A radiotherapy plan dose distribution prediction method based on domain adaptation transfer learning

Abstract

This invention relates to the field of transfer learning technology and discloses a method for predicting radiotherapy plan dose distribution based on domain adaptive transfer learning. The method includes: identifying source and target domain medical institutions; collecting radiotherapy plan data from the source domain medical institutions; extracting radiotherapy plan features from the radiotherapy plan data to train an initial dose distribution analysis model for the source domain medical institutions; constructing a federated learning network for the source and target domain medical institutions; outputting a dose distribution analysis map of the initial dose distribution analysis model; identifying domain differences between the source and target domain medical institutions; defining the optimization objective of the initial dose distribution analysis model; performing domain adaptive transfer learning on the initial dose distribution analysis model to obtain a local dose distribution analysis model; and generating the target dose distribution for patients in the target domain medical institutions. This invention can improve the efficiency and accuracy of radiotherapy plan dose distribution prediction.

Description

Technical Field

[0001] This invention relates to a method for predicting the dose distribution of radiotherapy plans based on domain adaptive transfer learning, belonging to the field of transfer learning technology. Background Technology

[0002] Radiation therapy planned dose distribution refers to the detailed distribution of radiation dose within the patient's body (especially the tumor area and surrounding tissues) in three-dimensional space, calculated and generated when developing a radiotherapy plan based on the tumor's location, size, shape, and the sensitivity of surrounding normal organs. Predicting radiation therapy planned dose distribution allows for rapid and accurate estimation of the final radiation dose distribution within the patient's body before the formal implementation of the radiotherapy plan, enabling more precise, safer, and more personalized tumor radiotherapy, and improving patient prognosis and quality of life.

[0003] Traditional radiotherapy dose distribution prediction is mainly based on physics engines. It achieves the ideal dose distribution by repeatedly optimizing beam parameters (angle, energy, intensity, etc.). Because this method requires repeated optimization and calculation, the prediction process is complex and time-consuming, which limits the efficiency of quickly trying different schemes and making optimizations. Summary of the Invention

[0004] This invention provides a radiotherapy planning dose distribution prediction method based on domain adaptive transfer learning, the main purpose of which is to improve the efficiency and accuracy of radiotherapy planning dose distribution prediction.

[0005] To achieve the above objectives, this invention provides a radiotherapy planning dose distribution prediction method based on domain adaptive transfer learning, comprising:

[0006] The source and target medical institutions are identified, radiotherapy planning data of the source medical institutions are collected, and radiotherapy planning features of the radiotherapy planning data are extracted. The radiotherapy planning features include anatomical structural features, tumor location, and dose distribution, in order to train the initial dose distribution analysis model of the source medical institutions.

[0007] A federated learning network is constructed between the source domain medical institutions and the target domain medical institutions. Based on the federated learning network, the initial dose distribution analysis model is distributed to the target domain medical institutions. Using the local radiotherapy planning data of the target domain medical institutions, the dose distribution analysis map of the initial dose distribution analysis model is output.

[0008] By combining the dose distribution analysis map and the local radiotherapy planning data, the domain differences between the source domain medical institution and the target domain medical institution are identified;

[0009] Based on the domain differences, the optimization objective of the initial dose distribution analysis model is defined, wherein the optimization objective includes: dose analysis loss, domain adversarial loss and feature alignment loss, so as to perform domain adaptive transfer learning on the initial dose distribution analysis model to obtain a local dose distribution analysis model;

[0010] Based on the local dose distribution analysis model, the target dose distribution for patients in the target domain medical institution is generated.

[0011] Optionally, the radiotherapy plan features extracted from the radiotherapy plan data include:

[0012] The radiotherapy planning data is formatted to obtain standardized radiotherapy planning data;

[0013] The standardized radiotherapy plan data is denoised to obtain denoised radiotherapy plan data;

[0014] The denoised radiotherapy planning data is aligned to obtain aligned radiotherapy planning data.

[0015] Identify the key structures in the aligned radiotherapy planning data, and calculate the geometric and anatomical features of the key structures;

[0016] Based on the geometric features and the anatomical relationship features, the anatomical structural features and tumor location of the aligned radiotherapy planning data are determined;

[0017] Based on the aligned radiotherapy plan data, the target dose characteristics and organ-at-risk dose characteristics at the tumor location are calculated;

[0018] Based on the target area dose characteristics and the organ at risk dose characteristics, the dose distribution at the tumor location is determined;

[0019] The radiotherapy planning characteristics of the radiotherapy planning data are determined based on the dose distribution, the anatomical features, and the tumor location.

[0020] Optionally, training the initial dose distribution analysis model for the source domain medical institutions includes:

[0021] The radiotherapy plan features corresponding to the source domain medical institutions are divided into a training set and a validation set;

[0022] Generate the model framework of the source domain medical institutions;

[0023] Based on the training set, the model framework is trained to obtain a training analysis model;

[0024] Based on the validation set, calculate the mean squared error of the trained analysis model;

[0025] When the mean squared error is greater than the preset error threshold, the model learning parameters of the training analysis model are adjusted, and the process returns to the step of training the model framework based on the training set to obtain the training analysis model.

[0026] When the mean square error is less than a preset error threshold, the trained analysis model is used as the initial dose distribution analysis model for the source domain medical institution.

[0027] Optionally, constructing the federated learning network of the source domain medical institutions and the target domain medical institutions includes:

[0028] Identify the source domain client node of the source domain medical institution and the target domain client node of the target domain medical institution;

[0029] Construct the communication protocol and dynamic encryption protocol between the source domain client node and the target domain client node;

[0030] Based on the communication protocol and the dynamic encryption protocol, a central server is constructed for the source domain client node and the target domain client node;

[0031] Configure the client environment of the source domain client node and the target domain client node in a unified manner;

[0032] Based on the client environment, define the model training loop rules for the source domain client node, the target domain client node, and the central server;

[0033] Based on the model training loop rules, a federated learning network is constructed for the source domain medical institutions and the target domain medical institutions.

[0034] Optionally, the step of outputting the dose distribution analysis map of the initial dose distribution analysis model using the local radiotherapy planning data of the target domain medical institution includes:

[0035] Collect local radiotherapy plan data from medical institutions in the target domain;

[0036] Extract key information from the local radiotherapy plan data;

[0037] Convert the key information into input format information;

[0038] The input format information is input into the initial dose distribution analysis model to obtain dose distribution data;

[0039] Based on the dose distribution data, a dose distribution analysis map of the patients corresponding to the target domain medical institutions is generated.

[0040] Optionally, the step of combining the dose distribution analysis map and the local radiotherapy planning data to identify the domain differences between the source domain medical institution and the target domain medical institution includes:

[0041] Extract the model analysis dose from the dose distribution analysis map and the local planned dose from the local radiotherapy planning data;

[0042] The anatomical structures of the dose distribution analysis map and the local radiotherapy planning data are uniformly identified;

[0043] Based on the anatomical structure, the model-analyzed dose, and the local planned dose, the domain differences between the source domain medical institution and the target domain medical institution are calculated.

