值，并进行重复迭代，直至满足收敛条件；输出迭代完成后的最终最优的优化参数向量集，确定计划参数，计算并输出多叶准直器的计划叶片走位和剂量分布，能够同步优化放疗计划剂量学质量和执行准确度，实现计划设计自动化，较少依赖人工经验，能够节省临床人力和时间成本。
上述对本发明特定实施例进行了描述。在一些情况下，在本发明实施例中记载的动作或 步骤可以按照不同于实施例中的顺序来执行并且仍然可以实现期望的结果。另外，在附图中描绘的过程不一定要求示出的特定顺序或者连续顺序才能实现期望的结果。在某些实施方式中，多任务处理和并行处理也是可以的或者可能是有利的。
基于同一个构思，本发明实施例还提供了一种肿瘤放疗计划设计装置。附图8所示，肿瘤放疗计划设计装置包括：初始化单元、备选集获取单元、总成本计算单元、更新迭代单元以及计划输出单元。其中，
初始化单元，用于获取作为当前优化参数向量集的初始优化参数向量集，并计算当前成本函数值；
备选集获取单元，用于随机修正所述当前优化参数向量集，产生预设数量套备选优化参数向量集；
总成本计算单元，用于并行对所述预设数量套所述备选优化参数向量集进行计划参数优化，并分别计算各套所述备选优化参数向量集的总成本函数值；
更新迭代单元，用于根据所述总成本函数值确定当前最优及次优备选优化参数向量集，根据所述当前最优备选优化参数向量的总成本函数值，集抽样更新所述当前优化参数向量集和所述当前成本函数值并进行重复迭代，直至满足收敛条件；
计划输出单元，用于输出迭代完成后的最终最优的优化参数向量集，确定计划参数，计算并输出多叶准直器的计划叶片走位和剂量分布。
为了描述的方便，描述以上装置时根据功能分为各种模块分别进行描述。当然，在实施本发明实施例时可以把各模块的功能在同一个或多个软件和/或硬件中实现。
上述实施例的装置应用于前述实施例中相应的方法，并且具有相应的方法实施例的有益效果，在此不再赘述。
基于同一发明构思，本发明实施例还提供了一种电子设备，该电子设备，包括存储器、处理器及存储在存储器上并可在处理器上运行的计算机程序，所述处理器执行所述程序时实现如上任意一实施例所述的方法。
本发明实施例提供了一种非易失性计算机存储介质，所述计算机存储介质存储有至少一项可执行指令，该计算机可执行指令可执行如上任意一实施例中所述的方法。
图5示出了本实施例所提供的一种更为具体的电子设备硬件结构示意图，该设备可以包括：处理器501、存储器502、输入/输出接口503、通信接口504和总线505。其中处理器501、存储器502、输入/输出接口503和通信接口504通过总线505实现彼此之间在设备内部的通信连接。
处理器501可以采用通用的CPU(Central Processing Unit，中央处理器)、微处理器、应用专用集成电路(Application Specific Integrated Circuit，ASIC)、或者一个或多个集成电路等方式 实现，用于执行相关程序，以实现本发明方法实施例所提供的技术方案。
存储器502可以采用ROM(Read Only Memory，只读存储器)、RAM(Random AccessMemory，随机存取存储器)、静态存储设备，动态存储设备等形式实现。存储器502可以存储操作系统和其他应用程序，在通过软件或者固件来实现本发明方法实施例所提供的技术方案时，相关的程序代码保存在存储器502中，并由处理器501来调用执行。
输入/输出接口503用于连接输入/输出模块，以实现信息输入及输出。输入输出/模块可以作为组件配置在设备中(图中未示出)，也可以外接于设备以提供相应功能。其中输入设备可以包括键盘、鼠标、触摸屏、麦克风、各类传感器等，输出设备可以包括显示器、扬声器、振动器、指示灯等。
通信接口504用于连接通信模块(图中未示出)，以实现本设备与其他设备的通信交互。其中通信模块可以通过有线方式(例如USB、网线等)实现通信，也可以通过无线方式(例如移动网络、WIFI、蓝牙等)实现通信。
总线505包括一通路，在设备的各个组件(例如处理器501、存储器502、输入/输出接口503和通信接口504)之间传输信息。
需要说明的是，尽管上述设备仅示出了处理器501、存储器502、输入/输出接口503、通信接口504以及总线505，但是在具体实施过程中，该设备还可以包括实现正常运行所必需的其他组件。此外，本领域的技术人员可以理解的是，上述设备中也可以仅包含实现本发明实施例方案所必需的组件，而不必包含图中所示的全部组件。
所属领域的普通技术人员应当理解：以上任何实施例的讨论仅为示例性的，并非旨在暗示本申请的范围被限于这些例子；在本申请的思路下，以上实施例或者不同实施例中的技术特征之间也可以进行组合，步骤可以以任意顺序实现，并存在如上所述的本申请的不同方面的许多其它变化，为了简明它们没有在细节中提供。
本申请旨在涵盖落入本发明实施例的宽泛范围之内的所有这样的替换、修改和变型。因此，凡在本发明实施例的精神和原则之内，所做的任何省略、修改、等同替换、改进等，均应包含在本申请的保护范围之内。

