CN-121999982-A - Automatic optimization design method for radiotherapy plan based on machine learning

CN121999982ACN 121999982 ACN121999982 ACN 121999982ACN-121999982-A

Abstract

The invention discloses an automatic optimization design method of a radiotherapy plan based on machine learning, which relates to the technical field of radiotherapy, and comprises the steps of firstly collecting preprocessing radiotherapy core data to generate a training set, training a feature extraction model, a dose prediction and optimization adjustment model, processing data to be processed through the feature extraction model, screening 3 optimal similar cases by combining cosine similarity and extracting effective features, outputting an initial dose plan, constraint parameters and a deposition matrix through the dose prediction model based on feature vectors and the effective features, constructing an objective function, and adopting an improved ant colony algorithm to fuse conjugate gradient method iteration optimization of the radiation field parameters and photon flux. The invention combines clinical experience and individual difference, realizes efficient and accurate generation of radiotherapy plans, and improves plan quality stability and clinical suitability.

Inventors

LI YING
Sun Jiachi
LI WANLING

Assignees

盐亭县肿瘤医院

Dates

Publication Date: 20260508
Application Date: 20260128

Claims (7)

1. The automatic radiotherapy plan optimization design method based on machine learning is characterized by comprising the following steps of: s1, collecting radiotherapy core data, preprocessing the radiotherapy core data, and outputting a training data set; S2, constructing and training a feature extraction model, a dose prediction model and an optimization adjustment model based on the training data set; S3, processing the current data to be processed by adopting the feature extraction model, outputting feature vectors, determining similar cases based on cosine similarity between the feature vectors and the training data set, and acquiring effective features through the similar cases; S4, based on the feature vector and the effective feature, outputting an initial dose plan, constraint parameters and a deposition matrix by adopting the dose prediction model; S5, constructing an objective function by combining the initial dose plan, the constraint parameters and the deposition matrix, and outputting an optimized radiotherapy plan through iterative optimization of the optimization adjustment model.
2. The automatic radiotherapy plan optimizing design method based on machine learning according to claim 1, wherein the step S3 specifically comprises: s3.1, extracting a model by using the trained characteristics Processing current data to be processed Obtaining the feature vector Calculating And training data set Feature vectors of all cases in (3) Cosine similarity of included angle (L) Screening out 3 cases with highest similarity; S3.2, extracting effective characteristics which are directly related to radiotherapy plan generation and are subjected to clinical verification from historical complete data of 3 similar cases, wherein the effective characteristics comprise clinical constraint parameters, plan adaptation parameters and optimization experience parameters; S3.3, the parameters of 3 similar cases are arranged into a structured collection, Wherein A complete set of auxiliary parameters representing the i-th similar case.
3. The machine learning based automatic radiotherapy planning optimization design method of claim 2, wherein the clinical constraint parameters comprise PTV dose range OAR dose threshold CI threshold value HI threshold The plan adaptation parameters comprise the number of the fields Optimized section of angle of the field Key coefficient of deposition matrix The optimized experience parameters comprise weight coefficients of PTV and OAR in an objective function 。
4. The automatic radiotherapy plan optimizing design method based on machine learning according to claim 2, wherein the step S4 specifically comprises: S4.1, receiving the feature vector And structured collection Will be Clinical constraint parameters and plan adaptation parameters of similar cases are standardized and matched with Spliced into As a model for dose prediction Is input to the computer; s4.2, calling the dose prediction model Based on Outputting an initial dose plan : Prescription dose Based on tumor type And (3) with Prescription dose for similar cases Generating; ; Wherein, the For a reference prescribed dose corresponding to a tumor type, For the prescribed dose of the i-th similar case, Weighting the tumor characteristics of the current case; Parameters of the field of view Based on Number of medium-field intervals And angle optimization interval Generating the number of the fields of the current data to be processed Angle of the field Initial intensity map ; ; Wherein, the As the reference field angle corresponding to the current data to be processed, The field angle for the i-th similar case, Adapting weights for similar case experiences; s4.3 from The medium screening is clinically verified to be effective constraint parameters, and the constraint parameters are formed by combining general clinical standard optimization Including PTV dose ranges OAR dose threshold CI threshold value With HI threshold ; S4.4 based on The radiation field parameter in the system is calculated by adopting a pencil beam algorithm to calculate the unit intensity dose contribution of each beam to all voxels Construction of a deposition matrix ; ; ; Wherein, the For the radiation attenuation correction factor, For the voxel tissue attenuation coefficient, For the path length of the mth beam through the jth voxel, For the scattering radius of the beam within the voxel, As a function of the scatter dose distribution, For the total number of voxels, For the total number of beams 。
5. The automatic radiotherapy plan optimizing design method based on machine learning according to claim 4, wherein the step S5 specifically comprises: S5.1, receiving an initial dose plan Constraint parameters Deposition matrix Analyzing the parameters of the radiation field Will restrict the parameter amount Optimizing boundaries for an algorithm; s5.2, based on constraint parameter quantity Is fused to a deposition matrix Is used for constructing an objective function ; S5.3, planning with initial dose In order to optimize the starting point, the objective function of S5.2 is used as a judging core, an improved ant colony algorithm is started, the intensity of the radiation field, the position of the blade and the angle of the radiation field are optimized, the photon flux is optimized by a fusion conjugate gradient method, after each round of parameter adjustment, the actual dose of voxels is calculated by means of the deposition matrix logic of S5.1, the current dose distribution is deduced, iteration is stopped when the current dose distribution meets the clinical index of the core, and the optimal parameter combination is output, so that the optimized radiotherapy plan is obtained.
6. The automatic radiotherapy plan optimizing design method based on machine learning according to claim 5, wherein the step S5.2 specifically comprises: ; ; ; Wherein, the For the weight of the PTV, As a result of the weight of the OAR, Is a PTV voxel weight, For the OAR voxel weights, The dosage value is 1, the dosage is overrun, and 0 is compliance; for the photon flux of the mth beam, For the number of voxels of the PTV, For i number of OAR voxels, Is the clinical dose threshold for the ith OAR.
7. The automatic radiotherapy plan optimizing design method based on machine learning according to claim 5, wherein the improved ant colony algorithm specifically comprises: Pheromone updating mechanism: ; Wherein, the In order to update the post-pheromone concentration, In order to obtain the volatility coefficient, For the number of ants, Is the pheromone increment of the nth ant, In order to correct the coefficient of the score, Is the current optimal plan score. Path selection probability: ; Wherein, the Is a pheromone inspiring factor, In order for the heuristic factor to be desirable, In order for the value of the heuristic to be a value, The cost of dose loss for parameter adjustment, U is the non-optimized parameter set, A direction guide operator; CG algorithm photon flux optimization: wherein, the method comprises the steps of, For the flux of photons of the kth iteration, In order to optimize the step size, Gradient for the objective function; optimizing field intensity Leaf position of multi-leaf collimator Angle of the field Guiding ants to search optimal parameter combinations through a pheromone updating mechanism, and optimizing photon flux by fusion conjugate gradient method Improving the convergence efficiency of the dose distribution, and in each iteration, the method is based on With the current photon flux Calculating actual dose of voxels 。