[0044] Optionally, defining the optimization objective of the initial dose distribution analysis model based on the domain differences includes:

[0045] Analyze the difference characteristics of the domain differences, including differences in dose-volume histograms and differences in dose distribution;

[0046] Based on the aforementioned differences, the anatomical structure weights, dose level weights, and voxel importance weights of the initial dose distribution analysis model are determined.

[0047] The dose analysis loss of the initial dose distribution analysis model is determined based on the anatomical structure weights, the dose level weights, and the voxel importance weights.

[0048] Construct a domain classifier for the initial dose distribution analysis model;

[0049] Extract the input features of the initial dose distribution analysis model, and input the input features into the domain classifier to obtain the classifier analysis value;

[0050] Based on the classifier analysis values, the domain adversarial loss of the initial dose distribution analysis model is calculated;

[0051] Extract the source domain data features and target domain data features of the initial dose distribution analysis model respectively;

[0052] Calculate the source domain kernel matrix between the source domain data features and the target domain kernel matrix between the target domain data features, respectively.

[0053] Calculate the kernel matrix between the source domain data features and the target domain data features;

[0054] Based on the source domain kernel matrix, the target domain kernel matrix, and the kernel matrix, determine the feature alignment loss of the initial dose distribution analysis model;

[0055] The optimization objective of the initial dose distribution analysis model is determined based on the feature alignment loss, the dose analysis loss, and the domain adversarial loss.

[0056] Optionally, the step of performing domain-adaptive transfer learning on the initial dose distribution analysis model to obtain a local dose distribution analysis model includes:

[0057] Based on the optimization objective of the initial dose distribution analysis model, the initial dose distribution analysis model is trained using preset training data to obtain a trained dose distribution analysis model.

[0058] Calculate the main task output and domain classification output of the trained dose distribution analysis model;

[0059] Based on the main task output and the domain classification output, calculate the main task loss and domain adversarial loss of the training dose distribution analysis model;

[0060] When the loss of the main task of the model is greater than the preset loss threshold of the main task, the model parameters of the trained dose distribution analysis model are adjusted, and then the process of training the initial dose distribution analysis model with preset training data according to the optimization objective of the initial dose distribution analysis model is returned to obtain the trained dose distribution analysis model.

[0061] When the domain adversarial loss of the model is less than the preset domain adversarial loss threshold, after adjusting the classifier parameters of the corresponding domain classifier of the trained dose distribution analysis model, the process returns to the step of training the initial dose distribution analysis model with preset training data according to the optimization objective of the initial dose distribution analysis model to obtain the trained dose distribution analysis model.

[0062] When the model's main task loss is less than a preset main task loss threshold and the model's domain adversarial loss is greater than a preset domain adversarial loss threshold, the training dose distribution analysis model is used as the training dose distribution analysis model.

[0063] Optionally, generating the target dose distribution for patients in the target domain medical institution based on the local dose distribution analysis model includes:

[0064] The patient data corresponding to the patient is input into the local dose distribution analysis model. The patient data is processed through the convolutional layer of the local dose distribution analysis model to obtain high-level features and low-level detail features.

[0065] The high-level features and low-level detail features are fused through the feature fusion layer of the local dose distribution analysis model to obtain fused features;

[0066] Based on the fusion features, the dose distribution analysis map of the patient is calculated through the dose calculation layer of the local dose distribution analysis model;

[0067] Based on the dose distribution analysis diagram, the target dose distribution of the patient is output through the output layer of the local dose distribution analysis model.

[0068] To address the aforementioned problems, this invention also provides a radiotherapy planning dose distribution prediction system based on domain adaptive transfer learning, the system comprising:

[0069] The source domain model training module is used to identify source domain medical institutions and target domain medical institutions, collect radiotherapy plan data of the source domain medical institutions, and extract radiotherapy plan features from the radiotherapy plan data. The radiotherapy plan features include: anatomical structure features, tumor location, and dose distribution, in order to train the initial dose distribution analysis model of the source domain medical institutions.

[0070] The local data prediction module is used to construct a federated learning network between the source domain medical institution and the target domain medical institution, distribute the initial dose distribution analysis model to the target domain medical institution based on the federated learning network, and output the dose distribution analysis map of the initial dose distribution analysis model through the local radiotherapy planning data of the target domain medical institution.

[0071] The domain difference analysis module is used to identify the domain differences between the source domain medical institution and the target domain medical institution by combining the dose distribution analysis map and the local radiotherapy plan data.

[0072] A local model training module is used to define the optimization objective of the initial dose distribution analysis model based on the domain differences. The optimization objective includes dose analysis loss, domain adversarial loss, and feature alignment loss, so as to perform domain adaptive transfer learning on the initial dose distribution analysis model to obtain a local dose distribution analysis model.

[0073] The target dose distribution module is used to generate the target dose distribution for patients in the target domain medical institution based on the local dose distribution analysis model.

[0074] Compared to the problems described in the background art, the embodiments of the present invention can directly generate higher-quality initial plans by training the initial dose distribution analysis model of the source domain medical institution, reducing the number and magnitude of subsequent optimization adjustments and shortening the overall planning time. Optionally, the embodiments of the present invention can achieve cross-institutional data collaboration by constructing a federated learning network between the source domain medical institution and the target domain medical institution under the premise of strictly protecting data privacy, thereby improving the generalization ability, applicability, and overall performance of the initial dose distribution analysis model. The embodiments of the present invention can output the dose distribution analysis map of the initial dose distribution analysis model using the local radiotherapy planning data of the target domain medical institution, which can preliminarily determine the applicability and accuracy of the initial model on the target domain data, providing a basis for whether and how to fine-tune it. The embodiments of the present invention can combine the dose distribution analysis... By analyzing the radiation spectrum and the local radiotherapy planning data, identifying the domain differences between the source and target domain medical institutions can intuitively and quantitatively identify the systematic differences in dose distribution between the two institutions in similar cases. This embodiment of the invention, through domain-adaptive transfer learning of the initial dose distribution analysis model, obtains a local dose distribution analysis model that adapts to specific local environments. This overcomes the challenges posed by inconsistent data distribution while maintaining or even improving performance, achieving more accurate, efficient, and practical localized applications. Finally, by generating the target dose distribution for patients in the target domain medical institution based on the local dose distribution analysis model, this embodiment enhances the model's adaptability and robustness in different medical institution environments, thereby generating dose distributions for target domain patients more reliably and accurately. Therefore, the radiotherapy plan dose distribution prediction method based on domain adaptive transfer learning provided by this embodiment can improve the efficiency and accuracy of radiotherapy plan dose distribution prediction. Attached Figure Description

[0075] Figure 1 This is a flowchart illustrating a radiotherapy planning dose distribution prediction method based on domain adaptive transfer learning, provided in an embodiment of the present invention.

[0076] Figure 2 This is a schematic diagram of a module for implementing the radiotherapy planning dose distribution prediction method based on domain adaptive transfer learning, according to an embodiment of the present invention.