Claims (10)

  1. 一种肿瘤放疗计划设计方法，其特征是，所述方法包括：
    获取作为当前优化参数向量集的初始优化参数向量集，并计算当前成本函数值；
    随机修正所述当前优化参数向量集，产生预设数量套备选优化参数向量集；
    并行对所述预设数量套所述备选优化参数向量集进行计划参数优化，并分别计算各套所述备选优化参数向量集的总成本函数值；
    根据所述总成本函数值确定当前最优及次优备选优化参数向量集，根据所述当前备选优化参数向量集的总成本函数值，抽样更新所述当前优化参数向量集和所述当前成本函数值，并进行重复迭代，直至满足收敛条件；
    输出迭代完成后的最终最优的优化参数向量集，确定计划参数，计算并输出多叶准直器的计划叶片走位和剂量分布。
  2. 如权利要求1所述的方法，其特征是，所述获取作为当前优化参数向量集的初始优化参数向量集，并计算当前成本函数值之前，包括：
    收集多个部位肿瘤放疗的计划参数和对应的伽马通过率数据，建立训练数据集；
    基于所述训练数据集对伽马通过率预测模型进行训练，得到训练后的所述伽马通过率预测模型。
  3. 如权利要求1所述的方法，其特征是，所述并行对所述预设数量套所述备选优化参数向量集进行计划参数优化，并分别计算各套所述备选优化参数向量集的总成本函数值，包括：
    对所述预设数量套所述备选优化参数向量集并行进行计划参数优化，获取各套所述备选优化参数向量集的当前计划参数和当前剂量学指标；
    根据所述当前计划参数、所述当前剂量学指标以及目标剂量学指标应用多功能优化总成本函数计算各套所述备选优化参数向量集的总成本函数值。
  4. 如权利要求3所述的方法，其特征是，所述对所述预设数量套优化参数向量集进行计划参数优化，获取各套所述备选优化参数向量集的当前计划参数和当前剂量学指标，包括：针对任一套所述备选优化参数向量集，
    基于CT影像及结构确定射野角度或范围；
    根据所述射野角度应用基于梯度信息的优化算法优化计划参数，获取当前计划参数和当前剂量学指标，其中所述当前剂量学指标包括靶区均匀性数值、靶区适形性数值以及危及器官的体积剂量及平均剂量。
  5. 如权利要求3所述的方法，其特征是，所述根据所述当前计划参数、所述当前剂量学指标以及目标剂量学指标应用多功能优化总成本函数计算各套所述备选优化参数向量集的总成本函数值，包括：针对任一套所述备选优化参数向量集，
    根据所述当前计划参数应用训练后的伽马通过率预测模型进行预测，获取预测伽马通过 率，并根据所述预测伽马通过率应用计划执行准确度成本函数计算计划执行准确度成本函数值；
    根据所述当前剂量学指标和所述目标剂量学指标应用计划剂量学质量成本函数计算计划剂量学质量成本函数值；
    根据所述计划执行准确度成本函数值和所述计划剂量学质量成本函数值应用多功能优化总成本函数计算总成本函数值。
  6. 如权利要求1所述的方法，其特征是，所述根据所述当前备选优化参数向量集的总成本函数值，抽样更新所述当前优化参数向量集和所述当前成本函数值，并进行重复迭代，直至满足收敛条件，包括：
    抽样决策是否采纳所述当前最优或次优备选优化参数向量集；如果采纳所述当前最优或次优备选优化参数向量集，则将所述当前最优或次优备选优化参数向量集和对应的总成本函数值分别作为当前优化参数向量集和当前成本函数值；
    判断是否满足收敛条件，其中所述收敛条件为达到预设的最大迭代次数或最长迭代时间、或者成本函数值降幅收敛；
    如果不满足收敛条件，则返回所述随机修正当前优化参数向量集，产生预设数量套备选优化参数向量集的步骤。
  7. 如权利要求6所述的方法，其特征是，所述抽样决策是否采纳所述当前最优或次优备选优化参数向量集，包括：
    根据所述当前最优或次优备选优化参数向量集的所述总成本函数值以及所述当前成本函数值计算所述当前最优或次优备选优化参数向量集的接受概率；
    基于均匀分布随机产生0至1之间的随机数；
    如果所述随机数小于计算的所述接受概率，则确定采纳所述当前最优或次优备选优化参数向量集；否则不采纳所述当前最优或次优备选优化参数向量集。
  8. 一种肿瘤放疗计划的设计装置，其特征是，所述装置包括：
    初始化单元，用于获取作为当前优化参数向量集的初始优化参数向量集，并计算当前成本函数值；
    备选集获取单元，用于随机修正所述当前优化参数向量集，产生预设数量套备选优化参数向量集；
    总成本计算单元，用于并行对所述预设数量套所述备选优化参数向量集进行计划参数优化，并分别计算各套所述备选优化参数向量集的总成本函数值；
    更新迭代单元，用于根据所述总成本函数值确定当前最优或次优备选优化参数向量集，根据所述当前备选优化参数向量集的总成本函数值，抽样更新所述当前优化参数向量集和所 述当前成本函数值并进行重复迭代，直至满足收敛条件；
    计划输出单元，用于输出迭代完成后的最终最优的优化参数向量集，确定计划参数，计算并输出多叶准直器的计划叶片走位和剂量分布。
  9. 一种电子设备，包括存储器、处理器及存储在存储器上并可在处理器上运行的计算机程序，其特征是，所述处理器执行所述程序时实现如权利要求1-7中任意一项所述的方法。
  10. 一种计算机存储介质，其特征是，所述存储介质中存储有至少一可执行指令，所述可执行指令使处理器执行如权利要求1-7中任一项所述的方法。
PCT/CN2023/087510 2022-11-09 2023-04-11 肿瘤放疗计划设计方法、装置、电子设备及存储介质 WO2024098659A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US18/354,621 US20240157171A1 (en) 2022-11-09 2023-07-18 Tumor radiotherapy planning design method and apparatus, electronic device and storage medium