Description

Automatic optimization design method for radiotherapy plan based on machine learning Technical Field The invention relates to the technical field of radiotherapy, in particular to an automatic optimized design method of a radiotherapy plan based on machine learning. Background Conventional radiotherapy planning is a time-consuming and labor-consuming process, requiring a physical operator to constantly adjust the parameters of the radiation field and the optimized parameters in the planning optimization to find the optimal plan. In addition, experience differences between planning designers, time taken to plan, and clinical indicators of medical institutions are also relevant. The existing automatic planning method has the problems of multiple iteration times, poor optimization pertinence and difficulty in accurately adapting to individual case differences, and part of methods are easy to influence the dose distribution of non-target organs by globally adjusting the optimization conditions, so that the planning quality is unstable. Along with the development of radiotherapy technology, the requirements of clinic on the accuracy, individuation and efficiency of the plan are increasingly improved, and a method for realizing the automatic optimization of the efficient, accurate and individuation radiotherapy plan by integrating the machine learning technology is urgently needed. Disclosure of Invention In view of the above, the invention provides an automatic optimized design method of radiotherapy plans based on machine learning, which is used for solving the problems existing in the background technology. In order to achieve the above purpose, the present invention adopts the following technical scheme: a radiotherapy plan automatic optimization design method based on machine learning comprises the following steps: s1, collecting radiotherapy core data, preprocessing the radiotherapy core data, and outputting a training data set; S2, constructing and training a feature extraction model, a dose prediction model and an optimization adjustment model based on the training data set; S3, processing the current data to be processed by adopting the feature extraction model, outputting feature vectors, determining similar cases based on cosine similarity between the feature vectors and the training data set, and acquiring effective features through the similar cases; S4, based on the feature vector and the effective feature, outputting an initial dose plan, constraint parameters and a deposition matrix by adopting the dose prediction model; S5, constructing an objective function by combining the initial dose plan, the constraint parameters and the deposition matrix, and outputting an optimized radiotherapy plan through iterative optimization of the optimization adjustment model. Preferably, the step S3 specifically includes: s3.1, extracting a model by using the trained characteristics Processing current data to be processedObtaining the feature vectorCalculatingAnd training data setFeature vectors of all cases in (3)Cosine similarity of included angle (L)Screening out 3 cases with highest similarity; S3.2, extracting effective characteristics which are directly related to radiotherapy plan generation and are subjected to clinical verification from historical complete data of 3 similar cases, wherein the effective characteristics comprise clinical constraint parameters, plan adaptation parameters and optimization experience parameters; S3.3, the parameters of 3 similar cases are arranged into a structured collection, WhereinA complete set of auxiliary parameters representing the i-th similar case. Preferably, the clinical constraint parameter comprises a PTV dose rangeOAR dose thresholdCI threshold valueHI thresholdThe plan adaptation parameters comprise the number of the fieldsOptimized section of angle of the fieldKey coefficient of deposition matrixThe optimized experience parameters comprise weight coefficients of PTV and OAR in an objective function。 Preferably, the S4 specifically includes: S4.1, receiving the feature vector And structured collectionWill beClinical constraint parameters and plan adaptation parameters of similar cases are standardized and matched withSpliced intoAs a model for dose predictionIs input to the computer; s4.2, calling the dose prediction model Based onOutputting an initial dose plan: Prescription doseBased on tumor typeAnd (3) withPrescription dose for similar casesGenerating; ; Wherein, the For a reference prescribed dose corresponding to a tumor type,For the prescribed dose of the i-th similar case,Weighting the tumor characteristics of the current case; Parameters of the field of view Based onNumber of medium-field intervalsAnd angle optimization intervalGenerating the number of the fields of the current data to be processedAngle of the fieldInitial intensity map; ; Wherein, the As the reference field angle corresponding to the current data to be processed,The field angle for th