[0077] The objectives, features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0078] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0079] This application provides a method for predicting radiotherapy plan dose distribution based on domain adaptive transfer learning. The execution entity of this method includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the method provided in this application: a server, a terminal, etc. In other words, the method can be executed by software or hardware installed on a terminal device or a server device. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster.

[0080] Example 1:

[0081] Reference Figure 1 The diagram shown is a flowchart illustrating a radiotherapy planning dose distribution prediction method based on domain adaptive transfer learning according to an embodiment of the present invention. In this embodiment, the radiotherapy planning dose distribution prediction method based on domain adaptive transfer learning includes:

[0082] S1. Identify the source domain medical institutions and the target domain medical institutions, collect radiotherapy plan data from the source domain medical institutions, and extract radiotherapy plan features from the radiotherapy plan data. The radiotherapy plan features include: anatomical structure features, tumor location, and dose distribution, in order to train the initial dose distribution analysis model of the source domain medical institutions.

[0083] This invention clarifies that source domain medical institutions and target domain medical institutions can utilize the large amount of data in the source domain to quickly build an initial model, significantly reducing the time and data volume required for training the target domain from scratch. The source domain medical institutions refer to those that provide raw data for training the initial model. The target domain medical institutions refer to those that deploy trained (typically based on source domain data and potentially adjusted) dose prediction models to improve the efficiency and quality of their radiotherapy planning.

[0084] This invention provides a large amount of training data for subsequent model training by collecting radiotherapy plan data from the source medical institutions. The radiotherapy plan data refers to the collection of various information directly related to the treatment plan generated during the development and execution of radiotherapy for patients within the medical institution.

[0085] Optionally, the radiotherapy planning data of the source medical institution can be obtained by exporting it from the radiotherapy planning system.

[0086] This invention, through extracting radiotherapy plan features from the radiotherapy plan data, achieves the structuring and quantification of the plan data, thereby supporting the training and application of machine learning models. The radiotherapy plan features refer to information extracted from the original radiotherapy plan data that can quantitatively describe the key attributes or characteristics of the plan. The anatomical structural features refer to the morphology, location, and extent of various organs, tissues, bones, tumors, etc., within the patient's body, as presented by medical imaging (most commonly CT scans, sometimes including MRI, PET, etc.). The tumor location refers to the specific spatial coordinates and anatomical region of the tumor within the patient's body. The dose distribution refers to the spatial dose intensity pattern formed by the energy deposited in the patient's body (or body surface) by rays generated by radiotherapy equipment (such as a linear accelerator) during the formulation and evaluation of the radiotherapy plan.

[0087] As an embodiment of the present invention, the step of extracting the radiotherapy planning features from the radiotherapy planning data includes:

[0088] The radiotherapy planning data is formatted to obtain standardized radiotherapy planning data;

[0089] The standardized radiotherapy plan data is denoised to obtain denoised radiotherapy plan data;

[0090] The denoised radiotherapy planning data is aligned to obtain aligned radiotherapy planning data.

[0091] Identify the key structures in the aligned radiotherapy planning data, and calculate the geometric and anatomical features of the key structures;

[0092] Based on the geometric features and the anatomical relationship features, the anatomical structural features and tumor location of the aligned radiotherapy planning data are determined;

[0093] Based on the aligned radiotherapy plan data, the target dose characteristics and organ-at-risk dose characteristics at the tumor location are calculated;

[0094] Based on the target area dose characteristics and the organ at risk dose characteristics, the dose distribution at the tumor location is determined;

[0095] The radiotherapy planning characteristics of the radiotherapy planning data are determined based on the dose distribution, the anatomical features, and the tumor location.

[0096] The standardized radiotherapy plan data refers to radiotherapy plan data that has undergone standardized processing steps and meets specific standards, has a unified format, and is of controllable quality. The denoised radiotherapy plan data refers to data after noise and interference have been removed or reduced using filtering techniques. The aligned radiotherapy plan data refers to radiotherapy plan data after coordinate system alignment or registration. The key structures refer to specific anatomical regions of significant clinical and anatomical importance during radiotherapy, such as tumor target areas and vital organs. The geometric features refer to the quantitative description of the shape and spatial position of anatomical structures delineated by medical images (mainly CT scans). The anatomical relationship features refer to the quantitative description of the relative positions, distances, and contact situations of different anatomical structures in three-dimensional space within the radiotherapy plan. The target dose features refer to a series of quantitative indicators used to measure and describe the radiation dose received by the tumor target area in the radiotherapy plan. The organ-at-risk dose features refer to a series of quantitative indicators used to measure and describe the radiation dose received by normal tissues (organs at risk) in the radiotherapy plan.

[0097] Optionally, the denoised radiotherapy planning data can be obtained through filters, such as mean filtering, median filtering, etc.

[0098] Optionally, the geometric features of the key structure can be calculated using boundary detection algorithms, such as the Roberts Cross operator, the Sobel operator, and the Prewitt operator.

[0099] This invention, through training the initial dose distribution analysis model of the source medical institution, can directly generate higher-quality initial plans, reducing the number and magnitude of subsequent optimization adjustments and shortening the overall planning time. The initial dose distribution analysis model refers to a machine learning-based model used to generate, evaluate, or optimize the initial dose distribution corresponding to a radiotherapy plan.

[0100] As an embodiment of the present invention, training the initial dose distribution analysis model of the source domain medical institution includes:

[0101] The radiotherapy plan features corresponding to the source domain medical institutions are divided into a training set and a validation set;

[0102] Generate the model framework of the source domain medical institutions;

[0103] Based on the training set, the model framework is trained to obtain a training analysis model;

[0104] Based on the validation set, calculate the mean squared error of the trained analysis model;

[0105] When the mean squared error is greater than the preset error threshold, the model learning parameters of the training analysis model are adjusted, and the process returns to the step of training the model framework based on the training set to obtain the training analysis model.

[0106] When the mean square error is less than a preset error threshold, the trained analysis model is used as the initial dose distribution analysis model for the source domain medical institution.

[0107] The training set refers to a portion of the radiotherapy plan feature data from source domain medical institutions used to train the initial dose distribution analysis model. The validation set refers to a portion of the radiotherapy plan feature data from source domain medical institutions used to evaluate and validate the model's performance during training. The model framework refers to the computational structure used to construct the initial dose distribution analysis model. The training analysis model refers to the model used to evaluate and optimize the performance of the initial dose distribution analysis model during training. The mean squared error is an indicator used to measure the degree of difference between the model's output value and the true value in the validation set. The preset error threshold is a pre-set boundary value used to determine whether the error generated by the model is within an acceptable range. The model learning parameters are values ​​automatically adjusted by the model during training based on the learned data, such as learning rate, weights, and biases.

[0108] Optionally, the model framework of the source domain medical institution can be generated by linear regression, such as ridge regression, Lasso, neural networks, etc.