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202211399398.1A CN115938564A (zh) 2022-11-09 2022-11-09 肿瘤放疗计划设计方法、装置、电子设备及存储介质
CN202211399398.1 2022-11-09

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US18/354,621 Continuation US20240157171A1 (en) 2022-11-09 2023-07-18 Tumor radiotherapy planning design method and apparatus, electronic device and storage medium

Publications (1)

Publication Number Publication Date
WO2024098659A1 true WO2024098659A1 (zh) 2024-05-16

Family

ID=86655042

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2023/087510 WO2024098659A1 (zh) 2022-11-09 2023-04-11 肿瘤放疗计划设计方法、装置、电子设备及存储介质

Country Status (2)

Country Link
CN (1) CN115938564A (zh)
WO (1) WO2024098659A1 (zh)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115938564A (zh) * 2022-11-09 2023-04-07 中南大学湘雅医院 肿瘤放疗计划设计方法、装置、电子设备及存储介质

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105617536A (zh) * 2015-12-24 2016-06-01 上海联影医疗科技有限公司 旋转逆向调强放疗优化方法及装置
US20170091387A1 (en) * 2015-09-25 2017-03-30 Varian Medical Systems International Ag. Clinical goal treatment planning and optimization
CN110327554A (zh) * 2019-07-08 2019-10-15 南方医科大学 基于预测剂量分布引导的调强放疗计划优化方法及应用
US20210158929A1 (en) * 2019-07-16 2021-05-27 Elekta Ab (Publ) Parameter search in radiotherapy treatment plan optimization
CN115938564A (zh) * 2022-11-09 2023-04-07 中南大学湘雅医院 肿瘤放疗计划设计方法、装置、电子设备及存储介质