[0109] As another implementation, the mean square error of the training analysis model can be calculated using the following formula:

[0110]

[0111] in, This represents the mean square error. This indicates the number of data sets in the validation set. Indicates the first in the verification set The output values ​​of the training and analysis model are set of data. Indicates the first in the verification set The true value of the set of data.

[0112] S2. Construct a federated learning network between the source domain medical institution and the target domain medical institution. Based on the federated learning network, distribute the initial dose distribution analysis model to the target domain medical institution. Using the local radiotherapy planning data of the target domain medical institution, output the dose distribution analysis map of the initial dose distribution analysis model.

[0113] This invention, through the construction of a federated learning network between the source domain medical institutions and the target domain medical institutions, achieves cross-institutional data collaboration while strictly protecting data privacy, thereby improving the generalization ability, applicability, and overall performance of the initial dose distribution analysis model. The federated learning network refers to a machine learning architecture and computational framework that allows multiple participants (in this scenario, source and target domain medical institutions) to jointly train a machine learning model without sharing their respective original data.

[0114] As an embodiment of the present invention, the construction of the federated learning network of the source domain medical institutions and the target domain medical institutions includes:

[0115] Identify the source domain client node of the source domain medical institution and the target domain client node of the target domain medical institution;

[0116] Construct the communication protocol and dynamic encryption protocol between the source domain client node and the target domain client node;

[0117] Based on the communication protocol and the dynamic encryption protocol, a central server is constructed for the source domain client node and the target domain client node;

[0118] Configure the client environment of the source domain client node and the target domain client node in a unified manner;

[0119] Based on the client environment, define the model training loop rules for the source domain client node, the target domain client node, and the central server;

[0120] Based on the model training loop rules, a federated learning network is constructed for the source domain medical institutions and the target domain medical institutions.

[0121] In this context, the source domain client node refers to a medical institution client participating in federated learning, providing an initial or baseline model, and potentially contributing some training data. The target domain client node refers to a medical institution client participating in federated learning and utilizing it to improve the performance of the target domain model. The communication protocol refers to a set of predefined rules, standards, and conventions used to regulate how information is exchanged and communicated between various nodes in the federated learning network (such as source domain client nodes, target domain client nodes, and the central server). The dynamic encryption protocol refers to an encryption protocol that adjusts or changes in real time based on communication conditions, time, data characteristics, or communication context. The central server refers to a node responsible for coordinating and managing the entire distributed learning process. The client environment refers to all the hardware and software conditions and configurations required for each client node (such as source domain client nodes and target domain client nodes) participating in the federated learning network to run the federated learning task on its local device. The model training loop rules refer to a set of steps that regulate how the model undergoes multiple rounds of iterative training throughout the federated learning process.

[0122] Optionally, the dynamic encryption protocol between the source domain client node and the target domain client node can be constructed using a dynamic differential privacy algorithm. For example, the communication content, number of participating nodes, and potential attackers of the data transmitted between the source domain client node and the target domain client node can be determined using a dynamic differential privacy algorithm. Based on the communication content, the number of participating nodes, and the potential attackers, the leakage risk coefficient of the transmitted data can be analyzed. Based on the leakage risk coefficient, the dynamic encryption protocol between the source domain client node and the target domain client node can be constructed.

[0123] Optionally, the model training loop rules for the source domain client node, the target domain client node, and the central server can be defined by iterative optimization algorithms, such as gradient descent algorithm, federated averaging algorithm, etc.

[0124] This invention, through the federated learning network, distributes the initial dose distribution analysis model to medical institutions in the target domain, which can accelerate deployment, improve initial performance, support local personalized optimization, achieve knowledge sharing, and lower the barrier to entry, laying a good foundation for subsequent more accurate dose distribution analysis in the target domain that better meets local needs.

[0125] This invention, through local radiotherapy planning data from the target domain medical institution, outputs a dose distribution analysis map of the initial dose distribution analysis model. This allows for a preliminary assessment of the applicability and accuracy of the initial model on the target domain data, providing a basis for subsequent fine-tuning and determining the appropriate method. The dose distribution analysis map refers to a graphical tool used in radiotherapy planning to visually display the distribution of radiation dose within the target area.

[0126] As an embodiment of the present invention, the step of outputting the dose distribution analysis map of the initial dose distribution analysis model using the local radiotherapy plan data of the target domain medical institution includes:

[0127] Collect local radiotherapy plan data from medical institutions in the target domain;

[0128] Extract key information from the local radiotherapy plan data;

[0129] Convert the key information into input format information;

[0130] The input format information is input into the initial dose distribution analysis model to obtain dose distribution data;

[0131] Based on the dose distribution data, a dose distribution analysis map of the patients corresponding to the target domain medical institutions is generated.

[0132] The local radiotherapy planning data refers to all data related to the patient's radiotherapy plan generated by the target domain medical institution (such as a hospital or clinic) during radiotherapy. The key information refers to the core information necessary for dose distribution analysis using the model, such as target area and organ contour information, imaging data, and radiotherapy planning data. The input format information refers to the data after the extracted key information has been converted and arranged according to the input format required by the initial dose distribution analysis model. The dose distribution data refers to numerical data describing the deposition or distribution of energy (usually radiation energy) within the patient's body.

[0133] Optionally, key information from the local radiotherapy planning data can be extracted using DICOM RT standard parsing, such as Python's pydicom and dcm2jpg, Java's DcmObject, and C++'s DCMTK.

[0134] Optionally, the dose distribution analysis map of the patient corresponding to the target domain medical institution can be generated by image overlay and fusion technology, such as overlaying the dose distribution volume data and the anatomical structure image corresponding to the patient to obtain a dose distribution overlay map, and slicing and rendering the dose distribution overlay map according to the dose value corresponding to the dose distribution volume data to obtain the dose distribution analysis map.

[0135] S3. Combining the dose distribution analysis map and the local radiotherapy planning data, identify the domain differences between the source domain medical institution and the target domain medical institution.

[0136] This invention, by combining the dose distribution analysis map and the local radiotherapy planning data, identifies the domain differences between the source domain medical institution and the target domain medical institution. This allows for a direct and quantitative identification of systematic differences in dose distribution between the two institutions in similar cases. The domain differences refer to systematic differences in distribution, characteristics, statistical properties, or targets between data or models from different sources.

[0137] As an embodiment of the present invention, the step of identifying domain differences between the source domain medical institution and the target domain medical institution by combining the dose distribution analysis map and the local radiotherapy planning data includes:

[0138] Extract the model analysis dose from the dose distribution analysis map and the local planned dose from the local radiotherapy planning data;

[0139] The anatomical structures of the dose distribution analysis map and the local radiotherapy planning data are uniformly identified;

[0140] Based on the anatomical structure, the model-analyzed dose, and the local planned dose, calculate the domain differences between the source domain medical institution and the target domain medical institution:

[0141] The model-analyzed dose refers to the dose value extracted from the dose distribution analysis atlas and used for analysis. The local planned dose refers to the actual clinical radiotherapy dose distribution data formulated by the target domain medical institution for its patients. The anatomical structure refers to a region or organ defined in the radiotherapy plan that has specific physiological or anatomical significance. The voxel refers to a point in three-dimensional space that represents the smallest volumetric unit of an anatomical structure.

[0142] As another implementation method, based on the above technical solution, the domain difference can be calculated using the following formula:

[0143]

[0144] in, Represents domain differences, Indicates the total number of anatomical structures. This indicates the first in the local radiotherapy plan data. Volume of an anatomical structure (unit: ), Indicates the first The total number of voxels of each anatomical structure This indicates the first in the local radiotherapy plan data. The anatomical structure corresponds to the first Local planned dose of individual pigments, The dose distribution analysis spectrum represents the first [number] dose distribution analysis spectrum. The anatomical structure corresponds to the first Dosage analysis of individual components This indicates the first in the local radiotherapy plan data. Local planned doses for each anatomical structure, The dose distribution analysis spectrum represents the first [number] dose distribution analysis spectrum. Analyze the dosage using a model of an anatomical structure.

[0145] For example, there are two anatomical structures A and B, where A has a volume of 100. The ratio of the intersection to the union of the local planned dose and the model analysis dose in anatomical structure A is 0.7 (i.e., =0.7), A contains 3 voxels with local planned doses of 2, 2.2 and 1.9, and model analysis doses of 1.8, 2 and 2.1, respectively. The volume of B is 150. In anatomical structure B, the ratio of the intersection to the union of the local planned dose and the model analysis dose is 0.6 (i.e., =0.6), B contains 3 voxels, whose local planned doses are 1.5, 1.8 and 1.4, and the model analysis doses are 1.6, 1.7 and 1.5, respectively. Substituting these into the formula, the domain difference between the source domain medical institution and the target domain medical institution is 3.7.

[0146] Optionally, the model analysis dose of the dose distribution analysis spectrum can be extracted through dose-volume histogram analysis.

[0147] S4. Based on the domain differences, define the optimization objective of the initial dose distribution analysis model, wherein the optimization objective includes: dose analysis loss, domain adversarial loss and feature alignment loss, so as to perform domain adaptive transfer learning on the initial dose distribution analysis model to obtain a local dose distribution analysis model.

[0148] This invention, by defining the optimization objective of the initial dose distribution analysis model based on the domain differences, enables the model to better adapt to subtle differences in the distribution of data in the target domain, improving the model's portability and practicality across different medical institutions. The optimization objective refers to a series of loss functions or combinations of objective functions used in the domain adaptive transfer learning process to guide the initial dose distribution analysis model in adjusting its parameters to adapt to the data of the target domain medical institution. The dose analysis loss is a loss function used to measure the difference between the model-predicted analyzed dose and the locally planned dose in the target domain medical institution. The domain adversarial loss is a loss function used to reduce the difference in feature distribution between the source and target domains through adversarial training. The feature alignment loss is a loss function used to measure the difference in the distribution of data in the source and target domains in the feature space extracted by the model.

[0149] As an embodiment of the present invention, defining the optimization objective of the initial dose distribution analysis model based on the domain differences includes:

[0150] Analyze the difference characteristics of the domain differences, including differences in dose-volume histograms and differences in dose distribution;

[0151] Based on the aforementioned differences, the anatomical structure weights, dose level weights, and voxel importance weights of the initial dose distribution analysis model are determined.

[0152] The dose analysis loss of the initial dose distribution analysis model is determined based on the anatomical structure weights, the dose level weights, and the voxel importance weights.

[0153] Construct a domain classifier for the initial dose distribution analysis model;

[0154] Extract the input features of the initial dose distribution analysis model, and input the input features into the domain classifier to obtain the classifier analysis value;

[0155] Based on the classifier analysis values, the domain adversarial loss of the initial dose distribution analysis model is calculated;

[0156] Extract the source domain data features and target domain data features of the initial dose distribution analysis model respectively;

[0157] Calculate the source domain kernel matrix between the source domain data features and the target domain kernel matrix between the target domain data features, respectively.

[0158] Calculate the kernel matrix between the source domain data features and the target domain data features;

[0159] Based on the source domain kernel matrix, the target domain kernel matrix, and the kernel matrix, determine the feature alignment loss of the initial dose distribution analysis model;

[0160] The optimization objective of the initial dose distribution analysis model is determined based on the feature alignment loss, the dose analysis loss, and the domain adversarial loss.

[0161] The difference features refer to the specific, quantifiable differences in dose allocation and distribution between source domain data and local medical institution data. The dose-volume histogram difference refers to the systematic deviation in the dose-volume histograms between the source and target domains for the same type of anatomical structure. The dose distribution difference refers to the inconsistency in the distribution of dose values ​​in three-dimensional space between source and target domain data. The anatomical structure weight refers to the weight coefficients assigned to the difference between the model analysis result and the true dose distribution, related to a specific anatomical structure, when calculating dose analysis loss. The dose level weight refers to the weight coefficients assigned to the difference between the model prediction result and the true dose distribution, related to a specific dose value, when calculating dose analysis loss. The voxel importance weight refers to the weight coefficients assigned to the difference between the model prediction result and the true dose distribution, related to a single voxel in the image, when calculating dose analysis loss. The domain classifier refers to a neural network module that facilitates the fusion of source and target domain data features. The classifier analysis value refers to the analysis probability output by the domain classifier after classifying the input features. The source domain data features refer to the intrinsic feature representations extracted from the source domain data that can represent the data, such as DICOM images of the source domain institution, patient anatomy, and treatment plan parameters. The target domain data features refer to the intrinsic feature representations extracted from the target domain data that can represent the data, such as DICOM images of the target domain institution, patient anatomy, and treatment plan parameters. The source domain kernel matrix refers to the similarity between various features in the source domain dataset. The target domain kernel matrix refers to the correlation measure between various features in the target domain dataset. The kernel matrix refers to the similarity between various features in the target domain dataset and the source domain dataset.

[0162] Optionally, the anatomical structure weights, dose level weights, and voxel importance weights of the initial dose distribution analysis model can be determined by statistical difference analysis methods, such as mean squared error, structural similarity index (SSIM), etc.

[0163] Optionally, the domain classifier of the initial dose distribution analysis model can be constructed using machine learning classification algorithms, such as support vector machine, logistic regression, random forest, K-nearest neighbors, etc.

[0164] Optionally, the source domain kernel matrix between the source domain data features can be calculated using a kernel function, such as a linear kernel function, a polynomial kernel function, a Gaussian kernel function, etc.

[0165] This invention, through domain-adaptive transfer learning on the initial dose distribution analysis model, obtains a local dose distribution analysis model. This model allows the local dose distribution analysis to adapt to specific local environments, thereby maintaining or even improving performance while overcoming the challenges posed by inconsistent data distribution. This results in more accurate, efficient, and practical localized applications. The local dose distribution analysis model refers to the model obtained through a domain-adaptive transfer learning process based on the initial dose distribution analysis model.