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20170091387A1 (en) * 2015-09-25 2017-03-30 Varian Medical Systems International Ag. Clinical goal treatment planning and optimization
CN105617536A (zh) * 2015-12-24 2016-06-01 上海联影医疗科技有限公司 旋转逆向调强放疗优化方法及装置
CN110327554A (zh) * 2019-07-08 2019-10-15 南方医科大学 基于预测剂量分布引导的调强放疗计划优化方法及应用
US20210158929A1 (en) * 2019-07-16 2021-05-27 Elekta Ab (Publ) Parameter search in radiotherapy treatment plan optimization
CN115938564A (zh) * 2022-11-09 2023-04-07 中南大学湘雅医院 肿瘤放疗计划设计方法、装置、电子设备及存储介质

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
YANG XIAOYU; LI SHUZHOU; SHAO QIGANG; CAO YING; YANG ZHEN; ZHAO YU-QIAN: "Uncertainty-guided man–machine integrated patient-specific quality assurance", RADIOTHERAPY AND ONCOLOGY, ELSEVIER, IRELAND, vol. 173, 23 May 2022 (2022-05-23), Ireland , pages 1 - 9, XP087133564, ISSN: 0167-8140, DOI: 10.1016/j.radonc.2022.05.016 *
杨晓喻 等 (YANG, XIAOYU ET AL.): "基于元启发策略的肿瘤调强放疗自动计划方法 (A metaheuristics-based automatic planning method for intensity-modulated radiation therapy)", 中华放射医学与防护杂志 (CHINESE JOURNAL OF RADIOLOGICAL MEDICINE AND PROTECTION), vol. 43, no. 1, 25 January 2023 (2023-01-25) *

Also Published As

Publication number Publication date
CN115938564A (zh) 2023-04-07

Similar Documents

Publication Publication Date Title
Vandewinckele et al. Overview of artificial intelligence-based applications in radiotherapy: Recommendations for implementation and quality assurance
US11557390B2 (en) Radiotherapy treatment plan modeling using generative adversarial networks
US11996178B2 (en) Parameter search in radiotherapy treatment plan optimization
US20210339049A1 (en) Radiotherapy treatment plans using differentiable dose functions
AU2020312643B2 (en) Radiotherapy treatment plan optimization using machine learning
US20160129282A1 (en) Systems and methods for specifying treatment criteria and treatment parameters for patient specific radiation therapy planning
US11367520B2 (en) Compressing radiotherapy treatment plan optimization problems
Fiege et al. PARETO: A novel evolutionary optimization approach to multiobjective IMRT planning
WO2024098659A1 (zh) 肿瘤放疗计划设计方法、装置、电子设备及存储介质
US11738208B2 (en) Pareto optimal projection search (POPS) for automated radiation therapy treatment planning
CN107145723B (zh) 基于人工神经网络的医院流程管理系统
EP3395405B1 (en) System and method for radiation therapy treatment planning
CN105377368A (zh) 用于控制imrt递送复杂度的交互式基于剂量梯度的优化技术
Mayo et al. Establishment of practice standards in nomenclature and prescription to enable construction of software and databases for knowledge-based practice review
EP3673956B1 (en) Systems and methods for treatment planning using continuous quality metrics
Chow Artificial intelligence in radiotherapy and patient care
Wall et al. Impact of database quality in knowledge-based treatment planning for prostate cancer
Mashayekhi et al. Artificial intelligence guided physician directive improves head and neck planning quality and practice Uniformity: A prospective study
Kallis et al. Automated treatment planning framework for brachytherapy of cervical cancer using 3D dose predictions
US20240157171A1 (en) Tumor radiotherapy planning design method and apparatus, electronic device and storage medium
US11651847B2 (en) Dose volume histogram and dose distribution based autoplanning
Ripsman et al. Light Pareto robust optimization for IMRT treatment planning
Huang Developing automated tools in radiation therapy to streamline treatment planning and improve plan quality
Vaniqui et al. Treatment plan quality assessment for radiotherapy of rectal cancer patients using prediction of organ-at-risk dose metrics
Bakx et al. Evaluation of a clinically introduced deep learning model for radiotherapy treatment planning of breast cancer

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 23887361

Country of ref document: EP

Kind code of ref document: A1