[0166] As an embodiment of the present invention, the step of performing domain adaptive transfer learning on the initial dose distribution analysis model to obtain a local dose distribution analysis model includes:

[0167] Based on the optimization objective of the initial dose distribution analysis model, the initial dose distribution analysis model is trained using preset training data to obtain a trained dose distribution analysis model.

[0168] Calculate the main task output and domain classification output of the trained dose distribution analysis model;

[0169] Based on the main task output and the domain classification output, calculate the main task loss and domain adversarial loss of the training dose distribution analysis model;

[0170] When the loss of the main task of the model is greater than the preset loss threshold of the main task, the model parameters of the trained dose distribution analysis model are adjusted, and then the process of training the initial dose distribution analysis model with preset training data according to the optimization objective of the initial dose distribution analysis model is returned to obtain the trained dose distribution analysis model.

[0171] When the domain adversarial loss of the model is less than the preset domain adversarial loss threshold, after adjusting the classifier parameters of the corresponding domain classifier of the trained dose distribution analysis model, the process returns to the step of training the initial dose distribution analysis model with preset training data according to the optimization objective of the initial dose distribution analysis model to obtain the trained dose distribution analysis model.

[0172] When the model's main task loss is less than a preset main task loss threshold and the model's domain adversarial loss is greater than a preset domain adversarial loss threshold, the training dose distribution analysis model is used as the training dose distribution analysis model.

[0173] The preset training data refers to the dataset prepared before training the initial dose distribution analysis model, used for model learning and optimization. The trained dose distribution analysis model refers to a dose distribution analysis model that has been optimized to a certain extent after specific training steps. The main task output refers to the dose distribution map analyzed by the trained dose distribution analysis model based on the input information. The domain classification output refers to the result produced by the classifier used to determine the source domain of the input data. The model main task loss is an indicator used to measure the degree of difference between the analysis result and the true result when the model performs its core analysis task. The model domain adversarial loss is an indicator used to measure the accuracy of the domain classifier in the model in determining the source domain of the data. The preset main task loss threshold is a pre-set value used to determine whether the model's performance in the main task (i.e., dose distribution analysis) has reached the target level. The model parameters refer to values ​​that can be learned and adjusted within the model, such as model weights and model biases. The domain adversarial loss threshold is a pre-set numerical limit used to evaluate whether the model's domain generalization ability meets the requirements. The classifier parameters refer to values ​​that can be learned and adjusted within the domain classifier, such as classifier weights and classifier biases.

[0174] Optionally, the model main task loss of the training dose distribution analysis model can be obtained by calculating a structural similarity index. For example, the output local mean, output variance, and output covariance of the main task output can be calculated, and the true local mean, true variance, and true covariance of the training data corresponding to the training dose distribution analysis model can be analyzed. Based on the output local mean, output variance, output covariance, true local mean, true variance, and true covariance, the structural similarity index between the main task output and the training data can be calculated to determine the model main task loss of the training dose distribution analysis model.

[0175] Optionally, the model parameters of the training dose distribution analysis model can be adjusted by optimization algorithms, such as stochastic gradient descent algorithm, cosine annealing algorithm, etc.

[0176] S5. Based on the local dose distribution analysis model, generate the target dose distribution for patients in the target domain medical institution.

[0177] This invention, by generating the target dose distribution for patients in the target domain medical institution based on the local dose distribution analysis model, enhances the model's adaptability and robustness in different medical institution environments, thereby generating dose distributions for patients in the target domain more reliably and accurately. The target dose distribution refers to the distribution of radiation doses received at various points in three-dimensional space within the patient's body, calculated by the treatment planning system.

[0178] As an embodiment of the present invention, generating the target dose distribution for patients in the target domain medical institution based on the local dose distribution analysis model includes:

[0179] The patient data corresponding to the patient is input into the local dose distribution analysis model. The patient data is processed through the convolutional layer of the local dose distribution analysis model to obtain high-level features and low-level detail features.

[0180] The high-level features and low-level detail features are fused through the feature fusion layer of the local dose distribution analysis model to obtain fused features;

[0181] Based on the fusion features, the dose distribution analysis map of the patient is calculated through the dose calculation layer of the local dose distribution analysis model;

[0182] Based on the dose distribution analysis diagram, the target dose distribution of the patient is output through the output layer of the local dose distribution analysis model.

[0183] The convolutional layer refers to a specific layer that extracts features from the input patient data. High-level features are highly abstract and information-rich feature representations extracted from patient data (CT images and contours) through the convolutional layer, representing complex anatomical structures, spatial relationships, and patterns, such as the overall shape and contour of organs, the relative position of tumor regions to surrounding organs at risk, and spatial relationships between organs. Skip connection layers are basic visual elements extracted from patient data by the model, such as bone edges and tissue textures. The feature fusion layer is a module specifically responsible for combining and integrating feature representations from different sources with different characteristics. The fused feature refers to a new, more comprehensive feature representation obtained after processing and combining high-level and low-level detailed features through the feature fusion layer. The dose calculation layer utilizes the information extracted and integrated by all previous layers to transform it into a dose distribution map of the patient's internal body. The dose distribution analysis map is the preliminary, analytical dose distribution result generated by the model after completing the dose calculation layer.

[0184] Optionally, the high-level features and the low-level detail features can be obtained through high-level semantic paths and low-level detail paths, such as determining the dilation rate and kernel weights of the convolutional layer, and constructing the high-level semantic path and low-level detail path of the convolutional layer based on the dilation rate and kernel weights to extract the high-level features and low-level detail features of the patient data.

[0185] Compared to the problems described in the background art, the embodiments of the present invention can directly generate higher-quality initial plans by training the initial dose distribution analysis model of the source domain medical institution, reducing the number and magnitude of subsequent optimization adjustments and shortening the overall planning time. Optionally, the embodiments of the present invention can achieve cross-institutional data collaboration by constructing a federated learning network between the source domain medical institution and the target domain medical institution under the premise of strictly protecting data privacy, thereby improving the generalization ability, applicability, and overall performance of the initial dose distribution analysis model. The embodiments of the present invention can output the dose distribution analysis map of the initial dose distribution analysis model using the local radiotherapy planning data of the target domain medical institution, which can preliminarily determine the applicability and accuracy of the initial model on the target domain data, providing a basis for whether and how to fine-tune it. The embodiments of the present invention can combine the dose distribution analysis... By analyzing the radiation spectrum and the local radiotherapy planning data, identifying the domain differences between the source and target domain medical institutions can intuitively and quantitatively identify the systematic differences in dose distribution between the two institutions in similar cases. This embodiment of the invention, through domain-adaptive transfer learning of the initial dose distribution analysis model, obtains a local dose distribution analysis model that adapts to specific local environments. This overcomes the challenges posed by inconsistent data distribution while maintaining or even improving performance, achieving more accurate, efficient, and practical localized applications. Finally, by generating the target dose distribution for patients in the target domain medical institution based on the local dose distribution analysis model, this embodiment enhances the model's adaptability and robustness in different medical institution environments, thereby generating dose distributions for target domain patients more reliably and accurately. Therefore, the radiotherapy plan dose distribution prediction method based on domain adaptive transfer learning provided by this embodiment can improve the efficiency and accuracy of radiotherapy plan dose distribution prediction.

[0186] Example 2:

[0187] like Figure 2 The diagram shown is a functional block diagram of a radiotherapy planning dose distribution prediction system based on domain adaptive transfer learning according to the present invention.

[0188] The radiotherapy planning dose distribution prediction system 200 based on domain adaptive transfer learning described in this invention can be installed in an electronic device. Depending on the functions implemented, the radiotherapy planning dose distribution prediction system based on domain adaptive transfer learning may include a source domain model training module 201, a local data prediction module 202, a domain difference analysis module 203, a local model training module 204, and a target dose distribution module 205. The module described in this invention can also be referred to as a unit, which refers to a series of computer program segments that can be executed by the processor of an electronic device and can perform a fixed function, and are stored in the memory of the electronic device.

[0189] In this embodiment of the invention, the functions of each module / unit are as follows:

[0190] The source domain model training module 201 is used to identify the source domain medical institutions and the target domain medical institutions, collect radiotherapy plan data of the source domain medical institutions, and extract radiotherapy plan features from the radiotherapy plan data. The radiotherapy plan features include: anatomical structure features, tumor location, and dose distribution, so as to train the initial dose distribution analysis model of the source domain medical institutions.

[0191] The local data prediction module 202 is used to construct a federated learning network between the source domain medical institution and the target domain medical institution, distribute the initial dose distribution analysis model to the target domain medical institution based on the federated learning network, and output the dose distribution analysis map of the initial dose distribution analysis model through the local radiotherapy plan data of the target domain medical institution.

[0192] The domain difference analysis module 203 is used to identify the domain differences between the source domain medical institution and the target domain medical institution by combining the dose distribution analysis map and the local radiotherapy plan data.

[0193] The local model training module 204 is used to determine the optimization objective of the initial dose distribution analysis model based on the domain differences. The optimization objective includes dose analysis loss, domain adversarial loss, and feature alignment loss, so as to perform domain adaptive transfer learning on the initial dose distribution analysis model to obtain a local dose distribution analysis model.

[0194] The target dose distribution module 205 is used to generate the target dose distribution for patients in the target domain medical institution based on the local dose distribution analysis model.

[0195] In detail, the modules in the radiotherapy planning dose distribution prediction system 200 based on domain adaptive transfer learning described in this embodiment of the invention employ the same methods as described above. Figure 1 The method used is the same as the one described in the article on the prediction of radiotherapy planning dose distribution based on domain adaptive transfer learning, and can produce the same technical effect, so it will not be described in detail here.

[0196] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0197] Finally, it should be noted that in the above embodiments, each embodiment can be combined with each other or independent. Deleting any one of them will not affect the technical implementation of other embodiments. The above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for predicting radiotherapy planning dose distribution based on domain adaptive transfer learning, characterized in that, The method includes: The source and target medical institutions are identified, radiotherapy planning data of the source medical institutions are collected, and radiotherapy planning features of the radiotherapy planning data are extracted. The radiotherapy planning features include anatomical structural features, tumor location, and dose distribution, in order to train the initial dose distribution analysis model of the source medical institutions. A federated learning network is constructed between the source domain medical institutions and the target domain medical institutions. Based on the federated learning network, the initial dose distribution analysis model is distributed to the target domain medical institutions so as to output the dose distribution analysis map of the initial dose distribution analysis model through the local radiotherapy planning data of the target domain medical institutions. By combining the dose distribution analysis map and the local radiotherapy planning data, the domain differences between the source domain medical institution and the target domain medical institution are identified. This identification includes: extracting the model analysis dose from the dose distribution analysis map and the local planned dose from the local radiotherapy planning data; uniformly identifying the anatomical structures from the dose distribution analysis map and the local radiotherapy planning data; and calculating the domain differences between the source domain medical institution and the target domain medical institution using the following formula based on the anatomical structures, the model analysis dose, and the local planned dose: in, Represents domain differences, Indicates the total number of anatomical structures. This indicates the first in the local radiotherapy plan data. The volume of an anatomical structure Indicates the first The total number of voxels of each anatomical structure This indicates the first in the local radiotherapy plan data. The anatomical structure corresponds to the first Local planned dose of individual pigments, The dose distribution analysis spectrum represents the first [number] dose distribution analysis spectrum. The anatomical structure corresponds to the first Dosage analysis of individual components This indicates the first in the local radiotherapy plan data. Local planned doses for each anatomical structure, The dose distribution analysis spectrum represents the first [number] dose distribution analysis spectrum. Dosage analysis of a model of an anatomical structure; Based on the domain differences, the optimization objective of the initial dose distribution analysis model is defined, wherein the optimization objective includes: dose analysis loss, domain adversarial loss and feature alignment loss, so as to perform domain adaptive transfer learning on the initial dose distribution analysis model to obtain a local dose distribution analysis model; Based on the local dose distribution analysis model, the target dose distribution for patients in the target domain medical institution is generated.

2. The radiotherapy planning dose distribution prediction method based on domain adaptive transfer learning as described in claim 1, characterized in that, The radiotherapy plan features extracted from the radiotherapy plan data include: The radiotherapy planning data is formatted to obtain standardized radiotherapy planning data; The standardized radiotherapy plan data is denoised to obtain denoised radiotherapy plan data; The denoised radiotherapy planning data is aligned to obtain aligned radiotherapy planning data. Identify the key structures in the aligned radiotherapy planning data, and calculate the geometric and anatomical features of the key structures; Based on the geometric features and the anatomical relationship features, the anatomical structural features and tumor location of the aligned radiotherapy planning data are determined; Based on the aligned radiotherapy plan data, the target dose characteristics and organ-at-risk dose characteristics at the tumor location are calculated; Based on the target area dose characteristics and the organ at risk dose characteristics, the dose distribution at the tumor location is determined; The radiotherapy planning characteristics of the radiotherapy planning data are determined based on the dose distribution, the anatomical features, and the tumor location.

3. The radiotherapy planning dose distribution prediction method based on domain adaptive transfer learning as described in claim 1, characterized in that, The training of the initial dose distribution analysis model for the source domain medical institutions includes: The radiotherapy plan features corresponding to the source domain medical institutions are divided into a training set and a validation set; Generate the model framework of the source domain medical institutions; Based on the training set, the model framework is trained to obtain a training analysis model; Based on the validation set, calculate the mean squared error of the trained analysis model; When the mean squared error is greater than the preset error threshold, the model learning parameters of the training analysis model are adjusted, and the process returns to the step of training the model framework based on the training set to obtain the training analysis model. When the mean square error is less than a preset error threshold, the trained analysis model is used as the initial dose distribution analysis model for the source domain medical institution.

4. The radiotherapy planning dose distribution prediction method based on domain adaptive transfer learning as described in claim 1, characterized in that, The construction of the federated learning network for the source domain medical institutions and the target domain medical institutions includes: Identify the source domain client node of the source domain medical institution and the target domain client node of the target domain medical institution; Construct the communication protocol and dynamic encryption protocol between the source domain client node and the target domain client node; Based on the communication protocol and the dynamic encryption protocol, a central server is constructed for the source domain client node and the target domain client node; Configure the client environment of the source domain client node and the target domain client node in a unified manner; Based on the client environment, define the model training loop rules for the source domain client node, the target domain client node, and the central server; Based on the model training loop rules, a federated learning network is constructed for the source domain medical institutions and the target domain medical institutions.

5. The radiotherapy planning dose distribution prediction method based on domain adaptive transfer learning as described in claim 1, characterized in that, The step of outputting the dose distribution analysis map of the initial dose distribution analysis model using the local radiotherapy plan data of the target domain medical institution includes: Collect local radiotherapy plan data from medical institutions in the target domain; Extract key information from the local radiotherapy plan data; Convert the key information into input format information; The input format information is input into the initial dose distribution analysis model to obtain dose distribution data; Based on the dose distribution data, a dose distribution analysis map of the patients corresponding to the target domain medical institutions is generated.

6. The radiotherapy planning dose distribution prediction method based on domain adaptive transfer learning as described in claim 1, characterized in that, The optimization objective of the initial dose distribution analysis model, based on the domain differences, includes: Analyze the difference characteristics of the domain differences, including differences in dose-volume histograms and differences in dose distribution; Based on the aforementioned differences, the anatomical structure weights, dose level weights, and voxel importance weights of the initial dose distribution analysis model are determined. The dose analysis loss of the initial dose distribution analysis model is determined based on the anatomical structure weights, the dose level weights, and the voxel importance weights. Construct a domain classifier for the initial dose distribution analysis model; Extract the input features of the initial dose distribution analysis model, and input the input features into the domain classifier to obtain the classifier analysis value; Based on the classifier analysis values, the domain adversarial loss of the initial dose distribution analysis model is calculated; Extract the source domain data features and target domain data features of the initial dose distribution analysis model respectively; Calculate the source domain kernel matrix between the source domain data features and the target domain kernel matrix between the target domain data features, respectively. Calculate the kernel matrix between the source domain data features and the target domain data features; Based on the source domain kernel matrix, the target domain kernel matrix, and the kernel matrix, determine the feature alignment loss of the initial dose distribution analysis model; The optimization objective of the initial dose distribution analysis model is determined based on the feature alignment loss, the dose analysis loss, and the domain adversarial loss.

7. The radiotherapy planning dose distribution prediction method based on domain adaptive transfer learning as described in claim 1, characterized in that, The process of performing domain-adaptive transfer learning on the initial dose distribution analysis model to obtain a local dose distribution analysis model includes: Based on the optimization objective of the initial dose distribution analysis model, the initial dose distribution analysis model is trained using preset training data to obtain a trained dose distribution analysis model. Calculate the main task output and domain classification output of the trained dose distribution analysis model; Based on the main task output and the domain classification output, calculate the main task loss and domain adversarial loss of the training dose distribution analysis model; When the loss of the main task of the model is greater than the preset loss threshold of the main task, the model parameters of the trained dose distribution analysis model are adjusted, and then the process of training the initial dose distribution analysis model with preset training data according to the optimization objective of the initial dose distribution analysis model is returned to obtain the trained dose distribution analysis model. When the domain adversarial loss of the model is less than the preset domain adversarial loss threshold, after adjusting the classifier parameters of the corresponding domain classifier of the trained dose distribution analysis model, the process returns to the step of training the initial dose distribution analysis model with preset training data according to the optimization objective of the initial dose distribution analysis model to obtain the trained dose distribution analysis model. When the model's main task loss is less than a preset main task loss threshold and the model's domain adversarial loss is greater than a preset domain adversarial loss threshold, the training dose distribution analysis model is used as the training dose distribution analysis model.

8. The radiotherapy planning dose distribution prediction method based on domain adaptive transfer learning as described in claim 7, characterized in that, The step of generating the target dose distribution for patients in the target domain medical institution based on the local dose distribution analysis model includes: The patient data corresponding to the patient is input into the local dose distribution analysis model. The patient data is processed through the convolutional layer of the local dose distribution analysis model to obtain high-level features and low-level detail features. The high-level features and low-level detail features are fused through the feature fusion layer of the local dose distribution analysis model to obtain fused features; Based on the fusion features, the dose distribution analysis map of the patient is calculated through the dose calculation layer of the local dose distribution analysis model; Based on the dose distribution analysis diagram, the target dose distribution of the patient is output through the output layer of the local dose distribution analysis model.

9. A radiotherapy planning dose distribution prediction system based on domain adaptive transfer learning, characterized in that, The system includes: The source domain model training module is used to identify source domain medical institutions and target domain medical institutions, collect radiotherapy plan data of the source domain medical institutions, and extract radiotherapy plan features from the radiotherapy plan data. The radiotherapy plan features include: anatomical structure features, tumor location, and dose distribution, in order to train the initial dose distribution analysis model of the source domain medical institutions. The local data prediction module is used to construct a federated learning network between the source domain medical institution and the target domain medical institution. Based on the federated learning network, the initial dose distribution analysis model is distributed to the target domain medical institution to output the dose distribution analysis map of the initial dose distribution analysis model through the local radiotherapy planning data of the target domain medical institution. The domain difference analysis module is used to identify the domain differences between the source domain medical institution and the target domain medical institution by combining the dose distribution analysis map and the local radiotherapy planning data. The identification of the domain differences between the source domain medical institution and the target domain medical institution includes: extracting the model analysis dose from the dose distribution analysis map and the local planned dose from the local radiotherapy planning data; uniformly identifying the anatomical structures from the dose distribution analysis map and the local radiotherapy planning data; and calculating the domain differences between the source domain medical institution and the target domain medical institution using the following formula based on the anatomical structures, the model analysis dose, and the local planned dose: in, Represents domain differences, Indicates the total number of anatomical structures. This indicates the first in the local radiotherapy plan data. The volume of an anatomical structure Indicates the first The total number of voxels of each anatomical structure This indicates the first in the local radiotherapy plan data. The anatomical structure corresponds to the first Local planned dose of individual pigments, The dose distribution analysis spectrum represents the first [number] dose distribution analysis spectrum. The anatomical structure corresponds to the first Dosage analysis of individual components This indicates the first in the local radiotherapy plan data. Local planned doses for each anatomical structure, The dose distribution analysis spectrum represents the first [number] dose distribution analysis spectrum. Dosage analysis of a model of an anatomical structure; A local model training module is used to define the optimization objective of the initial dose distribution analysis model based on the domain differences. The optimization objective includes dose analysis loss, domain adversarial loss, and feature alignment loss, so as to perform domain adaptive transfer learning on the initial dose distribution analysis model to obtain a local dose distribution analysis model. The target dose distribution module is used to generate the target dose distribution for patients in the target domain medical institution based on the local dose distribution analysis model.

Images